Split (3)

train · 4.35k rows

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
The peptidoglycan cell wall is a defining structural feature of the bacterial kingdom . Curiously , some bacteria have the ability to switch to a wall-free or ‘L-form’ state . Although known for decades , the general properties of L-forms are poorly understood , largely due to the lack of systematic analysis of L-forms in the molecular biology era . Here we show that inhibition of peptidoglycan precursor synthesis promotes the generation of L-forms from both Gram-positive and Gram-negative bacteria . We show that the L-forms generated have in common a mechanism of proliferation involving membrane blebbing and tubulation , which is dependent on an altered rate of membrane synthesis . Crucially , this mode of proliferation is independent of the essential FtsZ based division machinery . Our results suggest that the L-form mode of proliferation is conserved across the bacterial kingdom , reinforcing the idea that it could have been used in primitive cells , and opening up its use in the generation of synthetic cells . The peptidoglycan ( PG ) cell wall is a major defining feature of bacterial cells and is present in all known major bacterial phyla , suggesting that the wall was present in the last common ancestor of the whole bacterial lineage ( Errington , 2013 ) . PG is composed of long glycan strands cross linked by short peptide bridges , forming a meshwork that covers the whole cell . The wall has a variety of important functions , including the following: maintenance of cell shape , protection from mechanical damage , and generation of turgor by restraining the outward osmotic pressure exerted on the cytoplasmic membrane . It is the target for our best antibiotics ( β-lactams , glycopeptides , etc ) , and fragments of the wall trigger important innate immune responses . The wall is assembled by polymerization and cross linking of a precursor molecule , termed lipid II , which is synthesized in the cytoplasm and then transferred to the cell surface for wall assembly ( Typas et al . , 2012 ) . Despite its importance , many bacteria , both Gram-positives and Gram-negatives , are capable of switching into a cell wall deficient state , called the ‘L-form’ ( Allan et al . , 2009 ) . Generally , L-forms were generated under osmoprotective conditions ( e . g . in the presence of 0 . 5 M sucrose ) by long term and repeated passage , sometimes for years , in the presence of β-lactam antibiotics that inhibit PG synthesis ( Allan , 1991 ) . However , the lack of reproducible and tractable model systems prevented the development of consensus views of the common properties of L-forms derived from different bacteria . We have recently undertaken a systematic analysis of the L-form transition in the experimentally tractable Gram-positive bacterium Bacillus subtilis . We have defined genetic pathways required to elicit a reproducible and rapid switch to the L-form state and identified genes required specifically for L-form growth in this organism ( Leaver et al . , 2009; Dominguez-Cuevas et al . , 2012; Mercier et al . , 2012 , 2013 ) . Our analysis of B . subtilis L-form growth led to two unexpected findings . First , that when dividing in the L-form state , B . subtilis becomes completely independent of the FtsZ ( tubulin ) based division machinery ( Leaver et al . , 2009 ) and the MreB ( actin ) cytoskeleton ( Mercier et al . , 2012 ) . Instead , the L-forms divide by a remarkable process of cell shape deformation , including blebbing , tubulation , and vesiculation , followed by spontaneous resolution ( scission ) into smaller progeny cells ( Kandler and Kandler , 1954; Leaver et al . , 2009 ) . We recently showed that L-form proliferation in B . subtilis simply depends on excess membrane synthesis , leading to an increase in the surface area to volume ratio ( Mercier et al . , 2013 ) . Upregulation of membrane synthesis can be driven directly , by mutations affecting the regulation of fatty acid synthesis , or indirectly , by shutting down PG precursor synthesis , which presumably depends on a regulatory circuit that we do not yet understand . To complicate matters , the growth of B . subtilis L-forms requires a second mutational change , most commonly affecting the ispA gene ( Leaver et al . , 2009 ) , which probably works by compensating for a metabolic imbalance that occurs when cells grow in the absence of wall synthesis ( Kawai and Mercier , unpublished ) . To date , we have restricted our attention to B . subtilis L-forms . In this study , we have shown that inhibition of PG precursor synthesis seems to be an efficient method to create stable L-forms from a range of diverse bacteria , including a Gram-negative Escherichia coli . We have also characterized several key properties of these L-forms , including their mode of proliferation , and we have found them to be strikingly reminiscent of B . subtilis L-forms , in the following ways: ( i ) mode of cell proliferation using cell shape deformation followed by a spontaneous formation of progeny cells; ( ii ) dispensability of the normally essential cell division machinery; and ( iii ) key role for the membrane synthesis rate in cell proliferation . The strikingly similar properties of L-forms from different bacterial lineages reinforces the idea that their mode of cell proliferation could have been used in primitive bacteria before the invention of the cell wall , and that they could be used in the generation of synthetic cells . We previously showed that excess membrane synthesis is required for L-form proliferation and that this can be achieved directly by upregulation of the fatty acid synthase ( FAS II ) system or indirectly by inhibition of PG precursor synthesis ( Figure 1A ) . We do not yet understand the basis for coupling of PG precursor and fatty acid synthesis but the effect on B . subtilis is shown in Figure 1B . Although inhibiting the PG precursor pathway was lethal on non-osmoprotective nutrient agar ( NA ) plates ( lipid II OFF , NA ) , growth of B . subtilis was restored on osmoprotective NA/supporting medium ( MSM ) plates ( lipid II OFF , MSM ) , via a switch to an L-form mode of proliferation ( Leaver et al . , 2009; Mercier et al . , 2013 ) . The gross morphological differences between walled and L-form B . subtilis are illustrated in Figure 1C . ( Note that B . subtilis L-form growth requires an additional mutation in a gene such as ispA [Figure 1—figure supplement 1] , for reasons that are not yet clear [Leaver et al . , 2009] . ) 10 . 7554/eLife . 04629 . 003Figure 1 . Inhibition of PG precursor synthesis induces L-form proliferation in bacteria . ( A ) Schematic model of peptidoglycan ( PG ) precursor ( lipid II ) synthesis in bacteria and its inhibition by the antibiotics fosfomycin ( FOS ) and D-cycloserine ( DCS ) . MurA , inhibited by the antibiotic FOS , and MurB catalyze the transformation of uridine diphosphate-N-acetylglucosamine ( UDP-GlcNAc ) into UDP-N-acetylmuramic acid ( UDP-MurNAc ) . The racemase Dal and the D-alanine ligase Ddl , both of which are inhibited by the antibiotic DCS , are required to generate D-Ala-D-Ala . This is incorporated into the UDP-MurNAc-pentapeptide , requiring MurC , MurD , MurE , and MurF enzymes . UDP-MurNAc-pentapeptide is transferred to undecaprenyl pyrophosphate by MraY , and the addition of GlcNAc is catalyzed by MurG to form lipid II . ( B ) Growth of Bacillus subtilis strain LR2 ( ispA Pxyl-murE-B ) streaked on L-form supporting medium ( MSM ) or nutrient agar ( NA ) plates in the presence ( lipid II ON ) or absence ( lipid II OFF ) of 0 . 5% xylose . ( C ) Phase contrast microscopy of B . subtilis LR2 cells grown on MSM plates in the presence ( left ) or absence ( right ) of 0 . 5% xylose . ( D–I ) Growth on plates ( D , F , H ) and corresponding phase contrast microscopy ( E , G , I ) of bacterial strains Staphylococcus aureus ATCC2913 ( D , E ) , Corynebacterium glutamicum ATCC13032 ( F , G ) , and Escherichia coli MG1655 ( H , I ) . ( D , F , H ) The different bacterial strains were streaked on MSM or NA plates in the absence ( lipid II ON ) or presence ( lipid II OFF ) of the antibiotics FOS ( D , H ) or DCS ( F ) . ( E , G , I ) Phase contrast microscopy of the different bacterial cells grown on MSM plates in the absence ( left ) or presence ( right ) of the antibiotics FOS ( E , I ) or DCS ( G ) . Scale bars , 3 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 04629 . 00310 . 7554/eLife . 04629 . 004Figure 1—figure supplement 1 . Bacillus subtilis L-form growth requires an additional mutation in a gene such as ispA . Growth of B . subtilis strain BS115 ( Pxyl-murE-B ) streaked on L-form supporting medium ( MSM ) in the absence ( lipid II OFF ) of 0 . 5% xylose . DOI: http://dx . doi . org/10 . 7554/eLife . 04629 . 00410 . 7554/eLife . 04629 . 005Figure 1—figure supplement 2 . Bacterial L-forms proliferate on β-lactams . ( A ) Growth of Staphylococcus aureus ( left ) , Corynebacterium glutamicum ( middle ) , and Escherichia coli ( right ) walled strains streaked on L-form supporting medium ( MSM ) in the presence of penicillin G ( S . aureus and C . glutamicum ) or ampicillin ( E . coli ) . ( B ) Growth of S . aureus ( top ) , C . glutamicum ( middle ) , and E . coli ( bottom ) L-forms streaked on MSM with the antibiotics fosfomycin ( S . aureus and E . coli ) or D-cycloserine ( C . glutamicum ) in the presence of penicillin G ( S . aureus and C . glutamicum ) or ampicillin ( E . coli ) . The different L-form strains were streaked several times under the same conditions every 3 days ( left to right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04629 . 00510 . 7554/eLife . 04629 . 006Figure 1—figure supplement 3 . Bacterial L-form cell wall reversion . ( A ) Cell wall reversion on L-form supporting medium ( MSM ) plates of L-forms of Escherichia coli strain TB28 grown on MSM plates containing fosfomycin and ampicillin . ( B ) Growth of the E . coli L-form reverted strain TB28 from panel ( A ) on nutrient agar ( NA ) . ( C ) Cell wall reversion on MSM plates of L-forms of Corynebacterium glutamicum grown on MSM plates with D-cycloserine and penicillin G . ( D ) Growth of the C . glutamicum L-form reverted strain from panel ( C ) on NA . DOI: http://dx . doi . org/10 . 7554/eLife . 04629 . 006 We wondered whether similar approaches could be used to elicit L-form growth in other bacteria . To simplify the experiments , we used biochemical inhibitors of the PG precursor pathway , fosfomycin ( FOS ) or D-cycloserine ( DCS ) , which inhibit the enzymes MurA and Ddl , respectively ( Figure 1A ) . We examined three different organisms: two Gram-positive organisms , the Firmicute Staphylococcus aureus ATCC29213 and the Actinobacterium Corynebacterium glutamicum ATCC13032 , and the Gram-negative organism , E . coli strain MG1655 . In all three cases , we were readily able to generate an L-form transition . S . aureus and E . coli were both susceptible to FOS at 400 μg/ml . C . glutamicum was resistant to FOS but susceptible to DCS at the same concentration . Figure 1 ( D , F , H ) shows that the growth of all three strains on NA was inhibited in the presence of the drug ( lipid II OFF ) . However , as observed in B . subtilis , growth of all three strains was efficiently restored under osmoprotective conditions ( lipid II OFF , MSM ) . Furthermore , phase contrast microscopy of the three treated cultures ( Figure 1E , G , I; OFF ) revealed the presence of large spheroidal cells strikingly similar to the L-forms of B . subtilis ( Figure 1C ) and quite different from the parental walled cells ( Figure 1E , G , I; ON ) , consistent with the idea that all three diverse organisms are able to switch to an L-form mode of proliferation on inhibition of the PG precursor pathway . We further showed that L-forms of the three different species could be successively propagated in the presence of high ( 500 μg/ml ) concentrations of β-lactam antibiotics ( Figure 1—figure supplement 2B ) in concentrations that are normally lethal in walled cells ( Figure 1—figure supplement 2A ) . Finally , on reactivation of PG precursor synthesis , the three different species readily reverted to their parental walled forms ( Figure 1—figure supplement 3 , Figure 4—figure supplement 1A–B ) , by de novo synthesis of the cell wall sacculus ( Kawai et al . , 2014 ) . In B . subtilis , proliferation in the L-form state renders the normally essential genes of the PG precursor pathway dispensable ( Leaver et al . , 2009; Mercier et al . , 2013 ) . Thus , to test whether FOS or DCS are sufficient to promote the full switch to an L-form mode of proliferation , we assessed whether the PG precursor pathway genes were essential in the genetically tractable bacterium E . coli . We first constructed plasmid pOU82-murA , which carried a copy of murA+ located on an unstable mini-R1 plasmid ( Gerdes et al . , 1985 ) , together with a lacZ gene encoding β-galactosidase . In the presence of this plasmid , we were then able to construct a chromosomal deletion of murA , which is an essential gene of the PG precursor pathway ( Figure 1A ) , giving strain RM345 ( murA::Kn , pOU82-murA ) . In walled cells , the presence of plasmid pOU82-murA was essential for growth of strain RM345 ( Figure 2A , bottom ) , as demonstrated by the uniform blue colonies on X-gal , while it was readily lost from the parental TB28 strain ( murA+ ) , giving many white colonies ( Figure 2A , top ) . Strikingly , the plasmid was also readily lost from strain RM345 when grown in the putatively L-form state , as indicated by the white colonies ( Figure 2B ) . To confirm the specific loss of the murA gene , we performed a multiplex PCR ( see ‘Materials and methods’ below ) on DNA purified from cells of strain RM345 grown in the presence of the cell wall or as L-forms . As shown in Figure 2C , in the walled state , the murA gene was readily detected ( lane 1 ) whereas it was not detected in the DNA from a white L-form colony ( lane 2 ) . 10 . 7554/eLife . 04629 . 007Figure 2 . E . coli L-forms proliferate independently of the peptidoglycan cell wall machinery . ( A ) Growth of the Escherichia coli strains TB28 ( top ) and RM345 ( ΔmurA , bottom ) containing the unstable plasmid pOU82-Amp-murA streaked on nutrient agar plates in the presence of X-gal . ( B ) L-form colonies of the E . coli strains RM345 ( ΔmurA , pOU82-Amp-murA ) on L-form supporting medium ( MSM ) plates in the presence of fosfomycin ( FOS ) and X-gal , after several repeated streakings on MSM plates in the presence of FOS . ( C ) Multiplex PCR of the ftsK , murA , ftsZ , and mreC genes from genomic DNA of the E . coli strains RM345 grown in the walled ( 1 ) or L-form ( 2 ) states , obtained from the strains in panel ( B ) . M represents the 100 bp DNA ladder . DOI: http://dx . doi . org/10 . 7554/eLife . 04629 . 007 Having created newly growing bacterial L-forms from different bacterial species , we wished to investigate their mode of cell proliferation using time lapse microscopy . For C . glutamicum , the L-forms grow readily under various conditions , including liquid media , and we were readily able to capture time lapse sequences that revealed a pattern of proliferative events very similar to those we described previously for B . subtilis ( Leaver et al . , 2009; Mercier et al . , 2013 ) . Figure 3A and Video 1 and Video 2 show typical time courses . In Figure 3A , the central cell underwent repeated shape deformations , with proliferative events generating separate cells after 200 and 315 min ( * ) . 10 . 7554/eLife . 04629 . 008Figure 3 . Mode of cell division of E . coli and C . glutamicum L-forms . ( A , B ) Corynebacterium glutamicum L-form strain grown in nutrient broth ( NB ) /L-form supporting medium ( MSM ) with D-cycloserine ( A ) , and Escherichia coli L-form strain RM345 ( ΔmurA ) grown on nutrient agar/MSM ( B ) , were observed by time lapse phase contrast microscopy . Elapsed time ( min ) is shown in each panel . Scale bars , 3 μm . Arrows represent the direction of protrusion formation and the asterisks ( * ) the daughter cells after division . See also Videos 1–4 . DOI: http://dx . doi . org/10 . 7554/eLife . 04629 . 00810 . 7554/eLife . 04629 . 009Video 1 . Time lapse series showing L-form cell growth of Corynebacterium glutamicum growing in nutrient broth ( NB ) /L-form supporting medium ( MSM ) with D-cycloserine ( DCS ) , from which the panels in Figure 3A were obtained . Phase contrast images were acquired automatically every 5 min for about 5 hr . Scale bar , 3 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 04629 . 00910 . 7554/eLife . 04629 . 010Video 2 . Time lapse series showing L-form cell growth of Corynebacterium glutamicum growing in nutrient broth ( NB ) /L-form supporting medium ( MSM ) with D-cycloserine ( DCS ) . Phase contrast images were acquired automatically every 5 min for about 3 hr 30 min . Scale bar , 3 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 04629 . 010 Unfortunately , in the case of S . aureus , we have so far been unable to grow them in liquid medium . On appropriate solid medium , although the L-form cultures clearly undergo substantial increases in biomass and the cells have a typical L-form morphology in still images , we have not yet been able to visualize specific division events by time lapse imaging . Growth of E . coli L-forms in liquid media has also been problematic . However , in this case , we have succeeded in capturing suitable time lapse data . Figure 3B ( and Video 3 and Video 4 ) show typical examples of an E . coli ΔmurA L-form strain grown on solid medium ( NA/MSM ) . Strikingly , the mode of cell proliferation is reminiscent of the Gram-positive B . subtilis and C . glutamicum L-forms . We observed a repeat cycle of cell deformation and cell protrusion formation at 105–160 min and 185–200 min ( arrows ) , each followed by a spontaneous division generating progeny cells after 170 min and 205 min ( * ) . 10 . 7554/eLife . 04629 . 011Video 3 . Time lapse series showing L-form cell growth of Escherichia coli strain RM345 ( ΔmurA ) growing in nutrient agar ( NA ) /L-form supporting medium ( MSM ) from which the panels in Figure 3B were obtained . Phase contrast images were acquired automatically every 5 min for about 4 hr . Scale bar , 3 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 04629 . 01110 . 7554/eLife . 04629 . 012Video 4 . Time lapse series showing L-form cell growth of Escherichia coli strain RM345 ( ΔmurA ) growing in nutrient agar ( NA ) /L-form supporting medium ( MSM ) . Phase contrast images were acquired automatically every 5 min for about 4 hr . Scale bar , 3 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 04629 . 012 It thus appears that the general features of the L-form mode of cell proliferation are conserved between Gram-positive and Gram-negative bacteria . Cell division of walled bacteria requires the assembly and function of a complex proteinaceous machinery built around the essential tubulin homologue FtsZ ( Adams and Errington , 2009 ) . We showed previously that in B . subtilis L-forms , remarkably the FtsZ protein and probably the whole cell division machinery become dispensable ( Leaver et al . , 2009 ) . We therefore tested the role of the cell division machinery in the newly created bacterial L-forms . For E . coli , we used the method described above for murA to construct an ftsZ deletion mutant complemented by plasmid pOU82-ftsZ ( strain RM349 , ftsZ::Kn , pOU82-ftsZ ) . When RM349 was grown in the walled state , FtsZ appeared essential , as judged by the blue only colonies ( Figure 4A , bottom ) . Once again , when induced to grow in the L-form state , FtsZ became dispensable , as characterized by the presence of white colonies on X-gal plates ( Figure 4B , top left ) . Multiplex PCR was used to confirm loss of the ftsZ gene only in the L-form cell DNA ( Figure 4C , lanes 1 and 2 ) . Additionally , using a similar strategy , we showed that both FtsZ and MurA proteins were simultaneously dispensable in L-forms ( Figure 4B , top right , and Figure 4C lanes 3 and 4 ) , as well as another essential cell division protein FtsK ( Figure 3B , bottom left and 4C lanes 5 and 6 ) , and the cytoskeleton proteins MreBCD ( Figure 4B , bottom right , and Figure 4C lanes 7 and 8 ) . 10 . 7554/eLife . 04629 . 013Figure 4 . Bacterial L-forms proliferate independently of the cell division machinery . ( A ) Growth of the Escherichia coli strains TB28 ( top ) and RM349 ( ΔftsZ , bottom ) containing the unstable plasmid pOU82-Amp-ftsZ streaked on nutrient agar plates in the presence of X-gal . ( B ) L-form colonies of the E . coli strains RM349 ( ΔftsZ , pOU82-Amp-ftsZ , top left ) , RM350 ( ΔmurA , ΔftsZ , pOU82-Amp-ftsZ , pSK122-Cm-murA , top right ) , RM61 ( ΔftsK , pSK122-Cm-ftsK , bottom left ) , and RM359 ( ΔmreBCD , pHM82-Kn-mreBCD ) on L-form-supporting medium ( MSM ) plates in the presence of fosfomycin ( FOS ) and X-gal , after several repeated streakings on MSM plates in the presence of FOS . ( C ) Multiplex PCR of the ftsK , murA , ftsZ , and mreC genes from genomic DNA of the E . coli strains RM349 ( 1 , 2 ) , RM350 ( 3 , 4 ) , RM61 ( 5 , 6 ) , and RM359 ( 7 , 8 ) grown in the walled ( 1 , 3 , 5 and 7 ) or L-form ( 2 , 4 , 6 and 8 ) states obtained from the strains in panel ( B ) . M represents the 100 bp DNA ladder . ( D ) Growth of the Staphylococcus aureus strain RNpFtsZ-1 ( erm-pSPAC-ftsZ , Pinho and Errington , 2003 ) streaked on MSM plates in the absence ( lipid II ON , left ) or presence ( lipid II OFF , middle and right ) of FOS , with ( +FtsZ , middle ) or without ( −FtsZ , left and right ) isopropyl β-d-1-thiogalactopyranoside . ( E ) Growth profiles of Corynebacterium glutamicum in MSM with ( L-form state; right , lipid II OFF ) or without ( walled state; left , lipid II ON ) D-cycloserine , and in the absence ( red ) or presence ( blue ) of cephalexin . DOI: http://dx . doi . org/10 . 7554/eLife . 04629 . 01310 . 7554/eLife . 04629 . 014Figure 4—figure supplement 1 . Staphylococcus aureus L-forms proliferate in the absence of the cell division machinery . ( A ) Cell wall reversion on L-form supporting medium ( MSM ) plates with isopropyl β-d-1-thiogalactopyranoside ( IPTG ) ( +FtsZ ) of L-forms of S . aureus strain RNpFtsZ-1 grown on MSM plates with fosfomycin ( FOS ) and without IPTG , obtained from Figure 4D , right . ( B ) Growth of the S . aureus RNpFtsZ-1 L-form reverted strain from panel ( D ) on nutrient agar plates with ( +FtsZ ) or without ( −FtsZ ) IPTG . ( C ) Growth of the S . aureus strain ATCC2913 ftsZR191P ( Haydon et al . , 2008 ) streaked on MSM plates with ( lipid II ON , left and middle ) or without ( lipid II OFF , right ) FOS , and in the presence ( left ) or absence ( middle and right ) of benzamide . DOI: http://dx . doi . org/10 . 7554/eLife . 04629 . 01410 . 7554/eLife . 04629 . 015Figure 4—figure supplement 2 . Corynebacterium glutamicum L-forms proliferate in the absence of the cell division machinery . Growth of the C . glutamicum strain streaked on nutrient agar ( NA ) ( lipid II ON ) in the absence ( left , −cephalexin ) or presence ( middle , +cephalexin ) of cephalexin , and on L-form supporting medium ( MSM ) plates with D-cycloserine and cephalexin ( right , lipid II OFF , + cephalexin ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04629 . 015 To examine whether the cell division machinery was essential in S . aureus , we took advantage of strain RNpFtsZ_1 ( Pinho and Errington , 2003 ) in which the ftsZ gene is controlled by an isopropyl β-d-1-thiogalactopyranoside ( IPTG ) inducible promoter . As expected , this strain was unable to proliferate without inducer in the presence of the cell wall ( Figure 4D , lipid II ON , −FtsZ ) . However , when the cells were switched into an L-form mode of proliferation ( lipid II OFF ) , no growth difference was detectable between the presence ( +FtsZ ) or absence ( −FtsZ ) of IPTG . To exclude the possibility that the strain picked up a suppressor mutation relieving the dependence of ftsZ expression on IPTG , we reverted the L-forms to the parental walled form ( by removing FOS ) and showed that the cells regained their dependence on IPTG ( Figure 4—figure supplement 1A and 1B ) . As an alternative way to test for dependence on the cell division machinery in S . aureus , we used the strain ATCC2913 ftsZR191P ( Haydon et al . , 2008 ) , which carries an amino acid substitution in FtsZ that renders the cells dependent on a benzamide antibiotic . Walled cells grow and divide normally in the presence of the antibiotic but the mutant FtsZ protein fails to support division in the absence of benzamide ( Figure 4—figure supplement 1C , lipid II ON ) . In accordance with the above results , growth in the absence of benzamide was restored when the cells were switched to the L-form state ( Figure 4—figure supplement 1C , lipid II OFF ) , again showing that S . aureus L-forms can proliferate independently of FtsZ and hence of the normal cell division machinery . Construction of conditional mutants of C . glutamicum is not as straightforward as for B . subtilis or E . coli , so , to test the requirement for the cell division machinery in C . glutamicum L-forms , we cultured the organism in the walled and L-form states in the presence of cephalexin , a specific inhibitor of the essential cell division protein FtsI ( Pogliano et al . , 1997 ) . As previously shown , cephalexin blocks cell division in normal walled cells ( Valbuena et al . , 2006 ) , leading to a severe growth defect ( Figure 4E , left and Figure 4—figure supplement 2 , middle ) . However , in the L-form mode of proliferation , no growth defect was observed ( Figure 4E , right and Figure 4—figure supplement 2 , right ) , again supporting the idea that L-form proliferation is independent of the normal cell division machinery . We recently showed that a minor reduction in fatty acid synthesis , that had no effect on walled cell growth or division , specifically abolished B . subtilis L-form proliferation ( Mercier et al . , 2013 ) . To investigate whether similar effects could be observed in the newly characterized bacterial L-forms , we assessed the effects of reductions in the rate of membrane synthesis on L-form proliferation . In E . coli , we used a non-essential fatty acid ( FA ) synthesis mutant fabH previously demonstrated to have a reduced rate of membrane synthesis ( Yao et al . , 2012 ) . As shown in Figure 5A , a fabH null strain proliferated in the walled state on NA/MSM plates ( lipid II ON ) , while no growth was detected following a switch to the L-form mode of proliferation ( lipid II OFF , middle ) . Importantly , this growth defect was restored by a fabH+ complementing plasmid ( Figure 5A , left ) . Similar results were obtained using cerulenin , an antibiotic that inhibits FA synthesis , which specifically inhibited L-form proliferation ( Figure 5—figure supplement 1 ) . Finally , to demonstrate whether FA synthesis regulation was essential for E . coli L-form proliferation , we constructed a strain bearing a double deletion of murA and fabH ( strain RM369 , murA , fabH::Kn pSK122-murA , pOU82-fabH ) bearing murA+ on an unstable mini-F plasmid and fabH+ on an unstable mini-R1 plasmid . This strain was grown in both walled and L-form states on NA/MSM plates with no direct selection for the plasmids . After DNA extraction , we assessed the presence of the murA and fabH genes using multiplex PCR . As expected , in the walled state , the murA gene was retained because PG synthesis is essential , but the fabH gene was lost , because E . coli apparently has a second activity capable of supporting the fabH function ( Yao et al . , 2012 ) ( Figure 5C , lane 1 ) . Strikingly , in the L-form state , the opposite was observed: murA was lost , while fabH was retained ( Figure 5C , lane 2 ) , supporting the idea that a higher rate of FA synthesis is required for proliferation of E . coli in the L-form state . 10 . 7554/eLife . 04629 . 016Figure 5 . Essential role of fatty acid synthesis in L-forms growth of E . coli and C . glutamicum . ( A ) Growth of Escherichia coli strains RM365 ( ΔfabH ) and RM366 ( ΔfabH , pCA24N-fabH ) streaked on L-form supporting medium ( MSM ) in the absence ( lipid II ON ) or presence ( lipid II OFF ) of fosfomycin ( FOS ) . ( B ) L-form colonies of the E . coli strain RM369 ( ΔmurA , pSK122-Cm-ftsK , ΔfabH , pOU82-Amp-fabH ) on MSM plates after several repeated streakings on MSM plates in the presence of FOS . ( C ) Multiplex PCR of the genes , murA , fabH , and mreC on genomic DNA of the E . coli strain RM369 grown in the walled ( 1 ) or L-form ( 2 ) state . Samples obtained from strains in panel ( B ) . M represents the 100 bp DNA ladder . ( D ) Growth of Corynebacterium glutamicum streaked on MSM in the absence ( lipid II ON ) or presence ( lipid II OFF ) of D-cycloserine ( DCS ) , and with ( cerulenin ) or without ( no ) 2 μg/ml of cerulenin . ( E ) Typical images of C . glutamicum L-forms after 16 hr of growth in MSM with DCS in the absence ( −cerulenin ) or presence ( +cerulenin ) of 2 μg/ml of cerulenin . Scale bars , 3 μm . See also Videos 5 and 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 04629 . 01610 . 7554/eLife . 04629 . 017Figure 5—figure supplement 1 . Specific inhibition of Escherichia coli L-forms growth by cerulenin . Growth of E . coli walled ( top ) and L-form ( bottom ) strains streaked on L-form supporting medium ( MSM ) in the absence ( lipid II ON ) or presence ( lipid II OFF ) of fosfomycin with different concentration of cerulenin ( 0 , 10 , and 20 μg/ml ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04629 . 017 To test whether the rate of FA synthesis is also important for C . glutamicum L-form proliferation , we streaked growing walled and L-form cells on NA/MSM plates in the presence of 2 μg/ml of cerulenin . In accordance with the results for E . coli ( above ) , partial inhibition of FA synthesis specifically inhibited L-form proliferation ( Figure 5D , left ) , with no effect on the walled cells ( Figure 5D , middle ) . Time lapse microscopy was used to assess the effects of reduction of FA synthesis on L-form proliferation . As shown in Figure 5E , left , and Video 5 , in the absence of cerulenin , cells grow and divide normally . Strikingly , in the presence of cerulenin , cells continued to grow but shape deformations did not occur , and the cells remained more or less spherical with no detectable division events ( Figure 5E , right , and Video 6 ) . 10 . 7554/eLife . 04629 . 018Video 5 . Time lapse series showing L-form cell growth of Corynebacterium glutamicum growing in nutrient broth ( NB ) /L-form supporting medium ( MSM ) with D-cycloserine ( DCS ) in the absence of 2 μg/ml of cerulenin . Phase contrast images were acquired automatically every 5 min for about 16 hr . Scale bar , 3 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 04629 . 01810 . 7554/eLife . 04629 . 019Video 6 . Time lapse series showing L-form cell growth of Corynebacterium glutamicum growing in nutrient broth ( NB ) /L-form supporting medium ( MSM ) with D-cycloserine ( DCS ) in the presence of 2 μg/ml of cerulenin . Phase contrast images were acquired automatically every 5 min for about 16 hr . Scale bar , 3 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 04629 . 019 Thus , as previously described for B . subtilis L-forms , regulation of membrane synthesis seems to have a pivotal role in the proliferation of diverse Gram-positive and Gram-negative L-forms . Historically , L-forms from diverse bacteria were generated using many different cell wall inhibitors , such as β-lactams , glycopeptides , and lytic enzymes ( Domingue and Woody , 1997 ) . The wide range of methods used to create L-forms has probably contributed to the heterogeneity in phenotypic properties and has made it difficult to define general properties for L-form bacteria ( Domingue and Woody , 1997; Allan et al . , 2009 ) . Included in the range of cells designated L-form ‘like’ have been cell types in which the PG synthesis machinery remained essential for proliferation ( e . g . the E . coli cells of Joseleau-Petit et al . , 2007 and Cambre et al . , 2014 ) . Given that we have now shown that E . coli can be converted into a state in which the cell wall precursor pathway can be deleted and cells become completely resistant to β-lactam antibiotics , we suggest that in future the term L-form be restricted to fully wall deficient cells . We previously showed that for B . subtilis , inhibiting an earlier step of the PG precursor pathway efficiently generates proliferating L-forms ( Leaver et al . , 2009; Dominguez-Cuevas et al . , 2012 ) . Perhaps surprisingly , this approach appears to have been tried only rarely in previous L-form work ( Schmid , 1984 , 1985 ) . We showed here that inhibition of the PG precursor pathway readily generates L-forms in diverse bacteria of both Gram-positive ( S . aureus and C . glutamicum ) and Gram-negative ( E . coli ) varieties . Furthermore , this method generated genuine cell wall-free proliferative bacteria , as their growth was not inhibited by high concentrations of β-lactam antibiotics and , more importantly , essential PG synthesis genes could be deleted ( at least for B . subtilis and E . coli ) . Finally , as the PG precursor synthesis pathway is almost ubiquitous in bacteria , it is reasonable that this method could be applied to a very wide range of bacteria . We do not yet understand why PG precursor synthesis inhibition efficiently promotes L-form proliferation from different bacteria . However , we recently uncovered that in B . subtilis , PG precursor synthesis inhibition triggers , by an unknown mechanism , induction of an excess of membrane synthesis , a key process for L-form cell division ( Mercier et al . , 2013 ) . Thus , as PG synthesis needs to be coordinated either with membrane synthesis or cell growth , it is plausible that PG precursor synthesis inhibition has general effects on the regulation of membrane synthesis in bacteria . Having created different types of bacterial L-forms , we identified several common and differentiated properties , as summarized in Table 1 . 10 . 7554/eLife . 04629 . 020Table 1 . General properties of bacterial L-formsDOI: http://dx . doi . org/10 . 7554/eLife . 04629 . 020Bacterial L-form strainBacillus subtilisStaphylococcus aureusCorynebacterium glutamicumEscherichia coliMode of inductionMurE-B repressionFOSDCSFOSSecondary mutation requiredYesn . d . n . d . n . d . Timing of induction24 hr3 days48h3 daysGrowth conditionSolid/liquidSolidSolid/liquidSolidCell wall reversionYesYesYesYesCell division machineryNot essentialNot essentialNot essentialNot essentialMode of cell proliferationVesicles blebbing , fission , tubulationn . d . Vesicles blebbing , fission , tubulationVesicles blebbing , fission , tubulationReferencesLeaver et al . , 2009 , Mercier et al . , 2013 , Kawai et al . , 2014DCS: D-cycloserine; FOS: fosfomycin; n . d . : not determined . ( i ) Growth conditions . We previously found that B . subtilis L-forms can proliferate in both solid and liquid media ( Leaver et al . , 2009 ) . Interestingly , although C . glutamicum L-forms shared the ability to proliferate under both conditions , S . aureus and E . coli L-forms only grew on an agar surface . Another recently characterized L-form , from the bacterium Listeria monocytogenes , was also reported to grow only under semi-solid conditions ( Dell'Era et al . , 2009 ) . Thus the ability of L-forms to proliferate under different growth conditions is dependent on as yet unknown inherent properties of each bacterial species . ( ii ) Genetic mutations . We previously showed that L-form proliferation in B . subtilis requires , in addition to inhibition of PG precursor synthesis , a mutation in a gene such as ispA ( Leaver et al . , 2009; Mercier et al . , 2013 ) . In the absence of such a mutation , no growth is detected on inhibition of PG precursor synthesis ( Figure 1—figure supplement 1 ) . Interestingly , the different bacteria tested here readily proliferated on inhibition of PG precursor synthesis , strongly suggesting that no ispA-like mutation is needed to promote L-form proliferation . Thus it appears again that the requirement for a secondary mutation to promote L-form proliferation will depend on the bacterium tested . ( iii ) FtsZ independent cell division . The most remarkable property observed in B . subtilis L-forms was a mode of cell division independent of the normally essential protein based machinery ( Leaver et al . , 2009 ) . Remarkably , S . aureus and E . coli L-forms share the ability to proliferate independently of FtsZ , and although definitive experiments are more difficult to perform in C . glutamicum , it appears that they will also share this property . Thus FtsZ independent proliferation is a common trait of bacterial L-forms , presumably reflecting their strange blebbing/tubulation mode of growth . We suggest that the ability to tolerate deletion of essential cell division genes such as ftsZ will be a useful operational test for the true L-form state . ( iv ) Cell wall reversion . We recently showed that B . subtilis L-forms are able to synthesis a cell wall sacculus de novo , followed by reversion to the parental walled form ( Kawai et al . , 2014 ) . Similarly , after reactivating PG precursor synthesis by removal of FOS or DCS , the three different bacterial L-forms tested here could also revert to their parental walled forms , suggesting that the ability to rebuild a cell wall sacculus de novo is also a common property of bacteria . We reported here that C . glutamicum and E . coli L-forms , at least , appear to proliferate by cell shape deformations followed by spontaneous scission events , in a very similar manner to the process we have described for B . subtilis ( Leaver et al . , 2009; Mercier et al . , 2012 , 2013 ) . Additionally , as previously observed for B . subtilis , a minor reduction in FA synthesis prevented the growth of both C . glutamicum and E . coli L-forms , supporting the idea that all three L-forms divide by a similar mechanism based on an increased ratio of surface area to volume synthesis . Thus it appears that evolutionarily divergent bacteria , with different envelope structures ( e . g . Gram-positive and Gram-negative ) , shape ( e . g . rod vs sphere ) , and different modes of cell wall extension ( e . g . lateral and apical ) have retained a common primitive mode of proliferation when forced to grow in the absence of a cell wall . Interestingly , this mode of proliferation is strikingly similar to the mode of proliferation of simple vesicle systems independent of protein based machineries ( Hanczyc et al . , 2003; Budin et al . , 2009; Terasawa et al . , 2012 ) . Therefore , our results strengthen the idea that the L-form mode of proliferation could have been used by a common ancestor of the bacteria prior to the invention of the cell wall , and are consistent with the notion that invention of the wall was a pivotal moment in the evolutionary divergence of the bacterial lineage ( Errington , 2013 ) . We report here a simple and possibly widely generalizable method with which bacteria can be switched to a cell wall-free mode of proliferation . Apart from its apparent importance for understanding an early step in the evolution of life , the simple mechanism of proliferation of L-forms may find application in attempts to design and engineer synthetic self-replicative systems , or minimal cells ( Caspi and Dekker , 2014 ) . The ability to delete and then restore normally essential genes in L-forms offers a powerful new model system with which to investigate important properties of the cell wall synthesis and cell division machineries , with implications for the discovery and development of novel antibacterials ( Bugg et al . , 2011; den Blaauwen et al . , 2014 ) . The bacterial strains and plasmids constructs used in this study are shown in Table 2 . DNA manipulations and E . coli DH5α transformation were carried out using standard methods ( Sambrook et al . , 1989 ) . The plasmids pOU82-murA and pSK122-murA contain the operon yrbA-murA . The plasmids pOU82-ftsZ and pOU82-fabH contain the ftsZ or fabH gene , respectively , fused to a constitutive E . coli promoter ( ttgacagctagctcagtcctaggtactgtgcta ) designed by John Anderson ( IGEM2006_Berkeley ) . The E . coli murA ( RM345 ) and ftsZ ( RM349 ) deletion mutant strains were created using the Lambda Red recombinase system with a derivate of pKD4 as a template ( Datsenko and Wanner , 2000 ) . Briefly , the strains TB28 containing pOU82-murA or pOU82-ftsZ and pKD46-sp were transformed by a PCR product containing the kanamycin cassette flanked by 40 nt homology regions , just upstream of the start and downstream of the stop codons , of the genes murA or ftsZ . Deletions were tested by PCR and backcrossed into fresh TB28 containing pOU82-murA or pOU82-ftsZ using P1 transduction . 10 . 7554/eLife . 04629 . 021Table 2 . Bacterial strains and plasmids used in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 04629 . 021StrainRelevant genotypeReferenceBacillus subtilis Bs115168CA ΩspoVD::cat Pxyl-murE ΩamyE:: ( tet xylR ) ( Leaver et al . , 2009 ) LR2Bs115 xseB* ( frameshift 22T > − ) a ( Mercier et al . , 2013 ) Escherichia coli TB28MG1655 ΔlacIZYA ( Bernhardt and de Boer , 2003 ) ND101fstK::Kn pSC101-fstKF-X Barre Lab unpublished RM61TB28 ΔftsK::kan pSK122-ftsKThis study RM345TB28 ΔmurA::kan pOU82-murAThis study RM349TB28 ΔftsZ::kan pOU82-ftsZThis study RM350TB28 ΔftsZ pOU82-ftsZ ΔmurA::kan pSK122-murAThis study RM359TB28 ΔmreBCD::cat pTK549 ( Kruse et al . , 2005 ) RM365TB28 ΔfabH::kan ( Baba et al . , 2006 ) RM366RM365 pCA24N-fabHThis study RM369TB28 ΔfabH pOU82-fabH ΔmurA::kan pSK122-murAThis studyStaphylococcus aureus WTATCC29213Laboratory collection S . aureus ftsZR191PATCC29213 ftsZR191P ( Haydon et al . , 2008 ) RNpFtsZ-1RN4220 Pspac-ftsZ erm ( Pinho and Errington , 2003 ) Corynebacterium glutamicum WTATCC13032Laboratory collectionPlasmidRelevant genotypeReference/originEscherichia coli pCA24N-fabHlacIq pT5-lac-fabH cat ( Kitagawa et al . , 2005 ) pOU82R1-replicon , bla lacZYA ( Gerdes et al . , 1985 ) pOU82-murAR1-replicon , bla lacZYA murAThis study pOU82-ftsZR1-replicon , bla lacZYA ftsZThis study pOU82-fabHR1-replicon , bla lacZYA fabHThis study pTK549R1-replicon , kan Pmre-mreBCD ( Kruse et al . , 2005 ) pSK112F-replicon , cat lacZYAF-X Barre Lab unpublished pSK112-ftsKF-replicon , cat lacZYA ftsKF-X Barre Lab unpublished pSK112-murAF-replicon , cat lacZYA murAThis studybla: b-lactamase; cat: chloramphenicol; erm: erythromycin; lacZ: β-galactosidase; kan: kanamycin; tet: tetracyclin . The different walled bacterial cells ( B . subtilis , E . coli , S . aureus , and C . glutamicum ) were grown on NA ( Oxoid Limited , UK ) and in Luria–Bertani broth . Bacterial L-forms were grown in osmoprotective medium composed of 2× MSM media , pH 7 ( 40 mM MgCl2 , 1 M sucrose , and 40 mM maleic acid ) , mixed 1:1 with 2× nutrient broth ( NB , Oxoid ) or 2× NBA ( NB with 2% agarose ) . When necessary , antibiotics and supplements were added to media at the following concentrations: FOS 0 . 4 mg/ml; DCS 0 . 4 mg/ml; penicillin G 0 . 5 mg/ml; ampicillin 50 μg/ml or 0 . 5 mg/ml; chloramphenicol 25 μg/ml; kanamycin 25 μg/ml; erythromycin 10 μg/ml; cerulenin 2 μg/ml , 10 μg/ml , or 20 μg/ml; xylose 0 . 5%; IPTG 1 mg/ml; and 1 μg/ml benzamide ( FtsZ inhibitor 8J [Adams et al . , 2011] ) . For multiplex PCR , E . coli walled and L-form genomic DNA samples were prepared using a standard phenol–chloroform extraction procedure . The primer couples were designed using MPprimer software ( Shen et al . , 2010 ) , using an open reading frame nucleotide sequence . Standard PCR reaction procedures were applied using GoTaq DNA Polymerase ( Promega , Madison , WI ) with a melting temperature of 56°C . For snapshot microscopy , the different bacterial walled and L-form cells were immobilized on microscope slides covered with a thin film of 1% agarose in NB/MSM . The cells were imaged on a Zeiss Axiovert 200M microscope controlled by Metamorph 6 ( Molecular Devices , Sunnyvale , CA ) with a Zeiss ×100 Plan-Neofluar oil immersion objective . For time lapse microscopy , C . glutamicum L-form cells were imaged in ibiTreat adherent , 35 mm sterile glass bottom microwell dishes ( ibidi GmbH , Munich , Germany ) . Briefly , an 0 . 1 ml sample of exponential phase C . glutamicum L-form was added to 0 . 5 ml of fresh NB/MSM and incubated in the microwell dish for 15 min . The cells were washed three times with NB/MSM , and 0 . 5 ml of fresh NB/MSM with DCS was finally added . For E . coli , L-form cells were immobilized on microscope slides covered with a thin film of 1% agarose in NB/MSM with FOS . The cells were imaged on a DeltaVision RT microscope ( Applied Precision , Issaquah , WA ) controlled by softWoRx ( Applied Precision ) with a Zeiss ×100 apo fluor oil immersion lens . A Weather Station environmental chamber ( Precision Control ) regulated the temperature of the stage . Pictures and videos were prepared for publication using ImageJ ( http://rsb . info . nih . gov/ij ) and Adobe Photoshop .	Bacterial cells are surrounded by a cell wall made of a molecule called peptidoglycan . This wall is important for many aspects of cell survival including the maintenance of cell shape and protection from mechanical damage . However , many bacteria are able to switch to a state in which they don't have a cell wall . Although this wall-free state was discovered several decades ago , little is known about its general properties because there isn't a quick and reliable method for making such bacteria . Recently , it has been shown that bacteria of the species Bacillus subtilis can rapidly switch to the wall-free state when the production of peptidoglycan is reduced . Here , Mercier et al . show that the same method also works for a wide range of bacterial species . The wall-free states of the various species share the same unusual way of dividing to produce daughter cells . Normally , bacterial cell division is a highly controlled process involving a protein called FtsZ that accumulates at the site of cell division . In bacteria without walls , on the other hand , cell division does not require FtsZ , but instead depends on the rate of production of new cell membrane . Excessive production of membrane leads to the cell changing shape , resulting in spontaneous separation into daughter cells . The results suggest that this form of cell division is conserved across all bacteria . It is possible that this is an ancient mechanism that may have been used by the ancestors of modern bacteria , before the evolution of the cell wall . In future , this simple form of cell division could prove useful the development of synthetic living cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "microbiology", "and", "infectious", "disease" ]	2014	General principles for the formation and proliferation of a wall-free (L-form) state in bacteria
The rostrocaudal ( head-to-tail ) axis is supplied by populations of progenitors at the caudal end of the embryo . Despite recent advances characterising one of these populations , the neuromesodermal progenitors , their nature and relationship to other populations remains unclear . Here we show that neuromesodermal progenitors are a single Sox2lowTlow entity whose choice of neural or mesodermal fate is dictated by their position in the progenitor region . The choice of mesoderm fate is Wnt/β-catenin dependent . Wnt/β-catenin signalling is also required for a previously unrecognised phase of progenitor expansion during mid-trunk formation . Lateral/ventral mesoderm progenitors represent a distinct committed state that is unable to differentiate to neural fates , even upon overexpression of the neural transcription factor Sox2 . They do not require Wnt/β-catenin signalling for mesoderm differentiation . This information aids the correct interpretation of in vivo genetic studies and the development of in vitro protocols for generating physiologically-relevant cell populations of clinical interest . The vertebrate rostrocaudal axis elongates in a rostral-to-caudal sequence by virtue of a population of progenitors at the caudal end of the embryo , in and near the primitive streak ( PS ) and later in the tail bud ( TB; reviewed in Wilson et al . ( 2009 ) ) . Clonal analysis has demonstrated the presence of axial progenitors that behave as dual-fated tissue stem cells of the neurectoderm and somitic mesoderm ( Tzouanacou et al . , 2009 ) , termed neuromesodermal progenitors ( NMPs ) . Population fate maps at early somite stages have identified two regions of dual neurectoderm and mesoderm fate: the dorsal layer of the bilayered Node-Streak Border ( NSB ) and the Caudal Lateral Epiblast ( CLE ) on either side of the PS . The NSB also gives rise to the majority of the neuromesoderm-fated chordoneural hinge ( CNH ) in the tail bud ( Cambray and Wilson , 2002; 2007 ) . The adjacent midline PS and rostral node ( RN ) are fated exclusively for mesoderm and notochord , respectively ( Figure 1A–D ) . The caudal-most tip of the CLE , and the adjoining caudal primitive streak , are solely fated for lateral/ventral mesoderm ( LVM ) ( Cambray and Wilson , 2007 ) . Thus cell fate in and around the PS is highly regionalised . 10 . 7554/eLife . 10042 . 003Figure 1 . Location of neuromesodermal and lateral/ventral mesoderm progenitors . ( A ) Diagrams of the E8 . 5 ( 2–5 s ) embryo showing the location of neuromesodermal and lateral/ventral-fated progenitors and the terminology used in this study . NSB , node-streak border; CLE , caudal lateral epiblast; RN , rostral node; St1 , rostral 1/5 of the streak; St5 , caudal 1/5 of the streak; L1-5 , lateral fifths of the lateral epiblast , with L1 corresponding to most rostral adjacent to St1 , and L5 adjacent to St5 . ( B ) DAPI-stained sections through the E8 . 5 ( 2–5 s ) embryo . ( Ba ) Transverse section illustrates dissected midline regions . ( Bb ) Magnified view of the NSB region shows the rostral , mid and caudal border ( termed Br , Bm and Bc ) . ( Bc ) Transverse section through the mid primitive streak shows the position of Lmid , Llat and Lmed ( mid , lateral and medial CLE , respectively ) . ( C ) At mid-gestation , NMPs are located in the chordoneural hinge ( CNH ) . The CNH is composed of two parts: the dorsal ( neural ) part ( termed CNH-N ) and the ventral ( notochordal ) part ( CNH-Noto ) . ( D ) Abbreviations of embryonic regions used in this manuscript . DOI: http://dx . doi . org/10 . 7554/eLife . 10042 . 003 Fate mapping studies in early somite-stage mouse and chick embryos indicate a rostrocaudal organisation of mesoderm progenitors within the PS . Midline mesoderm is produced by the rostral PS , while successively more caudal regions of the streak are fated for the paraxial and lateral mesoderm ( Cambray and Wilson , 2007; Brown and Storey , 2000; Catala et al . , 1996; Schoenwolf , 1992; Wilson and Beddington , 1996 ) . We have previously shown that cells fated exclusively for the medial somite in the NSB can be re-fated to more lateral regions of the somite on transplantation to the anterior primitive streak ( Cambray and Wilson , 2007 ) , indicating a degree of plasticity in paraxial mesoderm ( PXM ) fates . However , the extent of mesoderm plasticity in the PS and CLE has not been fully investigated . In particular , it is not known whether axial , paraxial and lateral mesoderm progenitors in the PS region are committed to these fates . Fate maps of the CLE in chick indicate that cells at the rostral or lateral edges of the CLE form more neurectoderm than those at caudal or medial positions , which tend to generate mesoderm ( Brown and Storey , 2000; Iimura et al . , 2007 ) . Furthermore , fate maps of mouse and chick have suggested that CLE progenitors are more transitory than those in the NSB , since they do not contribute to long axial stretches or extensively to the CNH ( Cambray and Wilson , 2007; Brown and Storey , 2000; Iimura et al . , 2007 ) . However , the potency of these populations has not been tested , leaving open the question of whether these regional differences are simply the product of differing cellular environments acting on a single cell type to determine fate , or whether multiple populations with restricted potency exist within the PS region . T , Tbx6 , Fgfr1 and Wnt3a are expressed in the PS region and required for correct mesoderm production , and loss of each of them leads both to shortened axes , and the ectopic production of neural tissue at the expense of somitic mesoderm ( Chapman and Papaioannou , 1998; Yamaguchi et al . , 1999; Yoshikawa et al . , 1997; Ciruna et al . , 1997 ) . This suggests that NMP maintenance is intimately linked with preserving a balance between neurectoderm and mesoderm production . Tbx6 expression in the midline PS represses Sox2 in mesoderm-fated cells , ensuring suppression of the neural transcription program ( Takemoto et al . , 2011 ) . Furthermore , in zebrafish , Wnt/β-catenin activation influences the decision of cells in both gastrula- and somite-stage embryos to enter neural or mesodermal lineages ( Martin and Kimelman , 2012 ) . More recently , lineage-tracing experiments showed that conditional deletion of Wnt3a or β-catenin in the T+ cell compartment leads to a switch of primitive streak progenitors towards a neural fate ( Garriock et al . , 2015 ) . However , constitutive Wnt/β-catenin activity in the T+ cell compartment is not sufficient to divert all neural progenitors to mesoderm fates: providing cells in the caudal progenitor region with a stabilised form of β-catenin results in an enlarged PSM domain , but does not lead to loss of neural cell production ( Aulehla et al . , 2008; Jurberg et al . , 2014 ) . Moreover , enhanced β-catenin activity does not necessarily compromise the presence of NMPs in the CLE ( Garriock et al . , 2015 ) . While these experiments point to an important role of Wnt signalling in axial progenitors , the promoters used do not specifically target NMPs . Grafting of precise NMP areas can provide a complementary approach that allows a direct assessment of the currently unresolved roles of Wnt signalling in NMPs and the caudal-most CLE . In this study , we investigate the heterogeneity , plasticity and commitment of NMPs and lateral/ventral mesoderm progenitors , and the mechanisms by which they choose between alternative fates . We find that NMPs are committed to neuromesodermal lineages and choose between retention as progenitors , and differentiation as either neurectoderm or mesoderm based on their location within the progenitor region; the latter choice is β-catenin dependent . We show that NMPs express low levels of T and Sox2 , and that during mid-trunk formation , Wnt/β-catenin signalling expands the number of Sox2+T+ NMPs and maintains the appropriate level of T in the NMP population . We further show that lateral/ventral mesoderm progenitors are exclusively mesoderm-committed yet show plasticity within the mesoderm lineage , and respond to distinct signalling and transcription factor cues from those that govern NMPs . The potency of NM-fated ( NSB , L1-3 , CNH ) and surrounding regions was examined by transplantation under the kidney capsule ( Figure 2A–C ) . Control grafts of embryonic day ( E ) 7 . 5 anterior ( rostral ) or posterior ( caudal , PS-containing ) parts of the late-streak or early headfold stage embryo formed large teratocarcinomas containing embryonal carcinoma ( EC ) cells and derivatives of all three germ layers including neural and non-neural ectoderm ( Beddington , 1983; Osorno et al . , 2012 ) . In contrast , E8 . 5 ( 2–6 somite ) grafts gave rise to smaller tissue masses containing only well-differentiated tissues and no EC cells . NSB , CLE ( L1-3 ) and most ( 4/5 ) CNH grafts gave rise only to neural and mesodermal derivatives , although one CNH graft included keratinised epithelium , possibly through contamination from neighbouring specified surface ectoderm cells . Grafts of the caudal CLE and neighbouring midline ( L/St5 ) , both of which produce LVM ( Cambray and Wilson , 2007 ) , produced very small growths devoid of neurectoderm . Rostral PS ( St1 ) grafts predominantly produced adipocytes , while rostral node ( RN ) grafts produced mainly neurectoderm . Therefore , NM-fated cells in ≥E8 . 5 embryonic regions are not pluripotent but restricted in potency to neural and mesoderm fates , while LVM-fated cells produce only mesoderm . 10 . 7554/eLife . 10042 . 004Figure 2 . The potency of NM-fated regions is restricted to neural and mesodermal lineages . Grafts of primitive streak and tail bud regions to the kidney capsule . ( A ) Masson’s trichrome-stained tumour sections derived from the indicated embryonic tissue regions . ( B ) Percentage of tumours that contain any of the scored tissues . ( C ) Average tumour area per stage . E7 . 5 cells give rise to larger tumours compared to tumours , derived from E8 . 5 or E10 . 5 grafts ( * , p<0 . 01 ) . a , adipose; b , bone; c , cartilage; ec , embryonal carcinoma cells; kw , keratin whorl; k , adult kidney; n , neural; m , skeletal muscle . DOI: http://dx . doi . org/10 . 7554/eLife . 10042 . 004 We recently showed that coexpression of Sox2 and T colocalises with regions of NM fate in the mouse E8 . 5 caudal region ( Tsakiridis et al . , 2014 ) . This combinatorial marker has also been reported to identify similar regions in zebrafish , chick and human ( Martin and Kimelman , 2012; Olivera-Martinez et al . , 2012 ) . However , the spatiotemporal correspondence between NMP activity and Sox2/T coexpression has never been rigorously tested in any organism . To determine whether Sox2+T+ cells coincide with NMP activity , we assessed the spatiotemporal expression of Sox2 and T throughout axis elongation . Wholemount in situ hybridisation and immunofluorescence showed that Sox2+T+ cells were first detected in the epiblast close to the NSB of the E7 . 5 ( late bud stage ) embryo ( Figure 3Ba , b ) . At E8 . 5 the expression of Sox2 transcripts and protein was strong in the differentiating neurectoderm , and weaker in the CLE , while T expression in the CLE was most prominent caudally in the PS ( Figure 3A , B ) . During trunk and tail formation , Sox2/T coexpression persisted in the CNH and surrounding areas ( E9 . 5–13 . 5; Figure 3B–D and 9Ca–c ) . At E13 . 5 , just before elongation had ceased , a few Sox2+T+ cells could still be detected , whereas once axial elongation was complete , Sox2+T+ cells had disappeared . At E14 . 5 , T expression was detected in the caudal notochord , while Sox2 expression was absent , consistent with the absence of the NT by this stage ( Figure 3D–F and Figure 3—figure supplement 1A ) . Thus , Sox2+T+ cells are found at all known locations and times that NMPs are present . 10 . 7554/eLife . 10042 . 005Figure 3 . Sox2+T+cells coincide with NMP regions . ( A ) In situ hybridisation for Sox2 and T . Abbreviations are the same as in Figure 1D . Noto , notochord; tbm , tail bud mesoderm . ( B–F ) Confocal sections of wholemount , immunostained embryos; DAPI-counterstain in grey . ( Ba ) Parasagittal section through a late bud ( LB ) stage embryo . ( Bb ) NSB magnified from Ba ( arrowheads , Sox2+T+ cells ) . ( Bc–l ) Transverse sections through the caudal progenitor area at E8 . 5 ( Bc–f ) and E9 . 5 ( Bg–l ) show double positive cells in the NSB and CLE . ( Ca–b ) Sagittal sections through the E10 . 5 tail bud . ( Da–c ) Sox2+T+ cells are detected in the E11 . 5 tail bud , but become sparser in the E12 . 5 and E13 . 5 ( up to 63s ) tail bud . ( E ) Overlapping Sox2 and T expression has all but disappeared in the E13 . 5 ( 65s ) tail bud in which somitogenesis is complete . ( F ) No Sox2+T+ cells were detected in the E14 . 5 tail tip . Inset , shows the overexposed red field ( Sox2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10042 . 00510 . 7554/eLife . 10042 . 006Figure 3—figure supplement 1 . Quantifying Sox2+T+cells during axis elongation . ( A ) Double in situ hybridisation for Sox2 and T at the end of axis elongation ( E13 . 5 , 65s ) . ( B ) Workflow for Sox2/T quantification and 3D reconstruction . ( C and F ) Total cells analysed per embryonic stage ( C ) or in 4-OHT-treated βcatCKO embryos ( F ) . Data is shown as the mean ± s . d . ( D and G ) Quantification of Sox2/T populations per embryonic stage ( D ) or in 4-OHT-treated βcatCKO embryos ( G ) . Data is shown as the mean ( s . d . ) with n , the number of embryos analysed . ( E and H ) Statistical analysis . ( E ) A standard , unpaired Student’s t-test was used to test the significance in Sox2+T+ cell number and the total cells analysed between consecutive stages . ( H ) P-values obtained through an unpaired Student's t-test ( with Welch’s correction ) comparing Sox2/T populations in E9 . 5 and E8 . 5 wildtype to E9 . 5 4-OHT-treated βcatCKO embryos ( * , p-value<0 . 05; ** , p-value<0 . 0001; ns , not significantly different ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10042 . 006 Use of an automated segmentation algorithm allowed us to further analyse and quantify T and Sox2 expression in confocal z-stacks throughout axis elongation ( Figure 3—figure supplement 1B and Materials and methods ) . The majority of Sox2+T+ cells described a ‘U’ shape in the epiblast at E8 . 5 including the NSB and L1-3 , coinciding with NMP locations ( Figure 4Aa , b ) . The total number of cells labelled with either marker increased up to E10 . 5 , decreasing thereafter , while Sox2+T+ cell numbers showed a similar profile but peaked at E9 . 5 . Sox2+T+ cells were rare at E13 . 5 , and undetectable by E14 . 5 ( Figure 4B , C , Figure 3—figure supplement 1C–E and Video 1 ) . Thus the number of putative NMPs peaks during mid-trunk formation . We further quantified the levels of each transcription factor per cell during the PS-to-TB transition at E8 . 5–10 . 5 . At E8 . 5 the levels of T were lower in the lateral than the medial CLE and the PS . T expression was high and ubiquitous in L/St5 . The highest levels of T were found in the notochord . Sox2 was detected in a rostral-to-caudal gradient , with higher expression levels found rostrally ( Figure 5A ) . Strikingly , Sox2+T+ cells were excluded from high Sox2 and high T expressing regions ( Figure 5A–C ) . 10 . 7554/eLife . 10042 . 007Figure 4 . Sox2+T+ NMPs peak at mid trunk formation . ( A ) 3D reconstruction of the E8 . 5 caudal region . ( Aa–b ) Dorsal view . ( Ac ) Frontal view along the PS and CLE . Colours show different thresholded Sox2/T populations . ( B–C ) Quantification of different Sox2/T populations in the caudal embryo shows a peak in overall cell labelling at E10 . 5 ( B ) . Sox2+T+ cell numbers are highest at E9 . 5 ( C ) . Data in graphs is shown as the mean ± s . d . See also Figure 3—figure supplement 1B–E and Video 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10042 . 00710 . 7554/eLife . 10042 . 008Video 1 . Sox2 and T expression domains during axis elongation . 3D reconstruction of the caudal progenitor region at different stages of axis elongation . ( 00:00 ) E8 . 5 ( 2-5s ) ; ( 00:38 ) E9 . 5; ( 01:02 ) E10 . 5; ( 01:26 ) E11 . 5; ( 01:47 ) E12 . 5; ( 02:08 ) E13 . 5 ( 62-63s ) ; ( 02:33 ) E13 . 5 ( 65s ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10042 . 00810 . 7554/eLife . 10042 . 009Figure 5 . NMPs express low levels of T and Sox2 . ( A–C ) 3D analysis showing the relative levels of Sox2 and T protein in the E8 . 5–10 . 5 caudal region . Sox2+T+ cells express low-to-medium levels of both transcription factors ( green cells in Ac , Ad , Bc and Cc represent Sox2+T+ cells , indicated by the area within the white dotted line ) . Colours represent the intensity range shown as arbitrary units ( AU ) . The lower threshold for positivity was calculated as before ( see Figure 3—figure supplement 1B ) , with the maximum corresponding to the highest intensity recorded in the z-stack . Asterisk , Sox2+T+ cells found in the dorsocaudal part of the gut; grey , segmented nuclear volumes negative for either transcription factor . DOI: http://dx . doi . org/10 . 7554/eLife . 10042 . 009 We observed small numbers of Sox2+T+ cells in regions not expected to harbour NMPs , such as the midline PS at E8 . 5 , suggesting that although NM progenitor identity and Sox2/T positivity substantially overlap , either this correspondence is not absolute , or previous grafting studies ( Cambray and Wilson , 2007 ) may not detect low proportions of NMPs . Using a Cited1-CreERT2-labelling system ( tracing mesodermal and hindgut lineages ) , Garriock et al . found that some ingressed PS cells behave similarly to cells in the epiblast when Wnt3a is lost ( Garriock et al . , 2015 ) , implying that some Sox2+T+ cells in the PS might retain NMP potential during early mesoderm differentiation . Interestingly , at E8 . 5 , the numbers of Sox2+T+ cells were highest in L1 , declining towards L3 . Furthermore , the medial CLE contained more Sox2+T+ cells than lateral CLE ( Figure 3Bc–f and 4A ) . Together with the graded rostral-to-caudal decline in Sox2 and increase in T , these results suggest cellular heterogeneity within the CLE . To test whether the differences in Sox2+T+ cell numbers and relative expression within the CLE reflect differences in fate , we used homotopic grafts of groups of ~100 GFP+ cells to refine the existing fate map of the CLE ( donor and host locations are shown in Figure 1A , B and in the diagrams of Figure 6B–D ) . We analysed cell fate in the Sox2+T+ rostral CLE in both rostral-to-caudal ( L1-3; Figure 6Aa–l and 6B ) and medial-to-lateral axes ( Lmed and Llat; Figure 6Am–t and 6C ) and also in the Sox2-T+ caudal CLE and streak ( L/St5; Figure 6Au–x and 6D ) . To check the accuracy of dissection , we performed immunohistochemistry on dissected L1-3 and L/St5 pieces . While Sox2+T+ cells were abundant throughout L1-3 regions ( n = 7 ) , no double positive cells were detected in L/St5 pieces ( n = 5; Figure 6—figure supplement 1 ) . In general , homotopic grafts incorporated well in cultured embryos ( 37 incorporated/42 grafts performed; Figure 6—figure supplement 2 and Figure 6—figure supplement 3 ) . 10 . 7554/eLife . 10042 . 010Figure 6 . Fate in the caudal lateral epiblast . ( A ) Representative examples of homotopic CLE grafts ( numbers identify individual embryos in B–D ) . Cell fate in the rostral CLE in rostral-to-caudal ( Aa–l ) and medial-to-lateral direction ( Am–t ) . Cell fate in the caudal-most lateral epiblast ( Au–x ) . Arrows show the notochord; white boxes the CNH . lvm , lateral and ventral mesoderm; nt , neural tube; pxm , paraxial mesoderm; tbm , tail bud mesoderm . ( B–D ) Diagrams indicate the graft type performed , the shading within them , the predominant prospective fate of each region ( pink , mesoderm; light blue , neuromesodermal; dark blue , neural; light brown , lateral/ventral mesoderm ) . Graphs display single grafted embryos and their contribution in the differentiated axis to the neurectoderm ( N ) and mesoderm ( PXM ) or both . Numbers below the bars indicate the graft series followed by an individual embryo identifier ( e . g . embryo 1 . 01 ) . Below the x-axis , graft cell contribution in the TB ( dorsal part of the CNH , ventral ( notochordal ) part of CNH and TBM ) , is represented by a black square , grey square or grey circle , respectively . ( B ) Fate in L1 , L2 and L3 . L1 grafts gave rise to two distinct contribution patterns , L1A ( axis only ) and L1AT ( axis and tail bud ) . ( C ) Fate in either the medial or lateral half of L1 or L2 . One graft ( embryo 2 . 06 ) contributed to the neural crest ( NC ) . ( D ) Grafts of L/St5 contributed entirely to the lateral/ventral mesoderm ( LVM ) . Unilateral versus bilateral PXM contribution is shown in Figure 6—figure supplement 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 10042 . 01010 . 7554/eLife . 10042 . 011Figure 6—source data 1 . Presomitic mesoderm contamination in CLE grafts . Additional information and analysis to show the majority of paraxial mesoderm in CLE grafts is derived from NMPs rather than contaminating PSM progenitors that were co-grafted . DOI: http://dx . doi . org/10 . 7554/eLife . 10042 . 01110 . 7554/eLife . 10042 . 012Figure 6—figure supplement 1 . Rostral CLE tissue contains Sox2+T+ cells . Graft donor tissue was examined for the presence of Sox2/T coexpressing cells . L1-3 and L/St5 pieces were dissected as described above . The underlying presomitic mesoderm was manually removed in the L1-3 grafts . ( A ) L1-3 donor tissue contains Sox2+T+ cells ( n = 7/7 ) . ( B ) L/St5 were devoid of coexpressing cells ( n = 5/5 ) . Blue , DAPI nuclear stain; green , Sox2; red , T . DOI: http://dx . doi . org/10 . 7554/eLife . 10042 . 01210 . 7554/eLife . 10042 . 013Figure 6—figure supplement 2 . Embryo numbers and section count in different grafting experiments . ( A ) Schematic diagram summarising the different grafting experiments conducted in this work and the embryos series they correspond to . Diagrams represent the primitive streak region at E8 . 5 , with coloured regions the donor tissue . In homotopic grafts , the colour indicates the predominant fate in that region ( pink , mesoderm; light blue , neuromesodermal; dark blue , neural; dark pink , lateral/ventral mesoderm ) . ( B ) Number of embryo grafts performed in this study , using AGFP7 donor cells . The number of embryos that was not scored ( and reasons thereof ) is also shown . ( C ) Number of sections scored for each graft series performed . DOI: http://dx . doi . org/10 . 7554/eLife . 10042 . 01310 . 7554/eLife . 10042 . 014Figure 6—figure supplement 3 . Homotopic and heterotopic grafts incorporate well into host embryos . ( A–C ) Immunohistochemical confirmation of differentiation markers in different graft series . Upper row , DAPI-counterstained images ( grey ) . Boxes , region magnified in lower row . Lower row , immunofluorescence channel . Position of grafted donor cells is outlined in white . ( A ) Sox2 , T , and Foxa2 expression in axis and TB sections of L1 homotopic grafts . Note lack of notochord-specific Foxa2 expression in graft-derived cells . ( B ) Sox2 and T expression in axis and TB sections of L2-3 heterotopic grafts ( to Br ) . ( C ) PDGFRβ expression in L/St5-derived cells , grafted either homo- or heterotopically to a wildtype host . ( D ) Table showing the number of embryos and sections stained for each marker . ( E ) Overview of marker expression in different regions of the axis . Note that none of the CLE-derived grafted cells expressed FoxA2 or high levels of T protein in the caudal notochord domain . DOI: http://dx . doi . org/10 . 7554/eLife . 10042 . 01410 . 7554/eLife . 10042 . 015Figure 6—figure supplement 4 . Fate of the CLE progenitors in the paraxial mesoderm . ( A-B ) Scoring of GFP+ cell contribution to the paraxial mesoderm in CLE homotopic grafts ( embryo series 1 and 2 ) . No PXM contribution was observed in L/St5 homotopic grafts . Graph format is the same as in Figure 6 . Pink bars show unilateral contribution; red bars bilateral contribution . ( C ) One grafted embryo ( embryo 2 . 06 ) showed contribution to the neural crest ( nc , white arrows ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10042 . 015 L1-3 homotopic grafts ( embryos 1 . 01–1 . 21; nsections = 613 ) were sectioned and scored for the presence of incorporated cells . Five of 11 L1 grafts ( L1-axis , ‘L1A’ grafts ) gave rise to differentiated regions of the axis and did not contribute to the TB . These contributed extensively and unilaterally to neurectoderm ( Figure 6Aa–d ) , and at lower frequency to mesoderm , predominantly ( in 31/36 sections ) unilaterally ( Figure 6—figure supplement 4A ) . This pattern resembled that of homotopic grafts to the immediately rostral Lateral Border ( LB ) region ( Cambray and Wilson , 2007 ) . The failure of L1A grafts to contribute to the TB shows that these cells are en route for exit from the progenitor region , and the unilateral mesoderm contribution suggests that this mesoderm was already formed and carried alongside the CLE graft rather than ingressing through the PS . This is in agreement with previous reports that the presomitic mesoderm underlying the CLE contributes only to short stretches of axial tissue ( estimated ~6 somites ) ( Nicolas et al . , 1996; Tam , 1986; 1988 ) ( see also Figure 6—source data 1 ) . The remaining six L1 grafts ( L1-axis/tail , ‘L1AT’ ) , as well as 9/10 L2-3 homotopic grafts , contributed from a variable rostral limit as far as the tail bud ( Figure 6Ae–l ) . These grafts all contributed extensively and bilaterally to mesoderm , the latter indicating that the mesodermal component had passed through the PS ( Figure 6—figure supplement 4A ) . Nearly all ( 15/16 ) L1AT and L2-3 grafts contributed to the TBM , but only L1AT grafts consistently colonised the CNH , indicating that L1 contains putative NMPs ( Figure 6B ) . Strikingly , only L1 grafts showed significant neural contribution in the axis , while both L1 and L2-3 grafts contributed to mesoderm . Thus , most neural-fated progenitors are in the rostral-most CLE and the boundary between short-term ( LB and L1A ) and long-term ( L1AT ) progenitors lies within the L1 region . Grafts to test the mediolateral fate of CLE cells ( embryos 2 . 01–2 . 08; nsections = 196; Figure 6Am–t , 6C ) did not show any major differences in neural or mesoderm contribution in L1/2med grafts compared to the L1-3 grafted tissue . However , when donor cells were grafted in a more lateral position ( L2lat ) , their descendants colonised mainly the lateral and dorsal neural tube . In contrast , contribution to the mesoderm was low , largely unilateral and found at the rostral limit of the labelled part of the axis ( in 10/12 sections; Figure 6—figure supplement 4B ) , suggesting this contribution might have originated largely from co-grafted committed presomitic mesoderm cells at the start of culture . In one grafted embryo ( embryo 2 . 06 ) , donor cells contributed to the neural crest ( Figure 6—figure supplement 4C ) . Thus , while NM fate is found in both medial and lateral CLE , cells in lateral positions are more likely to adopt neural ( particularly dorsolateral ) fates . Lastly , we assessed the fate in the caudal extreme of the CLE ( L/St5; embryos 3 . 01–3 . 04; nsections = 103; Figure 6Au–x , D ) . All homotopic L/St5 grafts displayed consistent LVM contribution to a small stretch of the post-hindlimb axis . While homotopic L2-3 grafts displayed low but reproducible LVM contribution distributed throughout the labelled stretch of axis ( 20% of sections in 80% of embryos; Figure 6Aj and Figure 10—figure supplement 1A ) , LVM contribution was very rare in L1 grafts ( <1% of sections in 9% of embryos; Figure 10—figure supplement 1A ) . Thus , L/St5 is the only region to contribute exclusively to LVM ( Figure 6Au–w , Figure 10—figure supplement 1A and Cambray and Wilson ( 2007 ) ) . These homotopic grafts show that neural and mesodermal fate is more highly regionalised than previously appreciated . In the rostral CLE ( L1-3 ) , cells in the caudal and medial parts of the CLE are more likely to adopt paraxial mesodermal fates , while the rostral and lateral CLE favours neural fates . These fates correlate with the relative levels of Sox2 and T , higher levels of which appear to predict neural/ neuromesodermal and mesodermal fates respectively . The caudal CLE ( L/St5 ) , which expresses no Sox2 , adopts exclusively lateral/ventral mesoderm fates . The regional diversity of fates in the CLE prompted us to test whether CLE cells were inherently biased towards neural or mesodermal lineages . Since the NSB contributes robustly to both neural and mesodermal lineages ( Cambray and Wilson , 2007 ) we used this site to test the plasticity of CLE cells by performing CLE-to-NSB grafts . As for homotopic grafts , heterotopic grafts incorporated well in the axis ( 57 incorporated/61 grafts performed ) . Strikingly , the relative contribution to neural or mesodermal fates in the axis varied extensively between grafts of rostral CLE ( L1-3 ) to NSB ( embryos 4 . 01–4 . 09; Figure 7A–B ) . 10 . 7554/eLife . 10042 . 016Figure 7 . Plasticity of L1-3 cells on heterotopic grafting to the NSB . ( A ) Representative examples of heterotopic CLE grafts ( numbers identify individual embryos in B ) . Three different contribution patterns were observed: mainly neural ( Aa-d ) , neuromesodermal ( Ae–h ) and mainly mesodermal ( Ai–l ) . ( B–D and F ) Scoring of GFP+ cell contribution in CLE-to-NSB heterotopic grafts . Graph format is the same as in Figure 6 . Highlighted numbers indicate embryos in which the graft fate has changed from that predicted by the fate map upon heterotopic grafting . ( B ) Grafts of L1 , 2 or 3 to the NSB . ( C ) Grafts of medial or lateral halves of L1 or L2 to the NSB . Note that these embryos are also scored with respect to their rostrocaudal origin in D . ( D ) Grafts of L1 , 2 or 3 to the NSB . Note that some of these embryos are also scored according to their mediolateral origin . For example embryo 5 . 13 received a graft of cells from medial L1 , and showed high neural contribution and is therefore highlighted as ‘changed fate’ in C but not in D . ( E ) Graft position at the start of culture ( to Br , Bm or Bc ) . Donor origin is shown in green . The crown ( arrowhead ) of the node ( dotted line ) was used as a landmark for placing GFP+ donor cells ( green ) . ( F ) Grafts of L1 , 2 or 3 to different aspects of the NSB shows the fate of rostral CLE cells is dependent on the environmental cues of the host . DOI: http://dx . doi . org/10 . 7554/eLife . 10042 . 016 Separating L1 from L2-3 ( embryos 5 . 01–5 . 18 ) tested whether this observed variation was influenced by the origin of the grafted tissue . However , both L1-to-NSB and L2-3-to-NSB grafts appeared indistinguishable , with about half of the embryos in each class containing neural contribution of differing extents . We further tested whether an intrinsic mediolateral bias could instead explain the variation seen above . Medial or lateral L1 or L2 pieces were grafted to the NSB . However , these grafts did not show consistent differences in neural or mesodermal differentiation , and some grafts reversed their predicted fate ( Figure 7C–D ) . Thus neither the neural fate of L1 and Llat , nor the mesodermal fate of L2-3 were retained after heterotopic grafting to the NSB , and therefore could not account for the variable neural and mesodermal contribution seen in individual grafted embryos . Consequently , the fate differences seen in different CLE areas cannot be explained by ≥2 localised populations of restricted potential . The NSB forms the junction between the streak , containing mesoderm-fated cells , and the caudal part of the node , which , according to fate maps in chick ( Selleck and Stern , 1991 ) contains neural-fated cells . In chick , the exact location in the node influences cell fate . To test whether exact position within the NSB could predict neural versus mesodermal contribution , we grafted L1-3 cells to the rostral , middle , or caudal aspects of the NSB ( Br , Bm or Bc respectively; Figure 7E . See also Figure 1Bb ) . In contrast to the previous series , these grafts produced distinct and reproducible patterns of labelled cell distribution ( embryos 6 . 01–6 . 21 ) . Rostral CLE cells placed in either Br or Bm tended to contribute entirely , or predominantly , to neural tube , while grafts to Bc resulted in a high proportion of embryos with mesoderm-only contribution . In several embryos , the position within the NSB reversed the predicted fate bias of L1 towards neural and L2-3 towards mesodermal fates: 5/5 L2-3-to-Br grafts contributed to the NT , while embryo 6 . 21 , an L1 graft , gave rise exclusively to mesoderm ( Figure 7F ) . Thus our data indicate that assignment of rostral CLE cells to neural or mesodermal fates is dependent on environmental cues , with no detectable cell-intrinsic biases . This data therefore indicates a single NMP type capable of context-specific neural or mesodermal differentiation and retention in the progenitor region . Homotopic NSB grafts give rise to a short stretch of notochord ( Cambray and Wilson , 2007 ) but it was not clear from these grafts whether this was derived from the dorsal or ventral layer of the node . CLE-derived cells were detected in the caudal notochord ( Figure 6—figure supplement 3 ) . However , these cells did not express the notochord marker Foxa2 , nor the high levels of T expressed in neighbouring host notochord cells . Therefore , although they were able to enter the notochord domain , they were unable to differentiate correctly , suggesting that E8 . 5 NMPs are not notochord progenitors . This is consistent with fate mapping studies indicating that the posterior notochord is derived from nodal positive cells in the mesoderm adjacent to the node ( Brennan et al . , 2002 ) and that convergent extension is the main driver of notochord extension at early somite stages ( Yamanaka et al . , 2007 ) . Together , these results suggest that from E8 . 5 onwards the notochord may elongate primarily as a result of rearrangement of pre-existing notochord progenitors rather than de novo addition of cells from the epiblast layer . To further test the context specificity of NMPs , we compared the overall neural versus mesoderm and CNH contribution of all graft series . Grafting L2-3 progenitors to the midline NSB not only changed their respective neural versus mesodermal contribution ( shown as the percentage of all sections for a given series; Figure 8A ) , but also increased their CNH contribution ( from 20% to 42% of sections ) , suggesting that a midline position favours NMP maintenance throughout axis elongation . 10 . 7554/eLife . 10042 . 017Figure 8 . Dorsolateral bias of L1-3 cells is reset upon grafting to the midline . ( A ) Quantitative analysis of L1-3 graft contribution . Left , the percentage of neural ( N ) vs paraxial mesoderm ( PXM ) contribution in all scorable sections . Right , the percentage of embryos with dorsal ( neural ) CNH contribution . Labels show the embryo series and the graft performed ( with n , the number of embryos ) . ( B ) Representative examples of dorsolateral NT distribution of grafted cells to either one lateral side ( CLE to CLE ) or to the midline ( CLE to NSB ) . White lines show the extent of NT contribution . ( C ) Left , diagram shows the method used to score mediolateral NT contribution of donor cells ( green ) , expressed as a percentage of the NT perimeter . Right , graphs display the number of sections containing GFP+ cells at defined positions along the NT . Colours represent graft type ( blue , L1A homotopic grafts; red and orange , L1AT homotopic grafts; purple and green , L1-3 to NSB heterotopic grafts ) . Sections in the TB are represented separately from those anterior to the TB ( termed ‘axis’ ) . The average notochord position at the ventral midline is shown by dark grey ( axis ) and light grey ( TB ) shading . DOI: http://dx . doi . org/10 . 7554/eLife . 10042 . 017 Further examination of homotopic CLE grafts indicated that they contributed to more dorsolateral positions than grafts to the NSB . To quantify this observation , we measured the dorsoventral extent of GFP+ cell contribution in all scoreable sections of CLE homotopic and CLE-to-NSB heterotopic grafts ( Figure 8B–C ) . The frequency of GFP+ cell colonisation of any NT position from 0% ( dorsal righthand extreme ) through 50% ( ventral NT ) to 100% ( dorsal lefthand extreme ) showed graft-type specific differences . CLE homotopic grafts predominantly contributed to lateral regions of the NT in the axis , but shifted towards the midline in the tail bud ( Figure 6Aa–h and Aq–t ) , indicating that there is a net displacement of at least some of the CLE cells towards the midline as axis elongation proceeds , perhaps to replace cells that have exited to the mesoderm via the midline PS . CLE cells grafted to the NSB appeared to have erased their positional bias , as NT contribution was predominantly midline in both axis and TB sections ( Figure 8C ) ; a contribution pattern which strongly resembled NSB homotopic grafts previously described in ( Cambray and Wilson , 2007 ) . Thus , as positional information in neural progenitors can be reset , it argues for the existence of a single NMP identity whose fate and dorsoventral NT position is instructed by positional cues . To define the molecular basis of NMP fate choice , we examined the role of Wnt/β-catenin signalling . Inhibiting Wnt signalling can divert predominantly mesoderm-fated cells in early zebrafish embryos towards neural fates ( Martin and Kimelman , 2012 ) . Conditional deletion of Wnt3a or β-catenin in the T-expressing population , followed by lineage tracing , has confirmed that , in mouse , β-catenin is required for mesoderm differentiation in PS/TB cells ( Garriock et al . , 2015 ) . However the above study does not directly address the role of β-catenin in NMPs , which form a subset of the T+ population . Therefore we investigated the role of Wnt/β-catenin specifically in mouse NMPs . We utilised embryos carrying a conditional ( floxed ) β-catenin mutation , a conditionally active ( floxed stop ) GFP marker and a ubiquitously expressed tamoxifen-inducible Cre transgene ( termed here βcatCKO:sGFP ) , which delete β-catenin and activate silent GFP ( sGFP ) upon 4-hydroxytamoxifen ( 4-OHT ) treatment ( Figure 9—figure supplement 1A–C ) . We dissected either βcatCKO:sGFP or constitutively active GFP transgenic control ( AGFP7 ) L1-3 pieces and grafted them to the rostral streak ( St1-3 ) . Embryos grafted with βcatCKO:sGFP cells were treated with 4-OHT for the first 8 hr of a 48-hr culture period , which proved sufficient to delete β-catenin in the majority of cells ( Figure 9—figure supplement 1D ) . For an overview of the experimental set-up and the number of grafts performed , see Figure 9—figure supplement 2A–E . As expected , all control L1-3 to St1-3 AGFP7 grafts ( n = 4 ) contributed extensively to the PXM and TBM . With the exception of a few cells in a single embryo , the CNH was not colonised ( Figure 9—figure supplement 2F ) . Similarly , the 4-OHT-treated βcatCKO:sGFP donor cells that exited the PS early , and thus contributed to rostral parts of the axis , produced exclusively mesoderm . In contrast , those exiting later to caudal regions produced only neurectoderm ( nembryos=11; 9/9 incorporated grafts switched from PXM to neural fate , Figure 9A ) . GFP+ cells generally showed undetectable levels of β-catenin , but expressed the cell adhesion molecule N-cadherin at levels similar to neighbouring wildtype cells , suggesting that their cell adhesion properties remain intact ( Figure 9B ) . Graft-derived cells in the PXM correctly expressed Pax3 , while those in the NT expressed neural markers ( Figure 9B and Figure 9—figure supplement 2G ) , showing that β-catenin deletion does not preclude either neural or mesoderm differentiation after cells have exited from the NMP compartment . Thus , the decision of NMPs to form mesoderm depends absolutely on β-catenin , and in its absence , cells differentiate to neural derivatives . 10 . 7554/eLife . 10042 . 018Figure 9 . Canonical Wnt signalling mediates neuromesodermal fate decisions . ( A ) Representative examples of L1-3 to St1-3 grafts . ( Aa and Af ) Cells immediately after grafting . βcatCKO:sGFP grafted embryos were grown for 8 hr in the presence of 5µM 4-OHT after which the medium was replaced . Note grafted cells are initially GFP- and turned green upon CreERT2 activation and LoxP recombination . Dotted line , primitive streak . Arrowhead , grafted cell position . ( Ab and Ag ) Whole embryo after 48 hr ex vivo culture . ( Ac–e and Ah–i ) Rostral to caudal transverse sections . ( Aa–e ) L1-3 AGFP7 to St1-3 control grafts contribute only to the paraxial mesoderm , although one embryo showed minor contribution to the CNH and resembled some of the L1-3 to Bc grafts ( compare Figure 9—figure supplement 2F and Figure 7F ) . ( Ai–Aj ) L1-3 βcatCKO:sGFP to St1-3 grafts shows NMPs switch from mesoderm to neural fate in the absence of β-catenin . ( B ) Marker and β-catenin expression in graft-derived βcatCKO:sGFP cells . Sections correspond to a similar rostral level as shown in Aj . Upper row , DAPI-counterstained images ( grey/blue ) . Squares , region magnified in lower row . Lower row , immunofluorescence channel with donor cell position outlined in white . See Figure 9—figure supplement 2 for an experimental overview , the number of embryo grafts performed and additional immunostaining results . ( C ) Immunofluorescence of T ( green ) and Sox2 ( red ) in E9 . 5 WT ( n = 5 , Ca-c ) and βcatCKO embryos ( n = 5 , Cd-f ) after 24 hr in vitro growth in the presence of 5µM 4-OHT . Arrowheads , notochord in treated embryos . ( D–F ) Quantitation of labelled populations in βcatCKO and WT embryos ( * , p-value<0 . 05; ** , p-value<0 . 0001; ns , not significant ) . See also Figure 3—figure supplement 1 and Video 2 . ( G ) 3D analysis showing the relative levels of T protein in E9 . 5 WT ( Ga and Gb ) and βcatCKO tail buds ( Gc and Gd ) shown as arbitrary units ( AU ) of staining intensity ) . Arrows , notochord ( noto ) position; grey , T- cells; pxm , paraxial mesoderm; white dotted line , NMP area . ( H ) Quantification of T levels in E9 . 5 WT ( n = 2 ) and βcatCKO tail buds ( n = 4 ) . Left bars , WT; right bars , βcatCKO embryos . ( * , p-value<0 . 05; ns , not significant ) . See also Figure 9—figure supplement 3C . DOI: http://dx . doi . org/10 . 7554/eLife . 10042 . 01810 . 7554/eLife . 10042 . 019Figure 9—figure supplement 1 . Obtaining conditional β-catenin knock out embryos . ( A ) Schematic diagram of mouse cross to obtain βcatCKO:sGFP embryos . Mice , homozygous for both a silent GFP ( Gilchrist et al . , 2003 ) and floxed β-catenin alleles ( Brault et al . , 2001 ) were crossed with mice homozygous for both the ROSA26-CreERT2 and floxed β-catenin alleles ( referred to as βcatCKO ) . F1 offspring are homozygous for the floxed β-catenin and heterozygous for the sGFP and ROSA26-CreERT2 alleles ( referred to as βcatCKO:sGFP ) . ( B ) Different β-catenin alleles and primers used for genotype screening and 4-OHT pulse dose/time optimisation ( primers as in Brault et al . ( 2001 ) ) . ( C-D ) 4-OHT dose and pulse time optimisation for in vitro embryo culture . ( C ) GFP fluorescence in F1 embryo after 24 hr 4-OHT treatment ( 5 µM ) , cultured from E8 . 5 ( 2–6 s ) . ( D ) Left part of gel , PCR identification of wild-type ( WT ) , floxed and floxdel alleles from gDNA . Right part , PCR from gDNA of different F1 embryos , grown in vitro for 24 hr at indicated 4-OHT pulse doses/times . DOI: http://dx . doi . org/10 . 7554/eLife . 10042 . 01910 . 7554/eLife . 10042 . 020Figure 9—figure supplement 2 . Grafts using β-catenin knock out donor tissue . ( A–C ) Grafting scheme of L1-3 ( A ) or L/St5 ( B ) to St1-3 heterotopic grafts . Either AGFP7 control or βcatCKO:sGFP donor tissue was used to transplant into WT host embryos . ( C ) After receiving the graft , embryos were grown in vitro for 48 hr , of which the first 8 hr in the presence of 10 µM 4-OHT . ( D ) Diagram summarising experimental outcome for L1-3 or L/St5 to St1-3 heterotopic grafts . ( E ) Number of embryo grafts performed . ( F ) Exceptional L1-3 AGFP7 to St1-3 grafted embryo showing minor neural CNH contribution , shown by Sox2 immunostaining in the tail bud . ( G ) DAPI-counterstained images ( grey ) . Immunofluorescent staining for neural Sox2 , FoxA2 and Pax6 differentiation , and dermomyotome-specific Pax3 after 48 hr in vitro culture . ( H ) Example of a L/St5 βcatCKO:sGFP to St1-3 graft that contained non-integrated clumps ( arrowhead ) near the notochord ( n = 3/7 ) . These non-integrated cell clumps did not express Sox2 , T or FoxA2 . DOI: http://dx . doi . org/10 . 7554/eLife . 10042 . 02010 . 7554/eLife . 10042 . 021Figure 9—figure supplement 3 . Levels of T in wildtype and 4-OHT-treated βcatCKO embryos . ( A ) Boxplot for T intensity in notochord ( noto ) , presomitic mesoderm ( PSM ) and T- ( Neg ) cells ( n , number of nuclei sampled from wholemount stained embryos ) . The threshold for positivity was set at 7 AU , with the maximum corresponding to the highest intensity recorded ( 80 AU ) . T levels are ( intensity [i] ) : negative [i < 7] , level 1 [7 ≤ i < 12] , level 2 [12 ≤ i < 24] , level 3 [24 ≤ i < 36] , level 2 [36 ≤ i <48] , level 2 [48 ≤ i] . ( B ) 3D analysis showing T- cells ( Ba , Bd ) , those with different T levels ( Bb and Be ) and Sox2+T+ cells in E9 . 5 WT and βcatCKO tail buds . Asterisk , Sox2+T+ cells found in the dorsocaudal part of the gut . ( C ) P-values from an unpaired Student's t-test comparing T levels between E9 . 5 WT ( n = 2 ) and E9 . 5 4-OHT-treated βcatCKO ( n = 4 ) embryos ( * , p-value<0 . 05; ns , not significantly different ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10042 . 021 To determine the effect of β-catenin on Sox2+T+ putative NMPs we deleted β-catenin in whole ex-vivo cultured βcatCKO embryos between E8 . 5–9 . 5 and quantified Sox2 and T expression as before . Compared to wildtype controls , the notochord extended further caudally than in wildtype embryos and the size of the NMP region was reduced in 4-OHT-treated βcatCKO and Sox2+T+ cells appeared scattered throughout the CLE ( Figure 9C and Figure 9—figure supplement 3B ) . We observed a significant drop in Sox2+T+ cell numbers in E9 . 5 treated samples ( Figure 9C and Video 2 ) . Moreover , the number of Sox2+T+ cells in E9 . 5 β-catenin-deleted embryos was not significantly different from untreated E8 . 5 samples , indicating a failure to expand NMP numbers ( Figure 9D ) . We also analysed the Sox2 and T single-positive populations to determine whether either was disproportionately affected . We observed a significant drop in Sox2-T+ cell numbers ( Figure 9E–F and Figure 3—figure supplement 1F–H ) , consistent with the observation that T is a direct transcriptional target of Wnt signalling via β-catenin ( Yamaguchi et al . , 1999 ) . 10 . 7554/eLife . 10042 . 022Video 2 . The number of T+ cells is affected upon β-catenin removal . 3D reconstruction of the caudal progenitor region . ( 00:00 ) E9 . 5 WT sample and ( 00:28 ) example of a βcatCKO embryo at E9 . 5 , after it was cultured in vitro for 24 hr in the presence of 5µM 4-OHT . DOI: http://dx . doi . org/10 . 7554/eLife . 10042 . 022 To determine the relative effects of β-catenin depletion on the different progenitor populations in the PS , we compared the different levels of T expression in WT and 4-OHT-treated βcatCKO embryos . T immunostaining intensity per cell was subdivided into five levels , quantitated and plotted on the reconstructed 3D scaffold ( Figure 9G and Figure 9—figure supplement 3A–B ) . The number of cells expressing all but the highest levels of T was significantly reduced in 4-OHT-treated βcatCKO embryos ( Figure 9H and Figure 9—figure supplement 3C ) , whereas the T- fraction was significantly increased . Moreover , levels of T protein were dramatically downregulated in the area containing βcatCKO NMPs ( Figure 9Ga , Ga”” , Ga”’ , Gc , Gc”” and Gc”’ ) , as well as in their descendants in the PSM ( Figure 9Ga , Ga’ , Ga” , Gc , Gc’ and Gc” ) . Interestingly , the LPMP-derived ventral mesoderm of the cloaca of β-catenin-depleted embryos still retained higher levels of T compared to the PSM ( compare Figure 9Ga”” and Gc”” ) . Thus , Wnt signalling is required for NMP expansion , at least in part through the maintenance of T . Moreover , our data suggests the maintenance of T in LPMPs and their descendants is less dependent on β-catenin . Therefore , LPMPs may represent an alternative , β-catenin-independent route towards mesoderm formation . The plasticity exhibited by rostral CLE cells led us to examine whether the caudal extreme of the CLE ( L/St5 ) is similarly environment-sensitive . We grafted L/St5 cells to the rostral PS ( St1-3; embryos 7 . 01–7 . 03 ) ; a region that contributes extensively to paraxial mesoderm ( Cambray and Wilson , 2007 ) . In this location , L/St5 cells contributed robustly to both PXM and LVM . Similar to homotopic St1 grafts , L/St5 to St1-3 grafts contributed to the TBM but not CNH ( Figure 10Aa–e and Cambray and Wilson ( 2007 ) ) . However contribution to the TBM was limited ( Figure 10Ae , B , D and Figure 10—figure supplement 1B ) . We did not observe obvious differences in the mediolateral distribution of PXM descendants in any of the L/St5 grafts ( Figure 10—figure supplement 2 ) . Hence , these caudal-most CLE progenitors show plasticity within the mesoderm lineage . To test whether these cells depend on active β-catenin to prevent neural differentiation , 4-OHT-treated βcatCKO:sGFP L/St5 cells were grafted more rostrally in the PS . Unlike rostral CLE , β-catenin deleted L/St5 cells formed only PXM and LVM without NT differentiation ( Figure 10Af–j ) . Thus , β-catenin is not required for L/St5 cells to undergo mesoderm differentiation . To determine whether L/St5 progenitors are capable of forming neural tissue using regional cues similarly to the rostral CLE , we grafted L/St5 cells to the rostral border ( embryos 8 . 01–8 . 09 ) . All L/St5 to Br grafts showed exclusively mesodermal contribution except for a small patch in the NT of one embryo ( embryo 8 . 03; Figure 10Ak–o , C–E and Figure 10—figure supplement 1C ) . Thus , environmental cues can switch the exclusive LVM fate of at least some L/St5 cells to include PXM , but they do not adopt neural fates , even where L1-3 cells undergo predominantly neural differentiation . We therefore term this distinctly-committed population lateral/paraxial mesoderm progenitors ( LPMPs ) . We have shown above that Sox2 is absent in the caudal-most tip of the CLE ( Figure 3Bf and 4A ) . This transcription factor , together with Sox3 , is required for neural differentiation , and can reinstate neural differentiation when ectopically expressed in emergent PXM ( Takemoto et al . , 2011; Yoshida et al . , 2014 ) . To test whether Sox2 is sufficient to confer neural fate on L/St5 cells , dissected L/St5 regions were electroporated with a CAG-Sox2-T2A-tdTomato expression plasmid and grafted to the rostral aspect of the border , which favours neural fate in grafted L1-3 cells . Cells that did not take up the vector incorporated well and contributed to the PXM as before . In contrast , tdTomato-expressing cells formed non-integrated clumps along the axis ( Figure 10F ) . No evidence of neural differentiation was apparent in electroporated cells ( nembryos = 6 ) . Thus , ectopic expression of Sox2 is unable to override the inability of L/St5 cells to contribute to the NT . 10 . 7554/eLife . 10042 . 023Figure 10 . The caudal tip of the CLE shows paraxial mesoderm but not neural potency . ( A ) Representative examples of L/St5 heterotopic grafts ( numbers in brackets identify individual embryos depicted in ( B–C ) . ( Aa , Af and Ak ) Cells immediately after grafting . Arrowheads show grafted cell position . ( Ab , Ag and Al ) Whole embryo after 48 hr ex vivo culture . ( Ac–e , Ah–i and Am–o ) Rostral to caudal transverse sections . ( Aa–e ) L/St5 AGFP7 to St1-3 grafts shows LVM-fated cells can switch fate to form PXM . ( Af–j ) L/St5 βcatCKO:sGFP to St1-3 grafts show a similar contribution pattern to control grafts . Non-integrated clumps near the notochord were observed in 3/7 embryos ( see Figure 9—figure supplement 2H ) . ( Ak–o ) L/St5 to Br grafts show robust contribution to the paraxial mesoderm . ( B–C ) Grafting of L/St5 donor cells to the primitive streak ( B ) or to Br ( C ) . Graph format is the same as in Figure 6 . ( D–E ) Quantitative analysis of L/St5 graft contribution . Left , the percentage of PXM vs LVM ( in D ) or NT ( in E ) contribution in all scorable sections . Right , the percentage of embryos with TBM ( in D ) or dorsal CNH ( in E ) contribution . ( F ) Ectopic over-expression of Sox2 in L/St5 cells ( n = 6 embryos ) . ( Fa ) L/St5 cells of AGFP7 embryos were electroporated with CAG-Sox2-T2A-tdTomato plasmid before grafting to Br of WT hosts . ( Fb–c ) Electroporated cells immediately after grafting ( Fb ) and after 4 hr culture ( Fc ) . ( Fd–f ) Grafted embryo after 48 hr in vitro growth . Arrowheads show areas containing not well-integrated cells ( orange ) . Green cells were not electroporated at the start and contributed to the PXM and LVM as before . ( Fg–h ) Axis and TB sections show orange cells never contribute to the NT . ( G ) Summary of L1-3 and L/St5 heterotopic grafts . Coloured bars represent the contribution to different axial tissues . Contribution in the neural CNH and TBM is represented by a black square or a grey circle , respectively . N , neural; PXM , paraxial mesoderm; LVM , lateral/ventral mesoderm; non-N , non-neural clumps; TBM , tail bud mesoderm; N-CNH , dorsal ( neural ) part of the CNH . DOI: http://dx . doi . org/10 . 7554/eLife . 10042 . 02310 . 7554/eLife . 10042 . 024Figure 10—figure supplement 1 . Overall lateral/ventral and tail bud mesoderm contribution . ( A ) Percentage of total embryos and sections containing GFP+ cells in the lateral/ventral mesoderm . LVM contribution is very rare in L1 grafts . Note that the L2–3 grafts produced LVM at different sections along the axis , whereas L/St5 derived LVM was always found posterior to the hindlimb . ( B ) TBM contribution in all L/St5 grafts is limited . ( C ) Two L/St5 to Br sections illustrate the limited NT contribution in embryo 7 . 03 . DOI: http://dx . doi . org/10 . 7554/eLife . 10042 . 02410 . 7554/eLife . 10042 . 025Figure 10—figure supplement 2 . Somite contribution in heterotopic LPMP grafts . ( A ) Different L/St5 grafts performed . ( B ) Representative sections taken from grafts in A . Green , grafted cells; cyan , autofluorescence channel; arrows , notochord position; nt , neural tube; pxm , paraxial mesoderm . DOI: http://dx . doi . org/10 . 7554/eLife . 10042 . 025 Despite previous reports that Sox2+T+ positivity identifies NMPs , no study has previously comprehensively correlated this phenotype with the known spatiotemporal characteristics of NMPs . Our immunohistochemical analysis shows that Sox2+T+cells are found in all NMP-containing regions: the NSB , rostral CLE , and the ectoderm of the CNH during throughout axis elongation , and are undetectable once somitogenesis ceases , suggesting they are associated with NMP activity . Therefore , are Sox2+T+cells equivalent to NMPs ? Sox2+T+cells are found at low frequency in regions not expected to be NM potent: the E8 . 5 midline primitive streak , the region caudal to the CNH at E10 . 5 , the hindgut and notochord ( Cambray and Wilson , 2007; McGrew et al . , 2008 ) . Conversely , the rostro-lateral CLE ( L1-2lat ) is NM potent but contains very few Sox2+T+cells . Importantly , however , T transcripts extend more laterally in the epiblast than protein ( Figure 3A–B; Wilson et al . ( 1995 ) ) , suggesting that Llat may already be poised to accumulate T protein . We demonstrate that the levels of T and Sox2 correlate precisely with likelihood of mesodermal and neural fate respectively , while NMPs are excluded from high-expressing areas ( Figure 5 ) . Moreover , the level of T in NMPs is significantly reduced when the capacity for mesodermal differentiation is lost in βcatCKO cells ( Figure 9G ) . Finally , Sox2+T+ cells first appear at late neural plate stage , several hours before pluripotency is lost in the epiblast ( Osorno et al . , 2012 ) . Thus , the Sox2+T+ phenotype seems to overlap extensively with ( although may not absolutely define ) NMPs . Two further observations support the NMP identity of Sox2+T+cells . Firstly , the efficient and extensive incorporation of grafted cells , and the ability of these regions to robustly adopt the fate of their new environment in heterotopic grafts , argue that most grafted NSB and L1-3 cells are NMPs . Secondly , clonal analysis indicates that the number of NMPs present at E9 . 5 is approximately 2 . 8-fold greater than that at E8 . 5 ( 5 NM clones with a rostral limit of s10-20 , versus 14 with a rostral limit of s21-30 ( Tzouanacou et al . , 2009 ) ) . This increase is similar to the 2 . 4-fold increase in number of Sox2+T+ cells during this period ( 960 ± 85 versus 2338 ± 176; Figure 3—figure supplement 1 ) . Furthermore , in vitro data shows that single positive T+ and Sox2+ cells can be derived from Sox2+T+ cells ( Tsakiridis and Wilson , 2015 ) , and cell clusters comprising 60–80% Sox2+T+ cells derived from pluripotent populations in vitro contribute to neural and mesoderm lineages when engrafted in the E8 . 5 embryo ( Gouti et al . , 2014 ) . Taken together , this indicates that this marker combination is useful for prospectively identifying NMPs both in vivo and in vitro . Interestingly , within the Sox2+T+ population , we show clear differences in levels of Sox2 and T ( Figure 5A ) and these appear to reflect greater likelihood to adopt neural and mesodermal fates respectively . Together with the observation that all L1-3 regions show equivalent potential on grafting to the NSB , this suggests that the levels of each protein may report on the cells’ responses to environmental cues directing their fate . Our previous study identifying the NSB and CLE as NMP-containing regions ( Cambray and Wilson , 2007 ) suggested that the NSB contained long-term NMPs , while the CLE contained shorter-lived progenitors . However the data presented here argue for a single , adaptable NMP type , since rostral CLE cells show equal potential to contribute to the CNH , as well as both neural and mesodermal fates over long axial distances , in the context of the NSB environment . Thus the differences in NSB and CLE fate do not result from intrinsic differences between the populations , but rather from the different environmental influences that they are exposed to in vivo . Fate mapping of the CLE and NSB ( Cambray and Wilson , 2007 ) shows that only L1 and the NSB contribute significantly to the neurectoderm , and this almost always occurs along with mesoderm contribution , whereas a larger area , L2–3 , contributes almost exclusively to mesoderm . This is consistent with single-cell labelling experiments HH4-9 chick embryos , which show that very few cells in the node region are fated solely for neural tube ( Selleck and Stern , 1991 ) . Clonal analysis in the mouse shows that medium-length mesoderm-only clones outnumber neuromesodermal clones ( 50 in mesoderm vs 8 NM; ( Tzouanacou et al . , 2009 ) . Since the volume of differentiated paraxial mesoderm is always greater than that of the neurectoderm , the overproduction of mesoderm relative to neurectoderm may begin to be implemented in the primitive streak/tail bud region in the NMPs themselves . Together with the correlation between the mid-trunk expansion of NMPs and the size of the somites ( Tam , 1981 ) , this implies that regulation of NMP proliferation/survival and differentiation may begin the process of determining the final proportions of neurectoderm and mesoderm in the embryonic axis . Whilst it has been shown that Wnt/β-catenin signalling is required for mesoderm formation at the expense of neurectoderm , our results demonstrate that this bias represents a fate choice in NMPs themselves ( Figure 11B ) . In zebrafish , downregulation of Wnt/β-catenin activity diverts cells fated for mesoderm towards neurectoderm , demonstrating that Wnt/β-catenin control of neural versus mesodermal fate choice is a vertebrate-wide phenomenon ( Martin and Kimelman , 2012 ) . The transcriptional downregulation of many Wnt pathway components in the emerging committed neurectoderm relative to the CLE is consistent with a role for Wnt downregulation in neural differentiation from NMPs ( Olivera-Martinez et al . , 2014 ) . In addition to the role of Wnt/β-catenin in NMP fate choice , we also uncover a novel function of Wnt/β-catenin signalling in the control of NMP cell numbers , since the expansion of Sox2+T+ cells between E8 . 5–9 . 5 depends on active Wnt/β-catenin ( Figure 9D and G ) . A decline in Wnt3a expression in the elongating tail ( Cambray and Wilson , 2007 ) occurs in tandem with the gradual loss of Sox2+T+ cells . Therefore it is likely that Wnt signalling is required for NMP maintenance . This suggests that the short axis of mutants with lowered Wnt3a levels ( Greco et al . , 1996; Takada et al . , 1994 ) is due to a failure to maintain NMPs . Since , we show that T is dramatically downregulated in the caudal epiblast when β-catenin is depleted , ( Figure 9G ) , this raises the possibility that T is necessary for NMP maintenance . Interestingly , a recent study ( Denans et al . , 2015 ) shows that posterior Hox gene activation in chick leads to downregulation of Wnt signalling , and of T expression , slowing of cell ingression from the ectoderm layer and shortening of the PSM . The equivalent late axial elongation period in mouse begins around the trunk/tail transition ( Gomez et al . , 2008 ) , when the number of NMPs begins to decline . Our data argues that the effects of posterior Hox genes on axial elongation operate in NMPs themselves via Wnt/β-catenin-mediated control of progenitor numbers ( Figure 11B ) . Despite strong evidence implicating Wnt/β-catenin in NMP fate choice and maintenance , several studies suggest that constitutive Wnt/β-catenin activity is not sufficient either to divert NMPs to mesoderm fates or to maintain NMPs . Providing the T+ caudal region with a stabilised form of β-catenin , or overexpressing Wnt3a in T+ or Cdx2+ progenitors , results in an enlarged PSM domain ( Garriock et al . , 2015; Aulehla et al . , 2008; Jurberg et al . , 2014 ) but not an obvious reduction in Sox2+T+ NMPs in the CLE ( Garriock et al . , 2015 ) . However in the converse experiment , when Wnt3a was deleted in the T+ caudal region , it was not clear to what extent Sox2+T+ NMPs were affected ( Garriock et al . , 2015 ) . Here we show that , when β-catenin is deleted , Sox2+T+ NMPs are significantly reduced in number , but are not completely eliminated from the caudal region . Taken together , these results show that elevated Wnt/β-catenin signalling alone is not enough to commit NMPs to a mesoderm fate , but its absence is sufficient to block mesoderm differentiation . This implicates additional signalling pathway ( s ) in driving NMPs towards mesoderm formation . Moreover , despite the requirement for Wnt signalling in expanding NMP numbers , at least some NMPs can tolerate both elevated and reduced levels of Wnt signalling , at least for short ( 24–48 hr ) periods . How can Wnt/β-catenin signalling have multiple distinct effects on NMPs during axis elongation ? Nuclear β-catenin staining in the zebrafish tail bud has shown that canonical Wnt signalling is only moderately active in NMPs and high in early differentiating mesodermal cells ( Bouldin et al . , 2015 ) . Since Wnt3a is expressed uniformly in the E8 . 5 caudal region ( Cambray and Wilson , 2007; Takada et al . , 1994 ) , it is unlikely that localisation of Wnt3a itself can provide the specificity to direct mesoderm formation from NMPs . It is therefore possible that Wnt modulators might provide this signal . R-spondin3 , an extracellular Wnt activator , is transcribed in the PSM ( Kazanskaya et al . , 2004 ) and has recently been shown to drive pluripotent cells towards skeletal muscle derivatives in vitro ( Chal et al . , 2015 ) . Moreover , the ability to both maintain progenitors and permit their differentiation may lie in its coordination with other signalling pathways . Fgf signalling , as well as steroid biogenesis and chondroitin sulphate biosynthesis were identified as differentially regulated between NMPs and committed neural progenitors in chick ( Olivera-Martinez et al . , 2014 ) , and it remains to be seen whether these play a role in coordinating the balance between neural and mesoderm differentiation in all vertebrates . There also appear to be temporal differences in cell responses to Wnt/β-catenin signalling . The data presented here and shown by Martin and Kimelman ( Martin and Kimelman , 2012 ) shows that prospective neurectoderm can be re-fated to mesoderm . However , mouse embryos lacking Wnt3a or β-catenin in the primitive streak are still able to form rostral paraxial mesoderm , ( Aulehla et al . , 2008; Takada et al . , 1994 ) , suggesting the mesoderm-inducing signals acting on the early progenitors of paraxial mesoderm must be different from those at later times . Previous studies have not distinguished whether Wnt/β-catenin signalling acts specifically in NMPs or is a generic inducer of mesoderm fate . We show that unlike NMPs , LPMPs in the caudal PS do not require Wnt/β-catenin signalling for mesoderm differentiation . Moreover , our data suggests that LPMPs are less dependent on Wnt/β-catenin signalling for T maintenance , and that additional signalling could be an important regulator of this population . Bmp4 and Bmp7 are expressed in the caudal CLE ( Arkell and Beddington , 1997; Fujiwara et al . , 2002 ) and have been implicated in the upregulation of T during mesoderm differentiation from embryonic stem cells ( Suzuki et al . , 2006 ) . Together with the inability of forced Sox2 expression to direct integration in the NT , our results indicate that LPMPs are regulated differently from NMPs . We also show that LPMPs can participate in PXM production , consistent with clonal data showing that mesoderm-restricted clones can contribute to both PXM and LVM ( Tzouanacou et al . , 2009 ) . We did not observe an obvious bias in mediolateral distribution in PXM formed from LPMPs . However , it will be interesting to determine whether any downstream mesoderm lineages are affected in the absence of β-catenin . Our fate map shows that L/St5 cells normally never contribute to the PXM , suggesting that LPMPs choose lateral mesoderm fates over paraxial ones based on their position . Here we show NMP numbers peak through the actions of Wnt/β-catenin signalling during trunk-to-tail transition . This role of β-catenin in NMP maintenance/expansion may be cell-autonomous , although it is also possible that it acts in neighbouring notochord progenitors . Wnt/β-catenin signalling is high in this region throughout axis elongation and is required in the notochord itself for notochord integrity and tail growth ( Ukita et al . , 2009 ) . Abnormalities in the caudal notochord are also present in embryos depleted for β-catenin between E8 . 5–9 . 5 ( Figure 9Cd–f ) , and the distance between the NMP area and the caudal notochord apparently increased ( Figure 9G ) . Despite the structural aberrations in βcatCKO embryos , the level of T in the notochord is far less affected compared to other tail bud domains ( Figure 9Gd ) , suggesting T expression is uncoupled from Wnt/β-catenin signalling in these cells . Notochordal cells are found close to NMPs in the NSB but not in the CLE , arguing against the idea that the presence of the notochord environment is absolutely essential for NMP identity . However , since the NMP pool is only maintained over long axial distances adjacent to the caudal notochord tip ( Figure 3C-F and Cambray and Wilson ( 2002 ) ; McGrew et al . ( 2008 ) ) , NMP maintenance may be indirectly dependent upon the notochord . This notochordal ‘niche’ could include extracellular matrix molecules , and secreted factors including Wnts . Therefore , Wnt/β-catenin signalling may act indirectly to maintain the NMP environment . Similarly LPMPs are likely to interact with NMPs during the time of trunk-to-tail transition . Gdf11 binds Tgfbr1 and activates Isl1 expression in the emerging LVM during trunk-to-tail transition and provides signals that induce hindlimb bud and cloaca formation ( Jurberg et al . , 2013 ) . Ectopic Isl1 activation in the whole tail bud , including NMPs , leads to complete loss of NM-derived trunk vertebrae and spinal cord ( Jurberg et al . , 2013 ) , while loss of Gdf11 function increases the interlimb area by 5 to 6 somites ( Jurberg et al . , 2013; McPherron et al . , 1999 ) . This suggests that Gdf11 acts via LPMPs to attenuate signals that expand the NMP population and , together with posterior Hox gene reduction of Wnt signalling ( Denans et al . , 2015; Young et al . , 2009 ) , control the timing of progenitor number reduction in the tail . In conclusion , we have shown two separate progenitor types exist in the primitive streak , characterised by differential plasticity and commitment , as well as by the mechanisms by which they choose between alternative fates . Interestingly , both NMPs and LPMPs exhibit potency beyond their normal fates , yet this potential is intrinsically constrained in each case . Thus , besides serving as a benchmark for generation of differentiated cell types in vitro , these progenitor types can form a paradigm for testing how the interplay between intrinsic competence and extrinsic signals is set up and maintained during cell differentiation . Wildtype , outbred MF1 and transgenic mice ( AGFP7 ( Gilchrist et al . , 2003 ) , βcatCKO and βcatCKO:sGFP ( Tsakiridis et al . , 2014 ) ) , were maintained on a 12-hr-light/12 hr-dark cycle . F1 βcatCKO:sGFP embryos were obtained as shown in Figure 9—figure supplement 1A . For timed matings , noon on the day of finding a vaginal plug was designated E0 . 5 . Dissection and in vitro culture of early somite stage embryos was performed as described ( Copp and Cockroft , 1990 ) . Embryos were cultured for 48 hr from E8 . 5 ( 2-5s ) to E10 . 5 ( 30-35s ) stage . Dissected PS regions shown in Figure 1A , corresponding to about 100–150 cells were dissected and grafted as described ( Cambray and Wilson , 2007 ) with the exception that for CLE dissection , underlying mesoderm was manually dissected away from the ectoderm using hand-pulled solid glass needles . Since the L5 region is small , we could not exclude that a part of the midline ( St5 ) was included in L5 grafts . Therefore , this region is termed L/St5 . An overview of all grafting experiments is shown in Figure 6—figure supplement 2A . In homotopic grafts , L1-3 cells often colonised a short stretch of rostral somites unilaterally on the grafted side . In more caudal sections of the axis , somite contribution in L1AT , L2-3 and L1-2med homotopic grafts were often bilateral ( 17/21 grafts ) . Therefore , it seems some of the grafted cells are able to encroach on the midline and produce mesoderm . In contrast , most L1AT and L2lat homotopic grafts ( 6/8 grafts ) colonised the paraxial mesoderm to one side only ( see also Figure 6—figure supplement 4 ) . Importantly , unilateral PXM contribution might have arisen from pre-existing presomitic mesoderm that was co-grafted along with the NMPs of the CLE . This is likely not to be the case for the majority of the PXM contribution observed ( see Figure 6—source data 1 ) . We estimated this presomitic mesoderm contamination to be less than 20% of the total section number . In addition , bilateral PXM contribution is present in almost all homotopic CLE grafts , indicating that cells must have encountered the midline PS at some point in their history ( Nicolas et al . , 1996; Eloy-Trinquet and Nicolas , 2002 ) , despite being grafted lateral to the streak ( Figure 6—source data 1 ) . The bilateral PXM contribution of most grafts also argues that most grafted cells are derived from CLE , which experiences a net movement towards the midline PS , while PSM progenitors would normally move away from it ( Psychoyos and Stern , 1996; Voiculescu et al . , 2007 ) . Thus , the bulk of the PXM contribution in CLE grafts has most likely descended from NMPs rather than from pre-existing PSM progenitors . To verify grafted cell locations , embryos were imaged immediately following grafting in a fluorescent dissecting microscope ( Figure 7E , 9A , 10A and F ) . Kidney capsule grafts were performed as described previously ( Beddington , 1983 ) , and processed and scored as in ( Osorno et al . , 2012 ) . Syngeneic kidney capsule grafts were performed in CBA or 129 mice . In E7 . 5 ( late streak/early headfold ) control grafts , 1 to 2 regions consisting of anterior or posterior halves of the embryonic region were transplanted , while up to 3 E8 . 5 ( 2–6 s ) regions were grafted under the kidney capsule . In preliminary experiments , little difference in size between grafts of different numbers of regions was noted , nor was there a significant difference between the size of E8 . 5 tumours recovered at 4 and 6 weeks . Therefore , data from these grafts was pooled . Tumours were fixed in 4% paraformaldehyde for 1 to 7 days depending on size , processed and stained as described ( Bancroft and Gamble , 2002 ) . The tumour size was measured as the average surface area ( ± s . d . ) in sections . Eight grafting experiment types were performed . In series 1 ( embryos 1 . 01–1 . 20 ) , control homotopic grafts of L1 , 2 or 3 tissue ( without distinction between the three areas ) were grafted to a recorded position in the L1 or L2-3 area of the CLE . In series 2 ( embryos 2 . 01–2 . 08 ) either a more medial or lateral piece of the L1-2 epiblast was homotopically grafted . In series 3 ( embryos 3 . 01–3 . 04 ) , L/St5 tissue was grafted to the same position of stage-matched embryos . In series 4 ( embryos 4 . 01–4 . 09 ) , L1 , 2 or 3 tissue ( without distinction between the three areas ) was grafted to the NSB . Series 5 ( embryos 5 . 01–5 . 18 ) contained grafts of individual areas of L1 , 2 or 3 , divided either by their rostrocaudal or mediolateral position , or both , to the NSB . In series 6 ( embryos 6 . 01–6 . 21 ) , defined L1 , 2 or 3 regions were grafted to subregions within the NSB . The crown of the node was used as a landmark: Br grafts were positioned immediately rostral to the crown , Bc grafts immediately caudal to it , with grafts to Bm inserted at the crown itself . Series 7 ( embryos 7 . 01–7 . 03 ) and series 8 ( embryos 8 . 01–8 . 09 ) contained grafts of L/St5 to the primitive streak ( St1-3 ) and the rostral part of the border ( Br ) respectively . All embryos were processed and scored as described below . Embryos were photographed in wholemount and then fixed , mounted and sectioned in a vibratome ( Series 1000 , The Vibratome Company or VT1000M , Leica ) at 50µm as described ( Cambray and Wilson , 2007 ) . Sections were counterstained with TO-PRO®-3 Iodide ( Life Technologies , Carlsbad , CA ) and visualised using an Olympus BX61 compound microscope with Optigrid confocal optics ( Qioptiq , Waltham , MA ) . In the axis rostral to the tail bud , contribution of GFP+ cells in each section to neural tissue , paraxial mesoderm , lateral/ventral mesoderm , axial mesoderm , endoderm or surface ectoderm was noted , along with any self-differentiated or non-integrated tissue . In the tail bud , contribution to the dorsal and ventral ( notochordal ) parts of the CNH and the tail bud mesoderm was scored . No graft-derived cells contributed to either endoderm or surface ectoderm . Figures 6 , 7 and 10 display the scored sections containing integrated graft-derived cells for each individual embryo , whereas Figures 8 and 10 also show the contribution per tissue type for each graft cohort , shown as the percentage of the total number of scorable sections . An additional six grafts were performed to test grafted cell identity using cryostat embedding , followed by immunohistochemistry ( 2x L1 homotopic , 2x L2-3 to Br , 1x L/St5 homotopic and 1x L/St5 to St1-3 ) . These grafts had similar contribution patterns , but since not all sections were scored , they were not included in the above series . All L1-3 and L/St5 βcatCKO:sGFP to St1-3 grafted embryos were cryosectioned , examined for their contribution pattern , but not serially scored . L/St5 cells were micro-dissected from E8 . 5 ( 2–6 s ) AGFP7 embryos as described above . Several L/St5 tissue pieces ( ~100 cells each ) were transferred to a glass electroporation well containing 1µg/mL CAG-Sox2-T2A-tdTomato plasmid in PBS ( Sigma , St . Louis , MO ) . To obtain this plasmid , the Sox2 ORF was cloned into pPyCAGIP ( Chambers et al . , 2003 ) , in which the IRES puromycin resistance casette was exchanged for a nucleotide sequence corresponding to a T2A self-cleaving peptide ( Szymczak et al . , 2004 ) and a tdTomato fluorescent reporter . Two 1 mm , L-shaped , gold tip Genetrode electrodes ( BTX Model 516 , BTX , Holliston , MA ) were placed at 3 mm distance and tissue pieces were electroporated using an Electro Square Porator ECM 830 ( BTX ) ( 30V , 5 pulses , 50 ms/pulse , 1s interval ) . The electroporated tissue pieces were subsequently grafted into wildtype stage-matched host embryos , as described above . Wholemount in situ hybridisation was performed as described previously ( Wilkinson , 1998 ) except that proteinase K treatment was empirically adjusted according to embryo size and stage ( time between 5–20 min ) . Dual-colour labelling of embryos was performed by simultaneously hybridising DIG-labelled Sox2 ( Avilion , 2003 ) and FITC-labelled T ( Herrmann , 1991 ) probes . Antibody incubations and color reactions were carried out sequentially with BCIP/NBT and BCIP/INT ( Roche , Switzerland ) . To inactivate the first AP enzyme , a heat inactivation step ( at 65°C in MABT ( 100 mM maleic acid , 150 mM NaCl , 0 . 1% ( v/v ) Tween-20 , pH 7 . 5 ) and an acid treatment ( 0 . 1 M glycine , pH 2 . 2 ) were carried out ( both 30 min ) after the first color reaction . After hybridisation , embryos were fixed , processed and sectioned as described below . Wholemount immunohistochemistry was performed as described previously ( Osorno et al . , 2012 ) . Embryos were fixed in 4% PFA in PBS at 4°C for 2 hrs ( HF ) or overnight ( >E8 . 5 embryos ) . Samples were costained against Sox2 ( Abcam , United Kingdom; ab92494; 1:200 ) and T ( R&D , Minneapolis , MN; AF2085; 1 µg/ml ) , followed by overnight incubation in PBS containing 4’ , 6-diamidino-2-phenylindole ( DAPI , Life Technologies ) . Confocal microscopy was performed after dehydration through a PBS/methanol series ( 10 min each ) , three 5 min washes in 100% methanol , clearing in 1:1 v/v methanol/BA:BB ( 2:1 benzyl alcohol:benzyl benzoate ) , and two washes in BA:BB . Selected embryos were embedded , sectioned and stained as described in ( Huang et al . , 2012 ) . Primary antibodies ( supplier , catalogue number and working concentration ) were as follows: anti-Sox2 ( Abcam; ab92494; 1:200 , Santa Cruz , Dallas , TX; sc-17320; 1 mg/ml or Merck Millipore , Germany; AB5603; 5 mg/ml ) ; anti-T ( R&D; AF2085; 1 µg/ml or Santa Cruz; sc-17743; 1 mg/ml ) ; anti-Foxa2 ( Santa Cruz; sc-6554; 1 mg/ml or R&D; AF2400; 1 µg/ml ) ; anti-GFP ( Abcam; ab13970; 10 µg/ml ) ; anti-Pax3 ( DSHB , Iowa City , IA; 1:20 ) ; anti-Pax6 ( DSHB; 1:20 ) ; anti-PDGFRβ ( Abcam; ab32570; 1:100 ) ; anti-β-catenin ( Sigma; C2206; 1:1000 ) ; anti-Active β-catenin ( Millipore; 05–665 , clone 8E7; 0 . 1 µg/ml ) ; anti-N-cadherin ( Sigma; C3865; 1:400 ) . Acquired confocal images were deconvoluted using Huygens software ( SVI , The Netherlands ) and saved as 8-bit tiff files . Due to the large size , images were cropped to get rid of most black areas , before the segmentation and co-expression analysis . For some samples , the brightness and/or contrast was increased in the blue channel ( DAPI ) to enhance weakly fluorescing nuclei , using ImageJ ( imagej . nih . gov/ij/ ) . The best parameters for nuclear segmentation were defined in cropped regions of greyscale DAPI z-stacks using FARSIGHT v0 . 4 . 5 software ( www . farsight-toolkit . org/wiki/Nuclear_Segmentation ) . These defined parameters were used on all original greyscale DAPI z-stacks and resulted in a segmented images ( in the blue field ) , in which each nuclear volume could be identified by a unique greyscale-based identifier ( Al-Kofahi et al . , 2010 ) . The segmented blue field ( DAPI ) was then overlaid with the red ( Sox2 ) and far red ( T ) field . Single-cell fluorescence quantification and 3D rendering was performed using a software program , designed by GB ( manuscript in preparation; for more information , see Davies et al . ( 2013 ) ) . The nuclear debris size ( the minimum size below which a ‘nucleus’ identified by the segmentation algorithm is too small to be real and is therefore considered as debris in the analysis ) was set as nuclear volume ≤2000 pixels . Initially , a threshold of detection was set for both fields using four internal controls for each individual sample . These controls were cropped and saved from the original z-stack to have ( a ) only Sox2+ cells , ( b ) only T+ cells , ( c ) both Sox2 or T single positive cells and ( d ) neither Sox2 nor T positivity within the region of interest ( ROI ) . The threshold for Sox2 and T positivity was defined by comparing how accurately the obtained quantification result could predict the expected values for each ROI ( pre-set as clearly separated or negative for both factors ) . The same software package and parameters were then used to analyse the sample . The statistical significance between developmental stages and Sox2/T fractions were tested with a standard unpaired Student's t-test , using Prism v6 . 0 software ( GraphPad , La Jolla , CA ) . To correct for differences in image acquisition , especially those at greater optical depths ( e . g . image acquisition was optimised to capture the Sox2+T+ population in E9 . 5 tails ) , an additional Welch correction was carried out to compare Sox2/T populations , though it did not alter the statistical significance; p-value<0 . 05 was considered significant ( * , p-value<0 . 05; ** , p-value<0 . 0001; ns , not significantly different ) . To calculate the margin of error that accompanies the use of any automated segmentation algorithm , we have compared our automated segmentation ( using FARSIGHT v0 . 4 . 5 software ) to a method involving manual identification of individual nuclei . For the latter , we designed a manual Region Of Interest ( ROI ) 'seeding' tool in the Icy software platform ( de Chaumont et al . , 2012 ) ( plug-in developed by GB ) , allowing us to manually seed ellipse-shaped ROIs in the centre of each cell . After all cells were identified , the 'Active Contours' plug-in was used to grow the original seed to the correct nuclear volume ( http://icy . bioimageanalysis . org/plugin/Active_Contours ) . This resulted in a segmented image that was processed as before . Six different embryonic regions were cropped from two different confocal stacks ( in E8 . 5 and E10 . 5 samples ) and their segmentation was compared for both methods . The degree of error at E8 . 5 is 15% ( n = 4 ) and 18% in E10 . 5 samples ( n = 2 regions ) . Our automated segmentation thus provides a reasonably accurate measure of the number of nuclei in any region of interest . Throughout this manuscript we therefore use the term ‘cells’ to refer to these segmented nuclear volumes for clarity . Whole embryo images were captured with a digital camera ( Qimaging , Canada ) attached to a Zeiss Stemi SV11 ( Zeiss , Germany ) , Nikon AZ100 ( Nikon , Japan ) or Leica M165 FC microscope ( Leica , Germany ) . Confocal images of wholemount immunostained embryos were acquired on a Leica SP8 or SPE , or Zeiss LSM510 Meta platform . An Olympus BX61 ( Olympus , Japan ) compound microscope with fluorescence optics was used to capture images of cryosections using Volocity software ( Perkin Elmer , Waltham , MA ) . Additional Optigrid confocal optics ( Qioptiq ) were used to image vibratome sections . Image processing was done using Adobe Photoshop ( Adobe Systems , San Jose , CA ) or ImageJ software ( imagej . nih . gov/ij/ ) .	Our bodies , like those of all animals with a backbone , form during embryo development in a head-to-tail sequence . This process is fuelled by populations of proliferating cells called progenitor cells , which are found in an early embryonic structure called the primitive streak , and later at the tail-end of the embryo . One of these populations – known as the neuromesodermal progenitors ( or NMPs ) – produces the animal’s spinal cord , muscle and bone tissue . However , it is not clear how this cell population is maintained or what triggers these cells to specialise into the correct cell type . It is even unclear whether NMPs are a single cell type or a collection of several types of progenitor , each with a slightly different propensity to make spinal cord or muscle and bone . Answering these questions could inform the future development of cell-replacement therapies for conditions such as spinal injuries . Wymeersch et al . used a range of techniques to identify , map the fate , and assess the developmental potential of progenitors in the primitive streak . This revealed fine-grained differences in the fates adopted by cells in the progenitor region . However , these regional differences were found to result from the progenitor cells’ extensive ability to respond to signals they receive from their environment , rather than being hard-wired into the progenitor cells . In fact , Wymeersch et al . detected only two distinct cell types: the NMPs and a new cell population termed lateral/paraxial mesoderm progenitors ( or LPMPs ) , which , unlike NMPs , do not form nerve cells . Further experiments investigated the molecular signals present in the environment of these progenitors that help to decide their fate . NMPs respond to an important developmental signal , called Wnt , by adopting a so-called mesoderm fate . This signal also induces NMPs to undergo a previously unknown phase of proliferation during the formation of the animal’s body . LPMPs , on the other hand , do not require Wnt to form mesoderm . These findings show that studies with embryos can identify new progenitor populations that might be clinically relevant , and reveal new ways in which a cell’s environment inside an embryo can determine its fate .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2016	Position-dependent plasticity of distinct progenitor types in the primitive streak
Retinal circuits detect salient features of the visual world and report them to the brain through spike trains of retinal ganglion cells . The most abundant ganglion cell type in mice , the so-called W3 ganglion cell , selectively responds to movements of small objects . Where and how object motion sensitivity arises in the retina is incompletely understood . In this study , we use 2-photon-guided patch-clamp recordings to characterize responses of vesicular glutamate transporter 3 ( VGluT3 ) -expressing amacrine cells ( ACs ) to a broad set of visual stimuli . We find that these ACs are object motion sensitive and analyze the synaptic mechanisms underlying this computation . Anatomical circuit reconstructions suggest that VGluT3-expressing ACs form glutamatergic synapses with W3 ganglion cells , and targeted recordings show that the tuning of W3 ganglion cells' excitatory input matches that of VGluT3-expressing ACs' responses . Synaptic excitation of W3 ganglion cells is diminished , and responses to object motion are suppressed in mice lacking VGluT3 . Object motion , thus , is first detected by VGluT3-expressing ACs , which provide feature-selective excitatory input to W3 ganglion cells . A diverse array of circuits in the retina processes signals from photoreceptors and parses information into spike trains of 20–30 types of retinal ganglion cells ( RGCs ) , each encoding distinct aspects of the visual scene ( Masland , 2012 ) . The most abundant RGC type in the mouse retina ( W3-RGC ) was recently shown to respond selectively to movements of small—in terms of size on the retina—objects ( Zhang et al . , 2012 ) . Detecting object motion is a challenging task as head , body , and eye movements frequently shift the retinal image ( Martinez-Conde et al . , 2004; Sakatani and Isa , 2007 ) . To distinguish movements of objects and the background , object motion sensitive ( OMS ) RGCs respond to differences in the timing of texture movements in their receptive field center and surround ( Olveczky et al . , 2003; Zhang et al . , 2012 ) . W3-RGCs share key properties with OMS RGCs in rabbit and salamander , but , due to stronger surround suppression , do not respond at the border of larger objects ( Zhang et al . , 2012 ) . This feature is reminiscent of local-edge-detector RGCs described in several species ( Levick , 1967; Zeck et al . , 2005; Roska et al . , 2006 ) . W3-RGCs , thus , appear to be in the intersection of OMS and local-edge-detector RGCs . Although postsynaptic inhibition and spike thresholds sharpen the tuning of W3-RGCs , similar to other OMS and local-edge-detector RGCs , key response properties appear to be inherited from their excitatory input ( van Wyk et al . , 2006; Baccus et al . , 2008; Russell and Werblin , 2010; Zhang et al . , 2012 ) . This suggests that feature selectivity arises presynaptic to W3-RGCs . Where and how object motion is first detected remains to be determined . Typically , RGCs receive excitatory input from bipolar cells ( Euler et al . , 2014 ) and inhibitory input from amacrine cells ( ACs ) . ACs are the most diverse class of neurons in the retina , encompassing 30–50 cell types ( MacNeil and Masland , 1998; Helmstaedter et al . , 2013 ) that serve task-specific functions in vision ( Dacheux and Raviola , 1986; Yoshida et al . , 2001; Euler et al . , 2002; Munch et al . , 2009; Grimes et al . , 2010; Chen and Li , 2012 ) . Although most ACs release γ-Aminobutyric acid ( GABA ) or glycine , a wide range of neurotransmitters and neuromodulators can be found in different cell types including one expressing the vesicular glutamate transporter 3 ( VGluT3 ) ( gene: Slc17a8 , protein: VGluT3 , AC: VG3 ) ( Fyk-Kolodziej et al . , 2004; Haverkamp and Wassle , 2004; Johnson et al . , 2004 ) . VG3-ACs are conserved from rodents to primates . Recent studies found that VG3-ACs respond to light increments ( ON ) and decrements ( OFF ) ( Grimes et al . , 2011 ) , show strong surround suppression , and following optogenetic or electrical stimulation can release glutamate ( Lee et al . , 2014 ) . However , what specific features of the visual world VG3-ACs detect and how , as well as their interactions with RGCs during sensory processing remain unknown . Here , we generate and obtain transgenic mouse lines to genetically label VG3-ACs and target them under 2-photon guidance for whole-cell patch-clamp recordings in retinal flat mount preparations . We find that VG3-ACs , like W3-RGCs , combine properties of OMS and local-edge-detector neurons and selectively detect movements of small objects . Using biolistics , we show that many excitatory synapses on W3-RGCs are apposed by boutons of VG3-ACs , and in whole-cell recordings from W3-RGCs , find that properties of their excitatory input match VG3-AC responses . Finally , we show amplitude and object motion preference of synaptic excitation and spike responses of W3-RGCs are reduced in VGluT3 knockout mice ( VGluT3−/− mice ) ( Seal et al . , 2008 ) . Thus , we identify VG3-ACs as object motion detectors , characterize the synaptic mechanisms underlying this computation , and show that VG3-ACs provide feature-selective excitatory input to W3-RGCs . To analyze the morphology of VG3-ACs , we generated bacterial artificial chromosome ( BAC ) transgenic mice expressing a ligand-activated Cre recombinase under control of regulatory sequences of the Slc17a8 gene ( VG3-CreERT2 mice ) and crossed them to a fluorescent reporter strain ( Ai9 ) ( Madisen et al . , 2010 ) . After tamoxifen injection , a subset of VG3-ACs expresses tdTomato in VG3-CreERT2 Ai9 mice ( Figure 1—figure supplement 1 ) . Neurites of VG3-ACs stratify broadly in the center of the inner plexiform layer ( Grimes et al . , 2011 ) , occupy medium-sized lateral territories ( Figure 1A and Figure 1—figure supplement 2 , 7662 ± 211 μm2 , n = 39 ) , and as a population , cover the retina approximately seven times ( coverage: 6 . 88 ) . To characterize light responses , we obtained VG3-Cre mice ( Grimes et al . , 2011 ) in which all VG3-ACs express Cre ( Figure 1—figure supplement 1 ) , crossed them to Ai9 , and targeted fluorescent somata in the inner nuclear layer ( INL ) for whole-cell patch-clamp recordings . Consistent with previous results , we find that VG3-ACs respond transiently to light increments and decrements , depolarizing to small and hyperpolarizing to large stimuli ( Lee et al . , 2014 ) ( Figure 1B , C ) . Voltage-clamp recordings revealed that this switch in response polarity is caused by a combination of pre- and post-synaptic surround inhibition ( Figure 1D–F ) . Excitatory and voltage-response receptive fields are well fit by Difference-of-Gaussians models ( Enroth-Cugell and Robson , 1966; McMahon et al . , 2004 ) , whereas a single Gaussian is sufficient to describe the monotonic rise of inhibition with stimulus size . For all components , OFF responses exceed ON responses , and for voltage responses and excitatory inputs , ON receptive fields are larger in diameter than their OFF counterparts . We characterized the temporal tuning of VG3-AC responses and underlying synaptic inputs in more detail using white noise stimuli ( Figure 1—figure supplement 3 ) . 10 . 7554/eLife . 08025 . 003Figure 1 . Morphology and receptive field properties of VG3-ACs . ( A ) Orthogonal maximum intensity projections of a confocal image stack through a representative VG3-amacrine cells ( ACs ) labeled in VG3-CreERT2 Ai9 mice . The fluorescent signal is colored to reflect depth in the inner plexiform layer ( IPL ) . Inset bar graph shows the mean ± SEM territory size of VG3-ACs ( n = 39 ) measured as the area of the smallest convex polygon to encompass their arbors in a z-projection . ( B , D , E ) Representative voltage ( B , black ) , excitatory postsynaptic current ( EPSC ) ( D , red ) , and inhibitory postsynaptic current ( IPSC ) ( E , blue ) responses to a stimulus in which luminance in a circular area of varying size is square-wave modulated ( 2 s ON , 2 s OFF , transitions indicated by ‘arrows’ ) . Stimuli were presented in pseudorandom order centered on the soma of the recorded cell . Each response trace is annotated with the radius of the stimulus eliciting it . The resting membrane potential of VG3-ACs in our recordings was −38 ± 1 . 2 mV ( n = 26 ) . ( C , F ) Summary data of the spatial ON ( open circles ) and OFF ( filled circles ) sensitivity profiles of VG3-ACs for voltage responses ( C , black , n = 26 ) and excitatory ( F , red , n = 38 ) and inhibitory ( F , blue , n = 38 ) conductances . Solid lines show fits of Difference-of-Gaussian ( for voltage and excitation ) and single Gaussian ( for inhibition ) models to the data . Receptive field diameters determined from fits to voltage responses were: ON-center 73 . 4 ± 8 . 5 μm , OFF-center 40 . 9 ± 4 . 2 μm , p < 0 . 002 , ON-surround 290 . 2 ± 25 . 5 μm , OFF-surround 213 . 3 ± 11 . 3 μm , p < 10−3 . Receptive field diameters for excitatory inputs were: ON-center 137 ± 15 . 8 μm , OFF-center 83 . 1 ± 10 . 2 μm , p < 0 . 005 , ON-surround 206 . 4 ± 14 . 3 μm , OFF-surround 189 . 8 ± 23 . 2 μm , p > 0 . 4 . Diameters of inhibitory center-only receptive fields were: ON 258 ± 24 . 7 μm , OFF 148 . 2 ± 12 . 3 μm , p < 10−4 . Response amplitudes to OFF stimuli exceeded those to ON stimuli for voltage ( at 100 μm , p < 10−9 ) , excitation ( at 100 μm , p < 10−11 ) , and inhibition ( at 100 μm , p < 10−5 ) . ( G ) Schematic illustration of split-field stimuli . The receptive field center is divided evenly ( left ) or in a biased manner ( right ) into two regions in which intensity is modulated by phase-shifted sine waves . ( H , I ) Representative EPSC traces and summary data ( n = 6 , p < 0 . 05 ) for even ( top ) and biased ( bottom ) split-field stimulation . ( J ) Schematic illustration of counter phase stimulation of surround regions . The receptive field surround is divided in bars of different size and their intensity is modulated by phase-shifted sine waves . ( K ) Representative IPSC traces to counter phase stimulation of bars of 25 μm ( middle ) and 50 μm ( bottom ) widths . ( L ) Summary data illustrating change in F2 power of inhibition as a function of bar widths . See also Figure 1—figure supplement 1 and Figure 1—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 08025 . 00310 . 7554/eLife . 08025 . 004Figure 1—figure supplement 1 . Distribution and specificity of VG3-Cre and VG3-CreERT2 labeling . ( A–F ) Maximum intensity projections of confocal image stacks from the INL of VG3-Cre mice crossed to a reporter strain expressing the red fluorescent protein tdTomato ( Ai9 , tdT , A–E ) stained for vasoactive intestinal peptide ( VIP , A ) , tyrosine hydroxylase ( TH , B ) , choline acetyltransferase ( ChAT , C ) , Calretinin ( D ) and VGluT3 ( E ) , and of VG3-CreERT2 Ai9 mice stained for VGluT3 ( F ) . ( G ) Density recovery profiles for tdTomato ( n = 17 retinas ) and VGluT3 ( n = 10 retinas ) signals in images like that shown in ( A ) from VG3-Cre Ai9 mice showed that VG3-ACs are arranged in regular mosaics with characteristic exclusion zones in their density recovery profiles ( average densities VGluT3: 898 ± 34 cells/mm2 , n = 10 retinas , tdTomato: 948 ± 37 cells/mm2 , n = 17 retinas , effective radii of exclusion zones VGluT3: 18 . 1 ± 1 . 8 μm , tdTomato: 16 . 9 ± 0 . 4 μm ) ( Rodieck , 1991 ) . The density of VG3-ACs was not significantly different between dorsal , ventral , nasal , and temporal quadrants of the retina ( data not shown ) . ( H ) Conditional probabilities illustrating the specificity ( blue bars ) and completeness ( red bars ) of genetic labeling in VG3-Cre and VG3-CreERT2 mice . While Cre expression is highly specific in the INL , ectopic expression was observed in a small subset of cells in the ganglion cell layer ( GCL ) in VG3-Cre ( Figure 4 ) but not VG3-CreERT2 mice . DOI: http://dx . doi . org/10 . 7554/eLife . 08025 . 00410 . 7554/eLife . 08025 . 005Figure 1—figure supplement 2 . VG3-ACs stratify in sublaminae 2 and 3 of the IPL . ( A ) Representative maximum intensity projection of a confocal image stack of the IPL acquired in a retinal vibratome slice stained for Calretinin ( blue ) and VGluT3 ( red ) . Inner nuclear and GCLs are bordering the top and bottom , respectively , of this image . ( B ) Intensity profiles ( mean ± SEM , n = 10 retinas ) of these labels show that VG3-ACs stratify in sublamina 2 and 3 ( Figure 1—figure supplement 2 and Figure 1—figure supplement 3 ) of the IPLs , the boundaries of which are marked by Calretinin ( Wassle , 2004 ) . Note the potential strategic importance of this laminar position . We find that OMS responses of VG3-ACs depend on convergent input from transient rectified ON and OFF bipolar cells . Figure 1—figure supplement 2 and Figure 1—figure supplement 3 contain terminals of OFF and ON bipolar cells , respectively ( Wassle et al . , 2009; Helmstaedter et al . , 2013 ) . Furthermore , bipolar cells with transient responses stratify near the center of the IPL ( i . e . , Figure 1—figure supplement 2 and Figure 1—figure supplement 3 ) , whereas axons of bipolar cells with sustained responses stratify closer to its borders ( Roska and Werblin , 2001; Baden et al . , 2013; Borghuis et al . , 2013 ) . Finally , a recent study identified gradients in the linearity of glutamate release across the IPL with more rectified bipolar cells stratifying towards the middle ( Borghuis et al . , 2013 ) . Stratification in Figure 1—figure supplement 2 and Figure 1—figure supplement 3 , thus , positions VG3-ACs ideally to recruit input from transient rectified ON and OFF bipolar cells , as well as to form synapses with dendrites of W3-RGCs . The strategic importance of this laminar position is further corroborated by the observation that in all species examined , VG3-ACs and LED RGCs stratify near the center of the IPL ( Berson et al . , 1998; Roska and Werblin , 2001; Famiglietti , 2005; van Wyk et al . , 2006; Zhang et al . , 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08025 . 00510 . 7554/eLife . 08025 . 006Figure 1—figure supplement 3 . Temporal receptive fields of VG3-ACs . To characterize temporal receptive fields of VG3-ACs , we presented white noise stimuli ( refresh rate: 30 Hz , RMS contrast: 40% ) to receptive field centers ( voltage and excitation ) or surrounds ( inhibition ) and adapted a principal-component-based approach to recover linear filters describing temporal sensitivity to ON and OFF stimuli , respectively ( ‘Materials and methods’ ) ( Greschner et al . , 2006; Gollisch and Meister , 2008 ) . ( A , E , I ) Linear ON and OFF filters constructed from voltage ( A , black ) , excitation ( E , red ) , and inhibition ( I , blue ) traces of representative VG3-ACs . ( B , C , F , G , J , K ) Peak times ( B , F , J ) and biphasic indices ( C , G , K , ON: \|trough\|/peak , OFF: peak/\|trough\| ) of ON and OFF filters . Dots show data from individual cells and circles ( error bars ) indicate mean ( ± SEM ) of the population . Peak times of ON and OFF filters were not significantly different for voltage ( B , black , n = 9 , p > 0 . 2 ) , excitation ( F , red , n = 9 , p > 0 . 4 ) , and inhibition ( J , blue , n = 9 , p > 0 . 08 ) . However , ON filters were more biphasic than OFF filters for excitation ( G , red , p < 0 . 002 ) and inhibition ( K , blue , p < 0 . 002 ) , but not voltage responses ( B , black , p > 0 . 1 ) . ( D , H , L ) Temporal frequency tuning functions calculated from Fourier amplitudes of ON ( left panels ) and OFF ( right panels ) filters for voltage ( D , black ) , excitation ( H , red ) , and inhibition ( L , blue ) responses show response suppression at high- and low-stimulus frequencies and illustrate the higher sensitivity of VG3-ACs and their synaptic inputs to OFF compared to ON stimuli . Circles ( error bars ) show mean ( ± SEM ) of the population . DOI: http://dx . doi . org/10 . 7554/eLife . 08025 . 006 The arbor size of VG3-AC neurites suggests that they receive input from >50 bipolar cells , which each sample a smaller region of visual space ( Wassle et al . , 2009; Morgan and Kerschensteiner , 2011 ) . Third order neurons that receive convergent rectified input from bipolar cells can detect motion and other changes in patterns with structure on the scale of bipolar cells' receptive fields even when the average luminance across their own receptive fields does not change ( i . e . , nonlinear spatial integration ) ( Victor and Shapley , 1979; Demb et al . , 2001; Schwartz et al . , 2012 ) . To evaluate input rectification of VG3-ACs , we presented sinusoidally modulated split-field stimuli in which the receptive field center was divided evenly or in a biased manner ( Figure 1G ) . We then compared the power of excitatory postsynaptic current ( EPSC ) responses at once ( F1 ) and twice ( F2 ) the frequency of modulation ( 2 Hz , Figure 1H , I ) . Excitatory input to biased split-field stimuli is modulated primarily at the stimulus frequency and recapitulates the OFF-preference of VG3-ACs . The F2-dominant responses to even split-field stimulation indicate that this excitatory input is provided by rectified bipolar cells ( Grimes et al . , 2014 ) . To test whether inhibition , like excitation , is driven by rectified receptive field subunits and to measure their spatial extent , we sinusoidally modulated square-wave gratings with different bar widths in an annular region overlaying the receptive field surround ( Figure 1J ) . Inhibitory postsynaptic currents ( IPSCs ) show frequency-doubled ( F2 ) responses characteristic of rectified input ( Figure 1K , L ) . F2 power increases in a step-like fashion between bar widths of 25 μm and 50 μm , suggesting that bipolar cells are likely the cellular substrate for nonlinear subunits ( Victor and Shapley , 1979; Demb et al . , 2001; Schwartz et al . , 2012 ) . Thus , VG3-ACs receive rectified excitatory input from transient ON and OFF bipolar cells , and inhibition from ACs , which themselves appear to be driven by rectified input from possibly the same types of bipolar cells . The receptive field mechanisms described so far—convergence of transient ON and OFF inputs , strong surround inhibition , and rectified receptive field subunits—led us to hypothesize that VG3-ACs may selectively detect the movements of small objects . To test this hypothesis , we evaluated voltage responses and synaptic inputs of VG3-ACs in differential motion ( Baccus et al . , 2008; Zhang et al . , 2012 ) and edge detection stimulus paradigms ( Levick , 1967; van Wyk et al . , 2006 ) . When square-wave gratings overlaying center and surround regions of their receptive field were shifted separately or together ( Figure 2A ) ( Baccus et al . , 2008; Zhang et al . , 2012 ) , VG3-ACs depolarized robustly to differential motion in the center , but hyperpolarized to synchronous movements in center and surround ( i . e . , global motion ) and differential motion in the surround ( Figure 2B , C ) . Voltage-clamp recordings revealed that this response pattern is caused by preferential excitation during center-only motion and strong inhibition elicited whenever motion includes the surround ( Figure 2B , E , D ) . Since the average luminance in center and surround regions does not change in this stimulus , the observed responses provide further evidence that both are composed of rectified subunits , allowing VG3-ACs to detect when an object moves at a different time than the background irrespective of the precise spatial patterns involved , a defining feature of OMS neurons ( Olveczky et al . , 2003; Baccus et al . , 2008 ) . 10 . 7554/eLife . 08025 . 007Figure 2 . Detection of object motion by VG3-ACs . ( A ) Schematic illustrating texture motion stimuli . Two square-wave gratings ( bar width: 50 μm ) , one covering the center and one the surround region of the VG3-AC receptive field , are separated by a gray annulus . During stimulus presentation , both gratings move first together ( global ) and then separately ( differential center motion denoted by diffCe and differential surround motion by diffSu ) . ( B ) Representative voltage ( black ) , EPSC ( red ) , and IPSC ( blue ) traces recorded during presentation of the stimulus shown in ( A ) . ( C–E ) Summary data of voltage ( C ) , excitatory ( D ) , and inhibitory ( E ) response amplitudes to global , differential center ( diffCe ) , and differential surround ( diffSu ) motion stimuli . Dots show data from individual cells ( voltage n = 9 , excitation n = 22 , inhibition n = 22 , p < 10−7 for all comparisons ) , and circles ( error bars ) indicate mean ( ±SEM ) of the respective population . ( F ) Schematic illustrating a stimulus in which a narrow bar ( height: 200 μm ) is moved across the receptive field of a VG3-AC at a variety of speeds annotated in ( F ) and encoded by color saturation throughout . ( G , I ) Representative voltage ( G , black ) , EPSC ( I , red ) , and IPSC ( I , blue ) traces recorded in response to bars moving at different speeds . Time points when the leading ( LE ) and trailing edges ( TE ) of the bar are in the center of the receptive field are indicated . ( H , J ) Summary data of voltage ( H , black ) , excitation ( J , red ) , and inhibition ( J , blue ) response amplitudes . Circles ( error bars ) indicate the mean ( ±SEM ) of these data sets ( voltage n = 32 , excitation n = 8 , inhibition n = 5 ) . ( K ) Schematic illustrating a stimulus in which bars of varying height are moved across the receptive field of a VG3-AC at a constant speed ( 400 μm/s ) . Bar heights are encoded in by color saturation as indicated in ( K ) . ( L–O ) Representative voltage ( L , black ) , EPSC ( N , red ) , and IPSC ( N , blue ) traces . Summary data of voltage ( M ) , excitation ( O , red ) , and inhibition ( O , blue ) response amplitudes reveal suppression of edge responses for voltage and excitation , but not inhibition at greater bar heights ( comparison of 200 μm and 600 μm , voltage n = 32 , p < 10−7 excitation n = 8 , p < 0 . 001 , inhibition n = 5 , p > 0 . 14 ) . Circles ( error bars ) indicate the mean ( ±SEM ) of these data sets . See also Figure 1—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 08025 . 00710 . 7554/eLife . 08025 . 008Figure 2—figure supplement 1 . Responses of VG3-ACs are not direction selective . ( A ) Representative responses of a VG3-AC to a sine grating drifting in eight different directions ( frequency: 2 Hz ) . The polar plot in the center shows the Fourier amplitude of the response at the stimulus frequency ( F1 ) . ( B ) Polar plot summarizing responses of nine VG3-ACs ( mean ± SEM ) . ( C ) Histogram of direction selectivity index ( DSI ) values of these VG3-ACs . The DSI was calculated as the absolute amplitude of ∑F1 ( θ ) eiθ∑F1 ( θ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08025 . 008 Next , we tested the ability of VG3-ACs to detect contrast edges . Narrow dark bars moving across the receptive fields of VG3-ACs elicited transient depolarizations as leading and trailing edges cross the receptive field center . Robust edge responses were observed over a wide range of motion speeds ( Figure 2G , H ) and were indistinguishable for leading and trailing edges in spite of the opposite polarity of the associated temporal contrast . Responses of VG3-ACs are not directionally selective ( Figure 2—figure supplement 1 ) . Similar to voltage recordings , transient EPSCs and IPSCs were observed during edge transits ( Figure 2I ) . EPSCs and IPSCs were larger at leading than trailing edges and showed a preference for the faster speeds tested ( Figure 2J ) . However , the ratio of excitation and inhibition remained relatively constant across these conditions , which likely accounts for the greater constancy of edge responses observed in voltage recordings . Local edge detectors are typically stimulated robustly by narrow contrast edges , but not by expansive ones ( Levick , 1967; Roska and Werblin , 2001; Zeck et al . , 2005; van Wyk et al . , 2006 ) . We , therefore , recorded responses of VG3-ACs to bars of varying heights moving across their receptive field at a constant speed ( Figure 2K ) . Edge responses were strongly suppressed for bar heights above 200 μm ( Figure 2L , M ) . Trailing edge responses to the smallest bars were lower than leading edge responses , likely a consequence of the asymmetric size of VG3-ACs' ON and OFF receptive field centers ( Figure 1 ) . Similar to voltage responses , excitatory synaptic inputs showed edge response suppression for larger stimuli , whereas inhibition rose and plateaued with increasing bar height ( Figure 2N , O ) , corroborating that surround inhibition acts both pre- and post-synaptic in VG3-ACs . VG3-ACs , thus , show key response features of OMS and local-edge-detector neurons , and through the synaptic mechanisms outlined above , selectively detect movements of small objects . The data presented so far suggest that pre- and post-synaptic surround inhibition cancel responses of VG3-ACs to global scene shifts and movements of large objects . Surround suppression of the excitatory input to W3-RGCs was found to rely on spiking ACs ( Zhang et al . , 2012 ) . To test whether surround suppression of VG3-ACs is similarly mediated by spiking ACs and to assess its importance to feature detection , we tested the effect of the sodium channel blocker TTX on VG3-ACs' responses in differential motion and edge detection stimulus paradigms . In the presence of TTX , VG3-ACs depolarized to global motion as well as differential center motion , and hyperpolarizations observed during surround stimulation were abolished ( Figure 3A , B ) . Edge responses elicited by narrow bars moving across the receptive fields of VG3-ACs at a variety of speeds were not changed by TTX ( Figure 3C , D ) , but the suppression observed for larger bars was blocked ( Figure 3E , F ) . Surround inhibition , thus , is critical for the feature detection of VG3-ACs and appears to be mediated by spiking ACs . The use of spikes to signal surround motion likely improves temporal coincidence of inhibitory input with bipolar cell depolarization and excitatory input to VG3-ACs during coherent motion , leading to more effective cancellation of center signals by the surround . 10 . 7554/eLife . 08025 . 009Figure 3 . Spiking ACs mediate surround suppression of VG3-ACs . ( A ) Representative voltage traces recorded from a VG3-AC during presentation of the texture motion stimulus illustrated in Figure 2A in control conditions ( top , black ) and in the presence of TTX ( bottom , blue ) . ( B ) Bars ( error bars ) indicating mean ( ±SEM ) response amplitudes to different segments of the texture motion stimulus in control conditions ( left , black ) and in the presence of TTX ( right , blue ) . TTX abolishes the suppression by global and differential surround motion ( control n = 9 , TTX n = 5 , p < 10−4 for TTX vs control ) but does not affect the response to differential center motion ( p > 0 . 2 for TTX vs control ) . ( C , D ) Representative traces ( C ) and summary data ( D , mean ± SEM ) for voltage responses of VG3-ACs to narrow dark bars ( height: 200 μm ) moving at different speeds ( 100 μm/s , 200 μm/s , 400 μm/s , and 800 μm/s from top to bottom ) , encoded by matching color saturation in ( C ) and ( D ) in the absence ( left , black ) or presence ( right , blue ) of TTX . TTX did not significantly alter responses to narrow bars irrespective of their speed ( control n = 32 , TTX n = 7 , p > 0 . 1 for all comparisons ) . ( E , F ) Exemplary traces ( E ) and population data ( F , mean ± SEM ) for voltage responses of VG3-ACs to dark bars of different heights ( 50 μm , 100 μm , 200 μm , 400 μm , and 600 μm from top to bottom ) moving at 400 μm/s , indicated by matching color saturation in ( E ) and ( F ) , in the absence ( left , black ) or presence ( right , blue ) of TTX . Surround suppression is canceled by TTX leading to increased responses to larger bars ( at 600 μm , control n = 32 , TTX n = 7 , p < p < 10−4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08025 . 009 We next wondered how signals of VG3-ACs are used in the retina . Recent optogenetic experiments and paired recordings suggest that VG3-ACs can release glutamate and activate postsynaptic receptors on several RGC types , including W3-RGCs ( Lee et al . , 2014 ) . Whether and how VG3-ACs contribute to visual processing in these RGCs remains unknown . Given our results on object-motion detection , we focus here on VG3-ACs' connections with and influence on W3-RGCs . To obtain anatomical evidence for or against excitatory synapses between VG3-ACs and W3-RGCs , we biolistically labeled W3-RGCs with cytosolic cerulean fluorescent protein and PSD95 fused to yellow fluorescent protein ( PSD95-YFP ) in VG3-Cre Ai9 mice ( Figure 4A , B ) . PSD95-YFP selectively localizes to excitatory synapses on RGC dendrites ( Morgan et al . , 2008; Kerschensteiner et al . , 2009 ) . More than half of the PSD95-YFP puncta on W3-RGCs were apposed by VG3-ACs boutons , whereas few appositions with VG3-ACs were observed when PSD95-YFP puncta were randomly repositioned along the dendrites in Monte Carlo simulations ( Figure 4C , D ) . We next characterized spike responses and synaptic inputs of W3-RGCs with the same differential motion and edge detection stimuli used for VG3-ACs , revealing matching tuning properties of excitatory input to W3-RGCs with responses of VG3-ACs ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 08025 . 010Figure 4 . Anatomy and function of input from VG3-ACs to W3-RGCs . ( A ) Orthogonal projections of a confocal image stack through a representative W3-retinal ganglion cell ( RGC ) labeled biolistically with cyan fluorescent protein ( CFP ) . W3-RGCs were identified by their characteristic morphology ( Kim et al . , 2010; Zhang et al . , 2012 ) with small dendritic fields ( territory size: 10 , 783 ± 409 μm2 , n = 25 ) filled by densely branched neurites stratifying in the center of the IPL with a secondary arborization near the border between the inner plexiform and inner nuclear layers ( INLs ) . The fluorescent signal is colored to represent depth in the IPL . Inset bar graph shows the mean ± SEM territory size of W3-RGCs ( n = 15 ) measured as the areas of the smallest convex polygons to encompass their arbors in a z-projection . ( B , C ) Overview projections ( B ) and single plane excerpts ( C ) of a W3-RGC biolistically labeled with CFP ( red ) and PSD95-YFP ( green ) in a VG3-Cre Ai9 mouse ( tdTomato shown in blue ) . ( D ) Summary data indicating the fraction of PSD95-YFP puncta apposed by VG3-AC boutons ( ‘Materials and methods’ ) in the obtained images ( real ) or when positions of PSD95-YFP puncta were randomized within the synaptic layer ( random ) in Monte Carlo simulations ( n = 9 cells , p < 10−6 ) . Gray lines indicate data from individual cells; circles ( error bars ) show the mean ( ±SEM ) of the population . ( E–H ) Representative EPSC ( E ) and spike response ( G ) traces to the texture motion stimulus illustrated in Figure 2A recorded from W3-RGCs ( wild-type [WT] black , vesicular glutamate transporter 3 [VGluT3−/−] blue ) , and bar plots summarizing differences in excitatory conductance ( F ) and spike rates ( H ) during different segments of the stimulus in WT ( left , black ) and VGluT3−/− mice ( right , blue ) . Bars ( error bars ) indicate the mean ( ±SEM ) of the respective data sets . W3-RGC EPSCs in VGluT3−/− mice were significantly reduced compared to WT littermates during differential center motion ( WT n = 8 , VGluT3−/− n = 9 , p < 0 . 02 ) , but not global or differential surround motion ( p > 0 . 1 for both ) . A similar pattern was observed in the spike responses of W3-RGCs , which were decreased for differential center motion ( WT n = 13 , VGluT3−/− n = 9 , p < 0 . 001 ) , but not altered during global image motion ( p > 0 . 9 ) . ( I–L ) Representative EPSC ( I ) and spike response ( K ) traces , and summary data ( excitation in J , spikes in L , mean ± SEM ) recorded in W3-RGCs during stimulation with dark bars of different heights ( indicated by color saturation ) moving at 400 μm/s in WT ( left , black ) and VGluT3−/− mice ( right , blue ) . Whereas excitatory inputs and spike responses were reduced for narrow bars ( excitation at 100 μm , WT n = 6 , VGluT3−/− n = 6 , p < 0 . 03 , spikes at 100 μm , WT n = 14 , VGluT3−/− n = 8 , p < 0 . 01 ) , they did not differ significantly for bars of greater heights ( excitation at 600 μm , WT n = 6 , VGluT3−/− n = 6 , p > 0 . 6 , spikes at 600 μm , WT n = 14 , VGluT3−/− n = 8 , p < 0 . 2 ) . See also Figure 2—figure supplement 1 , Figure 4—figure supplement 1 and Figure 4—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 08025 . 01010 . 7554/eLife . 08025 . 011Figure 4—figure supplement 1 . Detection of object motion by W3-RGCs . ( A ) Representative spike rate ( black ) , EPSC ( red ) , and IPSC ( blue ) traces recorded from W3-RGCs during presentation of the texture motion stimuli ( illustrated in Figure 2A ) . W3-RGCs were either recorded under conventional infrared illumination and identified by characteristic responses in cell-attached recordings or targeted under 2-photon guidance in Isl2-GFP transgenic mice ( Triplett et al . , 2014 ) . In both cases , correct targeting was confirmed by intracellular dye filling and reconstruction of dendritic arborizations at the end of the recordings . ( B–D ) Summary data of spike rate ( B , black ) , excitatory ( C , red ) , and inhibitory ( D , blue ) response amplitudes to global , differential center ( diffCe ) , and differential surround ( diffSu ) motion segments . Dots show data from individual cells and circles ( error bars ) indicate mean ( ± SEM ) of the respective population . W3-RGCs remain mostly silent during global and differential surround motion stimulation , but show robust spike responses to grating movements restricted to their receptive field center ( n = 13 , p < 10−7 for diffCe vs global and vs diffSu ) . Differential motion sensitivity of WG3-RGCs appears to be inherited from their excitatory input , which during center-only motion exceeds that observed during global motion nearly fivefold , and which is suppressed from tonic levels during isolated surround stimulation ( n = 8 , p < 0 . 003 for diffCe vs global and vs diffSu ) . In addition , W3-RGCs receive stronger direct inhibition when grating movements include the surround ( n = 7 , p < 0 . 001 for diffCe vs global and vs diffSu ) ( E–J ) Representative spike rate ( E , black ) , EPSC ( G , red ) , and IPSC ( I , blue ) traces and summary data of spike ( F , black ) , excitatory ( H , red ) , and inhibitory ( J , blue ) response amplitudes to dark bars ( height: 200 μm ) moving at different speeds indicated by matching color saturation of example traces and summary data . Circles ( error bars ) represent the mean ( ± SEM ) of these data sets ( spikes n = 32 , excitation n = 8 , inhibition n = 5 ) . ( K–P ) Representative spike rate ( K , black ) , EPSC ( N , red ) , and IPSC ( O , blue ) responses elicited by bars of different heights moving at 400 μm/s and summary data of spike rate ( L , black ) , excitatory ( N , red ) , and inhibitory ( P , blue ) response amplitudes . Bar heights are encoded by matching color saturation of responses traces and summary data . Circles ( error bars ) represent the mean ( ± SEM ) of these data sets . Responses of W3-RGCs are progressively suppressed when bar heights increase above 100 μm ( n = 14 , p < 10−4 for 200 μm vs 600 μm bar heights ) . The spatial tuning of the excitatory input to W3-RGCs is similar to that observed in spike responses ( n = 6 , p < 0 . 02 for 200 μm vs 600 μm bar heights ) , whereas inhibition rises monotonically and saturates with increasing bar heights ( n = 6 , p > 0 . 4 for 200 μm vs 600 μm bar heights ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08025 . 01110 . 7554/eLife . 08025 . 012Figure 4—figure supplement 2 . Lamination patterns of cells and neurites are preserved in VGluT3−/− mice . ( A–P ) Maximum intensity projections of confocal image stacks of representative vibratome sections of WT ( A , C , E , G , I , K , M , O ) and VGluT3−/− ( B , D , F , H , J , L , N , P ) retinas . Sections were stained for choline acetyltransferase ( ChAT , A , B ) , protein kinase C alpha ( PKCα , C , D ) , synaptotagmin II ( Syt II , E , F ) , hyperpolarization activated cyclic nucleotide gated potassium channel 4 ( HCN4 , G , H ) , Tyrosine hydroxylase ( TH , I , J ) , Recoverin ( K , L ) , Melanopsin ( M , N ) , and Protein Kinase A regulatory subunit II beta ( PKARIIβ , O , P ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08025 . 01210 . 7554/eLife . 08025 . 013Figure 4—figure supplement 3 . Dendritic morphology of W3-RGCs is unchanged in VGluT3−/− mice . ( A , B ) Maximum intensity projections through 2-photon image stacks of representative W3-RGCs recorded in WT ( A ) and VGluT3−/− ( B ) mice . ( C ) Summary data of dendritic territories covered by W3-RGCs labeled biolistically or filled during patch-clamp recordings in WT ( black ) or VGluT3−/− ( blue ) retinas . DOI: http://dx . doi . org/10 . 7554/eLife . 08025 . 01310 . 7554/eLife . 08025 . 014Figure 4—figure supplement 4 . Schematic of object motion detection circuit . ( A ) Overview illustration of the object motion detection circuit . Somata of ON and OFF BCs are shown as open and filled ovals , respectively . Axons of these neurons converge onto the other components of the circuit: wide-field ACs ( wACs ) , VG3-ACs , and W3-RGCs . ( B ) Schematic of the inferred connectivity motif , repeated in ON and OFF layers of the object motion detection circuit . Excitatory and inhibitory synaptic output is shown by circles and triangles , respectively , and use of spikes ( wAC , ss ) or graded potentials ( BC , VG3 ) is indicated by different waveforms . DOI: http://dx . doi . org/10 . 7554/eLife . 08025 . 014 To test whether VG3-ACs provide excitatory input to W3-RGCs during visual stimulation , to compare the tuning of VG3- and non-VG3 inputs , and assess VG3-ACs' contribution to object motion signals sent to the brain , we recorded W3-RGCs in mice lacking VGluT3 ( VGluT3−/− mice ) ( Seal et al . , 2008 ) . Removal of VGluT3 , which in the retina is only expressed by VG3-ACs , affected neither gross morphological development of the retina ( Figure 4—figure supplement 2 ) nor dendritic patterns of W3-RGCs ( Figure 4—figure supplement 3 ) . EPSCs elicited by differential center motion were reduced by approximately 50% in W3-RGCs of VGluT3−/− compared to wild-type ( WT ) mice ( Figure 4E , F ) . By contrast , excitation during global motion and tonic excitation revealed by differential surround stimulation were unchanged . A similar pattern was observed in the spike responses of W3-RGCs , which were selectively decreased for differential center but not global motion stimuli ( Figure 4G , H ) . Similarly , EPSCs and spike responses evoked by edges of narrow moving bars ( height: 100 μm ) were suppressed across a wide range of speeds in VGluT3−/− mice ( Figure 4I–L and data not shown ) , whereas excitation and spike responses elicited by broader moving bars remained intact in VGluT3−/− mice ( Figure 4I–L ) . In agreement with anatomical results ( Figure 4B , C ) , VG3-ACs , thus , appear to provide approximately half of the excitatory input to W3-RGCs . Importantly , feature selectivity of this VG3-input is more sharply tuned than the excitatory input remaining in VGluT3−/− mice—likely provided by ON and OFF bipolar cells—and is required for normal spike responses of W3-RGCs . In the OMS circuit ( Figure 4—figure supplement 4 ) , VG3-ACs serve to amplify and sharpen the tuning of responses to object motion . Multi-tiered inhibition combined with delayed excitation , and successive threshold nonlinearities likely contribute to sharpening . Surround inhibition acts at three levels: bipolar axon terminals , VG3-ACs , and W3-RGCs ( Zhang et al . , 2012; Lee et al . , 2014 ) . Key features—transient ON and OFF input driven by rectified subunits—are similar at all three stages , arguing that inhibition is provided by a single AC type or a shared set of AC types , which remain to be identified . The added level of inhibition onto VG3-ACs compared to conventional pathways through bipolar cells likely contributes to the more complete surround suppression in the OMS circuit . Moreover , channeling of excitation through VG3-ACs introduces a delay not shared by the inhibitory input , which could improve cancellation of center signals by the surround , for example , during global image motion . The sequential arrangement of three thresholding nonlinearities—glutamate release from bipolar cells , glutamate release from VG3-ACs , and spike generation in W3-RGCs—likely further contributes to the increasing selectivity for narrow vs broad edges and differential center vs global texture motion at successive stages of the OMS circuit . Finally , our results support the notion that the diversity of AC types and circuit motifs in which they participate are integral to the diversity of features encoded in the signals the retinal sends to the brain ( Masland , 2012; Jadzinsky and Baccus , 2013 ) . We used homologous recombination to introduce the CreERT2 DNA recombinase into a BAC containing regulatory sequences of the Slc17a8 gene , which encodes the VGluT3 . The resulting construct was injected into pronuclei to generate VG3-CreERT2 mice . To induce Cre-mediated recombination in VG3-CreERT2 mice , adult animals ( i . e . , older than P21 ) were injected intraperitoneally with 1 mg tamoxifen for five consecutive days . Both VG3-CreERT2 and the noninducible VG3-Cre ( Grimes et al . , 2011 ) mice were crossed to a fluorescent reporter strain expressing tdTomato in a Cre-dependent manner ( Ai9 ) ( Madisen et al . , 2010 ) to enable anatomical reconstructions and targeted patch-clamp recordings from VG3-ACs . Contributions of VGluT3-mediated neurotransmission to visual processing in the retina were evaluated by comparing knockout mice ( VGluT3−/− ) ( Seal et al . , 2008 ) and WT littermates . Isl2-GFP transgenic mice ( Triplett et al . , 2014 ) were obtained from the Mutant Mouse Regional Resource Center ( MMRRC ) , to which they were donated by Dr Nathaniel Heintz and used for targeted patch-clamp recordings of W3-RGCs . All procedures in this study were approved by the Animal Studies Committee of Washington University School of Medicine ( Protocol #: 20140095 ) and performed in compliance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals . Mice were dark-adapted for >2 hr , deeply anesthetized with CO2 , killed by cervical dislocation , and their eyes removed . Retinas were isolated and flat mounted on membrane discs ( for anatomy: HABG01300 , Millipore , Billerica , MA; for physiology: Anodisc 13 , Whatman , Pittsburgh , PA ) . For patch-clamp recordings , enucleation and tissue preparation were carried out under infrared ( >900 nm ) illumination . For immunohistochemistry , tissue was fixed for 30 min in 4% paraformaldehyde in mouse artificial cerebrospinal fluid ( mACSFHEPES ) , washed for >10 min in phosphate-buffered saline ( PBS ) , washed in 10% sucrose in PBS for 1 hr at RT , washed in 20% sucrose in PBS for 1 hr at RT , washed in 30% sucrose in PBS overnight at 4°C , freeze-thawed three times , washed in PBS for >10 min , and incubated in 5% normal donkey serum for 2 hr prior to addition of primary antibodies . Vibratome sections ( 60 μm thick ) and retinal flat mounts were stained with rabbit anti-calretinin ( 1:1000 , Millipore ) , goat anti-ChAT ( 1:500 , Millipore ) , rabbit anti-recoverin ( 1:1000 , Millipore ) , rabbit anti-PKARIIβ ( 1:500 , BD Bioscience , San Jose , CA ) , rabbit anti-TH ( 1:1000 , Millipore ) , rabbit anti-HCN4 ( 1:500 , Neuromab , Davis , CA ) , rabbit anti-VIP ( 1:1000 , Immunostar , Hudson , WI ) , mouse anti-melanopsin ( 1:1000 , Advanced Targeting Systems , San Diego , CA ) , mouse anti-PKCα ( 1:500 , Sigma , Saint Louis , MO ) , and mouse anti-Znp1/SytII ( 1:1000 , ZIRC , Eugene , OR ) for 1 ( vibratome slices ) or 5 days ( flat mounts ) at 4°C . The tissue was then washed in PBS ( 3 × 30 min ) , incubated with DyLight 405- ( 1:100 , Jackson ImmunoResearch , West Grove , PA ) , Alexa Fluor 488- , Alexa Fluor 568- , and/or Alexa Fluor 633-conjugated secondary antibodies ( 1:1000 , Invitrogen , Grand Island , NY ) for 2 hr at RT ( vibratome slices ) or 2 days at 4°C ( flat mounts ) , washed again in PBS ( 3 × 30 min ) , and mounted in Vectashield mounting medium ( Vector Laboratories , Burlingame , CA ) for confocal imaging . Gold particles ( 1 . 6-μm diameter , BioRad , Hercules , CA ) were coated with plasmids encoding cytosolic cyan fluorescent protein ( CFP ) and PSD95 fused at its C-terminus to yellow fluorescent protein ( PSD95-YFP ) ( Morgan and Kerschensteiner , 2012 ) . Particles were delivered to RGCs from a helium-pressurized gun ( BioRad ) at approximately 40 psi ( Morgan and Kerschensteiner , 2011 ) . After shooting , retinal flat mount preparations in mACSFHEPES—containing ( in mM ) : 119 NaCl , 2 . 5 KCl , 2 . 5 CaCl2 , 1 . 3 MgCl2 , 1 NaH2PO4 , 11 glucose , and 20 4- ( 2-hydroxyethyl ) -1-piperazineethanesulfonic acid ( HEPES ) ( pH adjusted to 7 . 37 with NaOH ) —were incubated in a humid oxygenated chamber at 33–35°C for 14–18 hr . The tissue was then fixed for 30 min in 4% paraformaldehyde in mACSFHEPES and washed PBS ( 3 × 10 min ) before mounting and imaging . VG3-ACs , W3-RGCs , and patterns of connections between them were reconstructed from confocal and 2-photon imaging stacks acquired on Fv1000 laser scanning microscopes ( Olympus , Tokyo , Japan ) using 60× 1 . 35 NA oil immersion or 20× 0 . 9 NA water immersion objectives . Synaptic connectivity was analyzed in image stacks with voxel size 0 . 103 μm ( x/y-axis ) –0 . 3 μm ( z-axis ) , whereas neurite territories were measured in image stacks with voxel size 0 . 206 μm ( x/y-axis ) –0 . 5 μm ( z-axis ) . Whole-cell patch-clamp recordings from VG3-ACs in the INL and W3-RGCs in the ganglion cell layer were obtained in the dorsal halves ( Wei et al . , 2010; Wang et al . , 2011 ) of flat-mounted retinas continuously superfused ( 6–8 ml/min ) with warm ( 33–35°C ) mACSFNaHCO3 containing ( in mM ) 125 NaCl , 2 . 5 KCl , 1 MgCl2 , 1 . 25 NaH2PO4 , 2 CaCl2 , 20 glucose , 26 NaHCO3 , and 0 . 5 L-Glutamine equilibrated with 95% O2/5% CO2 . In some experiments , the following pharmacological agents were added to mACSFNaHCO3 individually or in combinations ( ‘Results’ ) and bath-applied: L-2-Amino-4-phosphonobutyric acid ( L-APB , 50 μM , Tocris , Bristol , United Kingdom ) , 1 , 2 , 5 , 6-tetrahydropyridine-4-yl-methylphosphinic acid ( TPMPA , 50 μM , Sigma ) , gabazine ( 5 μM , Tocris ) , strychnine ( 500 nM , Sigma ) , and tetrodotoxin ( TTX , 1 μM , Sigma ) . Current-clamp recordings were performed with an intracellular solution containing ( in mM ) : 125 K-gluconate , 10 NaCl , 1 MgCl2 , 10 ethylene glycol tetraacetic acid ( EGTA ) , 5 HEPES , 5 Adenosine triphosphate disodium salt ( ATP-Na2 ) , and 0 . 1 Guanosine triphosphate disodium salt ( GTP-Na2 ) ( pH adjusted to 7 . 2 with KOH ) . The intracellular solution used in voltage-clamp recordings contained ( in mM ) : 120 Cs-gluconate , 1 CaCl2 , 1 MgCl2 , 10 Na-HEPES , 11 EGTA , 10 TEA-Cl , and 2 Qx314 ( pH adjusted to 7 . 2 with CsOH ) . Alexa 488 or 568 were added ( 0 . 1 mM ) to both intracellular solutions . Patch pipettes had resistances of 4–7 MΩ ( borosilicate glass ) . All reported voltages were corrected for liquid junction potentials . For voltage-clamp recordings , series resistance ( 10–15 MΩ ) was compensated electronically by ∼75% . Signals were amplified with a Multiclamp 700B amplifier ( Molecular Devices , Sunnyvale , CA ) , filtered at 3 kHz ( 8-pole Bessel low-pass ) and sampled at 10 kHz ( Digidata 1440A , Molecular Devices ) . EPSCs were isolated by clamping the voltage of the recorded cell to the reversal potential for Cl− ( −60 mV ) , the main permeant ion of inhibitory transmitter receptors , whereas IPSCs were recorded at the reversal potential of currents through excitatory transmitter receptors ( 0 mV ) . In current-clamp recordings , no bias current was injected . Fluorescent VG3-ACs were targeted under 2-photon guidance ( excitation wavelength: 900 nm ) in VG3-Cre Ai9 mice . W3-RGCs were either recorded under conventional infrared illumination ( >900 nm ) or targeted under 2-photon guidance in Isl2-GFP mice . Correct targeting was confirmed by monitoring entry of Alexa dyes ( 488 or 568 ) included in the intracellular solution from the recording pipette into the soma during break in and by reconstructing the morphology of neurite arbors at the end of each recording . Custom stimuli written in MATLAB ( The Mathworks , Natick , MA ) using Cogent graphics extensions ( John Romaya , Laboratory of Neurobiology at the Wellcome Department of Imaging Neuroscience , University College London ) were presented on an organic light-emitting display ( xOLED , eMagin , Bellevue , WA ) and focused onto the photoreceptors through the substage condenser of an integrated 2-photon patch-clamp setup . The average intensity of all stimuli was kept constant at ∼3000 R/rod/s or ∼2500 M/M-cone/s and their position centered on the soma of the recorded cell . To measure area response functions of VG3-ACs and W3-RGCs , the intensity of a circular area with varying radii was square-wave modulated at 0 . 125 Hz ( Michelson contrast: 96% ) . Circles of different size were presented in different pseudorandom sequences for each cell , and the first stimulus in the sequence repeated at its end to confirm stability of the recording . Temporal response functions and filter properties were analyzed using Gaussian white noise stimulation in which the intensity of a circular region over the receptive field center ( voltage , excitation ) or an annular region covering its surround ( inhibition ) was chosen at random from a normal distribution ( RMS contrast: 40% ) and updated at 30 Hz for 10 min . The properties of spatial integration were tested using sine-wave modulated ( 2 Hz ) contrast-reversing square-wave gratings ( varying spatial frequency ) masked to preferentially stimulate the center ( excitation ) and surround ( inhibition ) regions of a receptive field . To test the sensitivity of VG3-ACs and W3-RGCs to differential or coherent luminance-neutral motion stimuli in their receptive field center and surround , the respective parts of square-wave gratings ( bar width: 50–75 μm ) were moved separately or in unison . A gray annulus was included in the spatial layout of the stimulus to reliably separate movement in the center and surround . Edge detection properties were tested by moving dark and light bars of different heights ( 50–600 μm ) and constant width ( 800 μm ) across the receptive fields of VG3-ACs and W3-RGCs at a variety of speeds ( 100–800 μm/s ) . The order in which bars of different contrast , height , and speed were shown was randomized for each cell . Data were analyzed using programs written in MATLAB . Response amplitudes to a variety of stimuli—circles of varying size and contrast , bars of different size and contrast moving at a variety of speeds and differential or global motion of gratings—were measured as baseline-subtracted averages ( spike rate , conductance , or voltage ) during 100–200 ms time windows . To estimate spatial receptive field parameters , we fit Difference-of-Gaussians ( voltage , excitation ) or single Gaussian ( inhibition ) models to area response data ( Enroth-Cugell and Robson , 1966; McMahon et al . , 2004 ) . Receptive field sizes were measured as the diameter of the respective Gaussians at 1 SD . To analyze of VG3-ACs , we recovered separate ON and OFF filters from responses to Gaussian white noise stimulation . First , overlapping stimulus segments were weighted by their ensuing baseline-subtracted response ( voltage or conductance ) to compute the response-weighted stimulus ensemble , analogous to the spike-triggered stimulus ensemble frequently used in the characterization of RGC responses ( Chichilnisky , 2001 ) . We then identified the dimension of highest variance in the response-weighted stimulus ensemble by principle components analysis and separated response-weighted stimulus segments in two groups based on their projection onto this first principle component ( positive = ON , negative = OFF ) . ON and OFF segments were averaged to determine ON and OFF filters , respectively ( Figure 1—figure supplement 3 ) . The amplitudes of these linear filters were scaled such that their sum equaled the sum of the global response-weighted stimulus average . Similar procedures have previously been used to analyze the contributions of ON and OFF pathways to RGC spike trains ( Fairhall et al . , 2006; Greschner et al . , 2006; Gollisch and Meister , 2008 ) . To summarize temporal frequency tuning and sensitivity across several cells , Fourier amplitudes of ON and OFF filters were calculated . To characterize spatial integration , Fourier amplitudes of responses at once ( F1 , 2 Hz ) and twice ( F2 , 4 Hz ) the frequencies of sine wave modulation of contrast-reversing square-wave gratings over the receptive field center ( excitation ) or surround ( inhibition ) were analyzed . Edge detection tuning was measured from response amplitudes as leading and trailing edges of dark or bright bars of different heights moving at a variety of speeds crossed the receptive field center . Similarly , response amplitudes during selective ( center or surround ) or global ( center and surround ) luminance-neutral motion stimuli were compared to reveal differential motion sensitivity . Synaptic connectivity between VG3-ACs and W3-RGCs was analyzed in confocal image stacks of retinas from VG3-Cre Ai9 mice in which a sparse subset of RGCs ( incl . W3-RGCs ) expressed CFP and PSD95-YFP following biolistic transfection . Using local thresholding W3-RGC dendrites , PSD95 puncta and VG3-AC neurites were masked separately in Amira ( FEI Company ) . Excitatory synapses on W3-RGCs formed by VG3-ACs were defined as PSD95 clusters with a center of mass within 0 . 5 μm from a VG3-AC neurite . We confirmed that varying this distance from 0 . 25 to 1 μm did not qualitatively change the results . Given the size of synaptic puncta , this range implies overlap or direct apposition of signals from PSD95-YFP and tdTomato in VG3 neurites . To compare the input fraction ( i . e . , fraction of PSD95 puncta apposed by a VG3 neurite ) to chance levels , positions of PSD95 puncta were randomized within the synaptic layer in Monte Carlo simulations . Territories occupied by neurites of VG3-ACs and dendrites of W3-RGCs were measured by the area of the smallest convex polygon to encompass the respective arbors in a z-axis projection of image stacks acquired in flat-mounted retinas . To analyze soma distributions of VG3-ACs , we used a previously described algorithm to automatically identify cell positions ( Soto et al . , 2012 ) and calculated the density recovery profiles of these positions ( Rodieck , 1991 ) . Throughout this study , paired and unpaired t-tests were used to assess statistical significance of observed differences .	Animals can use their eyes to detect moving objects , which helps them to avoid predators and other threats , and to spot potential prey or allies . Visual information from the eyes is sent to the brain , which processes the information to form a coherent picture of how the objects are moving . This processing has to be able to account for movements of the head , eyes , and body—which can cause the image of an object on the retina within the eye to move even if the object itself remains stationary . Within the retina , light is converted into electrical signals by cells called rods and cones . A layer of cells called bipolar cells relay these signals to the ‘ganglion’ cells , which in turn pass them on to the brain . In mice , a type of ganglion cell called the W3 ganglion cell has been shown to respond selectively to small moving objects , but exactly how these cells acquire their motion sensitivity remained unclear . Kim et al . now reveal that cells called amacrine cells , which regulate the transfer of signals from the bipolar cells to ganglion cells , supply the information needed for motion detection . The mouse eye contains up to 50 different types of amacrine cells . One of these—called the VG3-amacrine cell—increases its activity whenever an object moves relative to its background , but decreases its activity whenever the object and background move together . The overall effect is that the cells respond selectively to the presence of small moving objects . Most amacrine cells regulate the transfer of signals within the retina by inhibiting the activity of ganglion cells . But , Kim et al . show that VG3-amacrine cells release a molecule called glutamate to activate W3 ganglion cells when a moving object is detected . These unusual and specialized cells are , thus , an essential component of a circuit in the nervous system that supports motion detection . It is possible that some other types of amacrine cells may also play specialized roles in the detection of other features in the visual world .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "short", "report", "neuroscience" ]	2015	An excitatory amacrine cell detects object motion and provides feature-selective input to ganglion cells in the mouse retina
Methicillin-resistant Staphylococcus aureus ( MRSA ) is a continued threat to human health in both community and healthcare settings . In hospitals , control efforts would benefit from accurate estimation of asymptomatic colonization and infection importation rates from the community . However , developing such estimates remains challenging due to limited observation of colonization and complicated transmission dynamics within hospitals and the community . Here , we develop an inference framework that can estimate these key quantities by combining statistical filtering techniques , an agent-based model , and real-world patient-to-patient contact networks , and use this framework to infer nosocomial transmission and infection importation over an outbreak spanning 6 years in 66 Swedish hospitals . In particular , we identify a small number of patients with disproportionately high risk of colonization . In retrospective control experiments , interventions targeted to these individuals yield a substantial improvement over heuristic strategies informed by number of contacts , length of stay and contact tracing . Antimicrobial resistance is a global concern in healthcare systems due to its substantial morbidity and mortality burden and the lack of effective treatment options ( CDC , 2013a; Magill et al . , 2014; WHO , 2018 ) . Among antibiotic-resistant agents , Methicillin-resistant Staphylococcus aureus ( MRSA ) emerges as one of the most widespread and virulent pathogens ( Grundmann et al . , 2006; Klevens et al . , 2007; Klein et al . , 2007; Jarvis et al . , 2012 ) and has been highlighted as a leading cause of healthcare-associated infections ( HAIs ) by the U . S . Centers for Disease Control and Prevention ( CDC ) ( CDC , 2013b ) . Initially confined to healthcare facilities , MRSA has since become increasingly prevalent in the broader population in both the United States and Europe ( Chambers , 2001; Naimi et al . , 2003; Zetola et al . , 2005; Hetem et al . , 2012; Kouyos et al . , 2013; Tosas Auguet et al . , 2016 ) . This entwined transmission among hospitals and the community has obscured understanding of the dynamics and persistence of MRSA . Further , MRSA can colonize patients without symptoms for years , during which it can be transmitted stealthily ( Cooper et al . , 2004a ) . These epidemiological features have greatly complicated its control and elimination . The prevalence of MRSA has large variations across different countries . In Europe , a general north-south gradient has been observed , with rare incidence in Scandinavian hospitals and much higher occurrence in Mediterranean hospitals ( Stefani and Varaldo , 2003; Tiemersma et al . , 2004; Johnson , 2011 ) . In particular , Sweden remains one of the few countries with a low prevalence of MRSA infection ( Stenhem et al . , 2006 ) . A substantial proportion of MRSA cases in Sweden has been imported from abroad due to traveling and healthcare contacts in foreign countries ( Stenhem et al . , 2010; Larsson et al . , 2014 ) . As a consequence , analyzing MRSA outbreaks in Sweden offers a good opportunity to study the hybrid dynamics of MRSA in hospital settings where both nosocomial transmission and importation occurs . To facilitate better control of MRSA in hospital settings , several critical questions need to be answered . First , what are the relative roles of nosocomial transmission and infection importation from the community ? Public health officials require an accurate assessment of the current force of infection within and into hospitals in order to deploy appropriate containment measures; however , with the increasing prevalence of MRSA in the community , disentangling HAIs from infections imported from the community has become difficult . Second , how many patients are colonized , and who and where are these high-risk individuals ? Effective control would benefit from accurate determination of asymptomatic colonization rates in the general population; failure to estimate and account for colonization may result in long-term control issues ( Cooper et al . , 2004a ) . Although colonized patients can be identified using sequencing methods ( Harris et al . , 2010; Long et al . , 2014 ) , the expense of these assays limits their application , particularly in underdeveloped countries where MRSA has become endemic . In light of this situation , mathematical modeling offers an alternative approach for locating individuals with a high probability of colonization and guiding the targeted deployment of laboratory testing ( Grundmann and Hellriegel , 2006; van Kleef et al . , 2013; Opatowski et al . , 2011 ) . However , this inference problem is again complicated by the unobserved stealth transmission dynamics that occurs in the highly complex time-varying contact networks of the real world ( Donker et al . , 2010; Vanhems et al . , 2013; Jarynowski and Liljeros , 2015; Obadia et al . , 2015a; Obadia et al . , 2015b; Rocha et al . , 2016; Nekkab et al . , 2017; Duval et al . , 2018 ) . To address these issues , here we develop an agent-based network model-Bayesian inference system for estimating unobserved colonization and importation rates from simple incidence records . We use this system to infer the transmission dynamics of the most commonly diagnosed MRSA strain , UK EMRSA-15 ( Grundmann et al . , 2010; Das et al . , 2013 ) , from multiple Swedish hospitals ( Materials and methods ) . Key features estimated include the number of infections acquired in hospital and imported from outside , as well as the locations of individuals with a high colonization probability . Such information is crucial for designing cost-effective control measures ( Cooper et al . , 2003; Cooper et al . , 2004b; Hubben et al . , 2011; Worby et al . , 2013 ) . In retrospective control experiments , decolonization of potentially colonized patients outperforms heuristic intervention strategies based on number of contacts , length of stay and contact tracing . These findings indicate that the model-inference system can inform effective , actionable and cost-effective measures for reducing nosocomial transmission . Observed incidence over a 6-year outbreak period is reported in Figure 1A . The relative date of diagnosis and associated ward of each observed infection were recorded; however , information on type of ward was not provided . Previous studies indicate that heterogeneity in infection risk exists across different types of ward ( Bootsma et al . , 2006 ) . For instance , patients in intensive care units ( ICUs ) typically suffer a higher risk of infection than those in non-ICU settings due to higher patient-healthcare-worker contact rates , high levels of antibiotic use , and high patient vulnerability to infection . In this study dataset , however , we observed no clear clustering of infections in certain wards . In Figure 1B , we use a raster plot to display the distribution of infections in 114 infected wards over time . Infections are distributed without noticeable clustering , presumably due to effective control measures taken in the study hospitals that maintain a low infection rate even in ICU settings ( Tiemersma et al . , 2004 ) . Distributions of some key statistics of the patient flow in infected wards differ from those in uninfected ones ( Figure 1C ) . Infected wards tend to have a higher number of inpatients , a longer average length of stay as well as a larger ward size . Intuitively , the number of MRSA infections in a ward should increase as patient-days within the ward increase . However , the average number of infections is not observed to increase linearly with patient-days , indicating that patient-days per ward alone cannot explain the observed patterns of infection ( Figure 1—figure supplement 1 ) . While these raw features provide a general understanding of MRSA transmission , they cannot be effectively employed to assess infection or colonization risk in a specific ward due to their largely overlapping distributions , which prevent a clear classification of risk . Instead , a quantitative analysis using mathematical modeling is needed . In hospital settings , MRSA transmission between colonized/infected patients and susceptible individuals is primarily mediated indirectly by healthcare workers ( Lowy , 1998; Temime et al . , 2009 ) . As a result , accurate representation of actual contact patterns is crucial for modeling MRSA transmission . Many previous studies have formulated transmission models using ordinary differential equations ( ODEs ) ( Cooper et al . , 2004a; Kajita et al . , 2007; D'Agata et al . , 2009 ) or stochastic processes ( Forrester et al . , 2007; Kypraios et al . , 2010 ) . To account for heterogeneity among different settings , several studies have included multiple facilities in a single-model construct , incorporating prior information on facility type in order to characterize and differentiate transmission dynamics ( Bootsma et al . , 2006; Forrester et al . , 2007 ) . These approaches were then generalized to permit connection among institutions at different scales ( hospitals , nursing homes and long-term healthcare facilities , or multiple wards or units within a facility ) with time-varying contact patterns ( Thomas et al . , 2018 ) . In this work , we model nosocomial MRSA transmission using an individual-level agent-based model ( Macal et al . , 2014; Assab et al . , 2017 ) . One major advantage of using an agent-based approach is that the heterogeneity of contact in different ward types can be accounted for within the model . For instance , even though we were provided no information about ward type , some of the heterogeneity among wards can be represented by the contact pattern specific to each ward , for example a longer length of stay in long-term healthcare units . Other aspects , however , are simplified; for example , due to the observed absence of infection clusters ( Figure 1B ) , we assume a uniform transmission rate across different wards within the model . This assumption is ultimately justified by the good agreement between inferred dynamics and observations ( as described later ) . In the model , transmission occurs on the substrate of a time-varying contact network , which is constructed using the actual hospitalization records from 66 hospitals in Stockholm County , Sweden . In this contact network , nodes represent uniquely labeled patients , connected by undirected links among individuals sharing a ward at a given time . The rationale behind this network construction approach is that , if two patients stay in the same ward simultaneously , the shared healthcare personnel may facilitate transmission between them . The structure of the contact network is relatively stable over time , as indicated by the degree distributions of the weekly aggregated networks ( Figure 1D ) . In particular , the degree distributions can be well fitted by a Weibull distribution , P ( k ) =abkb−1e−akb , where a=6 . 15×10-4 ( 95% CI: 5 . 96×10-4-6 . 34×10-4 ) , b=2 . 13 ( 95% CI: 2 . 12-2 . 14 ) ( R2=0 . 95 ) . The contact network is time-varying and exhibits high spatiotemporal complexity . The daily in-hospital patient number fluctuates between 4000 and 7000 during the study period ( Figure 1—figure supplement 2A–B ) . Patient hospitalization time , readmission time , and patient-to-patient contact time all follow heavy-tailed distributions , spanning several orders of magnitude ( Figure 1—figure supplement 2C ) . Moreover , about 100 connected components coexist in the contact network each week ( Figure 1—figure supplement 2D ) . Connections between different connected components change over time due to the transfer and readmission of patients . 128 , 119 patients were transferred from one ward to another during their stay , and another 280 , 506 patients were readmitted within 1 year of their previous discharge . These patient movements connect healthcare facilities that would otherwise be isolated in the network and facilitate long-range transmission across multiple hospitals . A direct consequence of this patient movement is difficulty tracking the indirect transmission path of MRSA across different hospitals . For instance , a patient located in one hospital can be involved in the transmission occurring in another when he/she moves across multiple facilities . Detailed analyses of the contact network structure and hospitalization traffic can be found in Appendix 1 . Model patients are classified into three categories: susceptible individuals who are free of MRSA ( S ) , colonized individuals who carry the bacteria asymptomatically ( C ) , and confirmed positive patients ( P ) . The model simulates two connected dynamics: nosocomial transmission and importation from the community . Here , the community is broadly defined as all locations outside the study hospitals , and may include households and healthcare facilities not covered in the study . Within hospitals , transitions between states ( S , C , P ) are governed by parameters that help define either interaction dynamics or the progression of infection . Specifically , a susceptible individual staying in a ward with a colonized person can become MRSA colonized with transmission probability β per day . In our model , we assume that patients within a ward have the same rate of contact with each other , presumably mediated by the shared healthcare workers in a ward . The transmission process is density-dependent , as the force of infection in a ward increases with the number of colonized patients within the ward ( Begon et al . , 2002 ) . Upon colonization , asymptomatic persons can return to the susceptible state at a spontaneous decolonization rate α , or they can test positive with an infection progression rate p . We assume infected patients will receive treatment , no longer spread bacteria , and return to state S with a recovery rate μ . Treatment is assumed to continue until infected patients are clear of MRSA . Given the exponential decay of infection probability , the characteristic treatment period is 1/μ days . Note that colonization only occurs between individuals connected by a link in the contact network , whereas decolonization , infection and recovery progress spontaneously , independent of the contact network . Outside the study hospitals , the transmission process is not explicitly simulated; instead , two additional parameters are introduced to represent transmission intensity . For patients who appear for the first time in hospital , we assume they belong to states C and P with probability C0 and I0 , respectively . As importation rates of colonized and infected patients depend on the time-varying MRSA prevalence outside hospitals , we assume the parameters C0 and I0 are time-dependent . Once patients appear in the contact network , the evolution of their states follows the dynamics as defined above . After discharge , we continue tracking the progression of colonized individuals; however , transmission outside the study hospitals is not represented . The flow of individuals between categories is illustrated schematically in Figure 1E . For a realistic scenario , disease-related model parameters may differ from person to person . To account for this variability during implementation , parameters , for example α , p and μ , for each individual are randomly drawn from uniform ranges obtained from prior literature ( Table 1 ) . The parameter ranges are enlarged slightly to cover the values reported in these works . Our main objective is to infer the three most important parameters governing transmission dynamics: the transmission rate β , the infection importation rate I0 and the colonization importation rate C0 . To infer epidemiological parameters in an agent-based model , we adapt an iterated filtering ( IF ) algorithm ( Ionides et al . , 2006; King et al . , 2008; Ionides et al . , 2011 ) . IF can be used to infer the maximum likelihood estimates ( MLEs ) of parameters in epidemic models and has been successfully applied to infectious diseases such as cholera ( King et al . , 2008 ) and measles ( He et al . , 2010 ) . Initially developed for ODE models , IF has subsequently been generalized for other model forms ( e . g . stochastic models ) using the plug-and-play approach ( He et al . , 2010 ) . Here , we adapt IF for agent-based models , leveraging an equation-free approach ( Kevrekidis et al . , 2003 ) that allows for mapping between the system-level observations ( e . g . weekly incidence ) used for the IF and the individual-level states evolved in the agent-based model ( Appendix 1 ) . In applying the IF , we perform multiple iterations using an efficient Bayesian filtering algorithm – the Ensemble Adjustment Kalman Filter ( EAKF ) ( Anderson , 2001 ) , which has been widely used in infectious disease forecast and inference ( Shaman and Karspeck , 2012; Yang et al . , 2015; Pei and Shaman , 2017; Pei et al . , 2018a; Kandula et al . , 2018 ) . Details of the IF implementation can be found in Materials and methods . Before applying the inference system to real-world data , we first need to validate its effectiveness . For the real-world data the inference targets are unobserved , so instead we test the inference system using model-generated synthetic outbreaks for which we know the exact values of the parameter . Although actual MRSA transmission dynamics cannot be fully described by the simplified agent-based model , performing synthetic tests provides validation that the inference system works if the transmission process generally follows the model-specified dynamics . To generate synthetic outbreak observations , we used the agent-based model to simulate weekly incidence during a one-year period ( 52 weeks ) , and then imposed noise to produce the observations used in inference ( See details in Appendix 1 ) . We ran 20 iterations of the EAKF within the IF framework . In Figure 2A , we display the inference results for the three parameters β , I0 and C0 at different iterations in one realization of the IF algorithm . The blue horizontal lines mark the target values used to generate the outbreak . The orange boxes show the distribution of posterior parameters ( 300 ensemble members ) after each iteration . The IF algorithm returns the stabilized ensemble mean as the MLEs of parameters . As a result of the stochastic nature of model dynamics and initialization of the inference algorithm , different runs of the IF algorithm usually return slightly different MLEs . To obtain the credible intervals ( CIs ) for the MLEs , we repeated the inference for 100 times ( see Materials and methods ) . The inferred mean values and 95% CIs for the parameters β , I0 and C0 are 9 . 00 , [8 . 07 , 9 . 68]×10-3 , 1 . 91 , [1 . 38 , 2 . 54]×10-3 and 7 . 18 , [5 . 84 , 8 . 70]×10-2 , with the actual values β=9×10-3 , I0=2×10-3 and C0=7 . 5×10-2 . The inference system thus accurately estimates β and I0 from noisy observations , and slightly underestimates C0 . In its implementation , the performance of the inference system depends on the sensitivity of the observations to each parameter . In the agent-based model used here , observed incidence is less sensitive to C0 due to the long period of colonization . As a consequence , estimates of C0 do not always exactly match the actual target and are here biased low . Nevertheless , this slight underestimation does not significantly affect the inferred dynamics . To demonstrate this insensitivity , we ran 1000 simulations using the inferred mean parameters and obtained distributions of weekly incidence from the stochastic agent-based model . The distributions of weekly incidence ( blue boxes ) are compared with the observed cases ( red crosses ) in Figure 2B . We also evaluated the agreement between the observed and simulated incidences in Figure 2B ( Figure 2—figure supplement 1; Analysis details are explained in Appendix 1 ) . The inferred dynamics fit the observed incidence well . The Matlab code for synthetic test on an example network is uploaded as an additional file . We repeated the above analysis for the colonized population ( Figure 2C ) and found that the numbers of unobserved colonized patients can also be well estimated by the inference system . Moreover , the inference system can distinguish the number of infections transmitted in hospital and imported from outside the study hospitals ( Figure 2D–E ) . More tests for alternate synthetic outbreaks and different observation frequencies were performed and are presented in Figure 2—figure supplements 2–5 . We next applied the inference system to the UK EMRSA-15 incidence data binned every 4 weeks . Because the UK EMRSA-15 transmission parameters are unlikely to remain constant over the entire 6-year outbreak cycle , we inferred model parameters year by year ( 52 weeks ) . Beginning with the first year , we ran the IF inference sequentially through each year . Between 2 consecutive years , the inferred results from the previous year were used to initialize the inference system in the next ( see Figure 3—figure supplement 1 ) . In Figure 3A , we present the distributions of the key parameters β , I0 and C0 for each year , generated from 100 independent realizations of the IF algorithm . The parameter estimates together with the associated 95% CIs are reported in Table 2 . All parameter values increased in the first 3 or 4 years , and then gradually decreased thereafter . The inferred parameters can be plugged back into the model to run simulations and obtain information addressing our questions of interest ( see Video 1 for an example ) . For instance , we performed 1000 model simulations using the inferred mean parameter values , and generated distributions of incidence from the stochastic agent-based model . These distributions are compared to observations in Figure 3B . All observations fall within the whisker range of Tukey boxplots ( see more analyses in Figure 3—figure supplement 2 ) . To further explore whether some of the key observed statistics can be reproduced using the inferred parameters , we display the distribution of the number of infected wards in Figure 3C . The observed number lies at the peak of the simulated distribution ( vertical dash line ) . The spatial distribution of infections among different wards can be characterized by the distribution of wards with a certain number of infections in an outbreak . In Figure 3D , we compare this distribution obtained from 1000 simulations with what we observed in the data ( red diamonds ) : the observed distribution agrees well with the simulated distributions . This close matching indicates that the model structure and inferred parameters can reliably reproduce the observed outbreak pattern in both space and time ( see also Figure 3—figure supplement 2 ) . In addition to generating a good model fit , the inference system also discriminates the burdens of nosocomial transmission and infection importation . Nosocomial and imported infections are distinguished by the location of MRSA colonization: if patients acquire MRSA in hospital , they are classified as nosocomial transmission cases; otherwise they are imported cases . Figure 3E compares the distributions of both types of infections generated from 1000 simulations: a substantial number of infections are inferred as importations . In clinical practice , the number of days between hospital admission and infection is usually used to distinguish hospital-acquired from community-acquired infections , typically with 48 hr used as the threshold . We performed this classification and compared the findings with our inference result . As shown in Figure 3—figure supplement 3 , the number of imported and nosocomial cases obtained from inference generally matches the classification result using days from admission to infection . Our findings indicate that , at its onset , during the first year of the outbreak , UK EMRSA-15 gradually invaded the hospital system from the community . Only sporadic nosocomial transmission occurred . With the accumulation of infected and colonized patients in the hospitals , a rise in nosocomial transmission occurred , reflected by an increase of the transmission rate β during the third year ( Figure 3A ) . Concurrently , both the infection and colonization importation rates , I0 and C0 , also experienced growth . This simultaneous rise may have been caused by household transmission initiated by asymptomatically colonized patients discharged from hospitals . After this growth phase , both transmission and importation rates were suppressed . Finally , the UK EMRSA-15 outbreak appears near eliminated in Swedish hospitals , represented by the inferred low values of all parameters . However , if control measures in hospital were to be relaxed , the colonized patients might spark another outbreak due to the lengthy colonization period , which highlights the need for asymptomatic colonization control in order to effect MRSA elimination ( Cooper et al . , 2004a ) . Asymptomatic colonization is a major issue hindering the control and elimination of MRSA in hospitals ( Cooper et al . , 2004a ) . Screening can identify colonized patients and evaluate the general colonization burden; however , it is an inefficient and costly measure that wastes resources that otherwise could be used to solve more urgent problems . As shown above , given the heterogeneity of contact among patients , levels of exposure to the hazard of colonization differ substantially . As a result , more efficient intervention strategies can be designed that leverage this individual-level heterogeneity . In Figure 4A , we display the inferred distribution of colonized patients in the Swedish hospitals over time . Colonized patient numbers peak in the middle of the record and decline thereafter . To determine who and where these high-risk individuals reside within the network , we can use the agent-based model to quantify colonization risk at the individual level . The distribution of individual colonization probability at T = 40 ( week 160 ) , generated from 104 simulations using inferred parameters , is displayed in Figure 4B . A clear heavy-tailed power-law distribution y∝x-2 . 13 emerges in which the colonization probability spans several orders of magnitude ( see Figure 4—figure supplement 1 and Appendix 1 for a rigorous statistical analysis of this distribution ) ( Clauset et al . , 2009; Muchnik et al . , 2013 ) . The complex spatiotemporal interaction patterns within the network give rise to a small number of patients with a disproportionately high risk of colonization . To examine how these individuals distribute among hospitals , we visualize the colonization probability in Figure 4C . High-risk patients tend to appear in densely connected clusters . Cost-effective interventions can be practiced by the targeted screen and decolonization of identified high-risk patients . In order to evaluate the effectiveness of such interventions , we performed a retrospective control experiment . Specifically , we used the inferred parameters in Figure 3A to run the model for 6 years to reproduce the outbreak . Every 4 weeks , we used currently available information ( as would be available in real time ) to estimate patient colonization probabilities ( see details in Materials and methods ) . The colonization probabilities estimated in real time are highly correlated with the results obtained using information from the whole course of the epidemic , shown in Figure 4C . During the model integration , every 4 weeks , we selected patients with an estimated colonization probability higher than a certain threshold for screening . If positive , these inpatients were decolonized . To assess the impact of decolonization success rate on intervention impact , two efficiencies , 100% and 75% , were tested , and we repeated the experiment 1000 times . The findings show that the proposed intervention strategy can avert considerable numbers of colonization and infection ( Figure 5A–B ) . Decreasing the decolonization threshold leads to a larger screened population ( as shown in the inset of Figure 5B ) , and thus reduces colonization and infection further . However , the marginal benefit becomes negligible below a certain threshold value , as the remaining colonized and infected patients are possibly caused by importation , which cannot be directly controlled by inpatient intervention . The decolonization success rate also plays an important role , as indicated by the increased colonization and infection for the lower success rate . The advantage of the proposed inference-based intervention can be better appreciated by examining its additional benefit over other heuristic control measures . Here , we compare the performance of the inference-based intervention with three alternative screening strategies informed by patient number of contacts , length of stay and contact tracing . For the former two , at each month , we ranked patients by their current total number of contacts ( i . e . cumulative number of connections in the time-varying network since admission ) or length of stay in a descending order , and created the screening list using the top-ranked patients . By varying the fraction of patients selected from the ranking ( from 0% to 5% ) , we can inspect the control results for different numbers of screened patients . For contact tracing , upon each observation of infection , we tracked patients who stayed in the same ward with an infected individual within a certain time window prior to the infection , and screened those possibly colonized patients in hospitals . Tracing time windows ranging from 1 day to 14 days were tested . The number of screened patients does not increase significantly with tracing times longer than 14 days . Note that , screening and decolonization are performed only within hospitals . If patients listed for screening have already discharged before the diagnosis of infection , they are screened upon their next re-admission . In Figure 5C–D , the average numbers of colonized and infected patients are compared based on the number of screened patients . Heuristic control measures relying on the number of contacts , length of stay and contact tracing all limit MRSA transmission; however , a substantial additional reduction in both colonization and infection can be achieved through inference-based intervention . On average , inference-based screening of approximately 0 . 89% ( 6 , 617/743 , 599 ) of all patients can avert up to 38% ( 121/315 ) of infections and 9% ( 1 , 610/17 , 810 ) of colonizations . In comparison , the other three methods given similar numbers of screened patients only reduced infections and colonizations by 21% and 4% ( number of contacts ) , 27% and 6% ( length of stay ) , and 28% and 6% ( contact tracing ) , respectively . The colonization probability obtained from inference quantifies individual systemic risk given the general situation of transmission , regardless of the specific location of undetected colonization . In contrast , screening based on contact tracing identifies colonized individuals related to observed infections; however , with an unknown amount of imported colonization , this approach may overlook a considerable number of colonized patients , who can sustain subsequent transmission . As a result , the inference-based intervention can identify and treat the pivotal individuals , or superspreaders ( Pei and Makse , 2013; Pei et al . , 2014; Pei et al . , 2017; Pei et al . , 2018b; Teng et al . , 2016 ) , who may otherwise transmit MRSA asymptomatically in the first place . This preventive approach is more effective than contact tracing in the presence of frequent importation , as it disrupts probable transmission pathways . In real-world hospital settings , the proposed inference-based intervention could be implemented and evaluated in real time: it only requires hospitalization records and ward information . In this work , we have developed an agent-based model-inference framework that can estimate nosocomial MRSA transmission dynamics in the presence of importation . Further , we have shown that these inferred dynamics can be used to quantify patient colonization risk and guide more effective interventions . The transmission dynamics generated using the agent-based model are intrinsically stochastic , that is , the observed record of UK EMRSA-15 infections is just one realization among an ensemble of all possible outcomes of an underlying highly stochastic process . In order to evaluate the general risk of MRSA transmission , key epidemiological parameters were inferred from the single observed realization . Previous studies have developed methods to infer transmission risk factors and reconstruct transmission paths using individual-level infection data for diseases such as H1N1 and MERS-CoV ( Cauchemez et al . , 2011; Cauchemez and Ferguson , 2012; Cauchemez et al . , 2016 ) . In particular , Bayesian data augmentation approaches have been applied to MRSA models ( Forrester et al . , 2007; Kypraios et al . , 2010 ) ; however , these approaches are not readily applicable to our dataset . The data assimilation scheme we developed here enables estimation of epidemiological parameters and key transmission information using aggregated incidence data . As demonstrated in the retrospective control experiment , assessment of individual colonization risk using aggregated data can be quite useful for preventing future MRSA transmission , especially when stealth importations are frequent . In this study , we omitted representation of heterogeneity across different wards . This simplification is valid for the study Swedish hospitals , as we observed no infection clusters and the model reproduced key statistics of observations well . However , in other settings , clustering analysis and ward information may be necessary before the application of the inference system . Should certain wards suffer a much higher rate of infection , a separate suite of parameters can be defined and inferred for these wards , using priors that better represent this more intense transmission . We also only considered transmission among patients staying in the same ward . In the future , more contact information such as healthcare workers shared by a group of patients could be incorporated into the contact network . In addition , as the community defined in the model may include non-sampled hospitals , inferred community risk may have been overestimated as it also included contributions from those healthcare facilities outside the network . We note that the reported inference results are obtained using only hospitalization records and UK EMRSA-15 case numbers . Should more data ( e . g . surgery or treatment records ) become available , this additional information could be incorporated into the agent-based model and used to refine the present results . Our model-inference framework provides a foundational platform for flexible simulation and inference of antibiotic resistant pathogens . In this study , we applied this system to Swedish hospitals with low MRSA prevalence . However , in the future , it could be used to provide actionable information for disease control in less developed settings where MRSA is endemic . In a highly interconnected area , transmission of antibiotic resistant pathogens from endemic regions to epidemic-free hospitals is more likely . This risk calls for containment measures in the general population and collaborative control efforts among multiple healthcare facilities ( Smith et al . , 2005; D'Agata et al . , 2009; Ciccolini et al . , 2014; Slayton et al . , 2015 ) . The dataset contains admission and discharge records of 743 , 599 distinct patients from 66 hospitals ( 271 clinics , 1041 wards ) in Stockholm County , Sweden ( Jarynowski and Liljeros , 2015; Rocha et al . , 2016 ) , spanning over 3500 continuous days during the 2000s . The exact dates and ward types are confidential for the protection of patient privacy . In total , 2 , 041 , 531 admission records were collected . The hospitalization dataset is quite comprehensive as the patients constitute over one third of the total 2 . 2 million population of Stockholm County . In addition , the dataset also contains individual diagnostic records of MRSA , which provide relative date of diagnosis and the strain of MRSA . Diagnosis was performed on patients with symptomatic infections as well as asymptomatic patients in contact with positive cases . A total of 991 positive cases from 172 different strains were confirmed , and the most prevalent strain was UK EMRSA-15 ( 289 cases ) . UK EMRSA-15 is present in 16 countries worldwide ( Grundmann et al . , 2010; Das et al . , 2013 ) . Here , we focus on this specific strain . Although the dataset spans over 3500 days ( nearly 10 years ) , we limit our study to a 6-year ( 300-week ) period with reported UK EMRSA-15 incidence . We display time series of 4-week incidence and cumulative incidence for UK EMRSA-15 in Figure 1A . We infer system epidemiological parameters using an iterated filtering ( IF ) algorithm ( Ionides et al . , 2006; King et al . , 2008; Ionides et al . , 2011 ) . This algorithm has been coupled with ODE models and used to infer latent variables associated with the transmission of cholera ( King et al . , 2008 ) and measles ( He et al . , 2010 ) . The IF framework is designed as follows: an ensemble of system states , which represent the distribution of parameters , are repeatedly adjusted using filtering techniques in a series of iterations , during which the variance of the parameters is gradually tuned down . In the process , the distribution of parameters is iteratively optimized per observations and narrowed down to values that achieve maximum likelihood . This approach is based on an analytical proof that guarantees its convergence under mild assumptions ( Ionides et al . , 2006 ) . In its original implementation , the data assimilation method used in IF is sequential Monte Carlo , or particle filtering ( Arulampalam et al . , 2002 ) . Here , due to the high computational cost of the agent-based model , we use a different efficient data assimilation algorithm - the Ensemble Adjustment Kalman Filter ( EAKF ) ( Anderson , 2001 ) . Unlike particle filtering , which requires a large ensemble size ( usually of the order O ( 104 ) or higher ) ( Snyder et al . , 2008 ) , the EAKF can generate results similar in performance using only hundreds of ensemble members ( Shaman and Karspeck , 2012 ) . Originally developed for use in weather prediction , the EAKF assumes a Gaussian distribution of both the prior and likelihood , and adjusts the prior distribution to a posterior using Bayes rule in a deterministic way such that the first two moments ( mean and variance ) of an observed variable are adjusted while higher moments remain unchanged during the update ( Anderson , 2001 ) . In epidemiological studies , the EAKF has been widely used for parameter inference and forecast of infectious diseases ( Shaman and Karspeck , 2012; Yang et al . , 2015; Pei and Shaman , 2017; Pei et al . , 2018a; Kandula et al . , 2018 ) . The implementation details of the EAKF are introduced in Appendix 1 . In this study , we focus on the inference of three transmission-related parameters: the nosocomial transmission rate β , the infection importation rate I0 and the colonization importation rate C0 . The initial prior ranges for these parameters are reported in Table 1 . Other disease-related parameters , for example the spontaneous decolonization rate α , the infection progress rate p , and the recovery rate μ , are drawn uniformly from ranges obtained from previous studies for each individual in the agent-based model ( see Table 1 ) . Should more specific information about these parameters become available , it may be possible in the future to better constrain the model with their incorporation into the system . In synthetic testing of the IF-EAKF algorithm , we use weekly incidence as observations . Given the parameter vector , 𝐳= ( β , I0 , C0 ) T , the IF-EAKF algorithm proceeds per the pseudo-code in Algorithm 1 . During the EAKF update , only the parameters β , I0 and C0 were adjusted; the microscopic state ( S , C or P ) in each ensemble member was set as the state at the end of previous time step and was not adjusted . Detailed explanation of the IF-EAKF system is provided in Appendix 1 . In each iteration of the IF , the standard deviation of each parameter is shrunk by a factor a∈ ( 0 , 1 ) ( or equivalently , the variance is discounted by a factor of a2 ) . In practice , the discount factor a can range between 0 . 9 and 0 . 99 ( Ionides et al . , 2006 ) . If a is too small , the algorithm may ‘quench’ too fast and fail to find the MLE; if it is too close to 1 , the algorithm may not converge in a reasonable time interval . We stop the IF algorithm once the estimates of the ensemble mean stabilize . The number of iterations required for this convergence was determined by inspecting the evolution of posterior parameter distributions , as in Figure 2A . Note that once the ensemble mean stabilizes , increasing the iteration time will not affect the MLE , although it can lead to a further narrowing of the ensemble distribution . Algorithm 1 only returns the MLEs for the parameters; however , it is also desirable to obtain CIs for those MLEs . For deterministic ODE models , Ionides et al . used ‘sliced likelihood’ to numerically estimate the Fisher information and standard errors ( SEs ) of MLEs ( Ionides et al . , 2006 ) . Here , for a highly stochastic system , evaluating the Fisher information numerically is challenging . As a result , we took another approach by running multiple realizations of the IF algorithm . In different runs , the MLEs are slightly different due to stochasticity in the agent-based model and in the initialization of the inference algorithm . In this work , we ran 100 independent realizations to generate the average MLEs of inferred parameters and their corresponding 95% CIs . Results from synthetic tests indicate that this approach is effective in calculating MLEs and quantifying their uncertainties . An alternative method to infer posterior parameters is to use Approximate Bayesian computation ( ABC ) ( Beaumont et al . , 2002 ) . ABC-based methods employ numerical simulations to approximate the likelihood function , in which the simulated samples are compared with the observed data . In a typical ABC rejection algorithm , large numbers of parameters are sampled from the prior distribution . For each set of parameters , the distance between simulated samples ( generated using the parameters ) and observed data is calculated . Parameters resulting in a distance larger than a certain tolerance are rejected , and the retained parameters form the posterior distribution . ABC methods can fully explore the likelihood landscape in parameter space . However , it requires large numbers of simulations , which may be prohibitive for the large-scale agent-based models considered here . In addition , a good choice of the tolerance in the rejection algorithm is needed . The IF algorithm , instead , is applicable to computationally expensive agent-based models , but may become trapped in the local optimum of the posterior distribution . In practice , this problem can be alleviated by exploring a larger prior parameter space and setting a slower quenching speed , that is , a smaller discount factor a . We report the inferred parameters and their corresponding 95% CIs for the synthetic tests in Table 3 . The actual parameters used to generate the synthetic outbreaks are also reported . Results are obtained from 100 independent realizations of the IF algorithm . To guarantee a fair comparison between the inference-based intervention and other heuristic strategies , we estimated the colonization probability using only real-time information available before control measures are effected . For instance , to estimate the colonization probability at the fifth month in the third year , we first infer the model parameters for the first 2 years , where we have data from the whole year , and then use the partial observation in the remaining 5 months to infer the model parameters for the third year . The inferred parameters are then used to generate 1000 synthetic outbreaks from the beginning , and the current colonization probability for each individual is calculated from these simulations . In the inset of Figure 5A , we show that the colonization probability estimated in real time is highly correlated with that obtained using information from the entire outbreak record . In practice , every 4 weeks , the estimated colonization probability and the decolonization list were updated . The inference-based intervention only uses information available at the time control measures are effected . As a consequence , it is a practical method that can be implemented in real time .	Antibiotic-resistant bacteria like the Methicillin-resistant Staphylococcus aureus ( MRSA ) can live in people for many years without making them sick . During this time , the bacteria can spread to others who come in contact with the MRSA-infected person . The number of people with stealth MRSA infections living in the community has been increasing . As a result , hospitals may not only be dealing with MRSA infections that originated onsite , but also cases imported from the community . That makes tracking and controlling MRSA infections in hospitals difficult . Now , Pei et al . show that computer modeling can help identify the role MRSA infections from the community play in hospital outbreaks and test ways to control them . In the experiments , data from an MRSA outbreak that occurred at 66 Swedish hospitals over 6 years were analyzed using statistical methods and computer modeling . This helped to identify patients who were likely colonized with MRSA within the hospital and those who had acquired it in the community . Next , Pei et al . used computer modeling to test what would have happened if these high-risk individuals had received interventions to prevent them from spreading MRSA in the hospital . This showed that targeting individuals at high-risk of a MRSA infection could reduce the spread of MRSA in the hospital . The computer models developed by Pei et al . may help researchers , clinicians and public health officials working to control the spread of antibiotic resistant bacteria . The model can improve our understanding of how antibiotic resistant bacteria spread in healthcare facilities and may enable the development of more effective strategies to control these pathogens . Infection-control strategies created with this system must first be tested in isolated , real-world settings to verify they work before they can be deployed broadly .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "epidemiology", "and", "global", "health", "computational", "and", "systems", "biology" ]	2018	Inference and control of the nosocomial transmission of methicillin-resistant Staphylococcus aureus
Hepatic platelet accumulation contributes to acetaminophen ( APAP ) -induced liver injury ( AILI ) . However , little is known about the molecular pathways involved in platelet recruitment to the liver and whether targeting such pathways could attenuate AILI . Mice were fasted overnight before intraperitoneally ( i . p . ) injected with APAP at a dose of 210 mg/kg for male mice and 325 mg/kg for female mice . Platelets adherent to Kupffer cells were determined in both mice and patients overdosed with APAP . The impact of α-chitinase 3-like-1 ( α-Chi3l1 ) on alleviation of AILI was determined in a therapeutic setting , and liver injury was analyzed . The present study unveiled a critical role of Chi3l1 in hepatic platelet recruitment during AILI . Increased Chi3l1 and platelets in the liver were observed in patients and mice overdosed with APAP . Compared to wild-type ( WT ) mice , Chil1-/- mice developed attenuated AILI with markedly reduced hepatic platelet accumulation . Mechanistic studies revealed that Chi3l1 signaled through CD44 on macrophages to induce podoplanin expression , which mediated platelet recruitment through C-type lectin-like receptor 2 . Moreover , APAP treatment of Cd44-/- mice resulted in much lower numbers of hepatic platelets and liver injury than WT mice , a phenotype similar to that in Chil1-/- mice . Recombinant Chi3l1 could restore hepatic platelet accumulation and AILI in Chil1-/- mice , but not in Cd44-/- mice . Importantly , we generated anti-Chi3l1 monoclonal antibodies and demonstrated that they could effectively inhibit hepatic platelet accumulation and AILI . We uncovered the Chi3l1/CD44 axis as a critical pathway mediating APAP-induced hepatic platelet recruitment and tissue injury . We demonstrated the feasibility and potential of targeting Chi3l1 to treat AILI . ZS received funding from NSFC ( 32071129 ) . FWL received funding from NIH ( GM123261 ) . ALFSG received funding from NIDDK ( DK 058369 ) . ZA received funding from CPRIT ( RP150551 and RP190561 ) and the Welch Foundation ( AU-0042–20030616 ) . CJ received funding from NIH ( DK122708 , DK109574 , DK121330 , and DK122796 ) and support from a University of Texas System Translational STARs award . Portions of this work were supported with resources and the use of facilities of the Michael E . DeBakey VA Medical Center and funding from Department of Veterans Affairs I01 BX002551 ( Equipment , Personnel , Supplies ) . The contents do not represent the views of the US Department of Veterans Affairs or the US Government . Acute liver failure ( ALF ) is a life-threatening condition of massive hepatocyte injury and severe liver dysfunction that can result in multi-organ failure and death ( Bernal and Wendon , 2013 ) . Acetaminophen ( APAP ) overdose is the leading cause of ALF in Europe and North America and responsible for more cases of ALF than all other aetiologies combined ( Bernal and Wendon , 2013; Jaeschke , 2015 ) . It is estimated that each week , more than 50 million Americans use products containing APAP and approximately 30 , 000 patients are admitted to intensive care units every year due to APAP-induced liver injury ( AILI ) ( Bernal and Wendon , 2013; Blieden et al . , 2014 ) . Although N-acetylcysteine ( NAC ) can prevent liver injury if given in time , there are still 30% of patients who do not respond to NAC ( Fisher and Curry , 2019 ) . Thus , identification of novel therapeutic targets and strategies is imperative . APAP is metabolized predominantly by cytochrome P450 2E1 ( CYP2E1 ) to a reactive toxic metabolite , N-acetyl-p-benzoquinone imine ( NAPQI ) . NAPQI causes mitochondrial dysfunction , lipid peroxidation , and eventually cell death ( Hinson et al . , 2004 ) . The initial direct toxicity of APAP triggers the cascades of coagulation and inflammation , contributing to the progression and exacerbation of AILI ( Hinson et al . , 2004 ) . In patients with APAP overdose , the clinical observations of thrombocytopenia , reduced plasma fibrinogen levels , elevated thrombin-antithrombin , and increased levels of pro-coagulation microparticles strongly suggest concurrent coagulopathy ( Stravitz et al . , 2013; Stravitz et al . , 2016 ) . Similarly , APAP challenge in mice causes a rapid activation of the coagulation cascade and significant deposition of fibrin ( ogen ) in the liver ( Groeneveld et al . , 2020; Sullivan et al . , 2012; Sullivan et al . , 2013 ) . With regard to the role of platelets in AILI , it is reported that in mice APAP-induced thrombocytopenia correlates with the accumulation of platelets in the liver and that platelet depletion significantly attenuates AILI ( Miyakawa et al . , 2015 ) . Two recent studies also demonstrate that persistent platelet accumulation in the liver delays tissue repair after AILI in mice ( Chauhan et al . , 2020; Groeneveld et al . , 2020 ) . These findings strongly indicate that hepatic platelet accumulation is a key mechanism contributing to AILI . However , little is known about the underlying molecular mechanism of APAP-induced hepatic platelet accumulation and whether targeting this process could attenuate AILI . Chitinase 3-like-1 ( Chi3l1 ) ( YKL-40 in humans ) is a chitinase-like soluble protein without chitinase activities ( Lee et al . , 2011 ) . It is produced by multiple cell types , including macrophages , neutrophils , fibroblasts , synovial cells , endothelial cells , and tumor cells ( Hakala et al . , 1993; Kawada et al . , 2007 ) . Chi3l1 has been implicated in multiple biological processes including apoptosis , inflammation , oxidative stress , infection , and tumor metastasis ( Lee et al . , 2009 ) . Elevated serum levels of Chi3l1 have been observed in various liver diseases , such as hepatic fibrosis , non-alcoholic fatty liver , alcoholic liver disease , and hepatocellular carcinoma ( Kumagai et al . , 2016; Lee et al . , 2011; Nøjgaard et al . , 2003; Wang et al . , 2020 ) . However , the biological function of Chi3l1 in liver disease is not clear . Our previous study revealed an important role of Chi3l1 in promoting intrahepatic coagulation in concanavalin A-induced hepatitis ( Shan et al . , 2018 ) . Given the importance of intrahepatic coagulation in the mechanism of AILI , we wondered whether Chi3l1 is involved in platelets accumulation during AILI . In the current study , we observed elevated levels of Chi3l1 in patients with APAP-induced ALF and in mice challenged with APAP overdose . Our data demonstrated a central role of Chi3l1 in APAP-induced hepatic platelet recruitment through CD44 . Importantly , we found that targeting Chi3l1 by monoclonal antibodies could effectively inhibit platelet accumulation in the liver and markedly attenuate AILI . C57BL/6J ( RRID:MGI:3849035 ) and Cd44-/- mice ( RRID:MGI:4941902 ) were purchased from the Jackson Laboratory . Chil1-/- mice were provided by Dr Jack Elias ( Brown University , Providence , RI , RRID:MGI:3846223 ) . All mouse colonies were maintained at the animal core facility of University of Texas Health Science Center ( UTHealth ) . C57BL/6J , not C57BL/6N , was used as wild-type ( WT ) control because both Chil1-/- and Cd44-/- mice are on the C57BL/6J background , determined by polymerase chain reaction ( PCR ) ( data not shown ) . Animal studies described have been approved by the UTHealth Institutional Animal Care and Use Committee ( IACUC ) . For APAP treatment , mice ( 8–12 weeks of age ) were fasted overnight ( 5:00 p . m . to 9:00 a . m . ) before intraperitoneally ( i . p . ) injected with APAP ( Sigma , A7085 ) at a dose of 210 mg/kg for male mice and 325 mg/kg for female mice , as female mice are less susceptible to AILI ( Munoz and Fearon , 1984 ) . Male mice have been the choice in the vast majority of the studies of AILI reported in the literature ( Ju et al . , 2002; Sullivan et al . , 2012 ) . Therefore , we used male mice in the majority of the experiments presented . In some experiments , APAP-treated mice were immediately injected i . p . with either PBS ( 100 μl ) or recombinant mouse Chi3l1 ( rmChi3l1 , 500 ng/mouse in 100 μl , Sino Biological 50929-M08H ) . Livers were harvested at time points indicated in the figure legends and immunofluorescence ( IF ) staining was performed using frozen sections to detect Mɸs and platelets using anti-F4/80 and anti-CD41 antibodies , respectively . Liver paraffin sections and sera were harvested at time points indicated in the figure legends . H&E staining and ALT measurement to examine liver injury were performed using a diagnostic assay kit ( Teco Diagnostics , Anaheim , CA ) . Mice were intravenously ( i . v . ) injected with Ctrl IgG ( Bioxcell InvivoMab , BE0087 , 100 μg/mouse , RRID:AB_1107782 ) or anti-podoplanin antibody ( Bioxcell InvivoMab , BE0236 , 100 μg/mouse , RRID:AB_2687718 ) in Chil1-/- reconstituted with rmChi3l1 at 16 hr prior to APAP treatment . WT mice were i . v . injected with Ctrl IgG ( BD Pharmingen , 553922 , 2 mg/kg , RRID:AB_479672 ) or CD41 antibody ( BD Pharmingen , 553847 , 2 mg/kg , RRID:AB_395084 ) to deplete platelets at 12 hr prior to APAP treatment . WT mice were i . v . injected with empty liposomes ( PBS , 100 μl/mouse ) or clodronate ( CLDN ) -containing liposomes ( 100 μl/mouse ) to deplete Kupffer cells ( KCs ) at either 9 or 40 hr prior to APAP treatment . CLDN-containing liposomes were generated as previously described ( Ju et al . , 2002 ) . To examine the therapeutic potential of anti-mouse Chi3l1 monoclonal antibodies ( mAbs ) , WT mice were injected ( i . p . ) with either Ctrl IgG or anti-mouse Chi3l1 antibody clones 3 hr after APAP administration . To examine the therapeutic potential of anti-human Chi3l1 mAbs , Chil1-/- mice treated with APAP were immediately injected ( i . p . ) with either PBS ( 100 μl ) or recombinant human Chi3l1 ( rhChi3l1 , 1 μg/mouse in 100 μl , Sino Biological 11227-H08H ) . After 3 hr , these mice were divided into two groups injected ( i . p . ) with either Ctrl IgG or anti-human Chi3l1 mAbs . The binding affinity between Fc-CD44 and His-Chi3l1 was measured using the Octet system eight-channel Red96 ( Menlo Park ) . Protein A biosensors and kinetics buffer were purchased from Pall Life Sciences ( Menlo Park ) . Fc-CD44 protein was immobilized onto protein A biosensors and incubated with varying concentrations of recombinant His-Chi3l1 in solution ( 1000–1 . 4 nM ) . Binding kinetic constants were determined using 1:1 fitting model with ForteBio’s data analysis software 7 . 0 , and the KD was calculated using the ratio Kdis/Kon ( the highest four concentrations were used to calculate the KD ) . H&E staining and immunohistochemistry ( IHC ) were performed on paraffin sections using the following antibodies: anti-human CD41 ( Proteintech , 24552–1-AP , 1:200 , RRID:AB_2879604 ) , anti-human CD68 ( Thermo Fisher , MA5-13324 , 1:100 , RRID:AB_10987212 ) , anti-human Chi3l1 ( Proteintech , 12036–1-AP , 1:100 , RRID:AB_2877819 ) . IF staining was performed on frozen sections using the following antibodies: anti-mouse CD41 ( BD Bioscience , Clone MWReg 30 , RRID:AB_395084 ) , mouse F4/80 ( Biolegend , 123122 , 1:100 , RRID:AB_893480 ) , anti-CD44 ( abcam , clone KM81 , ab112178 , 1:200 , RRID:AB_10864553 ) , anti-Chi3l1 ( Proteintech , 12036–1-AP , 1:100 , RRID:AB_2877819 ) , anti-podoplanin ( Novus , NB600-1015 , 1:100 , RRID:AB_2161937 ) , and anti-Clec-2 ( C-type lectin-like receptor 2 ) ( Biorbyt , orb312182 , 1:100 , RRID:AB_2891123 ) . Alexa 488-conjugated donkey anti-rat immunoglobulin ( Invitrogen , A-21208 , 1:1000 , RRID:AB_141709 ) was used as a secondary antibody for CD41 and CD44 detection . Alexa 488-conjugated goat anti-rabbit immunoglobulin ( Invitrogen , A-11034 , 1:1000 , RRID:AB_141709 ) was used as a secondary antibody for Clec-2 detection . Alexa 594-conjugated goat anti-rabbit immunoglobulin ( Invitrogen , A-11012 , 1:1000 , RRID:AB_141359 ) was used as a secondary antibody for Chi3l1 detection . Alexa 594-conjugated goat anti-hamster immunoglobulin ( Invitrogen , A-21113 , 1:1000 , RRID:AB_2535762 ) was used as a secondary antibody for podoplanin detection . Nuclei were detected by Hoechst ( Invitrogen , H3570 , 1:10 , 000 , RRID:AB_10626776 ) . Mice were prepared for intravital microscopy as previously described ( Da et al . , 2018 ) . Briefly , mice were anesthetized using pentobarbital and underwent tracheostomy ( to facilitate breathing ) and internal jugular catheterization ( for antibody administration ) followed by liver exteriorization as described by Marques et al . , 2015 with modifications . Mice were placed supine on a custom-made stage with the liver overlying a glass coverslip wetted with warmed saline and surrounded with wet saline-soaked gauze . Mice were kept euthermic at 37°C using radiant warmers and monitored with a rectal thermometer . Anesthesia was maintained using an isoflurane delivery device ( RoVent with SomnoSuite; Kent Scientific ) with 1–3% isoflurane delivered . Mice were i . v . injected with an antibody mixture in sterile 0 . 9% sodium chloride containing TRITC/bovine serum albumin ( Sigma; to label the vasculature; 500 µg/mouse , RRID:AB_2891111 ) , BV421-anti-F4/80 antibody ( to label Kupffer; 0 . 75 µg/mouse , RRID:AB_11203717 ) , and DyLight 649/anti-GPIbβ antibody ( emfret analytics; to label platelets; 3 µg/mouse , RRID:AB_2861336 ) for visualization . Mice were imaged on an Olympus FV3000RS laser scanning confocal inverted microscope system at 30 fps using a 60×/NA1 . 30 silicone oil objective with 1× and 3× optical zoom using the resonance scanner . This allows for simultaneous excitation and detection of up to four wavelengths . All animals were euthanized under a surgical plane of anesthesia at the end of the experiments . The images were then analyzed by a blinded investigator to assess platelet area . Eleven to fifteen 1 min fields of view ( 1× optical zoom ) were analyzed per mouse using FIJI/ImageJ software . Background noise was removed using a Gaussian filter ( one pixel ) for all channels prior to analysis . Vascular area was measured in each field using the region of interest selection brush in the TRITC ( albumin ) channel . The platelet area within the vascular ROI was then determined using threshold of the DyLight 649 ( platelet ) channel . Rabbit mAbs were generated using previously reported methods ( Deng et al . , 2018 ) . Briefly , two New Zealand white rabbits were immunized subcutaneously with 0 . 5 mg recombinantly expressed human Chi3l1 protein ( Sino Biological , Cat#11227-H08H ) . After the initial immunization , animals were given boosters three times in a 3-week interval . Serum titers were evaluated by indirect enzyme-linked immunosorbent assay ( ELISA ) and rabbit peripheral blood mononuclear cells ( PBMCs ) were isolated after the final immunization . A large panel of single memory B cells were enriched from the PBMCs and cultured for 2 weeks , and the supernatants were analyzed by ELISA . To isolate mouse Chi3l1 antibody , the rabbits were boosted twice more with mouse Chi3l1 before the memory B cell culture . The variable region genes of the antibodies from these positive single B cells were recovered by reverse transcription PCR ( RT-PCR ) and cloned into the mammalian cell expression vector as described previously ( Deng et al . , 2018 ) . Both the heavy and light chain constructs were co-transfected into Expi293 cell lines using transfection reagent PEI ( Sigma ) . After 7 days of expression , supernatants were harvested and antibodies were purified by affinity chromatography using protein A resin as reported before ( Deng et al . , 2018 ) . Data were presented as mean ± SEM unless otherwise stated . Statistical analyses were carried out using GraphPad Prism ( GraphPad Software ) . Comparisons between two groups were carried out using unpaired Student’s t-test . Comparisons among multiple groups ( n ≥ 3 ) were carried out using one-way ANOVA . p-Values are as labeled and less than 0 . 05 was considered significant . Platelets counts/mm2 was analyzed using ImageJ software . Serum samples from patients diagnosed with APAP-induced liver failure on day 1 of admission were obtained from the biobank of the Acute Liver Failure Study Group ( ALFSG ) at UT Southwestern Medical Center , Dallas , TX , USA . The study was designed and carried out in accordance with the principles of ALFSG and approved by the Ethics Committee of ALFSG ( HSC-MC-19–0084 ) . Formalin-fixed , paraffin-embedded human liver biopsies from patients diagnosed with APAP-induced liver failure were obtained from the National Institutes of Health-funded Liver Tissue Cell Distribution System at the University of Minnesota , which was funded by NIH contract # HHSN276201200017C . See Materials and methods for details for other methods . Mice were i . p . injected with Ctrl IgG ( BD Pharmingen , 559478 , 50 μg/mouse ) or anti-CD44 antibody ( BD Pharmingen , 553131 , 50 μg/mouse ) in Chil1-/- reconstituted with rmChi3l1 at 30 min prior to APAP treatment . Hepatic NPCs and hepatocytes were isolated as previously described ( Shan et al . , 2018 ) . In brief , mice were anesthetized and liver tissues were perfused with EGTA solution , followed by a 0 . 04% collagenase digestion buffer . Liver hepatocytes and NPCs were isolated by gradient centrifugation using 35% percoll ( Sigma ) . To further purify LSEC and Mɸs , LSEC and Mɸs fractions were stained with phycoerythrin ( PE ) -conjugated anti-CD146 ( for LSEC , Invitrogen , 12-1469-42 ) , and anti-F4/80 ( for Mɸs , Invitrogen , 12-4801-82 ) antibodies and positively selected using EasySep Mouse PE Positive Selection Kit ( Stemcell Technologies ) following manufacturer’s instructions . Each subset will yield a purity around 90% . Isolated Mɸs were cultured in DMEM with 10% fetal bovine serum and pre-treated with podoplanin antibody ( Bioxcell InvivoMab , BE0236 , 2 μg/ml ) for 30 min and then co-culture with washed platelets for 30 min . Unbound platelets were washed out and podoplanin and Clec-2 on Mɸs were stained . Mouse whole blood was collected with anti-coagulant ACD solution from Inferior vena cava . Platelets were further isolated by additional washes with Tyrode’s buffer . Isolated washed platelets were subjected to functional assay after incubation with PGI2 ( Sigma , P6188 ) for 30 min . Isolated liver NPCs were incubated with1 μl of anti-mouse FcγRII/III ( Becton Dickinson , Franklin Lakes , NJ ) to minimize non-specific antibody binding . The cells were then stained with anti-mouse CD45-V655 ( eBioscience , 15520837 ) , F4/80-APC/Cy7 ( Biolegend , 123118 ) , Ly6C-APC ( BD Pharmingen , 560595 ) , Ly6G-V450 ( BD Pharmingen , 560603 ) , CD146-PerCP-Cy5 . 5 ( BD Pharmingen , 562134 ) , CD44-PE ( BD Pharmingen , 553134 ) , anti-His-FITC ( abcam , ab1206 ) . In some experiments , cells were incubated with 2 μg rmChi3l1 for 2 hr before antibody staining . The cells were analyzed on a CytoFLEX LX Flow Cytometer ( Beckman Coulter , Indianapolis , IN ) using FlowJo software ( Tree Star , Ashland , OR ) . For flow cytometric analysis , CD45+ cells were gated to exclude endothelial cells , hepatic stellate cells , and residue hepatocytes . Within CD45+ cells , CD44+ cells that bind to Chi3l1 were back-gated to determine the cells types . Snap-frozen liver tissues were pulverized to extract liver proteins in STE buffer . Protein concentration was measured by BCA kit ( Thermo Scientific , 23225 ) following the manufacturer’s instructions . Proteins were extracted from NPCs lysates and incubated with 5 μg rmChi3l1 , followed by immunoprecipitation with 2 μg rabbit IgG ( negative control , Peprotech , 500-p00 ) or 2 μg anti-His tag antibody ( Abnova , MAB12807 ) . Dynabeads Protein G ( Invitrogen , 1003D ) were used to pull down antibodies-binding proteins . Immunoprecipitated proteins were subject to mass spectrometry analyses by the Proteomics Core Facility at UTHealth . Cd44-/- and WT mice were treated with APAP for 2 hr; 10 mg liver proteins were extracted from treated mice and incubated with 5 μg rmChi3l1 , followed by immunoprecipitation with 2 μg anti-CD44 antibody ( BD Pharmingen , 553131 ) . Dynabeads Protein G ( Invitrogen , 1003D ) were used to pull down antibody-binding proteins . Input and immunoprecipitated proteins were subject to Western blot analyses . Two microgram rhChi3l1 ( Sino Biological , His Tag , 11227-H08H ) or 2 μg GST protein ( His Tag ) as control were incubated with 2 μg human CD44 ( Sino Biological , Fc Tag , 12211-H02H ) and immunoprecipitated with 2 μg anti-His antibody ( Abnova , MAB12807 ) . Input and immunoprecipitated proteins were subject to Western blot analyses . Samples were prepared with loading buffer and boiled before loading onto SDS-PAGE gels . Nitrocellulose membranes ( Bio-Rad ) were used to transfer proteins . Primary antibodies used to detect specific proteins: anti-Chi3l1 ( Proteintech , 12036–1-AP , 1:1000 ) , anti-CD44 ( abcam , ab25340 , 1:500 ) , anti-β-actin ( Cell Signaling , 4970 , 1:1000 ) , anti-His ( Abnova , MAB12807 , 1:1000 ) , anti-cyp2e1 ( LifeSpan BioSciences , LS-C6332 , 1:500 ) , anti-APAP adducts ( Ju et al . , 2002 ) ( provided by Dr Lance R . Pohl , NIH , 1:500 ) . Secondary antibodies include goat anti-Rabbit IgG ( Jackson ImmunoResearch , 111-035-144 , 1:1000 ) , goat anti-Rat ( Jackson ImmunoResearch , 112-035-003 , 1:1000 ) . Total RNA was isolated from 1 × 106 cells using RNeasy Mini Kit ( Qiagen , Valencia , CA ) . After the removal of genomic DNA , RNA was reversely transcribed into cDNA using Moloney murine leukemia virus RT ( Invitrogen , Carlsbad , CA ) with oligo ( dT ) primers ( Invitrogen ) . Quantitative PCR was performed using SYBR green master mix ( Applied Biosystem ) in triplicates on a Real-Time PCR 7500 SDS system and software following manufacturer’s instruction ( Life Technologies , Grand Island , NY ) . RNA content was normalized based on amplification of 18S ribosomal RNA ( rRNA ) ( 18S ) . Change folds = normalized data of experimental sample/normalized data of control . The specific primer pairs used for PCR are listed in Table 1 . Although elevated serum levels of Chi3l1 have been observed in chronic liver diseases ( Kumagai et al . , 2016; Lee et al . , 2011; Nøjgaard et al . , 2003; Wang et al . , 2020 ) , modulations of Chi3l1 levels during acute liver injury have not been reported . Our data demonstrated , for the first time , that compared with healthy individuals , patients with AILI displayed higher levels of Chi3l1 in the liver and serum ( Figure 1A , B ) . Similarly , in mice treated with APAP , hepatic mRNA and serum protein levels of Chi3l1 were upregulated ( Figure 1C , D ) . To determine the role of Chi3l1 in AILI , we treated WT mice and Chi3l1-knockout ( Chil1-/- ) mice with APAP . Compared with WT mice , serum ALT levels and the extent of liver necrosis were dramatically lower in Chil1-/- mice ( Figure 1E , F ) . Moreover , administration of rmChi3l1 protein to Chil1-/- mice enhanced liver injury to a similar degree observed in APAP-treated WT mice ( Figure 1E , F ) . These data strongly suggest that Chi3l1 contributes to AILI . Thrombocytopenia is often observed in patients with APAP overdose ( Harrison et al . , 1990; Stravitz et al . , 2013; Stravitz et al . , 2016 ) . We hypothesized that this phenomenon may be attributed to the recruitment of platelets into the liver . We performed IHC staining of liver biopsies from patients with APAP-induced liver failure and found markedly increased numbers of platelets compared with normal liver tissues ( Figure 2A ) . Similarly , in mice treated with APAP , a marked increase of platelets in the liver was observed by intravital microscopy ( Figure 2B ) . It is reported that depletion of platelets prior to APAP treatment can prevent liver injury in mice ( Miyakawa et al . , 2015 ) . Our data demonstrated that even after APAP treatment , depletion of platelets could still attenuate AILI ( Figure 2C , D; Figure 2—figure supplement 1 ) . These data indicate a critical contribution of platelets to AILI . Given the role of Chi3l1 in promoting intrahepatic coagulation in concanavalin A-induced hepatitis ( Shan et al . , 2018 ) , we hypothesized that Chi3l1 might be involved in platelet recruitment to the liver during AILI . To examine this hypothesis , we detected platelets in the liver by IHC using anti-CD41 antibody . Comparing with WT mice , we observed much fewer platelets in the liver after APAP treatment ( Figure 2E ) . Moreover , administration of rmChi3l1 to Chil1-/- mice restored hepatic platelet accumulation similar to APAP-treated WT mice ( Figure 2E ) . These data suggest that Chi3l1 plays a critical role in promoting hepatic platelet accumulation , thereby contributing to AILI . To further understand how Chi3l1 is involved in platelet recruitment , we set out to identify its receptor . We isolated non-parenchymal cells ( NPCs ) from WT mice at 3 hr after APAP treatment and incubated the cells with His-tagged rmChi3l1 . The cell lysate was subjected to immunoprecipitation using an anti-His antibody . The ‘pulled down’ fraction was subjected to LC/MS analyses , and a partial list of proteins identified is shown in Supplementary file 1 . Among the potential binding proteins , we decide to further investigate CD44 , which is a cell surface receptor expressed on diverse mammalian cell types , including endothelial cells , epithelial cells , fibroblasts , keratinocytes , and leukocytes ( Ponta et al . , 2003 ) . Immunoprecipitation experiments using liver homogenates from APAP-treated WT and Cd44-/- mice demonstrated that the anti-CD44 antibody could ‘pull down’ Chi3l1 from WT but not Cd44-/- liver homogenates ( Figure 3A ) . Supporting this finding , interferometry measurements using rhChi3l1 revealed a direct interaction between Chi3l1 and CD44 ( Kd = 251 nM , Figure 3B ) . Moreover , we incubated rhChi3l1 with human CD44 and then performed immunoprecipitation with an anti-CD44 antibody . Data shown in Figure 3C confirmed that Chi3l1 directly binds to CD44 . Together , these results suggest that CD44 is a receptor for Chi3l1 . To investigate the role of CD44 in mediating the function of Chi3l1 , we treated Cd44-/- mice with rmChi3l1 simultaneously with APAP challenge . We found that rmChi3l1 had no effect on platelet recruitment or AILI in Cd44-/- mice ( Figure 3D–F ) . This is in stark contrast to restoring platelet accumulation and increasing AILI by rmChi3l1 treatment in Chil1-/- mice ( Figure 1E , F and 2E ) . However , these effects of rmChi3l1 in Chil1-/- mice were abrogated when CD44 was blocked by using an anti-CD44 antibody ( Figure 3—figure supplement 1A–C ) . Together , these data demonstrate a critical role of CD44 in mediating Chi3l1-induced hepatic platelet accumulation and AILI . CYP2E1-mediated APAP bio-activation to form NAPQI and the detoxification of NAPQI by glutathione ( GSH ) are important in determining the degrees of AILI ( Hinson et al . , 2004 ) . Although unlikely , there is a possibility that the phenotypes observed in Chil1-/- and Cd44-/- mice were due to the effects of gene deletion on APAP bio-activation . To address this concern , we compared the levels of GSH , liver CYP2E1 protein expression , and NAPQI-protein adducts among WT , Chil1-/- and Cd44-/- mice ( Figure 3—figure supplement 2A–C ) . However , we did not observe any difference , suggesting that Chi3l1 or CD44 deletion does not affect APAP bio-activation and its direct toxicity to hepatocytes . Moreover , although we used male mice performed all of the experiments , we observed a similar phenotype in female Chil1-/- and Cd44-/- mice as in male mice ( Figure 3—figure supplement 3 ) . To further identify the cell type on which Chi3l1 binds to CD44 , we incubated liver NPCs with His-tagged rmChi3l1 . We found that almost all CD44+Chi3l1+ cells were F4/80+ Mɸs ( Figure 3—figure supplement 2D ) . This finding suggested the possible involvement of hepatic Mɸs in platelet recruitment . We performed IHC staining of liver biopsies from AILI patients and observed co-localization of Mɸs ( CD68+ ) and platelets ( CD41+ ) ( Figure 4A ) . In the livers of APAP-treated mice , adherence of platelets to Mɸs was also observed by IHC ( Figure 4B ) and intravital microscopy ( Figure 2B ) . Quantification of the staining confirmed that there were higher numbers of platelets adherent to Mɸs than to liver sinusoidal endothelial cells ( LSECs ) after APAP challenge ( Figure 4B ) . To further investigate the role of hepatic Mɸs in platelet recruitment during AILI , we performed Mɸ-depletion experiments using liposome-encapsulated CLDN . We first followed a previously published protocol ( Campion et al . , 2008; Fisher et al . , 2013; Ju et al . , 2002 ) and injected CLDN around 40 hr prior to APAP treatment ( Figure 4C , ‘Previous Strategy’ ) . We examined the efficiency of Mɸ-depletion by flow cytometry analysis , which can distinguish resident KCs ( CD11blowF4/80+ ) from infiltrating Mɸs ( IMs , CD11bhiF4/80+ ) ( Holt et al . , 2008 ) . We found that compared with control mice treated with empty liposomes , there were actually more Mɸs , consisted of mainly IMs , in the liver of CLDN-treated mice ( Figure 4C ) . Consistent with the increase of Mɸs , there were also higher numbers of platelets in the liver of CLDN-treated mice ( Figure 4C ) . These findings suggest that although KCs are depleted using the ‘Previous Strategy’ , the treatment of CLDN induces the recruitment of IMs , resulting in higher numbers of Mɸs in the liver at the time of APAP treatment . As reported , this treatment strategy resulted in exacerbated AILI ( Figure 4E , F , ‘Previous Strategy’ ) , which had led to the conclusion in published reports that KCs play a protective role against AILI ( Campion et al . , 2008; Fisher et al . , 2013; Ju et al . , 2002 ) . However , alternatively the enhanced injury could be due to increased IMs and platelet accumulation . To better investigate the role of hepatic Mɸs in platelet recruitment , we set out to identify a time period in which both KCs and IMs are absent after CLDN treatment . We measured hepatic Mɸs by flow cytometry at various time points after CLDN treatment and established a ‘New Strategy’ , in which mice were injected with CLDN and after 9 hr treated with APAP . As shown in Figure 4D , at 6 hr after APAP challenge ( 15 hr after CLDN ) , both KCs and IMs were dramatically reduced . Interestingly , when compared to control mice treated with empty liposomes , CLDN-treated mice developed markedly reduced liver injury with nearly no platelet accumulation in the liver ( Figure 4D–F , ‘New Strategy’ ) . These data suggest that hepatic Mɸs play a crucial role in platelet recruitment into the liver , thereby contributing to AILI . To further understand how Chi3l1/CD44 signaling in Mɸs promotes platelet recruitment , we measured Mɸs expression of a panel of adhesion molecules known to be important in platelet recruitment ( Hamburger and McEver , 1990; Hitchcock et al . , 2015; Larsen et al . , 1989; Simon et al . , 2000 ) . Our data showed that podoplanin is expressed at a much higher level in hepatic Mɸs isolated from APAP-treated WT mice than those from Chil1-/- or Cd44-/- mice ( Figure 5A ) . Interestingly , rmChi3l1 treatment of Chil1-/- , but not Cd44-/- mice , markedly increased the podoplanin mRNA and protein expression levels in Mɸs ( Figure 5B , C ) . To examine the role of podoplanin in mediating platelet adhesion to Mɸs , we blocked podoplanin using an anti-podoplanin antibody in Chil1-/- mice reconstituted with rmChi3l1 . As shown in Figure 5D–F , blockade of podoplanin not only abrogated rmChi3l1-mediated platelet recruitment into the liver but also significantly reduced its effect on increasing AILI in Chil1-/- mice . Clec-2 is the only platelet receptor known to bind podoplanin ( Kerrigan et al . , 2012 ) . To further elucidate the role of podoplanin in mediating platelet adhesion to Mɸs , we isolated Mɸs from WT mice treated with APAP . After treating Mɸs with anti-podoplanin antibody or IgG as control , we added platelets . IF staining of podoplanin and Clec-2 showed that the Clec-2-expressing platelets only bound to IgG-treated , but not anti-podoplanin-treated Mɸs ( Figure 5—figure supplement 1 ) . Together , our data demonstrate that Mɸs recruit platelets through podoplanin and Clec-2 interaction , and that the podoplanin expression on Mɸs is regulated by Chi3l1/CD44 signaling . Although NAC greatly reduces morbidity and mortality from ALF due to APAP overdose , the death rate and need for liver transplantation remain unacceptably high . While elucidating the underlining biology of Chi3l1 in AILI , we also generated mAbs specifically recognizing either mouse or human Chi3l1 . We screened a panel of anti-mouse Chi3l1 monoclonal antibodies ( α-mChi3l1 mAb ) to determine their efficacies in attenuating AILI . We injected WT mice with an α-mChi3l1 mAb or IgG at 3 hr after APAP challenge . Our data showed that clone 59 ( C59 ) had the most potent effects on inhibiting APAP-induced hepatic platelet accumulation and attenuating AILI ( Figure 6A–C ) . To evaluate the potential of targeting Chi3l1 as a treatment for AILI in humans , we screened all of the α-hChi3l1 mAb we generated by IHC staining of patients’ liver biopsies ( data not shown ) and selected the best clone for in vivo functional studies . Because the amino acid sequence homology between human and mouse Chi3l1 is quite high ( 76% ) , we treated Chil1-/- mice with rhChi3l1 . We found that rhChi3l1 was as effective as rmChi3l1 in promoting platelet recruitment and increasing AILI in Chil1-/- mice ( Figure 6D–F ) . To our excitement , the α-hChi3l1 mAb treatment could abrogate platelet recruitment and dramatically reduce liver injury ( Figure 6D–F ) . Together , these data indicate that mAb-based blocking of Chi3l1 may be an effective therapeutic strategy to treat AILI and potentially other acute liver injuries . The current study unveiled an important function of Chi3l1 in promoting platelet recruitment into the liver after APAP overdose , thereby playing a critical role in exacerbating APAP-induced coagulopathy and liver injury . Our data demonstrate that Chi3l1 signals through CD44 on Mɸs to upregulate podoplanin expression and promote platelet recruitment ( Figure 7 ) . Moreover , we report for the first time significant hepatic accumulation of platelets and marked upregulation of Chi3l1 in patients with ALF caused by APAP overdose . Importantly , we demonstrate that neutralizing Chi3l1 with mAbs can effectively inhibit hepatic platelet accumulation and mitigate liver injury caused by APAP , supporting the potential and feasibility of targeting Chi3l1 as a therapeutic strategy to treat AILI . The elevation of serum levels of Chi3l1 has been observed in various liver diseases ( Kumagai et al . , 2016; Lee et al . , 2011; Nøjgaard et al . , 2003; Wang et al . , 2020 ) , but studies of its involvement in liver diseases have only begun to emerge . There are several reports describing a role of Chi3l1 in models of chronic liver injuries caused by alcohol , CCl4 , or high-fat diet ( Higashiyama et al . , 2019; Lee et al . , 2019; Qiu et al . , 2018; Zhang et al . , 2021 ) . However , the molecular and cellular mechanisms accounting for the involvement of Chi3l1 have yet to be defined . The present study unveils a function of Chi3l1 in promoting platelet recruitment to the liver during acute injury . We provide compelling data demonstrating that Chi3l1 acts through its receptor CD44 on Mɸs to recruit platelets , thereby contributing to AILI . Multiple receptors of Chi3l1 have been identified , including IL-13Rα2 , CRTH2 , TMEM219 , and galectin-3 ( Geng et al . , 2018; He et al . , 2013; Lee et al . , 2016; Zhou et al . , 2015; Zhou et al . , 2018 ) . The fact that Chi3l1 could bind to multiple receptors is consistent with a diverse involvement of Chi3l1 under different disease contexts . A recent study showed that Chi3l1 was upregulated during gastric cancer ( GC ) development and that through binding to CD44 , it activated Erk , Akt , and β-catenin signaling , thereby enhancing GC metastasis ( Geng et al . , 2018 ) . Our studies illustrated a novel role of Chi3l1/CD44 interaction in the recruitment of hepatic platelets and contribution to AILI . Our in vivo studies using Cd44-/- mice and anti-CD44 antibody provide strong evidence that CD44 mediates the effects of Chi3l1 . Our observation that Chi3l1 predominantly binds to CD44 on Mɸs , but not other CD44-expressing cells in the liver , suggests two possibilities which warrant further investigation . First , Chi3l1 may bind a specific isoform of CD44 that is uniquely expressed by Mɸs . Second , the Chi3l1-CD44 interaction requires binding of a co-receptor , which is expressed on Mɸs but not on other CD44-expressing cells in the liver . We identified hepatic Mɸs as a key player in promoting platelet recruitment to the liver during AILI . Given the involvement of platelets in AILI , this finding would suggest that hepatic Mɸs also contribute to liver injury . The role of hepatic Mɸs in AILI has been a topic of debate and the current understanding is confined by the limitation of the methods used to deplete these cells ( Campion et al . , 2008; Fisher et al . , 2013; Ju et al . , 2002; Laskin et al . , 1995; Michael et al . , 1999 ) . Several previous studies using CLDN to deplete Mɸs concluded that these cells play a protective role against AILI ( Campion et al . , 2008; Fisher et al . , 2013; Ju et al . , 2002 ) . However , in those studies , Mɸ-depletion was confirmed by IHC staining of F4/80 , which cannot distinguish KCs from IMs . Our laboratory and others had since developed a flow cytometric approach to detect and distinguish the two Mɸs populations . Using flow cytometry to monitor Mɸ-depletion , we found that the timing of CLDN treatment was critical . In the previously published reports , mice were treated with CLDN around 2 days before APAP challenge ( Campion et al . , 2008; Fisher et al . , 2013; Ju et al . , 2002 ) . Using this treatment regimen , IMs became abundant prior to APAP treatment , even though KCs were depleted . Without this knowledge , previous studies attributed the worsened AILI to the depletion of KCs . However , the advancement of knowledge on the recruitment of IMs and their contribution to acute liver injury offers an alternative interpretation that the worsened AILI is due to IM accumulation ( Chauhan et al . , 2020; Holt et al . , 2008; Mossanen et al . , 2016; Zigmond et al . , 2014 ) . In the current study , we analyzed KCs and IMs in the liver at various time points after CLDN treatment to identify a new strategy to achieve more complete hepatic Mɸ-depletion . Our data demonstrated that when both Mɸs populations were absent at the time of APAP treatment , platelet recruitment was abrogated and AILI was significantly reduced . During the preparation of this manuscript , a study was published describing that IMs could recruit platelets ( Chauhan et al . , 2020 ) . Together , these data suggest that hepatic Mɸs ( both KCs and IMs ) play a crucial role in promoting hepatic platelet accumulation , thereby contributing to AILI . Our data suggest that platelet-derived Clec-2 interacts with podoplanin expressed on Mɸs , resulting in platelet recruitment to the liver during the early phase of AILI . The role of podoplanin/Clec-2 interaction in platelet recruitment and thromboinflammation has been indicated in multiple inflammatory and infectious conditions ( Chauhan et al . , 2020; Hitchcock et al . , 2015; Kerrigan et al . , 2012 ) . Our data , for the first time , provide evidence that the podoplanin expression on Mɸs is regulated by the Chi3l1/CD44 axis . Future studies focusing on gaining molecular insight into such regulation are warranted . An increasing number of studies suggest that platelets play an important , but paradoxical role in liver injury . It has been proposed that they contribute to tissue damage during injury phase but promote tissue repair at later time points ( Chauhan et al . , 2016 ) . However , two recent studies of AILI demonstrate that persistent platelet accumulation in the liver significantly delays liver repair . One study described a podoplanin/Clec-2 interaction between platelets and hepatic IMs during tissue repair and demonstrated a detrimental role of such interaction through blocking the recruitment of reparative neutrophils ( Chauhan et al . , 2020 ) . Another study showed that AILI was associated with elevated plasma levels of von Willebrand factor , which prolonged hepatic platelet accumulation and delayed repair of APAP-injured liver in mice ( Groeneveld et al . , 2020 ) . These studies together with our finding that platelets drive tissue damage during early stage of AILI suggest that platelets may be a therapeutic target to treat acute liver injury . We observed hepatic platelet accumulation as early as 3 hr after APAP treatment in mice , prior to APAP-induced liver necrosis , indicating that platelets are likely to be the driver of AILI . Mitochondrial damage is a key event in APAP-induced cell necrosis , in which APAP triggers c-jun N-terminal kinase ( JNK ) activation in the cytosol and translocation of phospho-JNK to the mitochondria , resulting in oxidant stress and the mitochondrial permeability transition pore opening ( Saito et al . , 2010 ) . Others and our lab have reported that Chi3l1 can induce phosphorylation of JNK directly in either bronchial epithelial cells or LSEC line ( Shan et al . , 2018; Tang et al . , 2013 ) . However , whether Chi3l1 or Chi3l1-recruited platelets affects mitochondrial damage or mitochondrial JNK activation in hepatocytes warrants further investigation . During this study , we did compare the liver injuries among WT , and Chil1-/- , Cd44-/- mice in the recovery/regeneration stage of AILI ( data not shown ) . Although ALT levels of WT mice were still slightly higher than both knockout strains of mice at 48 hr post-APAP , it is most likely due to high degrees of injury in WT at the initiation stage of AILI but not due to delayed repair . There were no differences in ALT levels at 72 hr post-APAP , again indicating that the Chi3l1/CD44 does not affect tissue recovery . Moreover , we compared ALT levels at 6 hr post-APAP and there were lower in Chil1-/- and Cd44-/- mice than WT mice ( data not shown ) , which were consistent with the data shown at 24 hr , indicating that Chi3l1/CD44 axis is involved in the initiation and injury phases of AILI . Our studies uncovered a previously unrecognized involvement of the Chi3l1/CD44 axis in AILI and provided insights into the mechanism by which Chi3l1/CD44 signaling promotes hepatic platelet accumulation and liver injury after APAP challenge . Taking our findings one -step further toward clinical application , we demonstrated the feasibility of targeting Chi3l1 by mAbs to attenuate AILI . There is an unmet need for developing treatments for AILI , as NAC is the only antidote at present . However , the efficacy of NAC declines rapidly when initiated more than a few hours after APAP overdose , long before patients are admitted to the clinic with symptoms of severe liver injury ( Larson et al . , 2005 ) . Our studies provide strong support for the potential targeting of Chi3l1 as a novel therapeutic strategy to improve the clinical outcomes of AILI and perhaps other acute liver injury conditions .	Acetaminophen , also called paracetamol outside the United States , is a commonly used painkiller , with over 50 million people in the United States taking the drug weekly . While paracetamol is safe at standard doses , overdose can cause acute liver failure , which leads to 30 , 000 patients being admitted to emergency care in the United States each year . There is only one approved antidote to overdoses , which becomes significantly less effective if its application is delayed by more than a few hours . This has incentivized research into identify new drug targets that could lead to additional treatment options . Acetaminophen overdose triggers blood clotting and inflammation , contributing to liver injury . It also causes a decrease in cells called platelets circulating in the blood , which has been observed in both mice and humans . In mice , this occurs because platelets accumulate in the liver . Removing these excess cells appears to reduce the severity of the damage caused by acetaminophen , but it remains unclear how the drug triggers their accumulation in the liver . In 2018 , researchers showed that a protein called Chi3l1 plays an important role in another form of liver damage . Shan et al . – including many of the researchers involved in the 2018 study – have examined whether the protein also contributes to acetaminophen damage in the liver . Shan et al . showed that mice lacking the gene that codes for Chi3l1 developed less severe liver injury and had fewer platelets in the liver following acetaminophen overdose . They also found that human patients with acute liver failure due to acetaminophen had high levels of Chi3l1 and significant accumulation of platelets in the liver . To test whether damage could be prevented , Shan et al . used antibodies to neutralize Chi3l1 in mice after giving them an acetaminophen overdose . This reduced platelet accumulation in the liver and the associated damage . These findings suggest that targeting Chi3l1 may be an effective strategy to prevent liver damage caused by acetaminophen overdose . Further research could help develop new treatments for acetaminophen-induced liver injury and perhaps other liver conditions .	[ "Abstract", "Introduction", "Materials", "and", "methods", "Results", "Discussion" ]	[ "medicine" ]	2021	Chitinase 3-like-1 contributes to acetaminophen-induced liver injury by promoting hepatic platelet recruitment
Classically , p53 tumor suppressor acts in transcription , apoptosis , and cell cycle arrest . Yet , replication-mediated genomic instability is integral to oncogenesis , and p53 mutations promote tumor progression and drug-resistance . By delineating human and murine separation-of-function p53 alleles , we find that p53 null and gain-of-function ( GOF ) mutations exhibit defects in restart of stalled or damaged DNA replication forks that drive genomic instability , which isgenetically separable from transcription activation . By assaying protein-DNA fork interactions in single cells , we unveil a p53-MLL3-enabled recruitment of MRE11 DNA replication restart nuclease . Importantly , p53 defects or depletion unexpectedly allow mutagenic RAD52 and POLθ pathways to hijack stalled forks , which we find reflected in p53 defective breast-cancer patient COSMIC mutational signatures . These data uncover p53 as a keystone regulator of replication homeostasis within a DNA restart network . Mechanistically , this has important implications for development of resistance in cancer therapy . Combined , these results define an unexpected role for p53-mediated suppression of replication genome instability . One of the most prominent hallmarks of cancer is genomic instability ( Hanahan and Weinberg , 2011 ) . As such , many DNA damage response or repair genes that restore genome stability are known tumor suppressors , including p53 , the guardian of the genome ( Kim et al . , 2015 ) . In breast cancers , p53 mutations are associated with more aggressive and triple negative breast cancers ( Turner et al . , 2013 ) . Similar to high serous ovarian cancers , these aggressive cancers respond to chemotherapy including platinum drugs and PARP inhibitors initially , but develop resistance thereafter ( Luvero et al . , 2014; Wahba and El-Hadaad , 2015 ) . First thought to be a proto-oncogene , the initial discovery of a gain-of-function ( GOF ) p53 mutant allele ( Lane and Crawford , 1979; Levine and Oren , 2009; Linzer and Levine , 1979 ) masked the loss of wild-type ( WT ) p53 function . Despite early discrepancies , only a decade later p53 was recognized as a tumor suppressor ( Baker et al . , 1989 ) . Loss of p53 function can occur either by deletion or by mutation . Mutations may also result in a GOF , typically enhancing transcription functions . To date , the most consistent defect for both null and GOF p53 mutants in cancers is the loss of p53 transcriptional responses to regulate apoptosis and cell cycle arrest . Genetic data of several separation-of-function p53 mutant mice suggest that there are additional p53 functions that contribute to tumor progression , which are transcription independent; Murine p53 R172P , corresponding to human R175P , retains much of its tumor suppression function despite loss of transcriptional induction and loss of apoptosis ( Liu et al . , 2004 ) . Similarly , p53 mutations in the transactivation domain and p53 acetylation mutations severely inhibit p53 induction of apoptosis and senescence , yet exhibit a mild and delayed tumor onset ( Li et al . , 2012; Zhu et al . , 2015 ) . p53 also has seemingly disparate cellular functions including during metabolism and epigenetic control , that is through its interaction with MLL3/4 histone methyltransferases ( Pfister et al . , 2015; Zhu et al . , 2015 ) , although the contribution of these functions to tumor suppression is not fully understood . For cancer , a prominent p53 function is to maintain genomic stability upon DNA damage as part of a damage response . Since DNA damage traditionally is most prominently considered in the context of double-strand break ( DSB ) lesions , many studies focus on putative p53 functions in DSB repair . Next to error-free repair of DSBs by homologous recombination ( HR ) involving BRCA1/2 and RAD51 , DSBs may also be repaired by non-homologous end joining ( NHEJ ) , or through secondary and typically mutagenic pathways of single strand annealing ( SSA ) mediated by RAD52 and micro-homology mediated end joining ( MMEJ ) involving POLθ ( Black et al . , 2016; Branzei and Foiani , 2008; Moynahan and Jasin , 2010; Wood and Doublié , 2016 ) . In response to DNA damage , ATM mediates p53 phosphorylation as part of a DNA stress response ( Saito et al . , 2002 ) , which is facilitated by PTIP ( Jowsey et al . , 2004 ) , a BRCT domain containing protein that is part of the MLL3/4 complex . Similar to its transcription function , discrepancies ensue in the molecular function of p53 during DNA repair . While indirect studies found p53 to inhibit error-free homologous recombination ( HR ) and spontaneous sister-chromatid exchange ( SCE ) , which somewhat paradoxically was proposed to promote genomic stability ( Bertrand et al . , 2004; Gatz and Wiesmüller , 2006 ) , loss of p53 does not change DSB repair rates by HR when measured in specific induced break assays ( Willers et al . , 2001 ) . Thus , the mechanism by which p53 promotes genomic stability associated with tumorigenesis remains contradictory . As previously hypothesized ( Cox et al . , 2000 ) , recent findings formalized that 2/3 of all mutations found across cancers are caused by errors occurring during proliferation ( Tomasetti et al . , 2017 ) , highlighting the critical importance of protective mechanisms during DNA replication . Intriguingly , early studies found p53 is activated at stalled replication forks ( Gottifredi et al . , 2001; Kumari et al . , 2004 ) , which are a source for genomic instability requiring distinct replication fork stability pathways ( Branzei and Foiani , 2010 ) . p53 interacts with BLM helicase at replication forks and represses HR in S-phase upon DNA damage , independent of its G1-S and transactivation activity ( Bertrand et al . , 2004; Janz and Wiesmüller , 2002; Saintigny and Lopez , 2002 ) . Hinting at a direct p53 replication function , recently p53 was found to interact with DNA POLι ( Hampp et al . , 2016 ) . Moreover , p53 deletion in U2OS cells was reported to slow unperturbed replication ( Klusmann et al . , 2016 ) , although this was suggested to require p53 transcription function , while p53 functions in DNA damage response are not . Here , we identify a critical role for p53 in balancing replication pathway homeostasis and show p53 suppresses replication genomic instability independent of transcription activation . We find p53 mutant alleles that separate transcription activation and replication restart functions and reveal a direct correlation between p53 replication and tumor progression functions . Importantly , we find p53 directly binds to ongoing and stalled DNA replication forks . Utilization of mutagenic RAD52/POLθ replication pathways increase for both GOF and p53 null alleles in a transcription independent manner , consistent with mutation signatures that we identify in p53 mutant breast cancers . Our results thus allow for an unexpected alternative hypothesis for acquisition of drug resistance in breast cancer cells due to p53 loss: mutant p53 boosts mutagenic RAD52/POLθ pathways , which increase deletion and point mutations that can lead to secondary resistance mutations . p53 is an ATM phosphorylation target , is activated at stalled replication forks ( Kumari et al . , 2004 ) and interacts with BLM helicase . As BLM helicase is implicated in replication restart ( Davies et al . , 2007 ) , as is ATM ( Trenz et al . , 2006 ) , we tested the role of p53 in DNA replication reactions when stalled with dNTP depleting hydroxyurea ( HU ) . Using single-molecule DNA fiber spreading ( Figure 1A ) , we assessed the number of stalled replication forks after low-dose replication stalling ( Figure 1A ) , as a test for defects in replication restart . We find a doubling of stalled forks in CRISPR/CAS9- engineered p53- null human HAP-1 cells compared to cells with wild-type ( WT ) p53 ( Figure 1B; 35% stalled forks in p53 null 18% WT p53 HAP-1 ) . This suggests a prominent role for p53 in the resumption of DNA replication after replication stress . Increased fork stalling is classically compensated for by increased new origin firing , as seen for CHK1 defects ( Petermann and Helleday , 2010; Petermann et al . , 2010 ) . Unexpectedly , we find that increased fork stalling in p53 null cells is accompanied by a decrease , rather than an increase , in new origin firing compared to both WT p53 ( Figure 1—figure supplement 1A; 20% newly fired origins in HAP-1 cells , respectively , and 9% newly fired origins in HAP-1 p53 null cells ) . Taken together , the data suggests that p53 defects in HAP-1 cells exhibit distinct and unconventional replication restart defects resulting in both decreased replication restart and decreased new origin firing . In tumors , p53 is deleted or mutated , the latter typically resulting in GOF ( Freed-Pastor and Prives , 2012 ) . To test whether p53 mutations alter replication restart , we investigated one of the most common GOF mutations in primary mouse embryonic fibroblasts ( murine p53 R172H corresponding to human p53 R175H ) ( Liu et al . , 2000 ) . These cells show both an increase in stalled replication forks , and fewer newly fired forks compared to WT p53 MEFs ( Figure 1C; 36% stalled forks in p53 R172H MEF and 15% in WT MEF and Figure 1—figure supplement 1B ) . These data thus uncover conserved defective outcomes for both null and GOF p53 mutations at stalled replication forks . Murine p53 R172P corresponds to a rare human polymorphism R175P , which results in loss of transcriptional activation and apoptosis resembling p53 null ( Liu et al . , 2004 ) . Yet , tumor development is markedly less severe in p53 R172P mice compared to p53 null mice , suggesting alternative mechanisms contribute to tumor progression besides transcriptional regulation of apoptosis with this mutation ( Liu et al . , 2004 ) . We therefore tested p53 R172P MEFs and find that they only show a moderate increase in fork stalling compared to p53 R172H ( Figure 1C ) . Thus , restart functions in p53 R172 mutant cells show an improved correlation with tumor suppression activity in vivo . To further examine the possible link of failed p53-mediated replication restart and cancer , we tested p53 S47 ( P47S ) , which is a breast-cancer pre-disposition polymorphism in African-descent populations ( Jennis et al . , 2016 ) . It is largely transcriptionally active including for p21 ( Felley-Bosco et al . , 1993; Jennis et al . , 2016 ) ; it remains proficient for apoptosis ( Felley-Bosco et al . , 1993; Jennis et al . , 2016 ) . Yet , p53 S47 mice are tumor prone , and p53 S47 contributes to breast cancer risk in African populations ( Jennis et al . , 2016; Murphy et al . , 2017 ) . We find that S47 slows cell growth similar to WT p53 ( Figure 1—figure supplement 1C ) consistent with intact cell-cycle check-point functions . Furthermore , we find that p53 S47 primary MEFs exhibit a loss-of-function ( LOF ) for replication restart , as measured by an increase in stalled forks upon HU treatment compared to WT MEF cultures ( Figure 1D; 31% stalled forks in p53 S47 MEF versus 15% in WT MEF ) . Moreover , p53 S47 MEFs resemble p53 null MEFs in their inability to restart forks ( Figure 1D; 34% stalled forks in p53 null MEF ) . Similar to p53 null HAP-1 and GOF R172H MEF cells , both p53 null MEF and p53 S47 MEFs show defects in new fork initiation ( Figure 1E; 21% newly fired origins in p53 WT MEF versus 9% and 8% in p53 S47 MEF and p53 null MEF ) . To test whether p53 S47 restart defects are conserved in human cells , we expressed human p53 S47 under doxycycline control in H1299 non-small cell lung carcinoma cells ( Figure 1—figure supplement 1D ) and examined stalled forks . We found an increase in stalled forks that resembles p53-null H1299 cells . Both p53 null and p53 S47 H1299 cells exhibit a substantial increase in stalled forks compared to WT p53-expressing H1299 cells ( Figure 1F; 13% stalled forks in p53 WT , 38% in p53 S47 , and 45% in null H1299 cells ) . Taken together , these results suggest that p53 promoted replication restart can be genetically separated from its transcription-activation function in cell cycle progression and apoptosis . Restart defects cause cellular sensitivity to replication stalling agents , as seen for cells with BLM defects ( Davies et al . , 2004 ) . We reasoned that in mutant p53 cells , this cellular phenotype so far may have been obscured by loss of apoptosis , which can override cellular sensitivity by inhibiting cell death . We therefore tested cellular sensitivity using the LOF mutant p53 S47 , which remains largely apoptosis proficient . We find that p53 S47-expressing H1299 cells are sensitive to replication stalling agents HU ( Figure 2A ) and mitomycin C ( MMC; Figure 2B ) . This cellular replication stress phenotype is masked when the p53 mutations additionally inactivate apoptosis and cell-cycle check-point functions , such as in p53 null H1299 cells ( Figure 2A and B ) or p53 null compared to WT mammary epithelial MCF10A cells ( Figure 2—figure supplement 1 ) . Collectively , the data with apoptosis-proficient p53 S47 implies that p53 functions in replication restart suppresses cellular sensitivity to replication stress . Separation-of-function mutations p53 R172P and S47 reveal a feasible correlation between loss of p53 restart function and tumor progression . We therefore tested whether p53’s replication restart function could contribute to genomic instability , which is a hallmark of cancer ( Hanahan and Weinberg , 2011 ) . Amongst others , unresolved stalled replication forks can result in DNA bridges , which convert to micronuclei , a mark of BLM-defective cells ( Hoffelder et al . , 2004 ) . We assessed genome instability by scoring micronuclei in p53 R172P , R172H and WT MEFs after arrest in cytokinesis immediately following replication stalling ( Figure 2C ) . By considering EdU-positive cells only and immediate cytokinesis arrest following HU , the experimental set up ensures that only micronuclei are scored that result from induced replication stalling during the preceding S-phase . This so greatly excludes contributions of canonical G1-related p53-function . Consistent with an intermediate phenotype for replication restart , primary MEF p53 R172P exhibited less micronuclei with replication stalling compared to p53 R172H ( Figure 2D; average of 8% micronuclei in p53 R172P and 26% in p53 R172H ) albeit considerably more than WT p53 MEFs ( none detected in WT ) . With replication stalling , human p53 S47-expressing cells showed a marked increase in cells containing micronuclei compared to WT p53-expressing H1299 cells ( Figure 2E average of 20% in p53 S47% and 5% in p53 WT H1299 cells ) . Micronuclei instability in p53 null H1299 cells similarly is significantly higher than in WT p53 H1299 cells ( Figure 2E; average of 12% ) . Taken together , the genomic instability data corresponds with restart defects rather than transcription functions found in the respective p53 mutants . The data suggest that p53 functions are required for resolving replication stress-dependent genome instability and by implication can contribute to tumor suppression . p53 has a prominent DNA-binding domain and RPA interaction sites ( Romanova et al . , 2004 ) . We reasoned that p53 could have a direct protein function during DNA replication . We therefore sought to test if p53 is present at active DNA replication forks using iPOND , which is an immunoprecipitation method of nascent EdU-labeled DNA ( Sirbu et al . , 2012 ) . We found that p53 associates with nascent-labeled DNA in HEK293 cells ( Figure 3A , IP fraction , E , 3 . 8 normalized to chromatin protein ORC2 ) . PCNA is associated with active replication forks and therefore found reduced in iPONDs with thymidine chase after the EdU pulse ( Dungrawala et al . , 2015 ) , ( Figure 3A ) . Similarly , we find that the p53 association with nascent DNA is reduced with a thymidine chase , suggesting p53 travels with the active replication fork ( Figure 3A ) . Replication stalling by HU retains p53 , but not PCNA at stalled forks ( Figure 3A ) . Taken together , the data suggests a direct role for p53 at active and stalled DNA replication forks . p53 interacts with chromatin remodeling complexes and is implicated in facilitating epigenetic alterations ( Pfister et al . , 2015; Zhu et al . , 2015 ) . We observed an atypical restart defect in p53 mutant and null cells with less newly initiated replication forks ( Figure 1E and Figure 1—figure supplement 1A and B ) . We hypothesized that replication restart and new replication fork firing may require local chromatin opening and epigenetic alterations mediated by p53 , which could explain the unusual decrease in new fork firing with p53 mutations . To investigate local protein changes specific to replication forks , we developed the SIRF assay ( in Situ Interactions at Replication Forks using PLA; Figure 3B ) . Specifically , we applied sensitive proximity ligation chemistry to detect interactions between nascent , EdU-labeled DNA and proteins within nanometer proximity . The signal is specific as elimination of EdU results in no signals ( Figure 3B ) . MLL3 promotes H3K4 histone methylation to mark open chromatin ( Ruthenburg et al . , 2007 ) . MLL3 association with replication forks in unperturbed cells is similar in p53 null and WT HAP-1 cells ( Figure 3C; 11 MLL3-bound replication sites per cell in WT p53 and 12 sites in p53 null HAP-1 cells ) . Upon replication stalling with HU , we see a marked increase in MLL3-bound replication sites in WT , but not in p53 null human HAP-1 cells ( Figure 3D; 17 MLL3-bound sites in WT and 13 in p53 null HAP-1 cells ) . These data suggest inefficient MLL3 recruitment to forks upon replication stalling in the absence of p53 . MLL3-mediated chromatin opening is implicated in MRE11 nuclease recruitment to stalled replication forks ( Ray Chaudhuri et al . , 2016 ) , a repair nuclease that is needed for efficient replication restart ( Trenz et al . , 2006 ) . We therefore examined the functional implications of reduced MLL3 recruitment to forks with p53 defects . Consistently , we find increased MRE11-bound replication sites in WT , but not in p53 null HAP-1 cells when challenged with HU ( Figure 3E and Figure 3—figure supplement 1; 30 MRE11-bound replication sites/cell in WT and 18 in p53 null HAP-1 cells ) . Collectively , these data suggest a mechanism whereby p53 promotes local chromatin responses that aid MRE11 recruitment to stalled forks , as necessary for replication restart ( Trenz et al . , 2006 ) . p53 is implicated in suppressing excessive repair by homologous recombination ( HR ) to balance genomic stability ( Bertrand et al . , 2004; Saintigny et al . , 1999; Sengupta et al . , 2004 ) . To further probe the underlying mechanism for genomic instability induced by aberrant p53 S-phase functions , we performed SIRF assays for local RAD51 recruitment to stalled forks as a surrogate marker for HR processes . From previous reports , we expected more RAD51 recruitment to forks in the absence of p53 ( Bertrand et al . , 2004; Gatz and Wiesmüller , 2006 ) . Instead , in p53 null HAP-1 cells , we find less RAD51 at local stalled replication forks ( Figure 4A; 24 RAD51-bound replication sites/cell in WT p53 and 19 in p53 null HAP-1 cells ) . In contrast , HCT116 cells expressing GOF p53 R248W show increased RAD51 fork-localization compared to WT-expressing HCT116 , as do LOF mutant p53 S47-expressing H1299 cells compared to cells with WT p53 ( Figure 4—figure supplement 1A and B ) . Due to these unexpected differences , we employed p53 null Saos-2 sarcoma cells in comparison to isogenic GOF mutant p53-expressing Saos-2 cells and U2OS cells , which are p53 WT sarcoma cells ( Figure 4B ) . p53 null Saos-2 show an increase in RAD51 SIRF signals , which is repressed with expression of mutant p53 GOF R175H and R273H . Together , these results do not support a correlation between RAD51 recruitment to forks and fork instability in these cells , but instead suggest alternative causation for the observed genomic instability in p53 mutant cells . Defects in p53 do not affect HR repair of induced double-strand breaks , however , they increase spontaneous sister-chromatid exchanges ( SCEs ) ( Willers et al . , 2001 ) , which are thought to occur at sites of stalled replication . Reports increasingly suggest that spontaneous SCE is independent of RAD51 and BRCA2 ( Bai and Symington , 1996; Clémence Claussin et al . , 2017; Ray Chaudhuri et al . , 2016 ) , but instead involves the single-strand annealing ( SSA ) protein RAD52 ( Thorpe et al . , 2006 ) . We therefore tested whether RAD52 recruitment is altered with p53 deletions . Using SIRF analysis , we find a marked increase of RAD52 bound to stalled forks in p53-null cells ( Figure 4C ) . As the signals were too abundant to be individually counted , we used the mean fluorescence intensity ( MFI ) as a quantitative readout ( Figure 4C , MFI of 494 in p53 null and 762 in p53 WT HAP-1 cells ) . Notably , we see stronger RAD52 recruitment at low compared to high concentration of HU ( Figure 4—figure supplement 1C ) . The former condition is less favorable for DSB formation , suggesting RAD52 recruitment to stalled forks is stronger than to bona fide DNA breaks . Strikingly , we find an increase in RAD52 recruitment in all p53-defective cell lines tested irrespective of the nature of the p53 defect . This includes HCT116 GOF p53 R248W ( Figure 4D ) , Saos-2 p53 null , Saos-2 GOF p53 R175H , Saos-2 GOF p53 R273H , and H1299 LOF p53 S47 cells ( Figure 4—figure supplement 1D and E ) compared to respective WT p53-expressing cells . These collective data unexpectedly uncover consistent replication fork pathway tipping toward mutagenic RAD52 processes in p53 defective cells . This pathway imbalance was not caused by transcriptional deregulations in p53-defective cells , as RAD52 protein levels remained unchanged with or without p53 , further supporting a transcription-independent function of p53 at stalled forks ( Figure 4E ) . Thus , the observed increased RAD52 recruitment to stalled forks is likely a consequence of defective replication restart . In p53-defective cells , we observed a stark RAD52 recruitment with low HU ( Figure 4—figure supplement 1C ) , which can lead to reversed replication forks ( Neelsen and Lopes , 2015 ) that provide free ends as substrates for DSB repair pathways . We reasoned that p53 may orchestrate reversed fork outcomes and so protect against aberrant double-strand end recognition by mutagenic DNA end pathways which may include SSA and micro-homology mediated end-joining ( MMEJ ) . POLθ is implicated in promoting error-prone MMEJ at replication-associated DNA ends ( Roerink et al . , 2014 ) , which includes collapsed or reversed replication forks . We therefore tested if DNA POLθ contributes to mutagenic events at imbalanced stalled forks . We find an increase of mutant p53 S47 association with POLθ in unchallenged H1299 cells , which is further enhanced with replication stalling ( Figure 5A , 24 associations per cell without and 38 with HU ) . Notably , WT p53-POLθ associations remain limited even with replication stalling ( Figure 5A , average of 6 associations without and 15 with HU ) , suggesting pathway tipping toward mutagenic MMEJ in p53-defective cells . To test if MRE11-dependent restart is responsible for suppression of error-prone POLθ recruitment , we inactivated the nuclease by inhibition with the specific MRE11 nuclease inhibitor PFM39 ( Shibata et al . , 2014 ) . Inhibition of MRE11 by PFM39 greatly increased WT p53-POLθ association with replication stalling ( Figure 5A , an increase from 15 to 50/cell average with PFM39 ) . This observation suggests that MRE11 inactivation can partially pheno-copy p53 deficiency at replication forks . In contrast , p53 S47-POLθ associations were only moderately increased with PFM39 ( Figure 5A , increase from 38 to 53 with PFM39 ) , where PFM39 likely blocks residual MRE11 activity in p53 S47-expressing cells . Of note , we find POLθ interactions with RAD52 increased in GOF p53 R248W HCT116 cells ( Figure 5—figure supplement 1A ) , giving rise to the possibility that POLθ may collaborate with RAD52 in p53-defective cells rather than acting in separate pathways . To further test pathway imbalance specific to local stalled forks and dependent on p53 status , we performed SIRF against POLθ in HAP-1 cells ( Figure 5B ) . Consistently , we find increased POLθ recruitment to stalled forks in p53-null HAP-1 cells ( Figure 5B ) . Similarly , inactivation of MRE11 nuclease in WT HAP-1 cells causes a significantly increase in recruitment of POLθ to stalled forks . Together these data uncover p53-MRE11 repression of mutagenic RAD52 and POLθ processes at replication forks . RAD52/SSA and POLθ/MMEJ pathways allow the prediction of specific mutation signatures; SSA predominantly results in larger deletion mutations , while MMEJ is signified by microhomology at repair junctions along with deletions ( Jasin and Rothstein , 2013; Wood and Doublié , 2016 ) . We therefore hypothesized that p53 replication-defective cancers may leave a telltale mutagenic pathway signature in vivo . We tested this by comparing COSMIC mutational signatures ( Stratton et al . , 2009 ) of p53 defective with p53 WT breast cancers reported in the TCGA database ( Figure 6 , Cosine similarity cutoff: 0 . 617; z-score >1 . 96 ) . Seven mutational signatures are increased in p53-defective breast cancers ( Figure 6B ) . However , of these seven signatures , only signatures 3 and signature 5 are significantly increased in p53 defective compared to WT p53 breast cancers ( Figure 6—figure supplement 1B ) ; COSMIC signature 3 is defined by larger deletion mutations ( >3 bp ) with microhomology at break junctions , consistent with expected RAD52 and POLθ mutation spectra . COSMIC signature 5 shows T > C transition mutations at ApTpN context with thus far unknown etiology . POLθ was reported to have a stark preference for T > C transition mutations ( error-rate of 42 × 10−4 , 4–40 fold higher than any other possible mutation rate ) as seen within a known CATCC hotspot ( Arana et al . , 2008 ) . Thus , our combined data uncovers the possibility of POLθ mediated origin of COSMIC signature 5 . Collectively , we find COSMIC signatures are in agreement with increased RAD52 and POLθ pathway usage in p53 mutant breast cancers . Rather than single gene predisposition or select environmental exposure , the strongest drivers for cancer incidence are DNA replication errors . This has been long hypothesized and recently formalized by showing replication errors comprise 2/3 of all mutations in cancers ( Tomasetti et al . , 2017; Tomasetti and Vogelstein , 2015 ) . This fundamental importance of replication fork maintenance is conserved in bacteria , where stress response and repair proteins primarily protect and stabilize DNA replication forks ( Cox et al . , 2000 ) . p53 is the ‘the guardian of the genome’ and the most frequently mutated tumor gene , but its functions in replication genome stability , which is the dominant source of tumor mutations , has been cryptic . The most studied p53 cellular function with regard to tumor-suppression has been its role in transcriptional activation of apoptosis and cell cycle checkpoint . Yet , p53 functions during the DNA damage response linked to genome integrity are transcription activation independent ( Bertrand et al . , 2004; Janz and Wiesmüller , 2002; Saintigny and Lopez , 2002 ) . Moreover , these classical p53 transactivation activities to promote apoptosis and cell cycle arrest are insufficient to fully explain p53’s role in tumor suppression . This is substantiated by reported p53 separation-of-function mutations , including tumor prone yet greatly transcription activation proficient p53 mutations , such as S47 . Conversely , several p53 mutant mice including p53 R175P show that inactivation of apoptosis and senescence by p53 transcription deregulation are insufficient for full inactivation of p53 tumor-suppression functions ( Brady et al . , 2011; Li et al . , 2012; Liu et al . , 2004 ) . Taken together , these observations point to p53 activities in addition to its transcription activation functions that critically contribute to its tumor suppressor function . Such additional functions may include metabolism and ferroptosis , a new cell death pathway ( Li et al . , 2012; Zhu et al . , 2015 ) . We here identify a new p53 function in suppressing genome instability at replication forks by promoting MLL3/MRE11-mediated replication pathway homeostasis . Importantly , this activity , which we show is independent of p53 transcription activation roles , avoids mutagenic RAD52/POLθ pathways likely acting at reversed forks ( Figure 7 ) . As replication mutations are thought to be the strongest cancer mutation driver and genome instability is associated with tumorigenesis , we propose that the here identified role of p53 as a replication homeostasis keeper to avoid genome instability provides a feasible novel additional p53 tumor suppression function . Moreover , the resulting understanding of p53-mediated genomic stability reconciles previous reports on apoptosis and p53 transactivation-independent roles of p53 for tumor suppression ( Phang et al . , 2015 ) . So far , the most consistent common defect to both GOF mutant p53 and p53 gene deletion is related to its transcription function in apoptosis and cell cycle arrest . These results revealing a p53 replication-restart function reconcile how GOF and null p53 have different cellular functions and phenotypes , yet can both cause genomic instability implicated for tumor etiology and progression . Supporting this concept , MRE11 impairment , which we show phenocopies p53 defects at stalled forks , promotes progression and invasiveness of mammary hyperplasia in mouse models similar to p53 inactivation ( Gupta et al . , 2013 ) . Separation-of-cellular p53 function studies have spanned from apoptosis , cell cycle arrest , epigenetics and stress response to metabolism . Yet , these seemingly separate functions may act together in the context of WT p53 to guard the genome , foremost from genotoxic replication stress . As such , the replication restart function identified here conceptually connects seemingly divergent p53 functions including stress response , genome stability and epigenetics . We propose that upon activation by replication stress , p53 orchestrates balanced error-free replication restart and suppresses genome instability , which is caused by excessive usage of mutagenic replication pathways when p53 is defective . Importantly , this model implies p53 promotes a replication homeostasis balance at forks for successful proliferation rather than a strict pathway control . If replication stress exceeds a threshold for proper genome maintenance , p53 may dissociate and , as a keystone replication stress regulator , induce cell death , including but not exclusively through apoptosis , as an added safeguard to avoid cellular dysplasia . We here find the African-descent tumor variant p53 P47S ( S47 ) to be a separation of function mutation defective in replication restart . p53 is phosphorylated by ATM at S46 , which is decreased in p53 S47 ( Jennis et al . , 2016 ) . Intriguingly , at the adjacent residues D48/D49 , p53 can directly interact with single-strand binding protein RPA ( Romanova et al . , 2004 ) , which is implicated in replication fork remodeling ( Neelsen and Lopes , 2015 ) . Specifically , RPA interaction mutations deregulate recombination reactions without affecting transactivation reactions ( Romanova et al . , 2004 ) . By proximity of these residues and phenotypical commonalities , we suggest that p53 P47S ( S47 ) may also affect RPA interactions . By extension , we propose that ATM phosphorylation of WT p53 may regulate such p53-RPA interactions for the purpose of fork remodeling , as a controlled process for restart balancing . Notably , we find p53 S47 exhibits cellular sensitivity to the DNA cross-linking reagent mitomycin C , which most prominently activates the Fanconi Anemia tumor suppressor pathway . The latest identified Fanconi Anemia tumor suppressor is the RFWD3 ubiquitin ligase that regulates p53 ( Feeney et al . , 2017; Inano et al . , 2017 ) . Furthermore , a Fanconi Anemia phenotypes causing patient mutation in RFWD3 leads to deregulation of RPA reactions at the replication fork ( Inano et al . , 2017 ) . We therefore propose that p53 could feasibly be a vital player in the Fanconi Anemia pathway through its replication function , and it will be exciting to decipher this relationship . We establish here that at forks , p53 controls MRE11 , a nuclease known to promote restart after replication stalling ( Trenz et al . , 2006 ) . Other prominent p53 collaborating proteins including BLM helicase ( Davies et al . , 2007 ) , ATM ( Trenz et al . , 2006 ) , and PARP-1 ( Bryant et al . , 2009 ) all promote replication restart . Our results thus implicate p53 as a potential keystone regulator of a greater restart network at stalled forks involving fork-reversal regulation players ( Figure 7 ) . PARP , BLM helicase and RPA promote and repress replication fork reversal ( Neelsen and Lopes , 2015 ) . While it is unclear whether fork reversal is required for normal replication restart , it is readily observed in cancer cells at low concentrations of replication stalling agents ( Zellweger et al . , 2015 ) . In the absence of fork reversal control and stabilization by p53 regulated players , our model suggests that RAD52/POLθ pathways hijack the free DNA end to invade replication ahead or behind the replication fork as an intramolecular reaction . This could in principle lead to deletion and insertion mutations with micro-homology and increased POLθ dependent point mutations . In support of this model , we find COSMIC cancer mutation signature three signified by larger deletion mutations with micro-homologies increased in cancers with p53 defects . Additionally , we found COSMIC signature five to be increased in p53-defective breast cancers , which shows T > C transition mutations at ApTpN context with so far unknown etiology . However , our analysis shows that this signature is consistent with POLθ-mediated mutations: POLθ shows a striking preference for T > C transition mutations ( error-rate of 42 × 10−4 , 4–40 fold higher than any other possible mutation rate ) as seen within a known CATCC hotspot ( Arana et al . , 2008 ) . Based on our data , we therefore suggest that POLθ-mediated mutagenesis may contribute to COSMIC signature 5 etiology . In mammary tumor suppression , p53 cooperates with BRCA1/2 ( Jonkers et al . , 2001; Ludwig et al . , 1997 ) , which is often associated with more aggressive and resistant tumors , although the mechanism of this collaboration has been elusive . Therapy resistance in BRCA-defective cells can arise through secondary mutations . Specifically , gene internal deletion and/or point mutations within the BRCA2 gene can restore reading frames of BRCA2 mutated stop codons in CAPAN-1 pancreatic and POE ovarian cancer cells ( Sakai et al . , 2009; Sakai et al . , 2008 ) . Interestingly , while BRCA2 peptide expression is restored , some of these new BRCA2 peptides conferring resistance have extensive internal deletion mutations . Such deletion mutation etiology is consistent with both RAD52 and POLθ pathways providing a specific and testable mechanism for development of resistance . Our model of RAD52/POLθ pathway increase at stalled replication forks promoting secondary mutation driving resistance is further supported by our understanding of tumor resistance biology . Triple negative breast cancers as well as serous ovarian cancers , which almost exclusively harbor p53 mutations , are often initially sensitive to therapy such as cis-platin drugs ( Luvero et al . , 2014; Wahba and El-Hadaad , 2015 ) , which cause replication stalling . After multiple treatments and opportunities for RAD52/POLθ-mediated mutagenic events at stalled forks , secondary mutations can become fixed and in turn promote survival and resistance . In this scenario , mutation fixation is not necessary to promote a proliferative advantage per se . Rather it could arise from stochastic and opportunistic replication stalling events promoted by the therapeutic drug dependent on the replication program of the tissue type , consistent with both a neutral mutation evolution theory ( Sottoriva et al . , 2015 ) and replication errors driving tumor etiology ( Tomasetti et al . , 2017 ) . We here identified a new p53 role for suppressing genome instability by orchestrating balanced replication fork homeostasis . Importantly , this role is derailed in both p53 null and GOF p53 mutants , which is the only LOF ascribed to both aside from transcriptional deregulation of apoptosis and cell cycle checkpoint . Our observations and concepts reconcile prevailing paradoxes of divergent p53 functions . They , furthermore , imply specific changes in strategies for cancer patient care: our model suggests that inhibition of RAD52/POLθ pathways as adjuvant therapy concomitant with initial conventional therapy could offer an actionable strategy for ameliorating aggressive tumor evolution and secondary mutations leading to resistance in p53-defective tumors . The finding that p53 is a key-protein in error-free replication restart may explain why p53 mutations are a dominant cause of cancer genome instability . HAP-1 parental and HAP-1 TP53 null ( Horizon Discovery ) cells were grown in Iscove’s modified Dulbecco’s medium ( Life Technologies ) supplemented with 10% fetal bovine serum ( Gemini Bio products ) and 100 units/ml Pen-Strep ( Life Technologies ) . H1299 small lung cell carcinoma cells expressing doxycycline inducible human WT and S47 mutant p53 constructs were previously described ( Jennis et al . , 2016 ) . H1299 and HEK293 cells were grown in Dulbecco’s modified Eagle medium supplemented with 10% fetal bovine serum and 100 units/ml Pen-Strep . P53 protein expression was induced by 0 . 5 μg/ml Doxycycline ( Sigma-Aldrich ) . MCF10A p53 null cells were obtained from Thermo Fisher Scientific and grown in DMEM/F12 medium supplemented with 5% horse serum ( Gibco ) , 20 ng/ml EGF ( Thermo Fisher Scientific ) , 0 . 5 μg/ml hydrochortisone ( Sigma-Aldrich ) , 100 ng/ml Cholera toxin ( Sigma-Aldrich ) , 10 μg/ml Insulin ( Sigma-Aldrich ) , 5 mM Hepes ( Gibco ) , 100 units/ml Pen-Strep . MEF harboring p53 mutations R172P , R172H were previously described ( Liu et al . , 2004 ) , and MEF harboring p53 mutations S47 and WT p53 and p53 null MEF were previously described ( Jennis et al . , 2016 ) , and obtained from the Guillermina Lozano lab and the Maureen Murphy lab , respectively . MEFs were grown in Dulbecco’s modified Eagle medium supplemented with 10% fetal bovine serum and 100 units/ml Pen-Strep and 2 mM glutamine . MEFs were generated from C57BL/6J mice with mixed sex background . HCT116 parental and CRISPR engineered mutant cells ( R248W/- ) were obtained from Thermo Fisher Scientific and grown in McCoy’s 5a media ( Lonza ) with 10% fetal bovine serum and 100 units/ml Pen-Strep . Saos-2 cells complemented with GOF p53 mutants were previously described ( Xiong et al . , 2014 ) , provided by Dr . Guillermina Lozano’s lab and maintained in DMEM ( Life Technologies ) with 10% fetal bovine serum and 100 units/ml Pen-Strep . Cell lines have been authenticated by short tandem repeat ( STR ) profile analysis and genotyping , and have been tested for Mycoplasm ( PCR ) . All cells were grown at 37°C and 5% CO2 . DNA fiber spreading experiments were performed as previously described ( Schlacher et al . , 2011 ) . Briefly , cells were pulsed with EdU ( 5–125 μM ) , CldU ( 50 μM ) or IdU ( 50 μM ) , washed with PBS , and then incubated with hydroxyurea ( 200–400 μM ) and CldU ( 50 μM ) for 4–5 hr as indicated . The cells were harvested , resuspended in PBS and lysed on a microscope slide with lysis buffer ( 20 mM Tris-Cl , 50 mM SDS , 100 mM EDTA ) . DNA was allowed to attach for 5 . 5 min before spreading by gravity . Slides were fixed in methanol/acetic acid ( 3:1 ) , before DNA denaturation with 2 . 5 N HCl and neutralization with PBS ( pH 8 , and subsequent pH 7 . 5 washes ) . Slides were blocked with 10% goat serum and 0 . 1% Triton X in PBS . IdU/CldU fibers were stained using standard immunostaining with antibodies against IdU ( BrdU , Beckton Dickinson , 1:100 ) and CldU ( BrdU , Novus Biological , 1:200 ) was performed before mounting slides with Prolong Gold ( Invitrogen , USA ) . IdU/CldU Fibers were imaged using a Nikon Eclipse Ti-U inverted microscope and analyzed using ImageJ software . Between 90 and 320 fibers were scored per experiment and number of stalled forks was calculated as the number of IdU tracts ( green only ) divided by the number of IdU tracts plus the number of IdU-CldU tracts ( green followed by red ) . The number of newly initiated forks was calculated as the number of CldU tracts ( red only ) divided by the number of IdU tracts plus the number of IdU-CldU tracts ( green followed by red ) plus the number of CldU tracts ( red only ) . Cells were pulse treated with EdU , washed two times with PBS and subsequently treated with HU ( 0 . 2 μM ) for 4 hr . Cells were fixed , permeabilized with 0 . 25% TritonX , and a click-iT reaction was performed using biotin azide ( Life Technologies ) according to manufacturer’s instructions . After incubation with primary antibodies , a Duolink proximity ligation assay ( Sigma-Aldrich ) was performed with mouse/rabbit detection red reagents according to the manufacturer's instructions . Slides were stained with DAPI and mounted with Prolong Gold before imaging using Nikon Eclipse Ti-U inverted microscope . Signals were analyzed using Duolink software , ImageJ and Nikon NIS elements , in addition to hand-counting of PLA signals . Data of repeated experiments were combined , and statistical analysis was performed using Prism6 software . H1299 cells were treated with 0 . 5 µg/ml doxycycline ( Sigma-Aldrich ) for 48 hr to induce expression of WT and mutant p53 and subsequently treated with 100 μM PFM39 ( synthesized by the MD Anderson Cancer Center pharmaceutical chemistry core facility according to [Shibata et al . , 2014] ) for 30 min , followed by 0 . 2 mM HU for 4 hr , as indicated . Cells were fixed , permeabilized and blocked as described above and incubated with antibodies against p53 and POLθ as indicated . Finally , a Duolink PLA ( Sigma-Aldrich ) was performed according to manufacturer's instructions . Slides were stained with DAPI and mounted with Prolong Gold before imaging using Nikon Eclipse Ti-U inverted microscope . Signals were analyzed using Duolink software , ImageJ and hand-counted . Data of repeated experiments were combined , and statistical analysis was performed using Prism6 software . For western blots , cells were treated with 0 . 3 mM HU for 4 hr , harvested and directly lysed in Laemmli buffer ( Bio-Rad ) , boiled for 5 min and loaded on SDS-PAGE gels . Antibodies used for immunoblots in SIRF , PLA and iPOND are as follows: MLL3 ( Abcam 1:100 ) , MRE11 ( Abcam 12D7 1:200 ) , RAD52 ( Santa Cruz F7 1:50 ) , POL θ ( Abcam 1:100 ) , RAD51 ( Abcam 14B4 1:200 ) , mouse biotin ( Sigma-Aldrich BN-34 1:100 ) , rabbit biotin ( Cell Signaling D5A7 1:200 ) , p53 ( Santa Cruz DO1 , 1:1000 ) , ORC2 ( Abcam , SB46 , 1:1000 ) and PCNA ( Santa Cruz , PC10 , 1:1000 ) . Cells were incubated with 50 μM EdU for 1 hr and subsequently with 0 . 2 mM HU and 2 μg/ml cytochalasin B ( Sigma-Aldrich ) for 5 hr . Cells were then collected , washed and treated with cytochalasin B for 20 hr to further capture arrested cells after division that previously were EdU labeled . Post incubation , cells were harvested and spun onto slides using a cytospin for 3 min at low acceleration setting . Cells were then fixed , permeabilized and click-iT reaction was performed with Alexa fluor 488 azide according to manufacturer’s instructions . Slides stained with DAPI and mounted with Prolong Gold before imaging using Nikon Eclipse Ti-U inverted microscope . EdU-positive cells and micronuclei were scored manually and using ImageJ software . Prism was used for statistical analysis of combined repeat experiments . Cell viability was determined using the colorimetric MTS assay . Cells ( 1–2 × 103 cells ) were seeded into 96-well plates for 24 hr and then exposed to varying concentrations of HU or MMC ( Sigma-Aldrich ) as indicated . After untreated control cells obtained ~80% confluence , the MTS assay was performed according to manufacturer’s instructions ( CellTiter 96 AQueous One Solution Cell Proliferation Assay , Promega ) . Experiments were performed in quadruplicate and repeated independently . Data was analyzed using Prism6 software and represents the mean ± standard error of the mean ( SEM ) . The mutation annotation file ( MAF ) for 992 samples was downloaded from BROAD TCGA GDAC website ( http://firebrowse . org/ ? cohort=BRCA&download_dialog=true , https://cancergenome . nih . gov ) . The mutation spectrum of each sample was estimated by calculating the fraction of 96 possible mutation substitutions defined in ( Alexandrov et al . , 2013 ) The cosine similarity score is computed for all pair-wise combinations of mutation spectrum of samples and 31 cosmic mutation signatures ( http://cancer . sanger . ac . uk/cosmic/signatures ) . Z-score is calculated based on the distribution of all cosine similarity score ( z_score=cos_score−mean ( cos_score ) sd ( cos_score ) ) . A z score greater than 1 . 96 indicates the sample could contain the corresponding cosmic signature . For SIRF assays , PLA signals were analyzed using Duolink Image Tool software and Nikon NIS elements software . A total of 50–300 nuclei were counted for each experimental condition . Data represents pooled experiments of two to four experiments . Signals were normalized to independent EdU-PLAs of the same condition ( Supplementary file 1 ) and a T-test to determine the Z-score and p-value for significance was performed using the following equation: z = [mean ( EdU-SIRF1 ) - mean ( EdU-SIRF2 ) ] - [mean ( SIRF1 ) - mean ( SIRF2 ) ]/ √[Variance ( EdU-SIRF1 ) /n + Variance ( EdU-SIRF2 ) /n + Variance ( SIRF1 ) /n + Variance ( SIRF2 ) /n] , whereby n is the number of measurements . The resultant p-values are indicated in the respective figures and figure legends . For DNA fiber assays , between 90 and 300 fibers were analyzed using ImageJ software . Unpaired Student t-test was performed using GraphPad Prism version six as indicated in the figures and figure legends . For genomic instability were analyzed using NIS elements software . Unpaired Student t-test was performed using GraphPad Prism version six to determine p value results as indicated in the figures and figure legends . For TCGA Computational Analaysis , Fisher Exact Test was calculated using GraphPad Prism software .	When a cell divides to make more cells , it duplicates its DNA to pass on an identical set of genes to the new cell . Copying DNA – also known as DNA replication – is a complex process that involves several steps . First , the double helix gradually unwinds and unzips to separate the DNA strands . This creates a molecule known as the ‘replication fork’ . Then , copies of each strand are created and proofread for errors . Eventually , the strands are sealed back together so that the helices contain one old and one new part . But sometimes errors sneak in during DNA replication , which can lead to mutations that may cause cancer . The higher the number of mutations , the bigger the chance is that the cancer becomes aggressive and resistant to therapy . Some of the most common mutations found in tumors happen in a protein called p53 . This protein is known to stop tumors from growing by selectively killing cells with mutations . When p53 is faulty , mutant cells no longer die and can grow uncontrollably to form tumors . However , its killing abilities do not fully explain how p53 protects cells from accumulating mutations that can cause cancer , and until now , it was not known if p53 also had any other roles . Now , Schlacher et al . discovered that p53 can protect the DNA from mutations . The experiments used normal cells and cancer cells from humans and mice , in which p53 was either blocked or modified . The experiments revealed that p53 plays an important role during DNA replication . When p53 is ‘healthy’ , it binds to the replication fork . This ensures that replication restarts properly after it has passed faulty patches of the DNA . The p53 protein also helps to organize the proteins involved in DNA replication . When p53 was absent or mutated , the DNA-repair protein that usually binds to the fork failed to attach properly . Instead , other proteins prone to make mutations took over the replication fork and created a pattern of mutations commonly found in tumors resistant to treatment . A next step will be to investigate p53’s role at damaged DNA replication forks and how it interacts with other proteins involved in DNA replication . To fully understand all roles that p53 plays in preventing tumor growth can help to find new ways to treat tumors with p53 defects or tumors that have become resistant to treatment .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression" ]	2018	p53 orchestrates DNA replication restart homeostasis by suppressing mutagenic RAD52 and POLθ pathways
Mutations in the polycystin proteins , PC-1 and PC-2 , result in autosomal dominant polycystic kidney disease ( ADPKD ) and ultimately renal failure . PC-1 and PC-2 enrich on primary cilia , where they are thought to form a heteromeric ion channel complex . However , a functional understanding of the putative PC-1/PC-2 polycystin complex is lacking due to technical hurdles in reliably measuring its activity . Here we successfully reconstitute the PC-1/PC-2 complex in the plasma membrane of mammalian cells and show that it functions as an outwardly rectifying channel . Using both reconstituted and ciliary polycystin channels , we further show that a soluble fragment generated from the N-terminal extracellular domain of PC-1 functions as an intrinsic agonist that is necessary and sufficient for channel activation . We thus propose that autoproteolytic cleavage of the N-terminus of PC-1 , a hotspot for ADPKD mutations , produces a soluble ligand in vivo . These findings establish a mechanistic framework for understanding the role of PC-1/PC-2 heteromers in ADPKD and suggest new therapeutic strategies that would expand upon the limited symptomatic treatments currently available for this progressive , terminal disease . The most common monogenetic disease in humans is ADPKD; a major nephropathy characterized by renal cysts that leads to urinary tract infection , hypertension , aneurysm , and ultimately end-stage renal disease ( ESRD ) ( Kathem et al . , 2014; Harris and Torres , 2009 ) . Mutations in one of two ciliary proteins , PC-1 ( encoded by PKD1 ) and PC-2 ( also known as TRPP1 and encoded by PKD2 ) , account for 85% and 15% of ADPKD-causing mutations , respectively ( Harris and Torres , 2009; Cornec-Le Gall et al . , 2014 ) . PC-1 is an 11 transmembrane ( TM ) protein of 4303 amino acids with a long extracellular N-terminus that contains multiple cell adhesion domains ( Yoder et al . , 2002; Hughes et al . , 1995 ) . PC-1 and its family members ( PC-1L1 , PC-1L2 , and PC-1L3 ) share the same 11-TM topology but vary in the size and adhesion domain composition of their N-termini ( Ishimaru et al . , 2006 ) . A unique feature of PC-1 family members , shared only by adhesion G protein-coupled receptors ( GPCRs ) , is the GPCR autoproteolysis inducing ( GAIN ) domain . This domain catalyzes peptide bond hydrolysis at a GPCR proteolysis site ( GPS ) that is positioned proximal to the first TM helix ( TM1 ) ( Araç et al . , 2012; Prömel et al . , 2013; Trudel et al . , 2016 ) . Autoproteolysis generates two fragments: an N-terminal fragment ( NTF ) and a multi-TM core ( Qian et al . , 2002 ) . The NTF contains multiple domains , including leucine-rich repeats , a C-type lectin ( CTL ) domain , immunoglobulin ( Ig ) -like polycystic kidney disease ( PKD ) repeat domains , a single low-density lipoprotein ( LDL ) receptor motif , and the receptor for egg jelly ( REJ ) domain ( Hughes et al . , 1995; Babich et al . , 2004; Moy et al . , 1996 ) . The function of the multi-TM core remains poorly defined . Autoproteolysis occurs early in the ER secretory pathway such that the N- terminal fragment remains non-covalently tethered ( Lin et al . , 2004; Wei et al . , 2007 ) . While mutations within the GPS site of PC-1 result in renal cyst formation , the full functional consequences of autoproteolysis are poorly understood ( Qian et al . , 2002; Yu et al . , 2007 ) . Recent studies suggest that PC-1 and PC-2 form a 1:3 heteromer ( Su et al . , 2018; Yu et al . , 2009 ) . Indeed , the cryo-EM structure of a PC-1/PC-2 heteromer reveals that the 11 transmembrane helices of PC-1 are further divided into two major domains: a peripheral TM1-TM5 complex and a core TM6-TM11 complex that interdigitates with three PC-2 subunits to form a TRP-like ion channel ( Su et al . , 2018 ) . In addition to this PC-1/PC-2 assembly , PC-2 can form homomeric channels , following a classic homotetramer of TRP channels ( Shen et al . , 2016; Grieben et al . , 2017; Wilkes et al . , 2017 ) . Attempts to exogenously express and record from PC-1 and/or PC-2 in whole cell recordings of mammalian cells ( Hanaoka et al . , 2000; Delmas et al . , 2004; Cai et al . , 2004 ) and reconstituted lipid bilayers González-Perrett et al . , 2001 have yielded divergent results , likely due to subcellular enrichment of PC-1 and PC-2 ( for a detailed discussion see Liu et al . , 2018; Kleene and Kleene , 2017 ) . In addition , despite recent advances in measuring ion channel activity in small subcellular organelles such as the cilium ( Kleene and Kleene , 2012; DeCaen et al . , 2013; Dong et al . , 2008 ) , it remains technically challenging to study ion channel activity in primary cilia due to their small size ( 10−15 L , approximately the size of a bacterium ) . This technical hurdle has hindered investigations of the physiological regulation of polycystins and how they lead to renal disease when mutated . The unknown role of PC-1 within the channel complex , in particular , has hindered progress in the field . Some studies suggest that PC-1 acts as a dominant-negative subunit of the heteromeric channel ( Su et al . , 2018 ) , whereas others suggest that it is a pore-forming subunit that fine-tunes ion selectivity ( Wang et al . , 2019 ) . Moreover , PC-1 has been shown to be dispensable for polycystin currents in ciliary membranes ( Liu et al . , 2018 ) . Here we develop cellular assays that allow us to probe the function of polycystin complexes in the plasma membrane and compare them to polycystin channels in their native primary cilia . Our heterologous system recapitulates the biophysical properties of cilia-localized polycystin channels and demonstrates how distinct PC-1 family members contribute to ion permeation . Importantly , we reveal that the N-terminal domain of PC-1 functions as a soluble ligand and is an indispensable activator of the polycystin complex , providing further evidence for chemosensory regulation of the polycystin complex ( Delling et al . , 2016 ) . Because robust assays to investigate the heteromeric polycystin complex are lacking , we developed an assay system to determine the contribution of PC-1 family members to polycystin function . We were able to target PC-1 to the plasma membrane of HEK293 cells ( Figure 1B ) and mIMCD3 cells ( Figure 1—figure supplement 1E ) by replacing its endogenous signal peptide with a strong Ig κ-chain secretion sequence followed by an HA tag ( sPC-1; Figure 1A , Table 1; Salehi-Najafabadi et al . , 2017; Zhu et al . , 2011 ) . The HA tag allowed us to quantify sPC-1 plasma membrane expression by measuring surface staining following addition of anti-HA antibodies to non-permeabilized cells . Co-expression of sPC-1 and PC-2 resulted in strong HA surface staining , whereas expression of sPC-1 alone resulted in only marginal surface staining , in both HEK293 cells ( Figure 1B and C ) and IMCD3 cells ( Figure 1—figure supplement 1D and E ) . Co-expression of PC-2 and PC-1 with an extracellular HA tag but without the endogenous leader sequence ( HAPC-1 ) failed to generate comparable HA surface staining , supporting the notion that an Ig κ-secretion sequence promotes membrane localization of the complex ( Figure 1B ) . To test whether PC-2 facilitates sPC-1 membrane localization by promoting ER exit or by co-migration to the plasma membrane , we inserted a FLAG epitope into the extracellular tetragonal opening for polycystins ( TOP ) domain of PC-2 ( PC-2FLAG ) ( Figure 1—figure supplement 1A , Table 1 ) . Robust surface HA and FLAG staining only occurred when both sPC-1 and PC-2FLAG were co-expressed , suggesting that they are shuttled simultaneously to the plasma membrane ( Figure 1—figure supplement 1B and C ) . Similarly , co-expression of sPC-1 with PC-2 containing a recently in an oocyte expression system characterized gain of function mutation of PC-2 ( PC-2F604P , Table 1; Arif Pavel et al . , 2016 ) resulted in comparable HA surface expression ( Figure 1B and C ) . Having established robust plasma membrane expression of sPC-1/PC-2 and sPC-1/PC-2F604P , we asked whether these proteins form functional channels using whole cell patch clamp recordings . Interestingly , only sPC-1/PC-2F604P produced constitutively active , outward rectifying currents ( 122 . 6 ± 23 . 0 pA/pF at +100 mV , n = 7 ) with a half-maximal activation voltage of +112 . 5 mV ( Figure 1D , E and G ) . sPC-1/PC-2 ( 5 . 1 ± 1 . 0 pA/pF , n = 11 ) , HAPC-1/PC-2F604P ( 5 . 0 ± 2 . 3 pA/pF , n = 15 ) and PC-2F604P alone ( 6 . 4 ± 2 . 1 pA/pF , n = 12 ) yielded negligible currents at +100 mV ( Figure 1F ) . To investigate the functional contribution of PC-1 to heteromeric polycystin complexes , we mutated three positively charged amino acids in the putative ion permeation pathway to the small , uncharged amino acid glycine ( R4100G , R4107G , and H4111G ) in order to facilitate cation permeation ( sPC-1pore; Figure 1H , Table 1 ) . In combination with PC-2F604P , currents of 71 . 1 ± 5 . 4 pA/pF ( n = 6 ) were generated at +180 mV ( sPC-1pore/PC-2F604P; Figure 1I ) . Although the recorded current density was ~50 pA/pF smaller than for sPC-1/PC-2F604P ( Figure 1J ) , the relative open probability at −100 mV was higher ( 0 . 3 ± 0 . 0 for sPC-1pore/PC-2F604P; 0 . 1 ± 0 . 0 for sPC-1/PC-2F604P ) , suggesting that the pore mutant resides in a more open state ( Figure 1K ) . Collectively , these results demonstrate that PC-1/PC-2 subunits form a functional channel when reconstituted in mammalian cells and support the notion that PC-1 subunits form part of the core channel complex . We sought to determine whether PC-1 subunits are involved in the regulation of the polycystin complex by investigating the functional role of the N-terminal domain . We reasoned that the N-terminus must be a critical determinant of polycystin function because ADPKD-causing mutations occur with high frequency in this region and therefore investigated a family member with a divergent N-terminus , PC-1L3 ( Figure 2A and Figure 2—figure supplement 1A ) . Substitution of PC-1L3’s signal peptide with the κ IgG secretion sequence ( sPC-1L3 , Table 1 ) resulted in strong surface HA staining when co-expressed with PC-2 or PC-2F604P ( Figure 2B and D ) . However , we were unable to record convincing currents in sPC-1L3/PC-2F604P-transfected HEK cells ( 3 . 5 ± 0 . 0 pA/pF , n = 12 ) , suggesting that , although present in the plasma membrane , the complex is inactive ( Figure 2C ) . To determine the structural domains in PC-1 subunits that are required for channel activation , we generated chimeras between PC-1 and PC-1L3: the N-terminus of PC-1 fused to the 11-TM core of PC-1L3 preceding the GPS cleavage site ( sPC-1L31NT ) and the opposing arrangement ( sPC-11L3NT ) ( Figure 2A , Table 1 ) . We also substituted the entire PC-1 N-terminus with the 26 amino acid-long N-terminus of the P2Y12 receptor and an extracellular HA tag ( PC-1ΔNT , Table 1 ) ; a modification known to increase surface expression of GPCRs without impairing functionality ( Liberles and Buck , 2006; Liebscher et al . , 2015 ) . All chimeras localized to the plasma membrane when co-expressed with PC-2F604P ( Figure 2B and D ) . PC-1ΔNT resulted in the strongest surface staining , likely due to the smaller size of this protein . Of the three chimeras , only sPC-1L31NT/PC-2F604P generated small but appreciable outward currents ( 35 . 9 ± 12 . 9 pA/pF , n = 5 ) . Currents in cells expressing sPC-11L3NT/PC-2F604P ( 9 . 1 ± 1 . 9 pA/pF , n = 10 ) and PC-1ΔNT/PC-2F604P ( 7 . 6 ± 1 . 5 pA/pF , n = 10 ) were negligible ( Figure 2C and E ) . Furthermore , the relative open probability of the sPC-1L31NT/PC-2F604P chimera at −100 mV ( 0 . 44 ± 0 . 0 ) was ~four-fold higher than sPC-1/PC-2F604P ( 0 . 1 ± 0 . 0 ) and similar to that for sPC-1pore/PC-2F604P ( 0 . 3 ± 0 . 0 ) ( Figure 2F ) . Interestingly , the predicted ion permeation region of PC-1L3 does not contain the positively charged amino acids found in PC-1 , offering a possible explanation for the greater open probability observed in the sPC-1L31NT/PC-2F604P chimeras ( Figure 2—figure supplement 1A ) . In terms of channel regulation , these results support the hypothesis that the N-terminus of PC-1 is a key determinant of polycystin complex activation . To gain a more detailed understanding of the role played by the N-terminus in polycystin complex activation , we replaced smaller regions of the PC-1L3 N-terminus with those from PC-1 . One chimera replaced the 108 AA CTL domain of PC-1L3 with the CTL of PC-1 ( sPC-1L31CTL; Figure 2A , Table 1 ) . sPC-1L31CTL/PC-2F604P generated robust outward currents ( 87 . 5 ± 9 . 0 pA/pF , n = 5 ) ( Figure 2C and G ) whereas the inverse chimera ( sPC-11L3CTL ) produced only minimal currents ( 9 . 1 ± 1 . 9 pA/pF , n = 10 ) . These data strongly suggest that the PC-1 CTL domain is essential to activate the polycystin complex . In mammalian cells , the PC-1 ectodomain undergoes at least one proteolytic cleavage event at the GPS site , suggesting that the N-terminus might be liberated as a secreted ligand . Indeed , it has been reported that PC-1 N-terminus-containing exosomes interact with primary cilia in kidney nephrons ( Hogan et al . , 2009 ) . We therefore investigated whether fragments of the PC-1 N-terminus can regulate the polycystin complex . We purified various PC-1 NTFs from HEK cell supernatant ( Figure 2—figure supplement 1C–1E ) and tested their ability to confer channel activity on PC-1ΔNT/PC-2F604P heteromers when included in the bath solution . Surprisingly , application of an NTF spanning the distal region AA 24–852 ( NTF24-852 ) , including the CTL domain between PKD I and II ( amino acids 415–531; Figure 2—figure supplement 1B and C ) , conferred the ability to carry outward currents on PC-1ΔNT/PC-2F604P heteromers ( 65 . 5 ± 37 . 2 pA/pF , n = 6 ) ( Figure 3A and C ) . Similarly , a shorter NTF containing the CTL domain ( NTF263-535; Figure 2—figure supplement 1B and C ) was also able to induce outward currents ( 61 . 4 ± 10 . 5 pA/pF , n = 4; Figures 3A , B and C ) . In both cases , the currents were smaller than those for sPC-1/PC-2F604P ( Figure 1F ) . Boiling NTF263-535 at 95°C for 10 min rendered it inactive ( Figure 3B ) . Next , we perfused HEK cells overexpressing PC-1ΔNT/PC-2F604P heteromers with NTF263-535 and measured acute changes in current density . While PC-1ΔNT/PC-2F604P channels remained inactive in the absence of NTF263-535 , acute application increased current density by 765 . 1 ± 185 . 05% ( from 5 . 9 ± 1 . 3 pA/pF to 59 . 1 ± 19 . 3 pA/pF , n = 7 ) within 10 s of application ( Figures 3D , F and G ) . However , non-transfected cells did not respond to NTF263-535 under comparable condition ( 4 . 6 ± 1 . 6 pA/pF to 6 . 8 ± 1 . 4 pA/pF , n = 9 ) ( Figure 3E and G ) . We then asked whether the removal of PC-1 N-terminus inactivates the polycystin heteromer . To test this possibility , we perfused sPC-1/PC-2F604P overexpressing HEK cells with 0 . 125% trypsin , assuming that proteolytic digestion of extracellular proteins also removes critical regions within the PC-1 N-terminus . As shown in Figure 3H perfusion with 0 . 125% trypsin decreased sPC-1/PC-2F604P dependent currents by 78 . 4 ± 4 . 9% ( from 52 . 0 ± 17 . 1 pA/pF to 8 . 9 ± 1 . 4 pA/pF , n = 6 , Figure 3H and J , and K ) . By contrast , untransfected cells did not respond to 0 . 125% trypsin ( 7 . 2 ± 1 . 9 pA/pF versus 4 . 6 ± 0 . 8 pA/pF , n = 5 ) . Taken together , these data suggest that the N-terminus including the CTL of PC-1 is necessary and sufficient to activate the polycystin complex . Next , we asked how the extracellular CTL domain in PC-1 might regulate activation of the entire channel complex . We reasoned that the TOP domain in PC-2 is well positioned to function as a surface receptor as it forms a bridge between the pore domain and voltage sensor-like domain , and is critical for channel activation ( Shen et al . , 2016; Vien et al . , 2020 ) . In addition , different glycosylation patterns of the TOP domain might underlie conformational changes in the homomeric PC-2 complex ( Wilkes et al . , 2017 ) . We observed prominent glycosylation of the TOP domain in PC-2 at position N375 , N362 , and N328 ( Figure 4A ) . Moreover , although mutation of N375 to glutamine did not affect surface localization of sPC-1/PC-2F604PN375Q ( Figure 4B and C ) , it rendered the channel inactive ( Figure 4D and E ) . Thus , glycosylation of the TOP domain in PC-2 appears to be required for CTL-dependent polycystin activation . Although the results so far strongly suggest that the PC-1 CTL domain is critical for activating heterologous polycystin complexes , native polycystin complexes reside in the specialized membranes of primary cilia ( Liu et al . , 2018; Kleene and Kleene , 2017; Vien et al . , 2020 ) . We , therefore , investigated whether the PC-1 N-terminus is able to activate polycystin complexes in the cilia of polarized mouse epithelial ( mIMCD-3 ) cells using excised ciliary membrane patch clamp recordings ( Figure 5—figure supplement 1B ) . Without any NTFs , we readily detected single channel openings at potentials more positive than +60 mV and more negative than −80 mV ( Figure 5A ) in 7 out of 38 cilia . After ablating PC-1 expression in mIMCD3 cells using CRISPR/CAS9 ( Figure 5—figure supplement 1A ) ciliary channels remained readily detectable in 4 of 13 cilia ( Figure 5B ) , in agreement with previous reports ( Liu et al . , 2018 ) . In addition , single channel conductance was comparable between wt mIMCD3 ( γ = 79 . 5 ± 0 . 1 pS ) and PC-1 knockout cilia ( γ = 82 . 3 ± 0 . 0 pS ) ( Figure 5—figure supplements 1G , H , 2A and B ) . However , we noted that the open probability of channels at negative membrane potentials in PC-1 knock out primary cilia ( 0 . 58 ± 0 . 01 ) was higher than in wt IMCD3 cilia ( 0 . 08 ± 0 . 01 ) ( Figure 5G ) . These data suggest that PC-2 forms a homomeric complex in the absence of PC-1 subunits with a higher open probability than that of PC-1/PC-2 heteromers . The results also support the idea that PC-1 subunits impede cation flux across the membrane , in agreement with our PC-1 pore mutant data ( Figure 1I–K ) . Ablation of PC-2 expression using CRISPR/CAS9 rendered primary cilia electrically silent ( n = 11 ) , confirming that PC-2 subunits are an essential component of polycystin channels ( Figure 5C; Liu et al . , 2018; Kleene and Kleene , 2017 ) . To test whether NTFs can activate the native polycystin complex , we recorded single channels in inside out patches pulled from primary cilia in the presence of 0 . 7 μg/mL ( equal to ~50 nM final concentration ) of NTF263-535 in the recording pipette . Although the percentage of electrically active cilia ( n = 4/13 ) remained largely unaffected , NTF263-535 increased the probability of single channels openings by ~five-fold and significantly prolonged the duration of channel openings at negative membrane potentials ( Figure 5E–G ) . Collectively , these findings demonstrate that PC-2 subunits are essential for functional polycystin channels in primary cilia , that PC-1 subunits also participate in the native heteromeric complex , and that NTF263-535 can activate this complex under physiological conditions . Having established that NTF263-535 can activate ciliary polycystin channels , we asked whether it could also trigger Ca2+ entry through the channel . We initially attempted single channel recordings from mIMCD-3 wt cilia under conditions in which Ca2+ is the sole charge carrier ( see Materials and methods ) . We were unable to measure currents , even at very negative membrane potentials ( −140 mV ) ( Figure 6A ) , in agreement with previously published results ( Liu et al . , 2018 ) . Surprisingly , however , the application of NTF263-535 revealed channel openings close to physiological membrane potentials ( Figure 6B ) . Single channel conductance was smaller when using Ca2+ instead of Na+ and K+ as the charge carrier ( 47 . 5 pS , Figure 6C ) . Voltage-dependent open probability and open time remained unchanged ( Figure 6D and E ) . We subsequently tested whether NTF263-535 can also induce changes in ciliary Ca2+ concentration ( [Ca2+]cilium ) using an improvement to our recently-published ciliary ratiometric Ca2+ sensors ( Delling et al . , 2016; Delling et al . , 2013 ) . We fused mScarlet and GCaMP6s to Arl13b ( Arl13b-sca-G6 ) and stably expressed the construct in both wt and PC-2 knockout mIMCD-3 cells . Cells growing on the bottom side of transwell inserts were imaged using TIRF microscopy ( TIRFM ) to detect localized changes in [Ca2+]cilium ( Ishikawa and Marshall , 2015 ) . Application of NTF263-535 generated multiple rapid spike-like elevations of [Ca2+]cilium ( ΔF/F > 0 . 35 , n = 35 cilia ) whereas untreated cells ( n = 22 ) and cells treated with boiled NTF263-535 ( n = 28 ) did not exhibit such fluctuations ( Figure 6F and G ) . In addition , NTF263-535 did not elicit [Ca2+]cilium fluctuations in PC-2 knockout cells ( n = 15; Figure 6H and I ) . These results show that polycystin channels conduct Ca2+ when activated by NTF263-535 . Our data identify PC-1’s CTL domain as an essential component for polycystin activation . Interestingly , the CTL domain is a hotspot for ADPKD-causing mutations ( Dong et al . , 2019 ) , supporting the critical importance of this domain ( Figure 2—figure supplement 1B ) . Although C-type lectins are the most diverse family of mammalian carbohydrate-binding proteins ( Keller and Rademacher , 2020 ) , little is known about the function of the CTL motif within PC-1 ( Sandford et al . , 1997 ) , but it has been reported to bind to carbohydrates in a calcium-dependent manner ( Weston et al . , 2001 ) . All PC-1 family members except PC-1L1 contain an N-terminal CTL , yet PC-1L3 CTL fails to activate the complex . Because the CTL amino acid sequence diverges between different PC-1s , we speculate that each binds to different carbohydrates . It is worth noting that other domains within the N-terminus are also implied in cell-cell recognition , including leucine-rich repeats ( LRRs ) , cell wall integrity and stress response component ( WSC ) , and REJ domain . Future studies addressing the specificity and mechanism of CTL binding , especially to carbohydrate moieties , will be of interest . Both PC-1 and PC-2 have a large extracellular TOP domain , which contains prominent glycosylation sites . We speculate that the CTL domain in the native PC-1/PC-2 complex interacts with glycans in the TOP domain , which allosterically regulates the ion permeation pathway and/or gating apparatus of the heteromeric polycystin channel ( Vien et al . , 2020; Figure 7 ) . In particular , the N-glycan attached to N375 of PC-2 is adjacent to a structure that extends above TM3 and TM4 of the voltage sensor-like domain , so it is conceivable that binding of the CTL domain to this glycan will initiate conformational changes in the voltage sensor-like domain , which could be transmitted to the pore to regulate ion permeation . Indeed , ablation of this glycan by mutation ( N375Q ) results in a complex that can traffic to the plasma membrane but that remains inactive . Signal peptides are cleaved from the nascent polypeptide early in the biosynthetic pathway , thus sPC-1 , which is ectopically targeted to the plasma membrane , closely resembles wt PC-1 ( Salehi-Najafabadi et al . , 2017 ) . Likewise , the effective membrane targeting sequence found in the N-terminus of P2Y12 has been successfully used to target adhesion GPCRs to the plasma membrane without compromising functionality ( Liebscher et al . , 2015 ) . The PC-2 F604P mutant has recently been characterized as a gain of function mutant in a Xenopus oocyte overexpression model ( Arif Pavel et al . , 2016 ) . It is thought that the F604P substitution within S5 induces conformational changes in the S4-S5 linker ( Zheng et al . , 2018 ) that lock PC-2 in an open state . Indeed , proline substitutions in the S5 segment of several other TRP family members result in constitutive channel activity ( Grimm et al . , 2007; Xu et al . , 2007; Zhou et al . , 2017 ) . A careful comparison of our ciliary recordings of endogenous polycystin channels with membrane-targeted polycystins reveals that both channel populations exhibit comparable biophysical characteristics , lending weight to the effectiveness of our plasma membrane expression system for studying polycystins . Our results are in agreement with previous reports of channel activity in the ciliary membrane of renal epithelial cells ( Liu et al . , 2018; Kleene and Kleene , 2017; DeCaen et al . , 2013; Flannery et al . , 2015 ) . Because no measurable electrical activity remains in mIMCD-3 cells following CRISPR ablation of PC-2 , we confirm that the PC-2 subunit is an indispensable component of ciliary currents in kidney epithelial cells ( Liu et al . , 2018; Kleene and Kleene , 2017 ) . We also find that primary cilia of mIMCD-3 cells still contain functional channels after CRISPR ablation of PC-1 . We speculate that currents recorded from PC-1 knockout cilia are due to homotetrameric PC-2 channels , although we cannot exclude a contribution from other PC-2-containing heteromeric channels such as PC-1L1/PC-2 , PC-1L3/PC-2 , or PC-2/TRPM3 ( Field et al . , 2011; DeCaen et al . , 2013; Flannery et al . , 2015; Bai et al . , 2008; Kleene et al . , 2019; Zhang et al . , 2013 ) . We noted two striking differences between channels in wt and PC-1 knockout mIMCD-3 cilia at voltages close to the resting membrane potential ( DeCaen et al . , 2013 ) . First , the native PC-1/PC-2 heteromeric channel in wt cilia showed very short channel openings compared to the presumably homotetrameric PC-2 channel , supporting the hypothesis raised by cryo-EM data that PC-1 reduces ion flux ( Su et al . , 2018 ) . Second , wt IMCD-3 cilia were electrically silent between membrane potentials of −60 mV and +40 mV whereas channels in PC-1 knockout cilia opened at −40 mV . In our experiments , soluble fragments containing the CTL motif not only restore channel function in polycystin complexes lacking the PC-1 N-terminus but also lengthen the open time duration of endogenous polycystins at negative membrane potentials . This strongly suggests that the PC-1 N-terminus is an agonist for polycystin activation . Our results provide the first insights into why ADPKD-causing mutations are enriched within both CTL and GAIN domains , the latter of which is critical for proteolytic processing and subsequent shedding of the CTL-containing ectodomain . Our results show that soluble fragments of PC-1 potentiate endogenous polycystin channels that already contain one PC-1 N-terminus . Given the predicted three PC-2:1 PC-1 stoichiometry of the heteromeric polycystin complex , we hypothesize that possibly all three PC-2 TOP domains need to be simultaneously engaged by CTL domains in order to trigger a concerted conformational change in their respective voltage sensor-like domains and therefore full channel activation . Alternatively , the tethered PC-1 N-terminus in a native heteromeric polycystin complex may be sterically constrained to participate in intramolecular interactions with its core complex , and thus might engage in intermolecular interactions , either between polycystin complexes on one cilium or as a cilia-cilia proximity sensor . Alternatively , as-yet-unknown proteins and/or small molecule ligands can bind to modules ( such as LRRs ) that are in proximity to or in the CTL domain itself . This might mask the carbohydrate-binding site on the CTL domain and thus add another layer of regulation to the polycystin complex . While our results show that CTL is required for channel activation , we cannot exclude the possibility that other more potent agonists of the polycystin complex exist . Future work will determine how the PC-1 N-terminus functions as a polycystin agonist in vivo , either as a soluble ligand or bound to vesicles such as exosomes . We speculate that transport of such ligands by fluids may underline the molecular chemosensory mechanism of primary cilia . Our data reveal that activation of endogenous polycystins by N-terminal fragments of PC-1 increases ciliary [Ca2+] in ~30% of cilia imaged . This is in agreement with previously published work in which only a fraction of cilia shows electrically active polycystin channels . It remains to be determined whether and how agonistic ciliary second messengers , which may only be stochastically present during ciliary recordings , add another layer of regulation to ciliary polycystin channels . Alternatively , there may be an as-yet-unidentified inhibitory signal in cilia that prevents activation of the polycystin complex in some ciliary recordings . hArl13B-EGFP was described previously . Arl13b-scarlet-GCaMP6s is an updated version of our previously characterized ciliary Ca2+ sensor Arl13b-mCherry-GECO1 . 2 in which mcherry and GECO1 . 2 have been replaced with mScarlet and GCaMP6s , respectively . κ IgG HA-PC-1L3 ( sPC-1L3 ) has been described previously ( Salehi-Najafabadi et al . , 2017 ) . sPC-1 was generated by using a unique BglII site within hPKD1 to replace the endogenous signal peptide with the 12 aa k IgG leader sequence followed by HA tag using PCR amplification and Gibson assembly ( NEB ) . HAPC-1 was generated by inserting an HA tag between the predicted endogenous signal peptide and mature polypeptide chain following the same protocol . For the construction of FLAG-PC-2 , several positions within the extracellular domains of PC-2 with negatively charged amino acids were tested . We identified one region at position 372 that tolerated the insertion of the FLAG sequence , based on surface trafficking . Chimera constructs of PC-1 and PC-1L3 were generated by substituting the ectodomain at the GPS cleavage site with the corresponding orthologue using PCR and Gibson assembly . All PC-1 and PC-2 combinations were cloned into pTRE3G-Bi vector ( Takara Bio ) . The MCSI site was used for PC-1 variants while the MCSII site was used for PC-2 variants . This ensured simultaneous translation of PC-1 and PC-2 from the same plasmid . All DNA sequences were confirmed by sequencing ( ELIM Bio ) . Rabbit anti-acetylated tubulin ( K40 ) ( D20G3 ) Cell Signaling Technology ( 5335 s ) ; rat anti-hemaglutinin ( HA ) , Roche ( 11867423001 ) ; mouse anti-FLAG M2 , Sigma-Aldrich ( F3165-1MG ) ; rat anti PC1 antibody ( E8 ) ( Baltimore PKD core center ) ; mouse anti PC2 ( YCE2 ) antibody , Santa Cruz Biotechnology ( SC-47734 ) . Cells were fixed with 4% formaldehyde , permeabilized with 0 . 2% Triton X-100 , and blocked by 2% FBS , 2% BSA , and 0 . 2% fish gelatin in PBS . Cells were labeled with the indicated antibody and secondary goat anti-rabbit , anti-rat or anti-mouse fluorescently labeled IgG ( Thermo Fisher ) and Hoechst 33342 ( Thermo Fisher ) . Confocal images were obtained using a Nikon spinning disk with a 63× oil immersion , 1 . 2 N . A . objective , or a 100× oil immersion , 1 . 4 N . A . objective at the UCSF Nikon Imaging Core . Images were further processed using ImageJ ( NIH ) . Tet inducible mIMCD3 and HEK293 cells were generated from parental lines ( obtained from ATCC ) according to the manufacture’s protocol ( Takara Bio ) . HEK or IMCD3 cells expressing the tet activator were transiently co-transfected with the indicated pTRE3G-Bi vector ( Takara Bio ) and the EGFP vector using Lipofectamine LTX for mIMCD3 or Lipofectamine2000 for HEK293 ( Thermo Fisher ) . Surface trafficking of PC-1 and PC-2 was measured by quantifying the amount of anti-HA or anti-FLAG antibody bound to the plasma membrane , respectively . 24 hr after transfection 1 ug/mL doxycycline was added and cells were incubated for an additional 24 hr . For surface staining , adherent live cells were incubated with opti-MEM containing either anti-HA antibody ( 1:100 ) or anti-FLAG antibody ( 1:100 ) for 20 min at room temperature to avoid internalization of antibodies . Cells were washed twice with opti-MEM and fixed with 4% PFA . After incubation with blocking buffer , cells were labeled with secondary goat anti-rat or goat anti-mouse antibodies conjugated to Alexa 647 . Several Z stacks of indicated groups were acquired using identical settings and 647 fluorescence was quantified on all GFP positive cells using ImageJ . This approach allowed an unbiased quantification of membrane inserted HA . In addition , expression of the proteins was confirmed in total staining using permeabilized cells . mIMCD3 cells with ablated PC-2 expression were described previously ( Kleene and Kleene , 2017 ) . Cells were tested for mycoplasma contamination regularly ( Lonza ) . Data are representative of at least three independent experiments . Recordings were performed using a multiclamp 200B ( Axon Instruments ) , digitized using a digidata 1324A ( Axon Instruments ) and recorded using pClamp software ( Axon Instruments ) . Whole cell configuration patch clamp data were filtered at 1 kHz and sampled at 10 kHz . Unless stated otherwise , the voltage step pulse was applied from −100 mV to 180 mV in 20 mV increments during 150 ms and holding potential was given at −60 mV . For the voltage ramp pulse , the same range of voltage steps was applied with 500 ms duration . The resistances of pipettes for whole cell and ciliary patch clamp were 6–8 MΩ and 18–24 MΩ , respectively . The tip of the pipette was further polished using Narishige MF-830 microforge equipped with a 100× Nikon objective . For patch clamp experiment , the extracellular solution consisted of ( mM ) : 145 Na-gluconate , 5 KCl , 2 CaCl2 , 5 MgCl2 , 10 HEPES , and adjusted to pH 7 . 4 using NaOH . The intracellular solution was used as follows: 90 NaMES , 10 NaCl , 2 MgCl2 , 10 HEPES , 5 EGTA , 100 nM free calcium adjusted by CaCl2 , and adjusted to pH 7 . 4 using NaOH . The free calcium was calculated using CaBuf software ( G . Droogmans , Leuven , Belgium ) . For calcium permeability test , the extracellular solution consisted of ( mM ) : 2 mM CaCl2 and 10 HEPES adjusted to pH 7 . 4 using Trizma base and adjusted to 295mOsm using mannitol . The intracellular solution consisted of ( mM ) : 10 HEPES and 1 EGTA adjusted to pH 7 . 4 Trizma base . Ciliary excised patch clamp data and whole cell configuration patch clamp were analyzed with Clampfit10 . 6 ( Axon Instruments/Molecular Devices ) , Origin8 ( Originlab ) , and Prism10 . 0 ( Graphpad ) . Data are shown as mean ± SEM , and n represents independent experiments for the number of tested cells in electrophysiology . For relative open probability of sPC-1-PC-2F604P , the data obtained at −80 mV tail pulse were fitted to a Boltzmann distribution using Origin8 ( Originlab ) . P o ( V ) = P-100 + ( P+180−P-100 ) / ( 1+exp[ ( V 1/2 −V ) /κ] ) where P-100 and P+180 are the open probabilities of the channel at the most negative potential ( −100 mV ) and the most positive potential ( +180 mV ) , respectively . V indicates the membrane potential , V1/2 is the half-maximal activation potential , and κ is the slope factor . For single channel open probability , Popen was calculated by:Popen=toT , where the total time that the channel presented in the open state and T is the total observation time . If a patch contains more than one of the same type of channel , Popen was computed by:Popen=toNT , where , N indicates the number of channels in the patch . We used the following equation to populate data . To=∑Lto , where , L indicates the level of the channel opening . The absolute probability of the channel being open NPo is computed by:NPo=ToTo+Tc , where , Tc indicates the total closed time IMCD3 cells expressing Arl13b-mScarlet-GCaMP6 in primary cilia were imaged as described previously ( Ishikawa and Marshall , 2015 ) . In brief , cells were grown in the bottom side of 24 mm transwell insert with 8 um mesh size allowing rapid exchange of fluids across the membrane . Cilia were observed under an inverted Nikon microscope equipped with an TIRF imaging setup . Images were acquired 1 min after applying 1 μg/mL PC-1 NTF into the transwell . This assay allowed fast imaging along the entire length of the cilium . In most cases , cilia were localized by 561 nm ( mScarlet ) excitation and imaged in 488 nm ( GCaMP6 ) and using the full CCD chip ( 200 ms exposure; five fps ) . Fluorescence was quantified and processed using ImageJ and Python . Constructs were designed to produce various PC-1 N-terminal fragments as secreted proteins in HEK293S GnT1-/- cells using the BacMam expression system . In brief , the endogenous signal sequence of PC-1 was replaced with the strong IgG kappa leader peptide followed by either a M1-FLAG epitope or a maltose-binding protein fusion that allows for affinity purification using anti-M1 Flag antibody or amylose resin , respectively . HEK293S GnT1-/- cells were transduced with baculoviruses when cell density reaches 1 – 2 × 106 cells/mL . Sodium butyrate was added to at a final concentration of 5 mM to enhance protein expression 8–12 hr post-transduction . Temperature was reduced to 30°C and the supernatant was harvested 5 days post-transduction . Various PC-1 N-terminal fragments were purified by affinity chromatography whereby secreted PC-1 fragments were captured by passing supernatant through affinity resins using gravity . The purified PC-1 fragments were then exchanged to a buffer containing 20 mM HEPES , 150 mM NaCl , pH 7 . 4 via extensive dialysis at 4C . PC-1 N-terminal fragments were then flash-frozen in liquid nitrogen and stored as aliquots at −80°C until use . Group data are presented as mean ± SEM . Statistical comparisons were made using unpaired student t-tests for electrophysiology and quantification of surface expression . The Mann-Whitney U-test was used for comparing peak area of Ca2+ signals and Fisher’s exact test was used for comparing # of Ca2+ peaks ( Prism ) . Statistical significance is denoted with an asterisk ( p<0 . 05; *p<0 . 01 ) .	On the surface of most animal and other eukaryotic cells are small rod-like protrusions known as primary cilia . Each cilium is encased by a specialized membrane which is enriched in protein complexes that help the cell sense its local environment . Some of these complexes help transport ions in out of the cell , while others act as receptors that receive chemical signals called ligands . A unique ion channel known as the polycystin complex is able to perform both of these roles as it contains a receptor called PC-1 in addition to an ion channel called PC-2 . Various mutations in the genes that code for PC-1 and PC-2 can result in autosomal dominant polycystic kidney disease ( ADPKD ) , which is the most common monogenetic disease in humans . However , due to the small size of primary cilia – which are less than a thousandth of a millimeter thick – little is known about how polycystin complexes are regulated and how mutations lead to ADPKD . To overcome this barrier , Ha et al . modified kidney cells grown in the lab so that PC-1 and PC-2 form a working channel in the plasma membrane which surrounds the entire cell . As the body of a cell is around 10 , 000 times bigger than the cilium , this allowed the movement of ions across the polycystin complex to be studied using conventional techniques . Experiments using this newly developed assay revealed that a region at one of the ends of the PC-1 protein , named the C-type lectin domain , is essential for stimulating polycystin complexes . Ha et al . found that this domain of PC-1 is able to cut itself from the protein complex . Further experiments showed that when fragments of PC-1 , which contain the C-type lectin domain , are no longer bound to the membrane , they can activate the polycystin channels in cilia as well as the plasma membrane . This suggests that this region of PC-1 may also act as a secreted ligand that can activate other polycystin channels . Some of the genetic mutations that cause ADPKD likely disrupt the activity of the polycystin complex and reduce its ability to transport ions across the cilia membrane . Therefore , the cell assay created in this study could be used to screen for small molecules that can restore the activity of these ion channels in patients with ADPKD .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2020	The heteromeric PC-1/PC-2 polycystin complex is activated by the PC-1 N-terminus
Complexes of specifically interacting molecules , such as transcription factor proteins ( TFs ) and the DNA response elements ( REs ) they recognize , control most biological processes , but little is known concerning the functional and evolutionary effects of epistatic interactions across molecular interfaces . We experimentally characterized all combinations of genotypes in the joint protein-DNA sequence space defined by an historical transition in TF-RE specificity that occurred some 500 million years ago in the DNA-binding domain of an ancient steroid hormone receptor . We found that rampant epistasis within and between the two molecules was essential to specific TF-RE recognition and to the evolution of a novel TF-RE complex with unique derived specificity . Permissive and restrictive epistatic mutations across the TF-RE interface opened and closed potential evolutionary paths accessible by the other , making the evolution of each molecule contingent on its partner's history and allowing a molecular complex with novel specificity to evolve . The relationship between gene sequence and molecular function is of central interest in both molecular biology and evolution . A useful construct for understanding this relationship is sequence space , an organized multidimensional representation of all possible genotypes of a biological system , each connected to its neighbors by edges representing changes in a single sequence site ( Maynard Smith , 1970 ) . Assigning functional information to each genotype yields a ‘topological map’ of the space , which depicts the total set of relations between sequence and function . As proteins evolve , they follow trajectories through sequence space , so the topology of the map also determines how mutation , drift , selection , and other forces can drive genetic evolution . The functional topology of sequence space depends strongly on the degree and type of epistasis , defined as genetic interactions between sequence sites , such that the effect of a mutation at one site depends on the state at others . Epistasis makes the space's topology rugged ( Gavrilets , 2004; Kondrashov and Kondrashov , 2015 ) in the sense that the functional effect of a mutation—a step in some direction along the map—depends on the genetic background in which it occurs . By causing the fitness effects of mutations to depend on the order in which they are introduced , epistasis can affect the probability that evolution will follow any given mutational trajectory under positive selection , purifying selection , or neutral drift ( Wright , 1932; Stadler et al . , 2001; Weinreich et al . , 2006; Poelwijk et al . , 2007; Ortlund et al . , 2007; Phillips , 2008; Bridgham et al . , 2009; Field and Matz , 2010; Salverda et al . , 2011; Breen et al . , 2012; Gong et al . , 2013; Harms and Thornton , 2014; Yokoyama et al . , 2014; Podgornaia and Laub , 2015; Tufts et al . , 2015 ) . Sequence space is vast , so an exhaustive functional mapping and characterization of epistasis for any protein or gene is impossible . Some studies have characterized libraries of genotypes in the sequence space immediately around present-day proteins ( Lunzer et al . , 2005; Fowler et al . , 2010; Hinkley et al . , 2011; Araya et al . , 2012; McLaughlin et al . , 2012; Tokuriki et al . , 2012; Jacquier et al . , 2013; Melamed et al . , 2013; Olson et al . , 2014; Bank et al . , 2015; Podgornaia and Laub , 2015 ) . Others have focused on smaller numbers of mutations that occurred during historical evolution , introducing them singly and in combination into extant proteins to understand their interactions and potential effects on evolutionary trajectories ( Weinreich et al . , 2006; Elde et al . , 2009; Bloom et al . , 2010; Gong et al . , 2013; Natarajan et al . , 2013 ) . Similar approaches have also been applied to reconstructed ancestral proteins in order to directly characterize how epistasis may have affected evolutionary history in the genetic backgrounds likely to have existed in the past ( Ortlund et al . , 2007; Bridgham et al . , 2009; Field and Matz , 2010; Harms and Thornton , 2010 , 2014; Lynch et al . , 2011; Yokoyama et al . , 2014; Wilson et al . , 2015 ) . Although the effect of epistasis on sequence space and evolution has begun to be characterized for individual molecules , many biological functions depend on physical interactions between molecules . Epistasis between sites across a molecular interface could play a key role in determining the functions and evolutionary potential of molecular complexes . An important but unexplored goal is therefore to functionally map the joint multidimensional sequence space that contains the combined genotypes of interacting molecules . The interactions between transcription factors ( TFs ) and the DNA response elements ( REs ) to which they bind exemplify this issue . TF-RE interactions regulate gene expression in virtually all biological processes ( Tjian and Maniatis , 1994; Lelli et al . , 2012 ) . Effective and precise gene regulation depends on the capacity of a TF to specifically bind its preferred RE targets with sufficient affinity and occupancy in a heterogeneous cellular environment ( Li et al . , 2008; Fisher et al . , 2012 ) . The genetic determinants of affinity and specificity of TF-RE complexes must lie in both molecules and the interactions between them: for example , an amino acid replacement that changes a TF's DNA specificity must affect affinity differently when combined with various RE genotypes ( Voordeckers et al . , 2015 ) . The joint sequence space of TFs and their REs has not been directly characterized , particularly in an evolutionary context . Previous work on TF-DNA recognition suggests that epistasis is likely to be important , as expected given the biophysical complexity of protein-DNA interfaces ( Dill , 1997; Stout et al . , 1998; Pabo and Nekludova , 2000 ) . For example , numerous studies have assessed the binding of a single TF to a library of REs , thus identifying the genetic states in the DNA that determine affinity ( Badis et al . , 2009; Portales-Casamar et al . , 2010; Stormo and Zhao , 2010; Payne and Wagner , 2014 ) ; in several cases , epistatic interactions between neighboring nucleotides in the RE are apparent ( Man and Stormo , 2001; O'Flanagan et al . , 2005; Moyroud et al . , 2011; Zhao et al . , 2012 ) . Other studies have addressed aspects of the TF's protein sequence space by investigating how amino acid variation in a TF affects RE binding ( Lynch et al . , 2008; Baker et al . , 2011; McKeown et al . , 2014; Perez et al . , 2014; Pougach et al . , 2014 ) , and here too there is some evidence of epistasis between residues ( Pabo and Nekludova , 2000 ) . We have little systematic knowledge , however , concerning the topology of joint TF-RE sequence space and how it may affect evolutionary processes . The total joint sequence space of multiple molecules is far too large to characterize comprehensively . It should be possible , however , to functionally map the small portion of that space defined by a specific historical change in function . Such a region contains all genotypes on all direct mutational paths between the molecules in a reconstructed ancestral complex and those in a descendant complex that has different specificity . Here we map the joint sequence space across an evolutionary transition in specificity for an ancient TF protein and the DNA REs to which it binds . This approach allowed us to identify sequence states in the DNA and protein that determine the affinity and specificity of binding , to characterize epistasis within and between the molecules , and to analyze the effects of intermolecular epistasis on the evolution of gene regulation and TF-RE interactions . The DNA binding domain ( DBD ) of steroid hormone receptors ( SRs ) are a model for exploring the sequence space of an evolving TF-RE complex . SRs are a class of ligand-activated TFs; they include a ligand-binding domain—which activates gene expression in the presence of specific sex or adrenal steroid hormones—and a DNA-binding domain , which binds as a dimer to palindromic REs consisting of two half-sites each six bases long ( Bentley , 1998; Bain et al . , 2007 ) . SRs group into two phylogenetic clades , each with a distinct DNA-binding specificity ( Figure 1A ) . The estrogen receptors ( ERs ) preferentially bind to estrogen RE ( ERE ) , which contain the half-site AGGTCA . The other clade—progestagen , androgen , mineralocorticoid and glucocorticoid receptors ( PAMGRs ) —preferentially bind to the SRE half-sites AGAACA ( SRE1 ) or AGGACA ( SRE2 ) ( Umesono and Evans , 1989; Lundback et al . , 1993; Beato and Sanchez-Pacheco , 1996; Welboren et al . , 2009 ) . Thus , the preferred RE half-sites for the two clades are identical except at two nucleotide positions ( underlined ) . 10 . 7554/eLife . 07864 . 003Figure 1 . Recognition helix ( RH ) substitutions change DNA-binding affinity and specificity . ( A ) Phylogenetic relationships of modern-day vertebrate SRs are shown , with ancestral proteins AncSR1 and AncSR2 marked . Each protein's preferred response element ( RE ) is shown: estrogen RE ( ERE; purple ) or steroid REs ( SRE1 , SRE2; light and dark green , respectively ) , with the half-site sequence of each . Gray box indicates evolutionary interval in which SRE specificity evolved ( McKeown et al . , 2014 ) . ( B ) Interface of steroid hormone receptor DNA-binding domains ( DBDs ) with their preferred RE half-sites . X-ray crystal structures of AncSR1 with ERE ( left , 4OLN ) and AncSR2 with SRE1 ( right , 4OOR ) . The RH is shown as a colored cylinder; sticks , side chains that differ between AncSR1 and AncSR2 . Colored surface , nucleotides that differ between REs . ( C ) Close-up of protein-DNA interface for AncSR1:ERE ( left ) and AncSR2:SRE1 ( right ) . In the DBD , the RH is shown as ribbon , with side chains of variable amino acids shown as sticks and Cα as spheres . In the RE , variable nucleotides are shown as sticks with backbone as cartoon . Atoms are colored by element . Dashed lines , polar interactions between variable amino acids and nucleotides . ( D ) Historical RH replacements change AncSR1's affinity for REs . Binding energies of AncSR1 ( left ) and AncSR1+RH replacements were measured using fluorescence polarization to single half-site REs containing all possible combinations of nucleotides at the sites that vary between ERE and SREs . ERE , SRE1 and SRE2 are highlighted in purple , light green and dark green , respectively . ΔGdissociation is the free energy of dissociation , calculated from dissociation constant ( Kd ) . Technical replicates ( dots ) with mean and SEM ( lines ) are shown . ( E ) RH replacements change the genetic determinants of affinity within the RE . Energy logos for AncSR1 ( left ) and AncSR1+RH ( right ) show the effects of nucleotide states on binding energy relative to the average across all REs tested; states with ΔΔGdissociation > 0 are associated with higher affinity binding . Main effects of nucleotides at variable positions 3 and 4 are shown , as is the epistatic effect of nucleotide combinations , defined as the excess effect beyond that predicted under additivity . The height of each state indicates the magnitude of their effect on binding energy; states are ranked from top to bottom by the magnitude of its effect . Each column's width shows the portion of variation in binding energy attributable to the effects of states in that column , calculated as the increase in the adjusted R2 when terms corresponding to those states are added to a linear regression model and fit to the experimental binding data . * , significant improvement in model fit ( likelihood ratio test , p < 0 . 05 Bonferroni-corrected ) . For complete explanation of linear modeling approach , see ‘Materials and methods’ . DOI: http://dx . doi . org/10 . 7554/eLife . 07864 . 003 The historical amino acid replacements that caused the two classes of extant SRs to evolve their distinct half-site specificities are already known ( McKeown et al . , 2014 ) . We previously reconstructed the ancestor of all SRs ( AncSR1 ) and the ancestor of all PAMGRs ( AncSR2 ) and assayed their affinities for ERE and SREs ( Figure 1A ) . We found that AncSR1 was ER-like , preferentially binding to ERE , and that AncSR2 was like PAMGRs , preferentially binding SREs . When just three of the 38 amino acid replacements that occurred during the evolutionary interval between AncSR1 and AncSR2 were introduced into AncSR1 , they fully recapitulated the shift in half-site preference from ERE to SREs . These three replacements ( glu25GLY , gly26SER , and ala29VAL , where lower and upper cases denote ancestral and derived states , respectively ) were the only changes on the DBD's ten-residue recognition helix ( RH ) , which inserts into the DNA major groove ( Figure 1B , C ) . The ancestral and derived states at these three sites are completely conserved in modern-day ERs and PAMGRs , respectively , so the inference of their states in AncSR1 and AncSR2 was unambiguous . Although other substitutions during this interval affected the DBD's non-specific affinity for DNA and the cooperativity of dimeric binding , only the RH substitutions affected half-site specificity , and they did so without affecting cooperativity ( McKeown et al . , 2014 ) . The ancestral complex AncSR1:ERE and the derived complex AncSR1+RH:SRE define an ancient evolutionary transition in TF:RE specificity , and the set of direct paths between these points constitutes a historically relevant region of the joint protein-DNA sequence space . Here we report on experiments to functionally map this region of the molecules' joint sequence space and to understand the implications of its topology for evolutionary processes . We experimentally measured the binding affinity of AncSR1-DBD variants containing all genotypic combinations of ancestral and derived amino acids in the RH in complexes with REs containing all possible combinations of nucleotides at the two variable positions in the half-site . We used these data to statistically estimate the main effects of every variable state in the DBD and the RE and of interactions between states within and between each molecule . This analysis allowed us to investigate how the causal determinants of binding changed as the regulatory complex evolved during the historical shift in function and to evaluate the plausibility that any pathway through the joint TF-RE space might have been followed under various evolutionary scenarios . We first characterized the RE-specificity of the ancestral and derived proteins AncSR1 and AncSR1+RH by measuring each protein's affinity for RE sequences containing all 16 combinations of possible nucleotides at the two sites that vary between ERE and SRE . We used a fluorescence anisotropy ( FA ) assay using labeled DNA probes , which provides direct and precise estimates of the free energy ( ΔG ) of binding ( Figure 1D , expressed as ΔGdissociation ) . Although the set of all possible REs is much larger , the preferred nucleotides at the other sequence positions in the RE did not change during SR evolution . We focused on differences in half-site affinity because the RH substitutions affected this phenomenon without changing the cooperativity of binding to palindromic REs ( McKeown et al . , 2014 ) . We found that several major changes occurred during the evolutionary transition from the ancestral to derived proteins . First , AncSR1 prefers ERE and AncSR1+RH prefers SREs , but both classes of RE are among the highest-affinity targets for both proteins ( Figure 1D ) . The derived preference for SREs therefore arose by reshuffling the protein's relative affinities among high- and moderate-affinity targets , rather than by radically increasing affinity for DNA sequences that were previously bound very poorly . Second , the RH substitutions dramatically impaired binding to the protein's best DNA targets in both absolute and relative terms; they reduced affinity for ERE by more than 2 kcal/mol while increasing affinity for SREs by a mere ∼0 . 2 kcal/mol ( Figure 1D ) . As a result , AncSR1's affinity for ERE is much higher than AncSR1+RH's affinity for SREs , and the difference between the best and next-best targets in AncSR1 is much larger than that in AncSR1+RH . Together , these effects make AncSR1+RH less specific for its preferred REs than AncSR1 . The RH substitutions therefore changed the protein's preferred DNA binding sites by exploiting a preexisting latent affinity for a moderate-affinity RE and dramatically reducing affinity for the ancestral RE . Exploitation of latent , moderate- or low-affinity interactions has been observed in the functional evolution of other molecules ( Khersonsky et al . , 2006; Coyle et al . , 2013; Pougach et al . , 2014 ) . We next used these measurements to quantitatively analyze the genetic factors within the RE's DNA sequence that determine affinity of AncSR1 and of AncSR1+RH . These determinants include the average ( or ‘main’ ) effects of each possible nucleotide at the variable RE sites , as well as epistatic interactions , which occur when a combination of nucleotides at multiple sites affects affinity differently from expectation based on each nucleotide's average contribution . We used a multiple linear regression model ( Stormo , 2011 ) that predicts a protein's free energy of binding to an RE as the sum of all main and epistatic effects at the two variable positions 3 and 4 ( for details , see ‘Materials and methods’ ) ; a linear model is appropriate because the dependent variable is the ΔG of reversible binding , which grows additively with the free energy of the factors that contribute to it ( Benson , 1968; Dill , 1997; Stout et al . , 1998 ) . We used global regression to estimate the values of the model's parameters that best predict each protein's measured affinity for all 16 RE sequences . This approach allows us to quantify the contribution to the total binding energy made by each individual state ( the main effects ) and every combination of states ( epistatic effects ) . It also allows us to estimate the portion of all variation in affinity explained by each main and epistatic effect ( expressed as the increase in adjusted R2 when a parameter for any class of effect is added to the model ) and to evaluate the statistical significance of the improvement in fit attributable to each class of parameters . We found that DNA affinity is determined by main effects of individual nucleotides and by epistatic interactions between them , but these determinants are radically different between AncSR1 and AncSR1+RH . AncSR1 prefers REs with G in position 3 ( G3 ) by 1 . 0 kcal/mole and those with T in position 4 ( T4 ) by 0 . 5 kcal/mole ( Figure 1E , Supplementary file 1 ) . A strong epistatic effect ( G3xT4 ) is also present , which further enhances affinity for the GT combination by 0 . 8 kcal/mole beyond that expected from the main effects of these two nucleotides . This epistatic interaction establishes the protein's specificity for ERE by generating a large energy gap between binding to GT and binding to other sequences containing G3 or T4 but not both ( Figure 1D ) . AncSR1's specificity is further affected by a strong negative epistatic determinant G3xA4 , which substantially reduces its affinity for SRE2 ( GA ) ; without this interaction , the protein's preference for G3 would have made SRE2 ( GA ) a high affinity target ( Figure 1E ) . The DNA determinants of binding by AncSR1+RH are very different ( Figure 1E , Supplementary file 1 ) . Unlike its ancestor , this protein's affinity is unaffected by main effects of the nucleotide at DNA position 3 . Further , the ancestral preference for T4 is replaced by a strong preference for A4 . The ancestral epistatic determinant G3xT4 is nearly abolished and replaced by two new positive epistatic interactions—A3xA4 and G3xA4—which together establish the protein's preference for its best targets SRE1 and SRE2 relative to other sequences that contain A4 . Consistent with the lower specificity of AncSR1+RH , the magnitude of all these determinants is smaller than those that determine AncSR1's target affinity . These data indicate that in both the ancestral and derived proteins , epistasis between the two variable nucleotide positions in the DNA target played a key role in establishing specificity of binding . The evolutionary effect of the RH substitutions was to erase all major ancestral DNA determinants of affinity and to establish novel determinants , both main and epistatic . We next sought to understand how each historical amino acid change in the RH altered the determinants of binding within the RE . We engineered variants of AncSR1 containing all combinations of ancestral and derived states at the three RH sites and measured each one's binding affinities for all 16 REs ( Figure 2 ) . All three RH amino acid replacements can be produced by single-nucleotide mutations; this set of proteins therefore comprises all direct pathways between AncSR1 and AncSR1+RH and represents all of the most parsimonious possible evolutionary histories . 10 . 7554/eLife . 07864 . 004Figure 2 . Protein intermediates between AncSR1 and AncSR1+RH are promiscuous or weak transcription factor proteins ( TFs ) . Binding energies of AncSR1 variants containing all combinations of ancestral and derived states at the RH sites with historical replacements are shown for all 16 REs as measured by fluorescence polarization . Single-replacement neighbors of AncSR1 are shown in the top row; two-replacement proteins are shown in the bottom row . ERE , SRE1 and SRE2 are highlighted with purple , light green and dark green bars , respectively . Dashed line , mean binding energy across all protein genotypes and all REs . Data points show three replicates; mean and SEM are shown with lines . DOI: http://dx . doi . org/10 . 7554/eLife . 07864 . 004 Every intermediate protein we tested preferred either ERE or SREs over all other REs; transiently preferred targets do not emerge along trajectories from AncSR1 to AncSR1+RH . Unlike the starting and ending states , however , all intermediates were either universally low-affinity or highly promiscuous TFs . Specifically , three low-affinity intermediates ( glu-gly-VAL , glu-SER-ala and glu-SER-VAL ) bound to their best targets far more weakly than either the ancestral or derived forms bound their preferred sequences , and they did not bind any REs with affinity greater than the global average of all 8 proteins with all 16 REs ( Figure 2 ) . Two others ( GLY-gly-ala and GLY-gly-VAL ) promiscuously bound all 16 REs with greater-than-average affinity , and they both bound many targets—16 and 7 , respectively—with affinity greater than AncSR1+RH's affinity for its best target . The remaining intermediate ( GLY-SER-ala ) , was moderately promiscuous , binding four REs—ERE , SREs , and one other—with similar and above-average affinities . We used these data to quantify the DNA determinants of TF affinity for each intermediate protein using the linear model described above . This allowed us to reveal how DNA specificity would have changed along any direct trajectory from AncSR1 to AncSR1+RH . We found that no single replacement is sufficient to abolish the ancestral DNA preferences or to establish the derived preferences . All of AncSR1's single-replacement neighbors maintain the major ancestral determinants of specificity—G3 , T4 , and G3xT4—but at reduced magnitude ( Figure 3 , Supplementary file 1 ) , consistent with the fact that all three of these proteins prefer ERE , but to a lesser degree than AncSR1 does ( Figures 1D , 2 ) . None of the single-replacement neighbors display any of the derived determinants of specificity ( main effect preference for A4 or epistatic preference for G3xA4 and A3xA4 ) . 10 . 7554/eLife . 07864 . 005Figure 3 . Each amino acid replacement contributes to the evolution of novel DNA specificity . For each protein intermediate in the sequence space between AncSR1 and AncSR1+RH , the energy logo depicts the main and epistatic effects of the RE nucleotide states and combinations on binding affinity by each TF ( for details , see Figure 1E ) . Vertices of the cube indicate protein genotypes; the number of amino acid differences from AncSR1 is indicated in the circle at each node . Edges represent single replacements between TF genotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 07864 . 005 After the second step of the mutational pathway from AncSR1 to AncSR1+RH , some of the derived determinants of specificity begin to appear , and the ancestral determinants are further weakened ( Figure 3 , Supplementary file 1 ) . In all cases , however , either the ancestral determinants are partially retained or the derived determinants are weak . Only upon addition of the third replacement are the derived determinants complete . All three amino acid replacements in the RH therefore contributed to the derived DNA specificity , and their effects depend on the background into which they are introduced . These analyses also point to a third-order form of epistasis: combinations of amino acid replacements uniquely affect binding to specific DNA sequences . The most apparent example of this phenomenon is the interaction between glu25GLY and ala29VAL in producing the derived preferences ( Figure 3 , Supplementary file 1 ) . Neither replacement increases relative affinity for REs containing A4 . But when the two changes are combined ( GLY-gly-VAL or GLY-SER-VA ) L , a strong A4 preference is established . Thus , the threefold combination of GLY25 , VAL29 , and nucleotide A4 increases affinity beyond that expected due to the main effects of any of these states or the pairwise interactions between them ( see also Supplementary file 1 ) . Taken together , our findings indicate that all direct paths from AncSR1 to AncSR1+RH involve one of two scenarios: losing high-affinity binding to the ancestral and all other REs followed by a gain of binding to new targets , or gaining very promiscuous high-affinity binding followed by a dramatic narrowing of targets . No paths involve an immediate transformation of the highly specific ancestral TF into a specific protein with a new but narrow set of targets . Instead , ancestral determinants of binding were weakened and derived determinants gradually strengthened as mutational combinations were assembled . We next quantitatively characterized the effects of each amino acid replacement on DNA affinity and specificity . For this purpose , we first expanded the regression model described above to incorporate variation in the TF's protein sequence and to determine the main and epistatic effects of each historical amino acid replacement on DNA affinity ( Supplementary file 2 ) . We first analyzed nonspecific effects on affinity averaged across all REs tested . We found that each amino acid replacement changed non-specific affinity , but they did so in different directions . glu25GLY increased average affinity by 1 . 3 kcal/mol ( Figure 4A , Supplementary file 1 ) , consistent with the observation that GLY-gly-ala is a promiscuous and high-affinity DBD ( Figure 2 , Supplementary file 1 ) . The other substitutions gly26SER and ala29VAL each reduced the average affinity of binding ( Figure 4A , Supplementary file 1 ) , explaining why proteins containing these states without glu25GLY are low-affinity proteins for all REs ( Figure 2 ) . Moderate epistatic interactions between the residues further modified their nonspecific effects on average affinity ( Figure 4A , Supplementary file 1 ) . When combined , the countervailing effects of the three replacements cause the final AncSR1+RH genotype to have an average affinity similar to that of AncSR1 . 10 . 7554/eLife . 07864 . 006Figure 4 . Epistasis across the protein-DNA interface: effect of historical replacements in the TF on DNA determinants of affinity in the RE . ( A ) Main and epistatic effects of RH replacement on DNA affinity . Bars indicate the mean change in binding energy caused by each amino acid change in the RH , averaged across all TF:RE combinations measured; epistatic effects represent the additional effect of pairs of replacements on average binding energy beyond that predicted by their main effects . Bar width depicts the portion of variation in binding energy attributable to each main or epistatic effect , calculated as the increase in the adjusted R2 of the fit to the experimental binding data when each term is added to a linear regression model . * , significant improvement in model fit ( likelihood ratio test p < 0 . 05 after Bonferroni correction ) . ( B ) Intermolecular epistasis . Energy logos indicate the effect of each amino acid replacement on the genetic determinants of binding within the RE . For each amino acid replacement , the size of each letter indicates the change the replacement causes in the main ( or epistatic ) effects of nucleotide states ( or combinations ) on relative binding energy . DOI: http://dx . doi . org/10 . 7554/eLife . 07864 . 006 Differences among proteins in their average affinity over REs cannot explain the shift in DNA specificity that occurred between AncSR1 to AncSR1+RH . To change specificity , an amino acid replacement must affect affinity for RE genotypes differently . Therefore , when amino acid replacements change specificity or preference for REs , epistasis across the protein-DNA interface must be involved . To statistically characterize this form of epistasis , we further expanded the linear models described above to simultaneously analyze the effects of the combined protein + DNA genotype on affinity . This approach allowed us to identify epistatic effects on affinity caused by combinations of amino acid replacements and nucleotide states . The magnitude of an epistatic effect across the molecular interface is the difference in the binding energy of complexes containing both a specific TF amino acid and a specific RE nucleotide from that predicted based on the average effects of each of those states . A replacement may also participate in third-order interactions with a combination of nucleotide states in the RE , if it specifically changes relative affinity for the combination more than expected from the first- and second-order effects together . We found that each amino acid replacement is involved in cross-interface epistatic interactions that contribute to eliminating the ancestral determinants of specificity and establishing the derived determinants . Although replacement glu25GLY increases affinity for all REs ( Figure 4A ) , it does not do so uniformly: it increases relative affinity for REs with nucleotide states A3 , A4 and G3xA4—thus specifically improving relative binding to SREs—and decreases relative affinity for REs containing the ancestral DNA determinants G3 and T4 , as well as the ERE-specific combination G3xT4 ( Figure 4B , Supplementary file 1 ) . This replacement's most dramatic effect is therefore to cause a specific increase in the protein's affinity for SREs relative to EREs . The other two replacements also participate in cross-interface pairwise and third-order epistatic interactions . Both gly26SER and ala29VAL reduce the ancestral determinants of binding G3 and T4 and increase the derived preference for REs with A4 . Both also are involved in third-order interactions , by which the amino acid changes further reduce affinity for the G3xT4 combination and increase affinity for the G3xA4 combination ( Figure 4B , Supplementary file 1 ) . As a result , although each of these replacements reduces binding affinity to every single RE ( Figure 4A ) , they do so most radically on ERE while only weakly affecting SREs ( Figure 4B ) . Taken together , these observations confirm that all three RH replacements contributed to the historical change in specificity via cross-interface epistatic interactions . Second-order intermolecular interactions allowed each replacement in the TF to shift specificity by differentially affecting affinity for REs containing specific individual nucleotide states . Third-order interactions allowed each replacement to have further effects on affinity when combined with pairs of nucleotides , beyond those expected based on any of the three sites' average effects plus all their pairwise interactions . We found that new specificity evolved without sign epistasis . Each replacement affected binding to all 16 REs in the same direction ( Figures 2 , 4 ) , but its effects on some REs were more extreme than on others . Because none of the replacements acted as a switch that impaired binding to some sequences while improving affinity for others , multiple replacements were necessary to achieve the derived specificity and affinity . Why would the shift in function not have occurred via a simpler genetic mechanism involving sign epistasis ? We speculate that there are few or no potential replacements that have pinpoint opposite effects on different REs . The interface between a TF and DNA is heterogeneous and densely packed , and all four possible nucleotides share many similarities in the physical properties they confer on the DNA surface to which a TF binds . Thus , amino acid changes are more likely to alter binding in a generic direction across all possible versions of the RE , but they do so more effectively when paired with some nucleotides than with others . Having mapped the free energy of binding—a fundamental biochemical property—across joint sequence space , we next sought to understand how evolution might proceed through this space under various evolutionary scenarios . If the relationship between biochemical affinity of a TF-RE complex and its function in the cell and organism were linear , then the ruggedness of the biochemical topology of sequence space would be identical to the ruggedness of the space's functional topology . It is highly unlikely , however , that all biological dependent variables—occupancy on DNA , gene regulatory output , phenotypic effect , and fitness—are linearly related the affinity of a TF-RE complex . A nonlinear relationship between affinity and biological function/fitness would introduce additional epistasis into the topology of sequence space and further change the kinds of paths that evolution is likely to follow under purifying selection , drift , and positive selection . The precise nature of this transformation depends strongly on biological context , is unknown . We therefore sought to gain preliminary insight into the plausibility of evolutionary trajectories through joint sequence space under two simple , biologically motivated scenarios . The first scenario was to consider how AncSR1 might have evolved novel recognition of SREs , thus establishing occupancy of a new set of target genes , while under purifying selection to maintain specific , high-affinity binding to at least one RE at every step in the trajectory . We based our analysis on the concept of a connected network of functional protein genotypes , which assumes that purifying selection makes mutational pathways through nonfunctional intermediates very unlikely ( Maynard Smith , 1970; Wagner , 2008 ) . Although the RH substitutions occurred after a gene duplication of AncSR1 , it is unlikely that either copy was released from purifying selection , because the ancestral function was conserved in both copies for tens or hundreds of millions of years before neofunctionalization occurred ( Bridgham et al . , 2008; Eick et al . , 2012 ) . We defined each protein's set of functional RE targets as those that fulfill two simple criteria . First , to achieve reasonable occupancy by the low concentrations of TF typical in cells ( Fisher et al . , 2012 ) , the RE must be bound with moderate to high affinity , which we defined as greater than the average affinity across all TF-RE complexes tested . Further , a potential DNA binding site must compete with other REs for occupancy by the same TF , so the second criterion we imposed is that the affinity constant for an RE must be within a factor of ten of that protein's best target ( Fisher et al . , 2012 ) . TF:RE complexes not meeting these criteria were classified as low-occupancy and therefore nonfunctional . Although these criteria are somewhat arbitrary , they provide a starting point for understanding how the biochemical epistasis we observed , along with a simple nonlinearity in the transformation of affinity into function , might affect the mutational paths available to the evolving TF-RE complex . We found that epistasis strongly structures possible evolutionary trajectories through sequence space . Starting from AncSR1 , only one replacement ( glu25GLY ) leads to a functional TF; the others yield proteins that do not effectively bind any REs and are therefore unlikely evolutionary intermediates ( Figure 5 ) . The glu25GLY mutation produces a high-affinity but extremely promiscuous TF that retains strong binding to the ancestral ERE while gaining high-affinity binding to 13 novel REs , including the SREs . This replacement opens up new evolutionary trajectories , because once it is in place , either of the other two amino acid changes become tolerable . glu25GLY therefore represents a permissive evolutionary mutation that broadens the network of other accessible replacements that the protein can explore . 10 . 7554/eLife . 07864 . 007Figure 5 . Accessible mutational pathways in the joint TF-RE sequence space . Each vertex of the cube represents a protein genotype between AncSR1 and AncSR1+RH; amino acid states at variable RH residues are shown; lower and upper case denote ancestral and derived states , respectively . Each protein's affinity for the 16 REs is shown using the color gradient from red to blue ( low to high ΔGdissociation ) . The genotype at the center of each cluster of REs is that protein's highest-affinity target . REs preferred by AncSR1 or AncSR1-RH are shown in triangles or squares , respectively; circles show other REs . TF-RE complexes with high affinity and occupancy ( binding energy greater than the average of all TF-RE complexes and affinity within tenfold of that TF's best target ) are outlined in bold . Blue lines represent amino acid replacements in the TF that maintain high-affinity/occupancy binding to a RE; green lines represent nucleotide substitutions in the RE that maintain high-affinity/occupancy binding to a TF . Nodes connected by blue or green lines represent the neutral network between the ancestral TF and its RE targets and the derived TF and its distinct targets . DOI: http://dx . doi . org/10 . 7554/eLife . 07864 . 007 Epistasis also affects the second and third steps in the protein's mutational trajectory . The two possible paths available to AncSR1+RH have different functional implications for specific RE recognition , depending on the order in which the remaining replacements are introduced . Adding ala29VAL first expands the set of occupied DNA targets to include all 16 REs; the final gly26SER mutation then radically narrows the TF's specificity , eliminating 13 REs as high-occupancy targets and leaving only SRE1 , SRE2 , and one additional target ( TA ) . Following the other pathway , incorporating gly26SER first narrows promiscuity somewhat but does not eliminate the ancestral targets , leaving four high-affinity REs ( ERE and the other ancestrally recognized target GG , plus the novel SRE1 and SRE2 ) ; the final replacement ala29VAL then eliminates the ancestral targets—rather than expanding the protein's promiscuity , as it would if introduced earlier—and yields a protein specific for the derived REs ( Figure 5 ) . These observations point to higher-order epistasis—two amino acids interacting with each other to differentially change regulation of specific DNA sequences—which causes the functional implications of trajectories through sequence space to depend on the order of sequence change among multiple molecules . Taken together , these data indicate that a derived TF could evolve to regulate a novel set of DNA targets completely distinct from those of the ancestral protein without losing the ancestral function until the final amino acid change . Because of epistasis—particularly the requirement for the permissive replacement glu25GLY for the other two replacements to be tolerated—only a small fraction of the possible pathways through protein sequence space pass through only functional TF intermediates on the way to the derived protein ( Figure 5 ) . Along both pathways , promiscuous intermediate genotypes gained recognition of the novel SREs , followed by further replacements that eliminated high-affinity binding to the ancestral RE and any transiently acquired targets . Our results are consistent with previous studies indicating an important role for permissive mutations in enabling a protein to tolerate other function-switching mutations that would otherwise be deleterious , particularly when a threshold relationship pertains between a biochemical property and fitness ( Ortlund et al . , 2007; Bloom and Arnold , 2009; Woods et al . , 2011; Gong et al . , 2013; Harms and Thornton , 2013 ) . By strongly increasing the protein's affinity for many REs , replacement glu25GLY moved the evolving AncSR-DBD well above the threshold for functionality , allowing the protein to tolerate other replacements that refined the protein's specificity while decreasing its generic RE affinity . The second scenario we examined was a process of mutually permissive neutral drift by a functional unit of one TF and one RE under purifying selection . Numerous studies have found that a TF and its RE at a regulatory element sometimes diverge in sequence from their ancestral states while maintaining a conserved regulatory association with each other ( True and Haag , 2001; Haag and True , 2007; Barriere et al . , 2012; Lynch and Hagner , 2015 ) . Using the same criteria as above to define a functional complex , we sought to understand whether a single complex of AncSR1:ERE could traverse the joint TF-RE sequence space through a series of changes in the protein and the DNA , reaching AncSR1+RH:SRE along a continuous neutral network without ever losing high-affinity , high-occupancy binding . We found that there are many pathways through the joint genotype space from the ancestral to the derived complex that maintain a functional complex at every intermediate step . These pathways are made possible by intermolecular epistasis , which causes replacements in the TF to alter constraints on evolution of the RE , and vice versa . At the starting point of AncSR1:ERE , both the TF and the RE genotypes are highly constrained by the requirement to maintain a high-affinity complex with each other: only a single mutation is available to the DNA ( from GT to GG ) , and only one replacement ( glu25GLY ) is available to the protein ( Figure 5 ) . Drift along each of these pathways has very different functional implications: if the G4 mutation in the RE were to occur first , the neutral network available to the protein would remain unchanged , with glu25GLY remaining as the only viable replacement on the trajectory towards AncSR1+RH . But if glu25GLY were to occur first , the set of viable mutations available to the RE would radically expand , with almost any RE genotype being tolerated . Reaplcement glu25GLY in the TF therefore acts as a permissive substitution for drift by the RE , allowing the DNA binding site to explore many new pathways—including mutation to SREs—while still maintaining an association with the TF . Thus , neutral substitutions in the protein can permit previously unavailable moves by the DNA through its neutral network . Conversely , neutral changes by the RE can permit amino acid replacements in the TF to occur that would otherwise have been unfavorable . For example , if the complex is in the state GLY-SER-ala:GT , adding the final replacement ala29VAL would lead to a nonfunctional complex . But if the RE were to drift first to one of the SREs , then the complex would be able to tolerate this amino acid change . Further , neutral substitutions in the RE can also restrict evolutionary pathways that were previously open to the TF , making certain amino acid changes inaccessible . For example , if the complex is in the state GLY-gly-VAL:GA , the TF can drift to AncSR1+RH via the gly26SER replacement; however , if the RE first drifts to GC , this replacement would abolish the TF:RE association , making it an unlikely evolutionary step ( Figure 5 ) . Permissive and restrictive mutations across the interface may act in a serial chain that reciprocally modifies the partner molecules' evolvability: a mutation in the RE may change the protein's capacity to tolerate a previously unavailable replacement , which then changes the RE's tolerance of a mutation that would otherwise have been unavailable , and so on . To follow one such example , replacement glu25GLY allows the ancestral ERE ( GT ) to drift to TT , which prevents gly26SER , which otherwise would have been available , and also permits the previously denied replacement ala29VAL; if ala29VAL does occur next , then mutation of the DNA to TA becomes available , which in turn is permissive for gly26SER . That replacement closes off many DNA mutations that were previously available , but the final mutation to SRE1 or SRE2 is open . Epistasis across the molecular interface therefore makes the evolution of the TF and the RE contingent upon the genotype—and therefore the prior evolutionary history—of its partner . A chain of serially contingent events may ensue , as permissive and restrictive mutations in the protein and DNA open or close evolutionary paths for mutations within the same molecule or its binding partner . Taken together , these findings indicate that epistasis across a molecular interface can allow interacting molecules to both evolve by drift to states that are incompatible with the ancestral versions of their partner . AncSR1:ERE is a highly specific complex that does not recognize SRE; it lies on a narrow peninsula in sequence space , with few mutations leading directly to other functional complexes . A series of permissive mutations in both partners , however , could have allowed the RE and the TF to drift together through numerous new genotypes , eventually reaching AncSR1+RH:SRE , which itself is a specific complex on a narrow peninsula that does not bind or connect directly to ERE . The nonlinear mapping we imposed from biochemical property to biological function/fitness added an additional form of epistasis to that we observed due to affinity alone . The criteria that we used to define this mapping are highly simplified . In reality , the selective effects of mutations that change TF:RE interactions are likely to depend on many factors , including the genomic context and role of other TFs in regulating a given target , interactions between the TF and other proteins , the genotype and function of other domains and sites in the TF , the physiological roles of the target genes , and the demographic characteristics of the population . How these factors affected the mapping of affinity onto fitness for the ancient molecular complex we study here is unknown; as a result , the mutational pathways most likely to have been followed during the evolutionary history of the SR DBD and its targets are also unknown . Despite this uncertainty about specific historical scenarios , however , it is clear that rampant epistasis was present at the most fundamental biochemical level , and even the simple biological criteria we imposed resulted in very strong impacts on the evolutionary accessibility of trajectories across sequence space . It therefore seems likely that epistasis will also structure evolutionary trajectories under more complex , realistic conditions that introduce further nonlinearities into the relationship between physical properties and selectable biological outcomes . Finally , we sought to understand the underlying biophysical mechanisms that cause variation in binding affinity among TF:RE pairs in this region of sequence space . To determine these mechanisms , we performed molecular dynamics ( MD ) simulations for AncSR1 , AncSR1+RH and all intermediate protein genotypes , each bound to every one of the 16 DNA sequences . We then measured hydrogen bonding and packing at the protein-DNA interface , which are known to contribute to high-affinity interactions between proteins and DNA ( von Hippel , 1994; Garvie and Wolberger , 2001; Coulocheri et al . , 2007; Rohs et al . , 2010; McKeown et al . , 2014 ) . For each protein in complex with all 16 REs , we used linear regression to analyze the statistical relationship between each biophysical parameter and affinity . We found that the number of hydrogen bonds formed across the protein-DNA interface is not positively correlated with the affinity of the TF-RE complex when all 128 combinations are examined ( Figure 6A , C ) . Thus , hydrogen bonding does not provide even a partial global explanation of affinity; however , it does explain , in part , affinity for a few specific genotypes . When we separately analyzed each protein's affinity across REs , we found that the number of hydrogen bonds was positively and significantly correlated with affinity for 2 of the 8 protein genotypes ( Figure 6C , Figure 6—figure supplement 1 ) , explaining at most 30% of the variation in affinity . These proteins contain residue glu25 , which in the crystal structure of AncSR1:ERE forms hydrogen bonds to specific nucleotide bases in the DNA major groove ( Figure 1C ) ( McKeown et al . , 2014 ) . 10 . 7554/eLife . 07864 . 008Figure 6 . Hydrogen bonding and packing efficiency do not explain TF-RE affinity . ( A ) The number of hydrogen bonds formed between atoms in the RH and atoms in the RE in molecular dynamic ( MD ) simulations is not positively correlated with the experimentally measured binding energy of TF-RE complexes . Each data point represents the number of hydrogen bonds formed by one of the 128 TF-RE pairs ( 8 variants of AncSR1 with 16 variant REs ) , each averaged over three replicate 50 ns simulations; error bars show SEM . Red line indicates best-fit linear regression model . For p-value and R2 , see panel C . ( B ) The efficiency of packing interactions across the RH-RE interface in MD simulations is not positively correlated with the experimentally measured binding energy of TF-RE complexes . In MD simulations , the number of protein-DNA atom pairs within 4 . 5 Å of each other was calculated for all 128 TF:RE complexes . Points and error bars show the mean and SEM over three replicate MD simulations . Red line indicates best-fit linear regression model; p-value and R2 are shown in C . ( C ) Correlation of hydrogen bonding and packing efficiency with binding energy for individual protein genotypes . For each TF , the experimentally measured binding energy for each of the 8 REs was regressed against either the number of hydrogen bonds formed from RH to RE or the efficiency of packing between RH and RE . The presence of positive ( blue ) , negative ( red ) , or non-significant ( NS ) correlations is indicated , along with the p-value of the correlation and the fraction of variation in binding energy explained by each dependent variable ( R2 ) . For full data sets and regressions , see Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 07864 . 00810 . 7554/eLife . 07864 . 009Figure 6—figure supplement 1 . Direct hydrogen bonding at the protein-DNA interface positively correlates with binding affinity for only 2 out of 8 protein genotypes . From MD simulations , the number of direct hydrogen bonds formed at the protein-DNA interface was calculated for each protein genotype across all 16 REs . The best-fit linear regression was determined for each data set . Blue and red lines show significant positive and negative correlations , respectively ( p < 0 . 05 ) . For best-fit linear regression terms , see Figure 6C . DOI: http://dx . doi . org/10 . 7554/eLife . 07864 . 00910 . 7554/eLife . 07864 . 010Figure 6—figure supplement 2 . Packing efficiency at the protein-DNA interface positively correlates with binding affinity for only 2 out of 8 protein genotypes . From MD simulations , the number of protein-DNA atom pairs within 4 . 5 Å of one another was calculated for each protein genotype across all 16 REs . The best-fit linear regression was determined for each data set . Of the 8 genotypes , 2 showed a positive correlation while the remaining 5 genotypes showed no significant correlation ( p < 0 . 05 ) . For linear regression terms , see Figure 6C . DOI: http://dx . doi . org/10 . 7554/eLife . 07864 . 010 Packing also does not provide a global explanation for variation in binding affinity . When all TF:RE combinations were analyzed , the efficiency of packing was not positively correlated with affinity ( Figure 6B , C ) . When analyzed separately , packing was significantly and positively associated with affinity for only two proteins , explaining at most 40% of the variance in a protein's affinity across REs ( Figure 6C , Figure 6—figure supplement 2 ) . There is no clear pattern to explain which protein genotypes manifest a correlation of packing with affinity . We conclude that no simple biophysical metric provides a general explanation of variation in affinity among TF-RE complexes . There may be common mechanisms for the effects of each amino acid replacement on affinity that are common to all REs: for example , replacement gly25GLY improved affinity for every RE , possibly by enhancing the entropic benefit of binding driven by the hydrophobic effect or decreasing the entropic cost of binding by introducing additional degrees of freedom in the protein backbone . But specificity—variation in affinity among REs—appears to be determined by biophysical interactions that are largely unique to each TF-RE combination . For example , replacement ala29VAL improves relative affinity for REs containing A4; in previously published ancestral X-ray crystal structures , the hydrophobic side chain of VAL29 packs against the hydrophobic methyl groups that are unique to the complementary T at position 4 in the RE , providing a likely explanation , at least in part , for this effect ( McKeown et al . , 2014 ) . These interactions only affect packing involving a small number of atoms , and they are present in only a small number of TF-RE pairs , so they do not have a statistically detectable effect on the relationship between packing and affinity across all REs . These results reinforce the importance of epistasis across the molecular interface . Differences in binding affinity among TF:RE complexes are typically determined by unique physical interactions between atoms on the protein and atoms on the DNA , and higher-order interactions sometimes involve more than two atoms across the interface . Linear modeling strategies have previously been used to statistically characterize the main and epistatic effects of variation in a DNA sequence on affinity for a specific TF , typically using high-throughput approaches that estimate affinity from measurements of occupancy ( Benos et al . , 2002; Stormo , 2011; Zhao et al . , 2012; Stormo et al . , 2015 ) . We extended these methods , using direct measurements of affinity for every combination of TF and RE genotypes in a defined region of sequence space , to identify the genetic determinants of protein-DNA affinity both within and between the two molecules . This strategy has both advantages and limitations . By simultaneously studying the effects of variation within the RE , within the TF , and between the two molecules were we able to describe in detail how changes in the TF protein affect the specificity of RE binding and vice versa . This approach can in principle be used to study interactions in joint sequence space for any complex of molecules , including those with more than two elements . It can also be applied to large datasets , larger numbers of sites , and higher-order interactions , so long as the data are sufficiently precise to allow robust quantitative analysis . Our experimental approach allowed us to obtain precise and direct measurements of affinity for each individual TF-RE complex , which were essential to our analysis . Statistical models of higher-order epistasis across a protein-DNA interface add many terms to the regression models previously used for individual proteins ( Zhao et al . , 2012; Stormo et al . , 2015 ) . As models become complex , estimates of higher-order determinants of affinity can become uncertain or biased if the estimates of affinity are noisy or imprecise; this problem can be exacerbated if affinities for some sequence states are estimated by interpolation rather than being measured ( Otwinowski and Plotkin , 2014 ) . By directly and precisely measuring the ΔG of binding for all TF:RE combinations , we were able to detect and estimate the magnitude and importance of higher-order epistatic effects with relatively high confidence . Further , because effects on the free energy of binding by independent factors are additive for reversible interactions , these measurements allowed us to use linear regression to rigorously estimate the magnitude and significance of main and epistatic effects on TF-RE recognition . Other metrics such as occupancy or entropy may require more complex descriptive models . Obtaining such precise and direct measurements of binding energy is time consuming and limits the number of complexes that can be studied . The 128 TF:RE combined genotypes whose affinity we measured represent a tiny slice of the vast joint sequence space of possible TFs and REs . Even within this small region of sequence space , however , we found rampant epistatic interactions within the DNA RE , within the TF protein , and between the RE and the TF . These epistatic interactions play key roles in determining the specificity of binding by each TF , and they strongly shape the functional topology of the joint sequence space and the capacity of each molecule to traverse this space under various evolutionary scenarios . As improving technical capacities allow larger tracts of joint sequence space to be analyzed and higher-level biological functions to be assessed , it seems quite likely that epistasis will be a major influence on the specificity and evolution of molecular complexes . Complexity and interdependence are often thought to act as constraints on evolution ( Lewontin , 1984; Bonner , 1988; Kauffman , 1993; Wagner and Altenberg , 1996; Orr , 2000; Schank and Wimsatt , 2001 ) . Previous studies of epistasis within a molecule have shown that interactions among mutations constrict the number of passable evolutionary trajectories through sequence space and make the outcomes of evolution contingent upon the prior occurrence of permissive mutations ( Weinreich et al . , 2006; Bridgham et al . , 2009; Podgornaia and Laub , 2015 ) . The joint sequence space of multiple interacting molecules contains more dimensions and therefore far more paths between functional joint genotypes than does either molecule's separate sequence space ( Gavrilets , 2004 ) . Our work shows that intermolecular epistasis is indeed rampant and blocks many of these additional pathways . But intermolecular epistasis also has an opposite effect , which introduces new degrees of freedom into the evolutionary process . Permissive epistatic mutations across the molecular interface open paths for the other molecule that would otherwise have been blocked . As a result , the absolute number of trajectories ( and ultimate endpoints ) available to either partner is larger than it would appear to be if only the ‘slice’ or profile through the joint sequence space represented by variability in a single molecule were considered . Under the scenario we examined , for example , an ERE regulated by an immutable AncSR1-DBD could not reach SRE without losing functionality , and AncSR1 could not reach AncSR1+RH if it regulated only immutable binding sites . But because amino acid replacements in the TF are permissive for mutations in the RE , and subsequent RE mutations are permissive for replacements in the protein , both members of the complex can explore regions of their own sequence space and reach genotypes that were not previously available to either . Molecular complexes are pervasive in biology . The extensive intermolecular epistasis that we observed among a small number of sites in a simple binary complex suggests that the reach of intermolecular epistasis may be vast . If the structure of that epistasis is anything like that present in the AncSR1-RE complex , then the web of contingent events that structures evolutionary trajectories is likely to be dense , with the evolutionary potential of any one molecule depending on prior events in other molecules . But if some of those events are permissive , as our results suggest they are likely to be , then the potential of evolution to generate new functional complexes by the combined action of mutation , drift and selection is greater than it may appear if the molecules are viewed only in isolation . Complexity and interdependence not only constrain evolution; they can also buy freedom for the parts of a system to reach new states , if the historical events are right . DBDs were cloned into the pETMALc-H10T vector ( Pryor and Leiting , 1997 ) ( a gift from John Sondek , UNC-Chapel Hill ) C-terminal to a cassette containing a 6xHis tag , maltose binding protein ( MBP ) and a TEV protease cleavage site . DBDs were expressed in BL21 ( DE3 ) pLysS Rosetta cells . Protein expression was induced by addition of 1 mM IPTG at A600 of 0 . 8–1 . 2 . After induction , cells were grown overnight at 15°C . Cells were harvested via centrifugation and frozen at −10°C overnight . Cells were lysed using B-PER Protein Extraction Reagent Kit ( Thermo Scientific ) . Lysate was loaded onto a pre-equilibrated 5 ml HisTrap HP column ( GE Fairfield , CT ) and eluted with a linear imidazole gradient ( 25 mM–1 M ) in 25 mM sodium phosphate and 100 mM NaCl buffer [pH 7 . 6] . The DBD was cleaved from the MBP-His fusion using TEV protease in dialysis buffer consisting of 25 mM sodium phosphate , 150 mM NaCl , 2 mM βME and 10% glycerol [pH 8 . 0] . The cleavage products were loaded onto a 5 ml HiPrep SP FF cation exchange column ( GE ) and eluted with a linear NaCl gradient ( 150 mM–1 M ) in 25 mM sodium phosphate buffer [pH 8 . 0] . DBDs were further purified on a Superdex 200 10/300 GL size exclusion column ( GE ) with 10 mM Tris [pH 7 . 6] , 100 mM NaCl , 2 mM βME , 5% glycerol . Protein purity was assayed after each purification by visualization on a 12% SDS-PAGE gel stained with Bio-Safe Coomassie G-250 stain ( Bio-Rad ) . DNA constructs were ordered from Eurofins Operon ( Huntsville , AL ) as HPLC-purified single stranded oligos with the forward strand labeled at the 5′-end with 6-FAM . Sequences of forward strands , with differences underlined , were as follows: CCAGGCCA , CCAGGGCA , CCAGCTCA , CCAGCACA , CCAGCCCA , CCAGCGCA , CCAGTTCA , CCAGTACA , CCAGTCCA , CCAGTGCA , CCAGACCA , CCAGAGCA , CCAGGTCA , CCAGAACA , CCAGGACA , CCAGATCA . Complementary reverse strands were also ordered . Forward and reverse strands were re-suspended in duplex buffer ( 30 mM Hepes [pH 8 . 0] , 100 mM potassium acetate ) to a concentration of 100 μM . Equimolar quantities of complementary forward and reverse strands were combined and placed in a 95°C water bath for 10 min then slowly cooled to room temperature . The double stranded product was diluted to 5 μM in water . Purified DBD was buffer exchanged using Illustra NAP-25 columns into 20 mM Tris [pH 7 . 6] , 130 mM NaCl and 5% glycerol . Protein concentration was determined by measuring absorbance at 280 nm , 320 nm and 340 nm and correcting for light scattering . A range of DBD concentrations was titrated in triplicate onto a black , NBS-coated 384 well plate ( Corning 3575 , Corning , NY ) . Labeled DNA was added to each well to achieve a final concentration of 5 nM in 91 μl total volume . Sample FP was read using a Perkin Elmer Victor X5 , exciting at 495 nm and measuring emission anisotropy at 520 nm . To determine K1 , we measured binding affinity to the half-site REs in triplicate and fit the data to a single-site binding model . The crystal structure of AncSR1 bound to ERE ( PDB: 4OLN ) was used as the starting point for all simulations . Historical substitutions and changes to the DNA RE sequences were introduced in silico ( Emsley and Cowtan , 2004 ) . Each system was solvated in a cubic box with a 10 Å margin , then neutralized and brought to 150 mM ionic strength with sodium and chloride ions . This was followed by energy minimization to remove clashes , assignment of initial velocities from a Maxwell distribution , and 1 ns of solvent equilibration in which the positions of heavy protein and DNA atoms were restrained . Production runs were 50 ns , with the initial 10 ns excluded as burn-in . The trajectory time step was 2 fs , and final analyses were performed on frames taken every 12 . 5 ps . We used TIP3P waters and the AMBER FF03 parameters for protein and DNA , as implemented in GROMACS 4 . 5 . 5 ( Duan et al . , 2003 ) . The zinc fingers were treated with a recently derived bonded potential for Cys-Zn interactions ( Hoops and Rindler , 1991; Lin and Wang , 2010 ) as previously described ( McKeown et al . , 2014 ) . Zinc finger partial charges were derived using the RED III . 4 pipeline ( Dupradeau et al . , 2010 ) as previously described ( McKeown et al . , 2014 ) . We extracted a tetrahedral Cys4 zinc finger from a 0 . 9 Å crystal structure ( Iwase et al . , 2011 ) , optimized its geometry with an explicit quantum mechanical calculation using the 6–31G** basis set ( Schuchardt et al . , 2007 ) , then derived partial charges using RESP ( Dupradeau et al . , 2010 ) . All quantum mechanical calculations were performed using the FIREFLY implementation of GAMESS ( Schmidt and Mohring , 1993; Granovsky , 2007 , ) . We verified that the zinc fingers maintained their tetrahedral geometry over the course of the simulations . Simulations were performed in the NTP ensemble at 300 K , 1 bar . All bonds were treated as constraints and fixed using LINCS ( Hess et al . , 1997 ) . Electrostatics were treated with the Particle Mesh Ewald model ( Darden and Pedersen , 1993 ) , using an FFT spacing of 12 Å , interpolation order of 4 , tolerance of 1e-5 , and a Coulomb cutoff of 9 Å . van der Waals forces were treated with a simple cutoff at 9 Å . We used velocity rescaled temperature coupling with a τ of 0 . 1 ps and Berendsen pressure coupling with a τ of 0 . 5 ps and a compressibility of 4 . 5e-5 bar−1 . Analyses were performed using VMD 1 . 9 . 1 ( Humphrey et al . , 1996 ) —with its built-in TCL scripting utility—as well as a set of in-house Python scripts ( adapted from scripts generously shared by Mike Harms , and available from Github: github . com/harmsm/md-analysis-tools ) .	Transcription factors are proteins that control which genes inside a cell are active by binding to specific short sequences of DNA called response elements . Small differences in a trancription factor's amino acid sequence or in a response element's DNA sequence can affect their ability to recognize each other . How these pairs of molecules recognize each other—and how they evolved to do so—are important questions in molecular biology and evolution . One way to understand these questions is to study ‘sequence space’ , an organized representation of all of the possible sequences of a molecule , each linked to its neighbors by single point mutations . As a molecule evolves , it follows one of many possible paths from its ancestral form to a later-day version . Some paths deliver improvements at every step , and some involve ‘neutral’ wanderings . Still other paths produce intermediate forms that work poorly or not at all; such paths are unlikely to be followed during evolution . By characterizing many different versions of a molecule and mapping their functions onto sequence space , scientists can better understand how biological molecules work and how evolution might have produced them . No one has previously explored the combined sequence space of two interacting molecules . Now Anderson , McKeown and Thornton have characterized the joint sequence space of a transcription factor that controls a cell's response to steroid hormones and the DNA response elements that it recognizes . Their experiments focused on the portion of sequence space between two ancient members of the transcription factor family which existed just before and just after a major shift in their ability to recognize different DNA sequences . To reconstruct these ancestral proteins , Anderson , McKeown and Thornton used computational methods to infer their most likely sequences based on those of hundreds of present-day members of the family and the relationships between them . These proteins were then tested in the laboratory to see how strongly each of them could bind to various DNA response elements . These proteins and their preferred response elements define the start and end of an ancient evolutionary journey through sequence space , which took place about 500 million years ago . Anderson , McKeown and Thornton then reconstructed all the possible steps on the paths between the two transcription factors and the two response elements . Every transcription factor was tested with every response element , and the information about how strongly they could bind was mapped onto the joint sequence space of the two molecules . The experiments revealed that mutations in either the DNA or the transcription factor had very different effects , depending on which other changes had already occurred elsewhere in the same molecule or in its partner . Geneticists call this phenomenon ‘epistasis’ . Because of epistasis , only a handful of paths connected the ancestral protein-DNA complex to the derived complex without passing through intermediate steps that functioned poorly . These few likely paths all involved ‘permissive mutations’—a change in the DNA that allowed the protein to tolerate a mutation that was previously detrimental , or vice versa . The findings show that the evolution of each molecule depended critically on chance events in the evolutionary history of its partner . By changing the evolutionary potential of the molecule it interacted with , the members of the complex wandered through sequence space together . This journey yielded two new molecules that now work specifically together , each with functions that are distinct from their ancestors' .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology" ]	2015	Intermolecular epistasis shaped the function and evolution of an ancient transcription factor and its DNA binding sites
Breathing in mammals is hypothesized to result from the interaction of two distinct oscillators: the preBötzinger Complex ( preBötC ) driving inspiration and the lateral parafacial region ( pFL ) driving active expiration . To understand the interactions between these oscillators , we independently altered their excitability in spontaneously breathing vagotomized urethane-anesthetized adult rats . Hyperpolarizing preBötC neurons decreased inspiratory activity and initiated active expiration , ultimately progressing to apnea , i . e . , cessation of both inspiration and active expiration . Depolarizing pFL neurons produced active expiration at rest , but not when inspiratory activity was suppressed by hyperpolarizing preBötC neurons . We conclude that in anesthetized adult rats active expiration is driven by the pFL but requires an additional form of network excitation , i . e . , ongoing rhythmic preBötC activity sufficient to drive inspiratory motor output or increased chemosensory drive . The organization of this coupled oscillator system , which is essential for life , may have implications for other neural networks that contain multiple rhythm/pattern generators . Coupled oscillator neural networks driving behavior are widespread , e . g . , for swimming ( Grillner , 2003 ) , and locomotion ( Goulding , 2009; Talpalar et al . , 2013 ) . Amongst complex and vital behaviors in mammals , breathing , an exceptionally reliable and continuous behavior throughout postnatal life , is one that we may be closest to understanding ( Feldman and Kam , 2015 ) . Not only do we know the location of the neural microcircuits that generate respiratory rhythm , but we also have direct , accurate and reliable behavioral measures of the output , i . e . , breathing . We hypothesize that the respiratory rhythm central pattern generator ( CPG ) in mammals is comprised of two oscillators ( Feldman et al . , 2013 ) : inspiratory rhythm originates in the preBötzinger Complex ( preBötC ) in the ventrolateral medulla ( Smith et al . , 1991 ) and active expiratory rhythm originates in the rostral medulla ventrolaterally adjacent to the facial nucleus ( parafacial lateral region; pFL ) ( Pagliardini et al . , 2011; Huckstepp et al . , 2015 ) . Two parafacial regions that potentially overlap and whose anatomical descriptions are incomplete and ambiguous , the retrotrapezoid nucleus ( RTN ) and parafacial respiratory group/embryonic parafacial ( pFRG/e-PF ) , contain overlapping subpopulations neurons that express the neurokinin-1 receptor ( NK1R ) , the homeobox gene Phox2b , and the glutamate transporter 2 ( VGlut2 ) ( Nattie and Li , 2002; Mulkey et al . , 2004; Onimaru et al . , 2008; Thoby-Brisson et al . , 2009; Ramanantsoa et al . , 2011 ) . Previously , we used unbiased descriptors to partition the ventral and lateral parafacial regions , designating them as parafacial lateral ( pFL ) and parafacial ventral ( pFV ) . These parafacial regions are chemosensitive ( Mulkey et al . , 2004; Onimaru et al . , 2008; Marina et al . , 2010; Onimaru et al . , 2012 ) , but can be functionally separated by: i ) their contribution to active expiration , with the pFV providing drive to expiration ( Huckstepp et al . , 2015; Silva et al . , 2016 ) and the pFL containing a presumptive expiratory rhythm generator ( Pagliardini et al . , 2011; Huckstepp et al . , 2015 ) , and; ii ) the presence of neurons with respiratory rhythmic behavior in pFL ( or presumptively equivalent areas ) ( Onimaru and Homma , 2003; Thoby-Brisson et al . , 2009; Pagliardini et al . , 2011 ) . In mammals at rest , during wakefulness and sleep , when active breathing movements are primarily inspiratory , generation of the underlying rhythm appears driven by the preBötC . As metabolic demand increases , e . g . , during exercise , the pFL appears to turn on to produce active expiration . Thus , while breathing is a unified act of inspiratory and expiratory airflow , we postulate this behavior results from the coordinated interaction of two anatomically and functionally distinct oscillators ( Mellen et al . , 2003; Janczewski and Feldman , 2006; Pagliardini et al . , 2011; Huckstepp et al . , 2015 ) . To understand the generation and control of respiration we need to determine how these two oscillators interact . For example , in adult rats the preBötC can generate inspiratory rhythm while the pFL is quiescent ( Pagliardini et al . , 2011; Huckstepp et al . , 2015 ) . Is the converse true ? To investigate their independent and interactive functions , we used a pharmacogenetic approach to selectively inhibit the preBötC and/or activate the normally quiescent pFL . We bilaterally transfected the preBötC with the Gi/o-coupled allatostatin receptors ( AlstR ) , which when activated by allatostatin ( Alst ) silences transfected preBötC neurons ( Tan et al . , 2008 ) . In the same rats , we bilaterally transfected the pFL with the Gq-coupled HM3D DREADD receptor ( HM3DR ) that when activated by clozapine-N-oxide ( CNO ) depolarizes ( Armbruster et al . , 2007 ) transfected pFL neurons . We slowed respiration in a controlled manner by titrating the dose of Alst applied to the AlstR-transfected preBötC , allowing us to examine the dynamics of this presumptive coupled oscillator system as we shifted the balance of activity from the preBötC to the pFL . Depressing preBötC activity resulted in quantal slowing of breathing , similar to slowing of breathing following opiate depression of preBötC activity in vitro and in juvenile rats ( Mellen et al . , 2003; Janczewski and Feldman , 2006 ) . As preBötC activity waned , burstlets appeared on inspiratory muscle electromyograms ( EMGs ) and in airflow , consistent with our postulate that burstlets , not bursts , in the preBötC are rhythmogenic ( Kam et al . , 2013 ) . In complementary experiments , we applied CNO to the HM3DR transfected pFL to activate this normally quiescent oscillator . We confirmed that pFL activation initiates active expiration ( Pagliardini et al . , 2011; Huckstepp et al . , 2015 ) . By combining these protocols , i . e . , silencing the preBötC while simultaneously driving the pFL , we removed some confounding factors from these initial experiments . Importantly , we found that active expiration could not be induced when preBötC inspiratory driven motor activity was suppressed and chemosensory drive was absent , indicating that in adult rats active expiration is driven by the pFL but requires an additional source of network excitation such as ongoing preBötC activity or chemosensory drive . The organization of this coupled oscillator system may have implications for other neural networks that contain multiple central pattern generators . We made two pairs of viral injections in each adult rat , i . e . , bilateral injections into the preBötC ( Figure 1A , B ) and into the pFL ( Figure 1C , D ) . In histological sections the preBötC is defined as the neurokinin-1 receptor ( NK1R ) dense area ventral to the semi-compact nucleus ambiguous ( Figure 1A , B ) and the pFL is defined as the area ventral to the lateral edge of the facial nucleus , juxtaposed to the spinal trigeminal tract ( Figure 1C , D ) ( Huckstepp et al . , 2015 ) . In representative 40 µm sections: from preBötC injection sites , 161 ± 42 , representing 82 ± 6% , neurons expressed GFP ( n=3 ) ; from pFL injection sites , 112 ± 31 , representing 81 ± 3% , neurons expressed mCitrine ( n=3 ) . Transfection sites ranged from ~350–600 µm in diameter . As the responses to Alst and CNO did not differ between the largest and smallest injections sites , the effects of receptor activation were due to silencing or driving of the preBötC and pFL respectively and not to spread of virus to neighboring regions . We found no labeling of neurons in regions of the medulla other than within the injection sites ( data not shown ) . In particular , we found no fluorescent reporters , i . e . , eGFP or mCitrine , in the Bötzinger complex ( BötC; Figure 1E , F ) . 10 . 7554/eLife . 14203 . 003Figure 1 . Transfection of neurons in preBötC and pFL . ( A ) Localization of preBötC viral injections . Transverse view of medulla at Bregma -12 . 7 mm , green circle shows location of AlstR-GFP expressing neurons . Dashed blue box indicates location of immunocytochemistry shown in Bi . ( B ) Histological analysis of preBötC . ( Bi ) preBötC injection site: neurons identified by NeuN staining ( blue ) transfected with AlstR expressing GFP ( green ) , co-localized with NK1R ( red ) . ( Bii ) Expanded from Bi ( dashed white boxes ) . ( C ) Localization of pFL viral injections . Transverse view of medulla at Bregma -11 . 1 mm , green circle shows location of HM3DR-mCitrine expressing neurons . Dashed blue box indicates location of immunocytochemistry shown in Di . ( D ) Histological analysis of pFL . ( Di ) pFL injection site: neurons identified by NeuN staining ( blue ) transfected with HM3D receptor ( HM3DR ) expressing mCitrine ( green ) , co-localized with NK1R ( red ) . ( Dii ) Expanded micrographs from merged figures in Di ( dashed white boxes ) : NeuN ( blue ) , mCitrine ( Green ) , and NK1R ( red ) . ( E ) Histological analysis of medulla at the level of the Bötzinger Complex ( BötC ) . Transverse view of medulla at Bregma -12 . 0 mm . Dashed blue box indicates location of immunocytochemistry shown in Fi . ( Fi ) No transfected neurons were found in the BötC: neurons identified by NeuN staining ( blue ) transfected with AlstR or HM3DR co-expressing GFP ( green ) , colocalized with NK1R ( red ) . ( Fii ) Expanded from Fi ( dashed white boxes ) . cNA – compact nucleus ambiguous , scNA – semi-compact nucleus ambiguous , SP-5 – spinal trigeminal tract , 12n – hypoglossal nucleus , 7n – facial nucleus , Py – pyramidal tract , MeV – medial vestibular nucleus , MeVMC – medial vestibular nucleus: magnocellular part , MeVPC - medial vestibular nucleus: parvocellular part , SOL – nucleus of the solitary tract , ROb – raphe obscurus , RP – raphe pallidus , RM – raphe magnus , ml – medial lemniscus , RVLM – rostral ventrolateral medulla , IOM - inferior olive , medial nucleus , IOP - inferior olive principle nucleus , ICP - inferior cerebellar peduncle ( restiform body ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14203 . 003 Hyperpolarizing preBötC AlstR-transfected neurons reduced f , ultimately progressing to apnea ( Tan et al . , 2008 ) . In anesthetized adult rats at rest transfected with AlstR in the preBötC ( n=8; Figure 2 ) , Alst injected bilaterally into the preBötC initially decreased f ( 42 iqr 20 to 22 iqr 11 s; p=0 . 008 ) , increased TI ( 0 . 4 iqr 0 . 1 to 0 . 5 iqr 0 . 2 s; p=0 . 04 ) and TE ( 1 . 0 iqr 0 . 7 to 3 . 2 iqr 1 . 0 s; p=0 . 008 ) , decreased VT ( 4 . 6 iqr 0 . 7 to 2 . 4 iqr 0 . 8 mL; p=0 . 008 ) and ∫DiaEMG ( 25 iqr 13 to 15 iqr 4 arbitrary units ( a . u . ) ; p=0 . 8 ) , and did not alter ∫GGEMG ( 8 . 6 iqr 9 . 8 to 11 . 1 iqr 13 . 7 a . u . ; p=0 . 008 ) . When rats became mildly hypercapnic ( ETCO2 increased from 39 . 7 iqr 5 . 4 to 43 . 4 iqr 10 . 6 mm Hg; p=0 . 008 ) , they exhibited expiratory-modulated ∫AbdEMG ( 0 . 3 iqr 0 . 2 to 9 . 7 iqr 9 . 3 a . u . , e , g . , Figure 2Aii , Bii ) , the signature of active expiration ( Pagliardini et al . , 2011 ) . After inspiratory activity ceased ( 10 . 4 iqr 2 . 7 mins post Alst injection ) and rats became severely hypercapnic ( ETCO2 57 . 8 iqr 22 . 3 mm Hg ) , ∫AbdEMG briefly became tonic ( 17 iqr 5 s . ; Figure 2Aiii , Biii ) before becoming silent , i . e . , no activity . 10 . 7554/eLife . 14203 . 004Figure 2 . Hyperpolarizing preBötC neurons reduced ventilation and induced active expiration , but eventually resulted in apnea . ( A ) Effect of Alst application to left preBötC ( unilateral , first red arrow and dashed line ) then right preBötC ( bilateral , second red arrow and dashed line ) ; gray arrow and dashed line mark onset of mechanical ventilation . ( B ) Expanded traces from A , indicated by shaded epochs ( i-iii ) : ( Bi ) Activity at rest . ( Bii ) and ( Biii ) Activity following bilateral Alst injection . ( C ) Comparison between ventilation in rats at rest ( Rst ) and following Alst: ( Ci ) before Alst had taken full effect ( ≈Bii ) . ( Cii ) After Alst had taken full effect ( ≈Biii ) . Lines connect data from individual experiments , box and whisker plots show combined data . Data are normalized to highest parameter , i . e . , f , TI , TE , VT , ∫GGEMG , ∫DiaEMG , or ∫AbdEMG , value regardless of whether it belonged to control or Alst group . f – frequency , TI – inspiratory period TE , – expiratory period , VT – tidal volume , ∫GGEMG – integrated genioglossus electromyogram , ∫DiaEMG – integrated diaphragm electromyogram , ∫AbdEMG – integrated abdominal electromyogram . DOI: http://dx . doi . org/10 . 7554/eLife . 14203 . 004 The decrease in f was not continuous ( n=8; Figures 2Bii , 3A–C ) , but instead quantal ( Mellen et al . , 2003 ) . In a representative example ( Figure 3C ) , kernel density plot estimations determined the optimal bandwidth , i . e . , bin size , to be 0 . 44 s and revealed a multimodal distribution with 3 peaks: control respiratory periods were ~2 . 1 s , and after Alst injections into the preBötC , respiratory periods increased by quantal multiples of this baseline , i . e . , 4 . 2 and 6 . 3 s ( Figure 3Ci ) . During longer periods , low amplitude activity , which was not present during eupnea , appeared at times when normal breaths were expected ( n=8; Figure 3A , B red arrows ) . This low amplitude DiaEMG and airflow activity are postulated to represent the inspiratory motor outflow manifestation of preBötC burstlets ( Kam et al . , 2013 ) , i . e . , low levels of rhythmogenic neural population activity in the preBötC that under normal conditions occur in the absence of motor output , being transmitted to motoneuron pools ( see Figure 7A in Kam et al . , 2013 ) . Ultimately , inspiratory activity ceased , i . e . , no phasic ∫DiaEMG or ∫GGEMG , and expiratory-related ∫AbdEMG , i . e . , active expiration , terminated , resulting in apnea ( n=8; Figure 2; p=0 . 008 for all variables ) . 10 . 7554/eLife . 14203 . 005Figure 3 . Hyperpolarizing preBötC neurons leads to quantal slowing of breathing and burstlet-like activity in ∫DiaEMG . ( A ) Burstlet-like activity in airflow and ∫DiaEMG traces ( red arrows ) . ( B ) Traces at different time points ( top to bottom ranging from -5 to +10 min ) after Alst infusion showing burstlet-like activity . ( C ) Quantal slowing of breathing . ( Ci ) Raster plot of respiratory period before and after Alst . ( Cii ) Kernel density estimations determined the optimal bandwidth , i . e . , bin size , of 0 . 44 s , revealing a multimodal distribution with respiratory periods at quantal intervals of ~2 . 1 s . DOI: http://dx . doi . org/10 . 7554/eLife . 14203 . 005 We tested for nonspecific effects of Alst injections in non-transfected rats ( n=8; Figure 4 ) . Injection of Alst into the preBötC did not alter f ( 47 iqr 12 to 54 iqr 12 s; p=0 . 1 ) , TI ( 0 . 3 iqr 0 . 0 to 0 . 3 iqr 0 . 0 s; p=0 . 1 ) , TE ( 1 . 0 iqr 0 . 3 to 0 . 8 iqr 0 . 3 s; p=0 . 3 ) , VT ( 4 . 7 iqr 1 . 3 to 4 . 7 iqr 1 . 3 mL; p=0 . 3 ) , ∫DiaEMG ( 34 iqr 13 to 34 iqr 14 a . u . ; p=0 . 5 ) , ∫GGEMG ( 14 iqr 11 to 13 iqr 10 a . u . ; p=0 . 8 ) , nor induce expiratory-related ∫AbdEMG ( 0 . 3 iqr 0 . 1 to 0 . 3 iqr 0 . 1 a . u . ; p=0 . 3 ) . As there were no effects of Alst in the absence of AlstRs , the cessation of inspiratory activity and termination of expiratory-related ∫AbdEMG following injections of Alst into the preBötC of transfected rats resulted from inactivation of AlstR-transfected preBötC neurons . 10 . 7554/eLife . 14203 . 006Figure 4 . Alst in absence of AlstRs does not affect breathing . ( A ) Effect of Alst application to left preBötC ( unilateral , first red arrow and dashed line ) then right preBötC ( bilateral , second red arrow and dashed line ) . ( B ) Expanded traces from A , indicated by shaded epochs ( i–ii ) : ( Bi ) Activity at rest . ( Bii ) Activity following bilateral Alst injection . ( C ) Comparison between ventilation in rats at rest ( Rst ) and following Alst . Lines connect data from individual experiments , box and whisker plots show combined data . Data are normalized to highest parameter , i . e . , f , TI , TE , VT , ∫GGEMG , ∫DiaEMG , or ∫AbdEMG , value regardless of whether it belonged to control or Alst group . Abbreviations defined in Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 14203 . 006 As we previously found nonspecific effects of CNO at a concentration of 100 µM ( Huckstepp et al . , 2015 ) , we tested for nonspecific effects of 90 µM CNO in non-transfected rats ( n=8; Figure 5 ) . 90 µM CNO did not alter f ( 54 iqr 11 to 53 iqr 9 s; p=0 . 6 ) , TI ( 0 . 3 iqr 0 . 0 to 0 . 3 iqr 0 . 0 s; p=0 . 2 ) , TE ( 0 . 8 iqr 0 . 3 to 0 . 8 iqr 0 . 2 s; p=0 . 6 ) , VT ( 4 . 6 iqr 1 . 5 to 4 . 7 iqr 1 . 6 mL; p=0 . 5 ) , DiaEMG ( 33 iqr 13 to 34 iqr 14 a . u . ; p=1 . 0 ) , GGEMG ( 13 iqr 12 to 13 iqr 13 a . u . ; p=0 . 1 ) , nor induce expiratory-related ∫AbdEMG ( 0 . 3 iqr 0 . 1 to 0 . 3 iqr 0 . 1 a . u . ; p=0 . 3 ) . There were no effects of 90 µM CNO in the absence of HM3DRs . 10 . 7554/eLife . 14203 . 007Figure 5 . CNO in absence of HM3DRs does not affect respiration . ( A ) Effect of CNO applied to ventral surface ( pink shaded area ) . ( B ) Expanded traces from A , indicated by shaded epochs ( i-–ii ) : ( Bi ) Activity at rest . ( Bii ) Activity in presence of CNO . ( C ) Comparison between ventilation in rats at rest and in presence of CNO . Lines connect data from individual experiments , box and whisker plots show combined data . Data are normalized to highest value for that parameter , i . e . , f , TI , TE , VT , ∫GGEMG , ∫DiaEMG , or ∫AbdEMG regardless of whether it belonged to control or CNO group . Abbreviations defined in Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 14203 . 007 We predicted that depolarizing pFL neurons would elicit active expiration , similar to the effect of their disinhibition or optogenetic photoactivation ( Pagliardini et al . , 2011 ) . In anesthetized adult rats at rest transfected with HM3DRs in the pFL ( n=8; Figure 6 ) , CNO ( 90 µM ) increased f ( 37 iqr 12 to 42 iqr 7 s; p=0 . 04 ) , did not alter TI ( 0 . 3 iqr 0 . 1 to 0 . 3 iqr 0 . 1 s; p=0 . 8 ) decreased TE ( 1 . 3 iqr 0 . 6 to 1 . 1 iqr 0 . 3 s; p=0 . 02 ) , increased VT ( 3 . 9 iqr 1 . 3 to 4 . 2 iqr 1 . 3 mL; p=0 . 02 ) , ∫DiaEMG ( 25 iqr 12 to 27 iqr 12 a . u . ; p=0 . 02 ) , and ∫GGEMG ( 7 . 6 iqr 6 . 6 to 8 . 4 iqr 5 . 6 a . u . ; p=0 . 04 ) , and induced expiratory-related ∫AbdEMG ( 0 . 3 iqr 0 . 2 to 3 . 0 iqr 4 . 0 a . u . ; p=0 . 02 ) . Thus depolarizing pFL neurons altered respiration in a similar manner to optogenetic photoactivation ( Pagliardini et al . , 2011 ) for all measured variables common to both studies . We conclude that depolarization of HM3DR-transfected pFL neurons resulted in active expiration . 10 . 7554/eLife . 14203 . 008Figure 6 . Depolarizing pFL neurons elicits active expiration . ( A ) Effect of CNO applied to ventral surface ( pink shaded area ) . ( B ) Expanded traces from A , indicated by grey shaded epochs ( i-ii ) : ( Bi ) Activity at rest . ( Bii ) Activity in presence of CNO . ( C ) Comparison between ventilation in rats at rest and in presence of CNO . Lines connect data from individual experiments , box and whisker plots show combined data . Data are normalized to highest value for that parameter , i . e . , f , TI , TE , VT , ∫GGEMG , ∫DiaEMG , or ∫AbdEMG regardless of whether it belonged to control or CNO group . Abbreviations defined in Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 14203 . 008 The preBötC can generate inspiratory rhythm while the pFL is quiescent ( Pagliardini et al . , 2011 ) . Can the pFL generate a respiratory rhythm when the preBötC is quiescent ? If active expiration is independent of inspiratory activity , then active expiration driven by activation of HM3DRs in pFL should persist after cessation of inspiratory activity resulting from injection of Alst in the preBötC . In anesthetized adult rats with an active expiratory breathing pattern induced by CNO , injection of Alst into the preBötC led to a significant decrease in respiratory activity ( n=8; p=0 . 008 for all variables; Figure 7 ) , ultimately resulting in apnea , with no rhythmicity in VT , ∫DiaEMG , ∫GGEMG , and ∫AbdEMG . After rhythmic inspiratory activity ceased , ∫AbdEMG continued tonically for a short duration ( 15 iqr 12 s; Figure 7A , Bii ) before disappearing , which was not different from the duration of tonic ∫AbdEMG following the onset of apnea in the absence of CNO ( Figure 2Biii; p=0 . 7 ) . When inspiratory activity stopped , rats were severely hypercapnic ( ETCO2 64 . 6 iqr 14 . 5 mm Hg; Figure 7Bii ) , this was not different from the hypercapnia at the onset of apnea in the absence of CNO ( ETCO2 57 . 8 iqr 22 . 3 mm Hg; p=0 . 8 ) . Rats were mechanically ventilated to restore CO2 and O2 levels to within the normal range ( Figure 7 ) . To assess the state of the respiratory CPG the mechanical ventilator was intermittently turned off . Shortly after allatostatin injections into the preBötC , no respiratory activity was seen following cessation of mechanical ventilation , and rats were placed back on the ventilator . As the time from allatostatin injection in the preBötC increased , removal from the ventilator led to brief periods of spontaneous breathing . When spontaneous breathing occurred , the return of inspiratory activity always preceded the return of expiratory abdominal activity; active expiration did not return until VT reached 3 . 2 iqr 1 . 7 mL , ∫DiaEMG reached 13 . 1 iqr 8 . 2 a . u . , and ∫GGEMG reached 5 . 2 iqr 4 . 2 a . u . ( n=8; Figure 8 ) . That is , at low levels of reinitiated inspiratory activity there was no active expiration , which appeared only after inspiratory activity reached a higher value . If rats were able to spontaneously breathe without ventilation , they were allowed to do so . However , most early periods of spontaneous breathing deteriorated , and when rats became apneic again they were re-ventilated; in 5 of 6 rats spontaneous breathing occurred during ventilation , and in the remaining rat , spontaneous breathing following removal from the ventilator was sustained . When spontaneous breathing occurred during ventilation the return of inspiratory activity always preceded the return of expiratory abdominal activity . Therefore , following silencing of the preBötC with Alst in the presence of CNO to drive the pFL , at no time during the re-initiation of spontaneous breathing , either with or without mechanical ventilation , did active expiration occur in the absence of inspiratory activity and chemosensory drive . 10 . 7554/eLife . 14203 . 009Figure 7 . Hyperpolarizing preBötC neurons leads to apnea , and loss of active expiration even with activation of pFL . ( A ) Integrated traces from a single experiment showing effect of Alst injection to left preBötC ( unilateral , first red arrow and dashed line ) then right preBötC ( bilateral , second red arrow and dashed line ) , in presence of CNO ( pink shaded area ) ; gray arrow and dashed line mark onset of mechanical ventilation . ( B ) Expanded traces from A indicated by shaded epochs: ( Bi ) In presence of CNO only . ( Bii ) Following Alst in presence of CNO . ( C ) Comparison between ventilation in rats in presence of CNO and following Alst in presence of CNO . Lines connect data from individual experiments , box and whisker plots show combined data . Data are normalized to highest value for that parameter , i . e . , ∫GGEMG , ∫DiaEMG , or ∫AbdEMG regardless of whether it belonged to CNO or CNO with Alst group . Abbreviations defined in Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 14203 . 00910 . 7554/eLife . 14203 . 010Figure 8 . Following apnea , in presence of CNO , active expiration only returns after inspiratory activity returns . ( A ) Integrated traces from one experiment showing effect of removing rat from ventilator ( dark grey shaded area ) , in presence of CNO ( pink shaded area ) , following induction of apnea by microinjection of Alst into the preBötC of preBötC-AlstR transfected rats . Intervening period between the two traces , whilst ventilation was ongoing and continuous , has been removed ( double break ) , so traces can be expanded . ( B ) Expanded traces from A ( indicated by black dashed lines ) , showing how measurements were taken for inspiratory parameters when active expiration returned . ( C ) Inspiratory parameters when active expiration returned . Dots represent individual experiments , box and whisker plots show combined data . ( Ci ) Tidal volume ( VT ) ( Cii ) ∫DiaEMG , ( Ciii ) ∫GGEMG . ( D ) Integrated traces from one experiment showing the return of inspiratory and expiratory activity during mechanical ventilation , in presence of CNO ( pink shaded area ) , following induction of apnea by microinjection of Alst into the preBötC of preBötC-AlstR transfected rats . Abbreviations defined in Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 14203 . 010 Our experimental design required long lasting activation of receptors at 4 separate loci , i . e . , preBötC bilaterally and pFL bilaterally . Optogenetics can have millisecond resolution , making them a valuable tool for studying breathing on a breath by breath basis , e . g . , ( Pagliardini et al . , 2011 ) . However opsins can rapidly ( >seconds ) desensitize and their activation requires placement of optic fibers to illuminate each transfected area , a particularly challenging problem for four regions in the brainstem . Therefore , we chose a pharmacogenetic approach to excite the pFL by activation of HM3DRs and inhibit the preBötC by activation of AlstRs . This way we were able to switch neurons in these regions on or off over a period of minutes by application of CNO to the ventral surface of the medulla , and careful titration of the dose and infusion rate of Alst injected into the preBötC . We note AAV2/5 can retrogradely transfect of some types of afferent neurons , e . g . , injections of AAV2/5 into the entorhinal cortex retrogradely labels a subset of dentate gyrus neurons , but not the dense afferent projections from other nuclei ( Aschauer et al . , 2013 ) . We assert that our results are unlikely to be confounded by retrograde transfection of distant neurons: i ) Other than the injection site , we found no labeling of medullary neurons following transfection of the pFL with AAV2/5 viruses that coexpress mCitrine with a DREADD receptor , in this ( Figure 1E , F ) or our previous ( Huckstepp et al . , 2015 ) study; ii ) CNO was applied directly to the ventral medullary surface , and the effective concentration for receptor activation would be limited to the first few hundred microns beneath the ventral surface , and; iii ) we find essentially the same effects of activation of neurons in the pFL as optogenetic photoactivation of the same neurons transfected with lentivirus ( Pagliardini et al . , 2011 ) , which is not retrogradely transported . Here we discuss the output of the preBötC and pFL , which ultimately form the final output of the respiratory network . Though we do not refer to other nuclei , we do not rule out their contribution to the control of respiration; for example following silencing of the preBötC , the increase in chemosensory drive to the respiratory oscillators , may come from increased drive from other respiratory-related nuclei , such as the pontine nuclei , i . e . , locus coreleus , parabrachial nucleus , and the Kolliker Fuse nucleus , the ventral respiratory group or other medullary nuclei , i . e . , RTN/pFV , medullary raphe , nucleus tractus solitarii , or even astrocytes . In addition , due to the necessity to perform injections into the medulla , we were unable to record directly from the preBötC or pFL . Therefore , the activity of the preBötC and pFL , was assessed by motor output recorded from respiratory muscles , which may not always reflect the activity of these oscillators , as they may have subthreshold activity , e . g . , ( Kam et al . , 2013 ) . Hyperpolarizing preBötC neurons initially decreased f , VT , and ∫DiaEMG ( Figures 2 , 3 , see also Tan et al . , 2008 ) . Breathing slowed in a quantal manner ( Mellen et al . , 2003 ) with missed inspiratory bursts , rather than through a gradual increase in period ( Figures 2Bii , 3A–C ) . Concurrently , as drive within the preBötC diminished and chemosensory drive increased , presumptive burstlets in the preBötC ( Kam et al . , 2013 ) appeared to be transmitted to motor pools , leading to low level inspiratory motor activity , i . e . , ∫DiaEMG and ∫GGEMG , resulting in minimal inspiratory airflow ( Figure 3A–B ) . Here , lowered drive in the preBötC from activation of the allatostatin receptor creates burstlets in the preBötC; under normal conditions these burstlet signals are not strong enough to drive motor activity . However , increased excitability in the premotor and motor network caused by increased chemosensory drive ( from decreased ventilation ) causes these normally subthreshold events to become suprathreshold , and thus these burstlets are transmitted to the motor output . During quantal slowing of breathing , inspiratory activity on the DiaEMG and GGEMG was interspersed with AbdEMG activity ( Figures 2A-B , 7A–B ) . Interestingly , even when expiratory activity was at its highest , low level preBötC activity , as seen as burstlets on the DiaEMG and GGEMG was still able to inhibit AbdEMG activity ( Figure 7Bii ) . These observations are consistent with our hypothesis that burstlets originate in the preBötC and , under atypical conditions can be transmitted to motoneuron pools to be seen as small events in muscle EMGs ( Kam et al . , 2013 ) . Following hyperpolarization of preBötC neurons by activation of AlstRs , active expiration appeared ( Figures 2 , 3 , 9Dii ) . Hypoxia and hypercapnia can themselves induce active expiration ( Iizuka and Fregosi , 2007; Huckstepp et al . , 2015 ) , whether changes in blood gases are sensed by pFL neurons , or by neurons driving the pFL , remains to be determined . In either case , active expiration could have been due to increased hypoxic and/or hypercapnic drive to the pFL as inspiratory movements waned ( Figures 2A–B , 7A–B , 9Dii ) . Alternatively , active expiration could be due to disinhibition of a conditional expiratory oscillator resulting from a loss of ( presumptive ) inhibitory preBötC drive ( Kuwana et al . , 2006; Morgado-Valle et al . , 2010 ) to the pFL ( Figure 9Dii ) . Subsequently , when inspiratory motor outflow disappeared and hypercapnia and hypoxia inexorably increased , active expiration ceased ( Figures 2 , 9Div , see also Tan et al . , 2008 ) , perhaps due to the inhibitory effect of severe hypoxia on expiratory motor output ( Sears et al . , 1982; Fregosi et al . , 1987 ) . At normal blood gas levels , active expiration may be seen in the absence of inspiratory motor activity resulting from lung inflation , e . g . , in juvenile rat ( Janczewski and Feldman , 2006 ) ( Figure 9C ) . Nonetheless , from the data presented here in the adult rat , we conclude a necessary role of preBötC excitatory drive in generating active expiration in normal breathing ( discussed further below ) . Depolarizing pFL neurons by activation of HM3DRs ( Figure 6 ) changes breathing in a manner similar to their disinhibition or optogenetic photostimulation ( Pagliardini et al . , 2011 ) . In the presence of inspiratory motor output , depolarizing pFL neurons led to substantial active expiration , but when inspiration ceased so did active expiration ( Figures 7 , 9Div ) . Tupal et al observed the absence of expiratory rhythm in perinatal mice lacking Dbx1 neurons and concluded that Dbx1-derived parafacial neurons are an essential component of an expiratory oscillator ( Tupal et al . , 2014 ) . Our data suggests an alternative interpretation , that suppressing preBötC neuron activity , either acutely as done here or perhaps genetically via Dbx1 deletion , is sufficient to prevent active expiration , without any need to invoke explicit perturbations of the pFL . As rats went from eupnea to apnea following suppression of preBötC activity , and after initiation of active expiration , ∫AbdEMG transitioned from phasic to tonic: ∫AbdEMG oscillated with a normal antiphase relationship to inspiration when preBötC drive and hypercapnia were moderate , but became tonic during apnea accompanied by severe hypercapnia ( Figures 2A–B , 7A–B , 9Diii ) . Similarly , in anesthetized cats as ventilation moves in the opposite direction from hypocapnic apnea to eupnea with a resultant increase in CO2 , expiratory motor activity transitions from tonic to phasic ( Sears et al . , 1982 ) . We suggest , like Sears et al , that near the transition from eupnea to apnea , expiratory activity is phasic due to periodic inhibition of tonic activity during inspiration , and becomes tonic once the phasic inspiratory inhibition is lost ( Figures 2Bii–iii , 7Bii , 9Diii ) . We suggest the loss of phasic inhibition is the result of silencing of inhibitory preBötC inspiratory neurons by Alst ( Kuwana et al . , 2006; Morgado-Valle et al . , 2010 ) , implying a significant role for inhibitory preBötC neurons in shaping expiratory output . Interestingly , tonic ∫ABDEMG activity was not seen during the transition from apnea to eupnea ( Figures 8Aii , 9Dv ) in contrast to its presence during the reverse transition from eupnea to apnea ( Figures 7Bii , 9Diii ) . Following apnea during ventilation to maintain blood gases , when pFL neurons were excited by activation of HM3DRs there were three phases to the re-initiation of breathing . Initially , soon after silencing the preBötC , no inspiratory or expiratory activity was seen , even in the absence of mechanical ventilation . Next , presumably as the effect of Alst on preBötC neurons was waning , no inspiratory or expiratory activity was seen during ventilation , but spontaneous breathing was present upon removal from the ventilator . Here , spontaneous breathing was likely due to increased chemosensory drive to the preBötC overcoming the waning hyperpolarizing effect of the Alst on preBötC neurons . Ultimately , as spontaneous breathing continued and chemosensory drive diminished , rats would once again become apneic and require mechanical ventilation ( Figure 8A–C ) . During this phase , upon removal from the ventilator , ∫AbdEMG did not return until inspiratory motor , and presumably preBötC , activity reached a threshold level ( Figure 8A–C ) . Finally , spontaneous breathing occurred during ventilation , once the effect of Alst on preBötC neurons had worn off ( Figure 8D ) . During this phase , inspiratory activity on the GGEMG and DiaEMG always returned before AbdEMG ( Figure 8D ) . At no time during mechanical ventilation , or during the periods where rats were briefly removed from the ventilator , did active expiration occur in the absence of inspiration . As blood gases were normal ( Figure 8 ) , the loss of expiratory activity is unlikely to be due to excessive hypoxia or hypercapnia . Thus there appears to be a minimum level of respiratory network excitability required for active expiration . In the absence of preBötC activity , this network excitability may be provided by chemosensory drive , accounting for the ∫AbdEMG activity seen shortly before the onset of apnea . However following recovery from apnea the increase in network excitability is provided by the preBötC either directly through its extensive excitatory projections throughout the respiratory network ( Tan et al . , 2010 ) , e . g . , the pFV that can modulate expiratory activity ( Marina et al . , 2010; Huckstepp et al . , 2015 ) , or indirectly when VT became large enough to induce mechanosensory feedback ( that can provide expiratory drive , e . g . , Remmers , 1973; Davies and Roumy , 1986; Janczewski and Feldman , 2006 ) . We have developed a new strategy using complementary pharmacogenetics for studying coupled oscillator systems . By independently altering the excitability of two anatomically and functionally separate respiratory oscillators , we have uncovered a fundamental interaction and further delineated their role within the breathing CPG . We conclude that though respiration results from the interaction of two distinct oscillators , the preBötC ( inspiration ) and the pFL ( expiration ) are not organized as symmetrical half centers . Instead in the adult rat , the preBötC is the dominant oscillator and the pFL is the subsidiary conditional oscillator that is normally suppressed at rest; whereas the preBötC can drive breathing alone , the pFL is unable to drive breathing ( including active expiration ) in the absence of an additional form of network excitation , i . e . , ongoing rhythmic preBötC activity sufficient to drive inspiratory motor output or increased chemosensory drive when changes in blood gases are below the threshold where hypoxia is sufficient to inhibit abdominal muscle recruitment . This hierarchy is established by the sensitivity of the system to each oscillator , even when preBötC drive is low , it is able to drive inspiratory bursts and inhibit expiration , whereas even when pFL drive is high , it requires a certain amount of network excitability from other sources to drive expiratory activity . This asymmetrical organization may be relevant to other neural networks that contain hierarchically organized coupled oscillators for pattern generation , such as may underlie the asymmetrical acts of flexion and extension in locomotion ( Grillner , 2003; Talpalar et al . , 2013 ) . The interactions of the preBötC and pFL change with development and maturation ( Figure 9 ) , and with state , represent another layer of complexity in understanding the neural control of breathing . Two different viruses were used: AAV-2/5 hSyn-HA-hM3D ( Gq ) -IRES-mCitrine ( 6 × 1012 vp/ml; HM3DR; UNC Gene Therapy Vector Core , Chapel Hill , NC ) , and; AAV-DJ synapsin-allatostatin receptor-GFP ( Huckstepp et al . , 2015 ) ( 5 . 8 × 1013 vp/ml; AlstR; SALK institute GT3 Core , La Jolla , CA ) . The viruses were stored at -80°C . For injection , viruses were held at 4°C and loaded into pipettes . All protocols were approved by the UCLA Chancellor’s Animal Research Committee . Male Sprague-Dawley rats ( 350–450 g ) were anesthetized by intraperitoneal ketamine ( 100 mg/kg; Clipper Distribution Co , St Joseph , MO ) , xylazine ( 10 mg/kg; Lloyd Inc , Shenandoah , IA ) , and atropine ( 1 mg/kg; Westward Pharmaceutical Co , Eatontown , NJ ) , supplemented with isofluorane ( 0 . 5–2%; Piramal Healthcare Ltd , India ) as required . Rats were placed prone in a stereotaxic apparatus ( Kopf Instruments , St , Tujunga , CA ) , with the head positioned with Bregma 5 mm below Lambda , on a heating pad to maintain body temperature at 37 ± 0 . 5°C . The dorsal medullary surface was exposed and pipettes placed stereotaxically into the preBötC or pFL . Coordinates were ( lateral , rostral , ventral from obex in mm ) : preBötC; ( 2 . 0 , 0 . 9 , 2 . 8 ) , and pFL; ( 2 . 5 , 1 . 8 , 3 . 5 ) . Viral solutions were pressure injected ( 100–150 nL ) with a Picospritzer II ( General Valve Corp , Fairfield , NJ ) controlled by a pulse generator . Pipettes were left in place for 3–5 min to prevent solution backflow up the pipette track . We injected AlstR-expressing AAV in preBötC and HM3DR-expressing AAV in pFL . Postoperatively , rats received buprenorphine ( 0 . 1 mg/kg; Reckitt Benckiser , UK ) intraperitoneally and meloxicam ( 2 mg/kg; Norbrook Inc , UK ) subcutaneously , and antibiotics ( 10 days; TMS: Hi-Tech Pharmacal , Amityville , NY ) and meloxicam ( 4 days; 0 . 05 mg/mL ) in their drinking water . Rats were allowed 3–6 weeks for recovery and viral expression , with food and water ad libitum . Anesthesia was induced with isofluorane and maintained with urethane ( 1 . 2–1 . 7 g/kg; Sigma , St Louis , MO ) in sterile saline via a femoral catheter . Rats were placed supine in a stereotaxic apparatus on a heating pad to maintain body temperature at 37 ± 0 . 5°C . The trachea was cannulated , and respiratory flow was monitored via a flow head ( GM Instruments , UK ) . A capnograph ( Type 340: Harvard Apparatus , Holliston , MA ) was connected to the tracheal tube to monitor expired CO2 , as a proxy of blood gases homeostasis . Paired electromyographic ( EMG ) wires ( Cooner Wire Co , Chatsworth , CA ) were inserted into genioglossal ( GG ) , diaphragmatic ( Dia ) , and oblique abdominal muscles ( Abd ) . Anterior neck muscles were removed , a basiooccipital craniotomy exposed the ventral medullary surface , and the dura was resected . A bilateral vagotomy was performed to remove confounding factors such as feedback from lung stretch receptors that can drive abdominal activity ( Janczewski and Feldman , 2006 ) , after which exposed tissue around the neck and mylohyoid muscle were covered with dental putty ( Reprosil; Dentsply Caulk , Milford , DE ) to prevent drying . As micturition is inhibited under anesthesia , rats bladders were expressed preceding and during the experiment to remove any risk of autonomic dysreflexia from bladder distension; to maintain fluid balance rats were given an IP injection of saline every time the bladder was expressed . Rats were left for 30 min for breathing to stabilize . At rest , spontaneous breathing consisted of alternating active inspiration and passive expiration . From a ventral approach allatostatin ( Alst; 10 μM; ~100-200 nL; Antagene Inc , Sunnyvale , CA ) in sterile saline was injected bilaterally into the preBötC to hyperpolarize neurons transfected by AlstRs . Coordinates were ( lateral from the basilar artery , caudal from the rostral hypoglossal nerve rootlet , dorsal from the ventral surface in mm ) : preBötC; ( 2 . 0 , 0 . 6 , 0 . 7 ) . Small adjustments were made to avoid puncturing blood vessels . Rats were ventilated for the duration of ensuing apnea . After breathing stabilized , clozapine-N-oxide ( CNO; 90 µM; Santa Cruz Biotechnology , Dallas , TX ) in sterile saline was applied to the ventral medullary surface to depolarize pFL neurons transfected with HM3DRs . Once breathing stabilized , rats received a second set of bilateral injections of Alst . Though rats were ventilated with room air for the duration of ensuing apnea , they were intermittently removed from the ventilator to assess spontaneous breathing , and drives to inspiration and expiration . Ventilation depths and speed were chosen to match end-tidal CO2 to that when the rat was spontaneously breathing room air . Once rats were spontaneously breathing and no longer required ventilation , CNO was removed from the ventral surface of the medulla and the medulla was washed in PBS . All 8 rats underwent the entire procedure , i . e . , there were 8 biological replicates , and were only exposed to each condition once , i . e . , there were no technical replicates . In age-matched rats not transfected with AlstR- or HM3DR-expressing AAVs , we injected Alst into the preBötC or applied CNO to the medullary surface to see if these protocols produced non-specific effects All 8 rats underwent the entire procedure , i . e . , there were 8 biological replicates , and were only exposed to each condition once , i . e . , there were no technical replicates . Rats were sacrificed by overdose of urethane and transcardially perfused with saline followed by cold ( 4°C ) paraformaldehyde ( PFA; 4% ) . The medulla was harvested and postfixed in 4% PFA overnight at 4°C , then cryoprotected in sucrose ( 30% ) in standard PBS ( 1–3 days at 4°C ) . PBS contained ( mM ) : NaCl 137 , KCl 2 . 7 , Na2HPO4 10 , KH2PO4 1 . 8 , pH 7 . 4 . Brainstems were transversely sectioned at 40 μm . Free-floating sections were incubated overnight in PBS containing 0 . 1% Triton X-100 ( PBT ) and 1° antibodies ( 1:500 ) : mouse anti-NeuN ( EMD Millipore , Billerica , MA ) , rabbit anti-neurokinin 1 receptor ( NK1R: EMD Millipore ) , and chicken anti-green fluorescent protein ( GFP: Aves lab , Tigard , OR ) . The tissue was washed in PBS , 6 times for 5 min and then incubated separately for 2–4 hr in a PBT containing 2° antibodies ( 1:250 ) : Donkey anti-mouse AlexFluor 647 , donkey anti-rabbit rhodamine red , donkey anti-chicken AlexFluor 488 ( Jackson ImmunoResearch Laboratories Inc , West Grove , PA ) . The tissue was washed in PBS , 6 times for 5 min . Slices were mounted onto polylysine-coated slides , dehydrated overnight at 22°C , and coverslipped using Cytoseal 60 ( Electron Microscopy Sciences , Hatfield , PA ) . Slides were analyzed using a fluorescent microscope with AxioVision acquisition software ( AxioCam2 , Zeiss , Germany ) . Sample sizes were calculated using Gpower 3 v3 . 1 . 9 . 2 ( http://www . ats . ucla . edu/stat/gpower/ ) ; using a ‘means: Wilcoxon signed-rank test ( matched pairs ) ’ test , with a desired power of 90% , at a 5% significance level , and an effect size of 1 . 15 ( calculated from the initial effect of Alst on respiratory frequency ) . Data were only included from animals where the preBötC and pFL were successfully targeted bilaterally , and no data were excluded from these animals . All statistical analysis was performed in Igor Pro ( WaveMetrics , Lake Oswego , OR ) . EMG signals and airflow measurements were collected using preamplifiers ( P5; Grass technologies , Rockland , MA ) connected to a Powerlab AD board ( ADInstruments , Australia ) in a computer running LabChart software ( ADInstruments ) , and were sampled at 400 Hz/channel . High pass filtered ( >0 . 1 Hz ) flow head measurements were used to calculate: tidal volume ( VT , peak amplitude of the integrated airflow signal during inspiration , converted to mL by comparison to calibration with a 3 mL syringe ) , inspiratory duration ( TI , beginning of inspiration until peak VT ) , expiratory duration ( TE , peak VT to the beginning of the next inspiration ) , and f ( 1/[TI+TE] ) . EMG data , expressed in arbitrary units ( A . U . ) , were integrated ( τ=0 . 05 s; ∫DiaEMG , ∫GGEMG , and ∫AbdEMG ) and peak amplitude of each signal was computed for each cycle . To obtain control values , all parameters except end-tidal CO2 ( ETCO2 ) , were averaged over 20 consecutive cycles preceding each experimental manipulation ( X¯control ) . After Alst , measurements were taken at 2 time points: i ) 20 cycles were averaged where only partial effects were seen , and; ii ) 20 points were averaged following apnea . After CNO , 20 cycles were averaged after breathing had stabilized . After Alst in the presence of CNO , 20 points were averaged following apnea . In the presence or absence of CNO , capnograph peaks were averaged for 10 cycles preceding Alst , and for 5–10 cycles preceding apnea . Following removal from ventilation , the amplitude of the inspiratory bursts and ∫DiaEMG and ∫GGEMG activity immediately preceding the first ∫AbdEMG burst ( Figure 8 , blue lines dashed lines ) were recorded to calculate inspiratory parameters at which active expiration returned . For each rat we obtained X¯control , and the average of 20 cycles during the stimulus ( X¯stimulus ) . X¯control values and their associated X¯stimulus values for each parameter in each rat were combined into a single data set . To facilitate graphical comparisons data were normalized to the highest value in the data set regardless of whether it belonged to X¯control or X¯stimulus group ( C in Figure 2 , 4–7 ) . Therefore the highest value in the data set , whether it be X¯control or X¯stimulus , was 1 . 0; except for measurements of VT , ∫DiaEMG , and ∫GGEMG following ventilation , which are displayed as absolute values . Recorded data were not normally distributed , and were therefore analyzed using non-parametric statistical test , and reported as median and interquartile range ( IQR ) . Statistical tests performed in Igor Pro ( WaveMetrics ) , are 2-sided Wilcoxon signed-rank tests with a significance level of p≤0 . 05 . Data are displayed as box and whisker plots for comparison of groups , and as line graphs for individual experiments . Kernel density estimations ( Parzen , 1962; Epanechnikov , 1969 ) , were used to determine the distribution of respiratory periods . After calculating the optimal bandwidth , i . e . , bin size ( Park and Marron , 1990; Sheather and Jones , 1991 ) , the data was smoothed ( Cao et al . , 1994 ) and plotted . The modality of kernel density plots , were used to assess baseline respiratory periods and whether breathing slowed by quantal integers of that baseline . Bandwidth selection , data smoothing , and kernel density plots were performed in Microsoft excel ( Microsoft Corporation , Redmond , WA ) using an add-in written by the royal society of chemistry ( http://www . rsc . org/Membership/Networking/InterestGroups/Analytical/AMC/Software/kerneldensities . asp ) .	Mammals breathe air into and out of their lungs to absorb oxygen into the body and to remove carbon dioxide . The rhythm of breathing is most likely controlled by two groups of neurons in a part of the brain called the brain stem . One group called the preBötzinger Complex drives breathing in ( inspiration ) , and normally , breathing out ( expiration ) occurs when the muscles responsible for inspiration relax . The other group of neurons – known as the lateral parafacial region – controls extra muscles that allow us to increase our breathing when we need to , such as during exercise . Huckstepp et al . set out to determine how these two groups of neurons interact with one another in anesthetized rats to produce a reliable and efficient pattern of breathing . The experiments provide further evidence that inspiration is mainly driven by the preBötzinger Complex . Whilst activity from the lateral parafacial region is needed to cause the rats to breathe out more forcefully than normal , a second low level of activity from another source is also required . This source could either be the preBötzinger Complex , or some unknown neurons that change their activity in response to the levels of oxygen and carbon dioxide in the blood or fluid of the brain . Further investigation is required to identify how these interactions go awry in diseases that affect breathing , such as sleep apneas .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Interactions between respiratory oscillators in adult rats
Decision making often involves a tradeoff between speed and accuracy . Previous studies indicate that neural activity in the lateral intraparietal area ( LIP ) represents the gradual accumulation of evidence toward a threshold level , or evidence bound , which terminates the decision process . The level of this bound is hypothesized to mediate the speed-accuracy tradeoff . To test this , we recorded from LIP while monkeys performed a motion discrimination task in two speed-accuracy regimes . Surprisingly , the terminating threshold levels of neural activity were similar in both regimes . However , neurons recorded in the faster regime exhibited stronger evidence-independent activation from the beginning of decision formation , effectively reducing the evidence-dependent neural modulation needed for choice commitment . Our results suggest that control of speed vs accuracy may be exerted through changes in decision-related neural activity itself rather than through changes in the threshold applied to such neural activity to terminate a decision . It is well known that the accuracy of many types of decisions suffers when there is pressure to decide quickly . Conversely , decision accuracy often improves if temporal demands permit longer deliberation . This balance between decision time and performance is referred to as the speed-accuracy tradeoff ( SAT ) . Examples of SAT are prevalent , with demonstrations for visual , auditory , olfactory , and memory tasks ( Green and Luce , 1973; Reed , 1973; Wickelgren , 1977; Luce , 1986; Reddi and Carpenter , 2000; Reddi et al . , 2003; Palmer et al . , 2005; Rinberg et al . , 2006; Ings and Chittka , 2008; Bogacz et al . , 2010 ) . Models based on the stochastic accumulation of evidence to a bound provide a unifying behavioral framework that can explain both decisions and the time taken to reach those decisions ( Stone , 1960; Laming , 1968; Link , 1992; Ratcliff and Rouder , 1998; Ratcliff and Smith , 2004; Palmer et al . , 2005 ) . In this class of models , the decision bound establishes a policy on the amount of evidence needed for decision commitment and thus determines the tradeoff between speed and accuracy . Providing confirmation of this idea , multiple studies have found that human SAT in a variety of decision tasks can be explained by changes in decision bound ( Reddi and Carpenter , 2000; Reddi et al . , 2003; Palmer et al . , 2005 ) . While providing insight into computation , these mathematical models of behavior leave open the question of neural implementation . Neuroimaging studies suggest that SAT may be implemented by changes in baseline neural activity in decision-related areas , with higher baseline responses when speed is given precedence over accuracy ( Forstmann et al . , 2008; Ivanoff et al . , 2008; van Veen et al . , 2008; Forstmann et al . , 2010; Wenzlaff et al . , 2011 ) . This suggests the possibility that a reduction of decision bound in behavioral models is implemented in the brain by increasing the starting level of neural activity in regions that accumulate evidence rather than by reducing the bound level required to terminate a decision . The only neurophysiological study of SAT to date reported that during a visual search task with different response deadlines , SAT was associated with a variety of effects in FEF , including changes in the baseline and gain of visual neurons , the duration of perceptual processing , and the rate of ramping responses of movement neurons ( Heitz and Schall , 2012 ) . Here we investigate the neural processes that may underlie SAT by recording from the lateral intraparietal area ( LIP ) of the rhesus monkey during a visual motion discrimination task . We focus on LIP because numerous studies suggest that neurons in this area exhibit the signatures of both evidence accumulation and a decision bound that leads to commitment for saccadic decisions ( Shadlen and Newsome , 2001; Roitman and Shadlen , 2002; Hanks et al . , 2006; Yang and Shadlen , 2007; Churchland et al . , 2008; Kiani and Shadlen , 2009; Rorie et al . , 2010; Churchland et al . , 2011; Bollimunta et al . , 2012 ) . We found that SAT was implemented in LIP by an evidence-independent signal that was added to responses during the same period that the neurons represent the accumulating evidence leading to a choice . The addition of this time-dependent , evidence-independent signal serves to reduce the amount of evidence needed to terminate a decision in the high speed regime . In both regimes , stronger motion was associated with improved choice accuracy and faster RTs , as previously shown ( Figure 2A; Roitman and Shadlen , 2002 ) . Behavior during the two regimes revealed a tradeoff between speed and accuracy . In the high accuracy regime , both monkeys exhibited longer RTs and improved performance ( Figure 2A ) . This observation is important because it indicates that the monkeys did not slow down merely by delaying the motor response , but rather , they used the time to accumulate more evidence from the stimulus . Similarly , if the monkeys had failed to pay attention to the early portion of the stimulus , an improvement in accuracy should not have accompanied the longer RTs . 10 . 7554/eLife . 02260 . 004Figure 2 . Decision accuracy and reaction time in two speed-accuracy regimes are explained by bounded evidence accumulation . ( A ) Data from two monkeys . Black symbols and lines show the high accuracy regime . Red symbols and lines show the high speed regime . The top graph shows the proportion of correct choices plotted as a function of motion strength . Decision accuracy improved with stronger motion in both regimes . The bottom graph shows mean RT ( ±SEM ) plotted as a function of motion strength . RTs decreased for stronger motion in both regimes . In the regime where the monkeys responded more slowly , their performance improved relative to the other regime . N = 3534 and 3307 trials for the high speed and high accuracy regimes , respectively , for monkey D and 4838 and 4733 trials for the same regimes for monkey E . ( B and C ) Bounded evidence-accumulation models . Noisy evidence is accumulated until it reaches a terminating threshold or bound , which establishes the time and sign of the decision . ( B ) Traditionally , this is represented as a single accumulation—or drift-diffusion process—between an upper and lower bound . ( C ) The model can be described equivalently as a race between two anticorrelated drift-diffusion processes , such that negative evidence for one process is positive evidence for the other . The decision is determined by the first accumulation to reach the bound . The traces in ( A ) show fits to the data using the bounded accumulation model . DOI: http://dx . doi . org/10 . 7554/eLife . 02260 . 004 A bounded accumulation model can explain the behavior in both regimes ( Figure 2B ) . The traditional depiction of this model ( Figure 2B ) would represent the accumulation of evidence as a random walk between two terminating bounds . We consider a version of this model that is more consistent with neurophysiology ( Figure 2C ) in which one process is replaced by two competing processes that accumulate moment-by-moment sensory evidence bearing on each alternative ( Usher and McClelland , 2001 ) . The choice is dictated by which process first reaches the bound ( Figure 2C ) . This two-process model explains the established neural representations of both saccadic choices in LIP for this task ( Mazurek et al . , 2003 ) . In the model , RT is determined by a combination of the time taken to reach the bound and a stimulus-independent non-decision time . As explained in ‘Materials and methods’ , the models in Figure 2B , C can be treated as mathematically equivalent . For both depictions , the distance from the starting point of evidence accumulation to the bound , termed the ‘excursion’ , determines how much evidence must be accumulated in order to commit to a choice . In theory , the excursion may change as a function of decision time , for example , if the bound collapses during the course of decision formation . However , for simplicity , we first consider the restricted set of models with static excursions . Monkeys traded off speed and accuracy by changing the amount of evidence that was needed to commit to a choice . In particular , SAT was explained by the bounded accumulation model with a change in the decision variable excursion . Overall , the excursion was reduced by 43% ± 2% in the high speed regime compared to the high accuracy regime ( monkey D: 58% ± 3%; monkey E: 23% ± 4% ) . Thus , less accumulated evidence was needed to commit to a choice in the high speed regime compared to the high accuracy regime . The change in SAT was not explained in a consistent way by the other model parameters . Overall , we observed a 33 . 4 ± 5 . 8 ms shorter non-decision time in the high speed regime compared to the accuracy regime , and this effect was not consistent between the two monkeys ( monkey D: 3 . 1 ± 5 . 2 ms longer in high speed regime; monkey E: 63 . 7 ± 9 . 1 ms shorter in high speed regime ) . Importantly , the RT elongation in the high accuracy regime exceeded any changes observed in non-decision time by a large margin ( mean RT elongation , monkey D: 176 ± 4 ms; monkey E: 159 ± 3 ms ) . Moreover , if changes in RTs were solely a result of altered non-decision time , there would be no associated changes in accuracy . In principle , changes in the SAT could be associated with changes in the signal-to-noise ratio of the sensory evidence , owing perhaps to a change in attentional state , which would lead to a change of sensitivity to motion strength ( parameter k of the model ) . We cannot rule this out , but the model fits provided inconsistent support for such a mechanism in the two monkeys . In the high speed regime , k changed by a factor of 1 . 36 ± 0 . 12 and 0 . 78 ± 0 . 07 , respectively , for monkeys D and E ( 1 . 01 ± 0 . 06 in combined data ) . We cannot rule out the possibility that some of the behavioral change was manifested by a competing process in which the monkey simply guessed quickly , so-called ‘fast guesses’ ( Ratcliff and Rouder , 1998 ) . A fast guess process could lead to faster and less accurate choices in the high speed regime . However , the level of accuracy for both monkeys at the highest motion strength in the high speed regime suggests that any such fast guess mechanism plays a relatively minor role because fast guesses would reduce accuracy for all stimulus strengths . Together , these considerations show that the change in SAT was achieved primarily via a process that alters the excursion required for the accumulating evidence to support termination . Traditionally , bounded accumulation models implement this change in excursion by altering the bound height . However , the brain is thought to employ competing accumulator processes for decision making ( Figure 2C; Usher and McClelland , 2001; Gold and Shadlen , 2007; Churchland et al . , 2008 ) . Accordingly , the brain can achieve a reduction in the excursion of accumulated evidence in two ways: ( 1 ) by lowering the bounds for both processes , or ( 2 ) by adding a signal to the accumulated evidence for both processes . These can be treated as mathematically equivalent: in both cases the effect of the change in the excursion is to alter the amount of accumulated evidence necessary to terminate the process . However , they imply different neural mechanisms . We recorded neural activity from 70 neurons while the monkeys performed the task ( 35 in the high speed regime and 35 in the high accuracy regime ) . In each recording session , we screened neurons based on whether they exhibited spatial selectivity during a delayed saccade task and persistent delay activity during a memory-guided saccade task . For the motion discrimination task , one of the choice targets was placed in the RF of each recorded neuron; the other choice target was placed in the opposite hemifield . In both speed-accuracy regimes , LIP neurons exhibited signatures of evidence accumulation . These features are illustrated in Figure 3 , and they are similar to previous observations ( Roitman and Shadlen , 2002; Churchland et al . , 2008 ) . Initially , the responses were elevated in the presence of the choice targets , one of which was in the neuron's RF . For the first 200 ms after motion onset , there was no relationship between stimulus strength and firing rate ( Figure 3A , B; Equation 1 , p>0 . 1 for both regimes in each monkey ) . Onset of the random dot motion ( outside the RF ) then induced a transient depression in the firing rate that recovered over a 200-ms period . This pattern was similar for speed and accuracy regimes , but the average firing rate in this epoch was 15% ± 1% lower in the high accuracy regime ( monkey D: 24% ± 2% lower in the high accuracy regime; monkey E: 8% ± 2% lower in the high accuracy regime; p<0 . 01 in all cases ) , as discussed in detail below . 10 . 7554/eLife . 02260 . 005Figure 3 . Comparison of neural responses accompanying decision formation in the two speed-accuracy regimes . ( A and B ) Population firing rates plotted as function of time from stimulus onset and sorted by stimulus strength . Lighter green colors correspond to stronger Tin motion; lighter orange colors correspond to stronger Tout motion; darker colors correspond to intermediate stimulus strengths . Averages depict combined data from both monkeys . In both regimes , the responses exhibit characteristic ramping profiles that depend on stimulus strength approximately 200 ms after stimulus onset . Prior to averaging , responses for each neuron were normalized based on the peri-saccadic response ( 0–50 ms prior to saccade initiation ) measured during a block of visually-guided delayed saccade trials . N = 35 neurons in each speed-accuracy regime . Error bars show SEM . For each condition , traces end at the point beyond which fewer than 50% of the trials would contribute to the averages , owing to the different RTs for each trial . ( C ) Buildup rate plotted as a function of motion coherence for the high accuracy regime ( black ) and the high speed regime ( red ) . Error bars show SEM for the estimated buildup rates for each data point . Line fits capture the relationship between buildup rate and motion strength . Overall , the buildup rates were shifted to larger values in the high speed regime , but the change in buildup rate per unit motion strength was similar in the two regimes . DOI: http://dx . doi . org/10 . 7554/eLife . 02260 . 005 After the transient depression in the firing rate , the average responses began to display characteristic ramping activity that depended on stimulus strength and direction ( Figure 3A , B ) . To characterize this relationship , we estimated the ‘buildup rate’ for each motion strength ( ‘Materials and methods’ ) . Stronger motion in the Tin direction led to larger buildup rates , and stronger motion in the Tout direction led to smaller buildup rates ( Figure 3C; Equation 2 , p<0 . 01 for each monkey in both regimes ) . This is consistent with the bounded accumulation framework , which posits that the drift rate of the decision variable should depend on the strength of the evidence . Interestingly , larger buildup rates were observed in the high speed regime ( Figure 3C; Equation 3 , p<0 . 05 for each monkey ) . Importantly , however , there was not a significant difference between the slope of the relationship between buildup rate and motion strength between the two regimes ( Equation 3 , p>0 . 1 for each monkey ) . This suggests that the sensory evidence influences the decision variable in the same way in both speed-accuracy regimes , consistent with the conclusions drawn from the fits to behavior . In both speed-accuracy regimes , LIP neurons also reflected a signature of decision commitment . Shortly before saccade initiation for Tin choices , rather than reflecting stimulus strength , responses attained a common level of activity ( Figure 4A; Equation 1 , p>0 . 3 for coherence effect 75 ms before the saccade for each monkey in both regimes ) . The coalescence of Tin responses is also evident when responses are grouped by RT rather than stimulus strength ( Figure 4B , C; Equation 4 , p>0 . 2 for RT effect 75 ms before the saccade for each monkey in both regimes ) . This overall pattern is consistent with the idea that LIP responses support the presence of a decision bound which when reached , terminates further deliberation and triggers commitment to a choice . 10 . 7554/eLife . 02260 . 006Figure 4 . Comparison of neural responses accompanying decision termination in the two speed-accuracy regimes . ( A ) Pre-saccadic responses plotted as a function of motion coherence for Tin choices . Firing rates were measured in a 50 ms window , centered 75 ms before the saccade ( accuracy and speed regime in black and red , respectively ) . The flat regression is consistent with the idea that the responses reach a common value before Tin choices , for all motion strengths . Normalization procedure is identical to Figure 3 . ( B and C ) Tin responses plotted as a function of time from the saccade and sorted by reaction-time quintile ( colors ) . Averages depict combined data from both monkeys . In both speed-accuracy regimes , responses are stereotyped before the saccade . N = 35 neurons in each speed-accuracy regime . Error bars show SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 02260 . 006 Interestingly , the level of this decision bound was undifferentiated by the speed-accuracy regime . Recall that the bounded evidence-accumulation model explained the behavioral data with a change in the excursion of the decision variable in the two regimes . The brain could implement a change in excursion by altering the threshold level of LIP firing rate necessary to commit to a choice . However , we did not observe significant differences between the firing rates in the two regimes within 200 ms of the saccade for either monkey ( Figure 5A; F test , Equation 5 , p>0 . 5 for each monkey ) . 10 . 7554/eLife . 02260 . 007Figure 5 . Speed-accuracy tradeoff alters LIP responses from the beginning of the decision process . ( A ) Population firing rates plotted for each monkey . Red and black traces correspond to the speed and accuracy regimes , respectively . On the left , responses are plotted as a function of time from motion onset averaged across both directions of motion and all motion strengths ( same normalization as in Figures 3 and 4 ) . Trials contribute data to the average only up to 100 ms before the saccade . Averages are shown up to the time when half of the 50% coherent motion trials contribute data . On the right , responses are plotted as a function of time from the saccade . Responses are averaged across all trials that ended in a Tin choice . Trials contribute data to the average only from 200 ms after motion onset onwards . Curve thickness ( shading ) shows SEM . N = 17 and 15 neurons for the high speed and high accuracy regimes , respectively , for monkey D . N = 18 and 20 neurons for the high speed and high accuracy regimes , respectively , for monkey E . ( B ) The relationship between the magnitude of the evidence-independent neural response and mean RT across experiments . The response statistic captures the evidence-independent component of the firing rate early in the decision process , expressed as a fraction of the average response from the neuron 75 ms before all Tin choices ( ‘Materials and methods’ ) . Larger values imply that less evidence-dependent signal is required to reach the bound . A value of 1 indicates that on average the evidence-independent signal alone would reach the ‘neural bound’ by 300 ms . Neurons from monkey D are shown as circles; neurons from monkey E are shown as crosses . Red symbols correspond to neurons recorded in the high speed regime; black symbols correspond to the high accuracy regime . Ellipses are drawn to the 50% confidence region based on the covariance matrix calculated for each speed-accuracy regime . The correlations in each regime were significant ( p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02260 . 007 Alternatively , the excursion could be modified by means of an evidence-independent signal that adds directly to the accumulated evidence . We looked for such a signal by averaging responses in each speed-accuracy regime across all stimulus strengths and both directions to extract the component of the response that was not dependent on the evidence . This evidence-independent component was significantly elevated in the high speed regime relative to the high accuracy regime ( Figure 5A ) . In one monkey , this difference began from before motion onset and persisted throughout the period of evidence accumulation ( p<0 . 001 ) . For the other monkey , a significant difference emerged 100 ms after motion onset ( p<0 . 01 ) and persisted at all time points thereafter ( p<0 . 001 ) . Thus , for both monkeys , LIP neurons undergo a smaller excursion in firing rate from the start to the end of the decision in the high speed regime compared to the high accuracy regime . It appears that an evidence-independent signal effectively pushes the neural representation of accumulated evidence closer to a threshold . This conclusion receives additional support from an analysis of individual neurons . Across all recorded neurons , there was a significant negative correlation between the estimated evidence-independent neural response and the mean RTs for the corresponding behavioral sessions ( Figure 5B , r = −0 . 57 , p<0 . 01 ) . Of course , this correlation is driven to a large extent by the changes in the mean values between the two regimes . Nonetheless , there is variation within each regime , and for neurons in each regime individually , the correlation was significant ( r = −0 . 40 and −0 . 63 for high speed and high accuracy regimes , respectively; p<0 . 05 in both cases ) . Beyond correlation , these changes in the evidence-independent neural signal explained the changes observed in the behavior between the two speed-accuracy regimes . One way to appreciate this is to fit the behavior in both regimes with the bounded accumulation model such that the only difference between the two regimes is established by the measured evidence-independent neural signals . Notice that in both regimes , the evidence-independent signal increased over time ( Figure 5A ) , consistent with urgency signals reported previously ( Ditterich , 2006; Churchland et al . , 2008; Cisek et al . , 2009 ) . For each monkey , we approximated the evidence-independent signals in each regime with hyperbolic functions ( Figure 6A ) , and used these neurally-derived ‘urgency functions’ to control the model's excursion for each regime . Importantly , the change in excursion between the two regimes is not fit to the behavioral data; it is measured from the neural data . In this way , the only free parameters in the model are the three that are shared between the two regimes ( ‘Materials and methods’ ) . This neurally-constrained model provided a satisfying account of the magnitude of the changes in performance and RT for the two regimes ( R2 = 0 . 99 for monkey D; R2 = 0 . 97 for monkey E; Figure 6B ) . 10 . 7554/eLife . 02260 . 008Figure 6 . The neurally-derived urgency signal explains the magnitude of the behavioral change between speed-accuracy regimes . ( A ) Hyperbolic ‘urgency’ functions fit to the neural responses from 200 ms after motion onset ( decision time = 0 ) to the point in time beyond which fewer than 50% of the trials contribute to the averages for the highest motion strength in each regime ( Figure 5A , Equation 6 in ‘Materials and methods’ ) . Left column is for monkey D; right column is for monkey E . ( B ) Neurally-constrained fits to behavior . Points show behavioral data from Figure 2 . The curves are neurally-constrained model fits to the data . All model parameters are shared ( i . e . , constrained to be equal ) between speed-accuracy regimes except for the difference in excursion , which is measured directly from the neural responses . DOI: http://dx . doi . org/10 . 7554/eLife . 02260 . 008 In traditional diffusion models , like the one depicted in Figure 2B , the neurally-derived , evidence-independent signal would be represented by a time-dependent change in the height of the bound ( Ditterich , 2006; Drugowitsch et al . , 2012 ) . In the high speed regime , the starting point of the bound is nearer to the origin , and the rate of collapse is more severe . In the race-like architecture ( Figure 2C ) , a change in bound height is equivalent to a change in the starting point of both accumulations , and a collapsing bound is equivalent to a time-dependent , monotonically increasing signal that is also added to both accumulators . The neural responses are consistent with the idea that both components comprise a singular , evidence-independent signal that adds to the evidence-dependent activity in LIP . Nonetheless , one might ask how well each component alone could explain the behavioral changes between the two speed-accuracy regimes . To test this , we fit the behavioral data using only the baseline or only the time-dependent component of the neurally-derived , evidence-independent signal . The model comparison , summarized in Table 1 , suggests that the time-dependent component of the signal is more important , although both components contribute to the fits in Figure 6D , especially for monkey D . This dissection is far less important than the rather remarkable observation that a pair of signals derived from the physiology in two speed-accuracy regimes and scaled identically , can explain the dramatic differences in the behavior . 10 . 7554/eLife . 02260 . 009Table 1 . Bayes Information Criterion ( BIC ) comparison of fits to behavioral data based on neurally-constrained modelsDOI: http://dx . doi . org/10 . 7554/eLife . 02260 . 009Modelmonkey Dmonkey EBaseline only664340Time-dependent only124174Full urgency signal71170Because the neurally-derived signals are incorporated in the model via a single scaling factor , all of the fits use the same number of free parameters . Lower BIC values indicate preferred models . The computations that underlie decision making must dictate not just what is decided but also when it is decided . Thus , the brain requires termination rules to end deliberation during a decision process . Just as the criterion in signal detection theory establishes a balance between false alarms and misses ( Green and Swets , 1966 ) , a termination rule for a decision process can establish a balance between speed and accuracy . We found that monkeys can control their speed-accuracy policy via a process that changes the amount of accumulated evidence needed to commit to a choice . This is consistent with psychophysical studies in humans that suggest the SAT is implemented as a change in the decision bound ( Reddi and Carpenter , 2000; Reddi et al . , 2003; Palmer et al . , 2005 ) . In neurophysiology , the bound is conceived as a threshold applied to the firing rates of neurons that represent the accumulated evidence for one or the other alternatives . Although we do not know which neurons in the brain are responsible for applying this threshold , a signature of the process is the stereotyped level of firing rate of LIP neurons just preceding a Tin decision . Interestingly , we found that the level of this threshold does not change when the monkey adopts a different speed-accuracy tradeoff . Instead of changing the level of the threshold , the brain changes the level of the starting point of the accumulation and adds a time-dependent signal to the accumulated evidence . The effect is to push the accumulated evidence closer to the fixed threshold . This dynamic , evidence-independent signal has been termed urgency ( Ditterich , 2006; Churchland et al . , 2008; Cisek et al . , 2009; Drugowitsch et al . , 2012 ) because it reflects the temporal cost associated with decision formation . It is mathematically equivalent to a collapsing bound because both mechanisms would alter the excursion of accumulated evidence from start to end of the decision , thereby allowing the brain to satisfy a mixture of desiderata related to time and evidence to terminate a decision . At the extremes of the mixture lie a temporal deadline or a fixed level of evidence ( e . g . , flat bound ) . Apart from the change in termination rule , other aspects of the decision making process were unaffected by the speed-accuracy regime . The fits to even the simplest diffusion model ( with flat bounds ) indicate that the conversion from motion to accumulated evidence ( e . g . , drift rate ) did not change consistently in the two regimes . This is further supported by the neural recordings . As adduced from the slopes of the graphs in Figure 3C , a change in motion coherence induces the same change in buildup rate in both regimes . The constant offset between these graphs reflects an evidence-independent component of the neural response—that is , the dynamic component of the urgency signal . This component combined with the change in baseline is mathematically equivalent to symmetrically collapsing bounds . Thus we were able to explain the difference in accuracy and RT in the two regimes by fixing all other parameters of the diffusion model to be identical between the two regimes ( Figure 6C ) . We simply introduced the urgency signals , measured physiologically , by converting from fraction of the firing rate ‘bound’ to the units of standard diffusion . The scaling itself ( α; Equation 8 ) was identical in the two regimes . Why would the brain add a signal to the accumulated evidence rather than adjusting the threshold ? One possibility is that it simplifies the task of the downstream structures responsible for decision termination ( Marshall et al . , 2012 ) , rendering it similar to operations involved in simpler motor control ( Hanes and Schall , 1996; Cisek and Kalaska , 2010 ) . The neural computations involved in sensing a signal threshold crossing must be affected by the fidelity ( i . e . , signal-to-noise ratio ) of the signal , and it is well known that fidelity is greater at higher firing rates ( because variance , not standard deviation , scales with mean firing rate ) . Indeed , it has been argued that some type of temporal smoothing is required to sense a threshold ( Mazurek et al . , 2003; Heitz and Schall , 2012 ) . By fixing the threshold to a common level , the mechanism would avoid the need to compensate for these changes . This architecture is supported by neuroimaging studies , which show an increase in baseline activity in decision-related areas of the human brain when speed is given precedence over accuracy ( Forstmann et al . , 2008; Ivanoff et al . , 2008; van Veen et al . , 2008; Forstmann et al . , 2010; Wenzlaff et al . , 2011 ) and with theoretical considerations that favor this implementation ( Marshall et al . , 2012; but see; Lo and Wang , 2006 ) . The origin of the urgency signal is unknown , but it may be related to other stimulus-independent signals observed in monkey LIP , which have been shown to represent reward and value ( Platt and Glimcher , 1999; Rorie et al . , 2010 ) , prior probability ( Hanks et al . , 2011; Rao et al . , 2012 ) and elapsed time ( Leon and Shadlen , 2003; Janssen and Shadlen , 2005 ) . In addition , when faster speeds are cued in a search task , a large stimulus-independent additive signal with a time-dependent component is evident in FEF responses , along with more complex changes in decision threshold ( Heitz and Schall , 2012; their Figure 3 ) . The tradeoff between speed and accuracy illustrates the flexibility with which the brain can establish policies on decision processes . Control of these policies endows individuals with the ability to reach different choices based on the same information ( Shadlen and Roskies , 2012 ) . We have revealed one potential neural mechanism by which a policy change can affect a decision process . We believe that studying how such policy control can go awry is an essential step in better understanding disorders of higher brain function , for instance by explaining what leads to more careful or more impulsive decision making when individuals are faced with the same circumstances . Two rhesus monkeys ( Macaca mulatta ) were trained to perform a motion discrimination task . Visual stimuli were presented on a computer monitor ( 75 Hz frame rate ) using the Psychophysics Toolbox for Matlab ( Brainard , 1997 ) . Trials began with the appearance of a single dot that the monkey was required to fixate . To minimize effects of anticipation on behavior and neural activity , random intervals between trial events were implemented using truncated exponential distributions , tmin + exprnd ( µ ) , where exprnd is a random sample from an exponential distribution with mean = µ . If the generated interval was longer than the maximum allowed time for the epoch ( tmax ) , the interval was resampled to ensure it was less than tmax . After a variable delay ( tmin = 300 ms , µ = 300 ms , tmax = 1 s ) , two bright red choice targets appeared at an equal distance from the fixation point and 180° apart . After another variable delay ( tmin = 300 ms , µ = 300 ms , tmax = 1 . 5 s ) , the random dot motion stimulus appeared in an aperture 5° in diameter that was centered at the fixation point . The dynamic random dot stimulus consists of sets of random dots , which are shown for one video frame and then updated 40 ms later ( e . g . , dots in frame 1 were updated in frame 4; dots in frame 2 were updated in frame 5 , and so forth ) . To update , each dot is either replaced at a random location , or displaced by 14 . 4 min arc ( i . e . , 6 . 0 deg s−1 ) with probability C ( expressed as a percent ) , termed the motion coherence or motion strength . The dot density was 16 . 7 dots deg−2 s−1 . For each trial , there were two possible directions of motion , differing by 180° . For one monkey ( monkey E ) , the directional axis of motion corresponded to that formed by the choice targets . For the second monkey ( monkey D ) , the directional axis of motion always corresponded to the horizontal meridian , but the choice target configuration was as described above , identical to the other monkey . Motion strength ( the percentage of coherently moving dots ) was chosen randomly from a set C ∈ {0% , 3 . 2% , 6 . 4% , 12 . 8% , 25 . 6% , 51 . 2%} . The monkey's task was to determine the direction of coherent motion , which it indicated by making a saccade to the appropriate choice target ( e . g . , rightward target for rightward motion , leftward target for leftward motion ) . The monkey received a liquid reward for all correct choices , and on a random half of the trials in which 0% coherence motion was displayed . The monkeys received no reward for the incorrect choices . The monkeys were trained to perform a RT version of the motion discrimination task in two different speed-accuracy regimes: a faster speed ( hence , lower accuracy ) or higher accuracy ( hence , slower speed ) one . Each regime was maintained for multiple consecutive daily sessions . For both monkeys , we started with the faster speed regime first , followed by the higher accuracy regime , and then switched back to the faster speed regime , with details as follows . The monkeys were initially trained on a version of the task that used an experimenter-controlled duration for the visual stimulus . Both monkeys achieved stable psychometric thresholds in this version of the task before proceeding to the RT version . In the RT version of the task , the color of the fixation point was changed from red to blue and the monkey was allowed to respond any time after motion onset . Previous training experience revealed a natural tendency of monkeys to respond with short latencies . We exploited this tendency by providing only a small incentive for accuracy; a time out penalty for incorrect choices was added to the standard inter-trial interval to discourage fast errors:tITI=t1+7 exp ( −0 . 5tR ) after errort1otherwisewhere tITI is the inter-trial interval , t1 is a constant , typically 1 s , and tR is the RT . With this approach , the monkeys established a stable speed-accuracy regime with well-behaved psychometric and chronometric functions . We refer to this as the high speed ( low accuracy ) regime . After collecting data from LIP neurons in multiple consecutive daily sessions in this regime , each monkey was switched to a higher accuracy regime . This was achieved through the conditional delay of reward . Specifically , reward was always given for correct choices , but it was delayed relative to motion onset . The duration of this delay was 600–800 ms . So , if the delay was 800 ms and the monkey responded correctly in 500 ms , it would have to wait an additional 300 ms before delivery of the reward . This eliminated any reward-based incentive to respond faster than the minimum delay to reward , encouraging longer deliberation times that would improve accuracy . After multiple daily sessions with this reward structure , the monkeys achieved a new stable speed-accuracy set point involving longer RTs and higher accuracy . Data were collected from LIP neurons in multiple consecutive daily sessions in this regime . Finally , the monkeys were moved back to the high speed regime . We performed a variety of manipulations to bring the monkeys to a similar speed-accuracy regime as observed in the initial high speed sessions . At first , the monkeys preserved their higher accuracy set point even after removal of the conditional delay of reward for multiple daily sessions . After the additional removal of the time out penalty for incorrect choices , one monkey ( monkey E ) slowly reached a faster speed set point over the course of 15 sessions . The other monkey ( monkey D ) required an additional manipulation . For 50% of trials , the motion stimulus was extinguished after 600 ms , so no further sensory information was available . After three sessions with this design , the monkey reached a faster speed set point . The task structure then reverted back to allowing the motion stimulus to be displayed indefinitely , and the monkey maintained the faster speed set point . After stable behavior was achieved in this regime , data were collected from 12 more LIP neurons ( 7 for monkey D , 5 for monkey E ) . For both monkeys , the behavior in these final fast speed sessions was similar to the earlier fast speed sessions that were collected before the high accuracy sessions ( mean RT difference <35 ms and mean accuracy difference <2 . 5% ) . The main effects on neural responses were similar as well , so data were combined across the two sets of high speed sessions . Behavioral data were fit using a bounded evidence-accumulation model , also known as bounded drift-diffusion ( Figure 2B ) . The model explains choice and decision time as an accumulation of momentary evidence to an upper bound or a lower bound , corresponding to the two direction choices . The momentary evidence gathered in each time step is drawn from a Gaussian distribution with unit variance for 1 s and mean µ determined by a linear transform of the motion strength: µ = kC , where C is the motion strength and k is a free parameter that scales the motion strength appropriately . The process terminates when the accumulated evidence , termed the decision variable , reaches ±B , the upper and lower decision bounds . The bound reached first by the accumulated evidence determines the choice , and the decision time is determined by how long it takes to reach that bound . RT is a combination of this decision time with an additional non-decision residual time , tnd , which accounts for visual and motion delays and other factors that are independent of the decision process . The basic model ( fits in Figure 2 ) uses three free parameters: k , B , and the mean tnd ( tnd¯ ) . This model can be described in a mathematically equivalent way in which one process is replaced by two competing processes with perfectly anti-correlated accumulations of moment-by-moment sensory evidence bearing on each alternative ( Figure 2C; Usher and McClelland , 2001 ) . In this version , the choice is dictated by which process first reaches the bound . A signal that adds an equivalent amount to both processes would be equivalent to a reduction of the bound of both processes by the same value . We refer to the distance between the bound and starting point of each process as the ‘excursion’ . In a single process version of the model , the excursion is controlled entirely by the heights of the decision bounds . A symmetric reduction of the magnitude of each bound is mathematically equivalent to an equally-sized additive signal in the two-process version of the model . To determine how the model accounts for the difference in behavior between the speed and accuracy regimes , we allowed all three parameters of the model to vary between the two regimes . As is well known , the simplified model , described above , cannot explain mean RT on error trials or the full shapes of the RT distributions for this task ( Ratcliff , 1978; Ratcliff and Rouder , 1998; Ditterich , 2006 ) . We therefore maximized the likelihood of the choice proportions ( binomial error ) and the mean and SD of the RT from correct choices ( Gaussian error ) . Each model's predictions for these values were derived from the analytic solutions for the boundary-crossing probability and first-passage times for a bounded drift-diffusion process ( Palmer et al . , 2005 ) . We recorded from 70 well isolated single neurons in the lateral intraparietal area ( LIP ) of two monkeys while they performed the reaction time ( RT ) motion discrimination task . In monkey D , 17 neurons were recorded in the high speed regime and 15 in the high accuracy regime . In monkey E , 18 neurons were recorded in the high speed regime and 20 in the high accuracy regime . Standard methods were used for extracellular recording of action potentials from single neurons ( e . g . , Roitman and Shadlen , 2002 ) . Neurons were selected using anatomical and physiological criteria . Registration of structural MRI scans to a standard cortical atlas ( Caret software; Van Essen et al . , 2001 ) was used to identify LIP and to direct the placement of recording electrodes . Registration of recording sites to structural MRI suggests that our neural recordings were obtained from the ventral subdivision of area LIP ( LIPv; Lewis and Van Essen , 2000 ) . Once the proper anatomic location was identified , we recorded from neurons during the motion task if they met two criteria: ( 1 ) spatially selective activity during visually-guided delayed saccade trials at both the time of visual onset and the time of the saccade , and ( 2 ) memory activity during the delay period for memory-guided saccade trials ( Gnadt and Andersen , 1988 ) . Both criteria were assessed qualitatively by the experimenter after finding a neuron with an isolated spike waveform . In the visually-guided delayed saccade task , each monkey maintained its gaze at a central fixation point while a target was displayed at a peripheral location . The monkey was required to maintain fixation for a delay taken from a truncated exponential as described above ( tmin = 300 ms , µ = 600 ms , tmax = 2 s ) . After this delay , the fixation point was extinguished , instructing the monkey to make a saccade to the location of the target . This task was used to find the response field ( RF ) of each neuron—that is , the region of space where a target would elicit an increased response for the neuron throughout this task . In the memory-guided saccade task , the peripheral target was flashed briefly ( 150 ms ) rather than maintained . Thus , the monkey was required to remember its location during the delay period . After the delay and the instruction to go , the monkey was required to make a saccade to the remembered location of the flashed target . When collecting neural data while the monkey performed the direction discrimination task , one target was placed in the center of the recorded neuron's RF and the other target was placed in the opposite hemifield . We refer to the target in the neuron's RF and its associated motion as Tin; we refer to the other target and its associated motion as Tout . All training , surgery and experimental procedures were in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals and were approved by the University of Washington Animal Care Committee . Population responses for motion discrimination trials were calculated by combining normalized responses across all neurons . Responses for each neuron were normalized by division by the mean peri-saccadic response ( 0–50 ms prior to saccade initiation ) measured during a block of visually-guided delayed saccade trials . The spikes from each trial were aligned with respect to two trial events . Spikes were aligned to motion onset and saccade initiation , respectively , for analyses of the beginning and end of the decision process . When aligning to motion onset , the time period within 100 ms before saccade initiation was excluded to reduce contamination of the averages with peri-saccadic activity . Similarly , when aligning to the saccade , the time period within 200 ms after motion onset was excluded to avoid contamination by the dip in firing rate that follows motion onset . Response averages ( running means ) were calculated after applying a symmetric ( non-causal ) boxcar filter to the spike trains ( boxcar width = 50 ms ) , appropriately normalized . For plotting purposes ( Figure 3A , B ) , motion strengths were binned into five groups: strong Tin ( 51 . 2% coh favoring Tin ) , medium Tin ( 25 . 6 , 12 . 8 , and 6 . 4% coh favoring Tin ) , weak ( 0 and 3 . 2% coh ) , medium Tout ( 25 . 6 , 12 . 8 , and 6 . 4% coh favoring Tout ) , and strong Tout ( 51 . 2% coh favoring Tout ) . All statistical tests were performed using the actual motion strengths; they were not influenced by the groupings of the trials . When plotting responses aligned to stimulus end ( Figure 4B , C ) , trials were sorted into quintiles based on RT . This was done separately for the responses for each neuron and then combined across corresponding quintiles . To estimate buildup rate during decision formation ( Figure 3C ) , we applied weighted regression to the mean firing rates for each motion strength and direction . The weighting is based on the time-dependent covariance of the running mean ( i . e . , the sample covariance of the boxcar filtered trials , divided by the number of trials ) , thereby correcting for temporal correlation induced by filtering and in the LIP responses themselves ( Churchland et al . , 2011 ) . The chosen analysis window was on the initial buildup , beginning 200 ms after motion onset and ending at the point in time when 50% of the trials were still included in the averages for each motion strength in the two speed-accuracy regimes , taking into account the RTs associated with each motion strength in each regime . The value of 200 ms is based on the established latency between stimulus motion ( near the fovea ) and the emergence of signals in LIP neurons that represent accumulation of evidence in favor of peripheral choice targets in the RFs of the neurons ( Roitman and Shadlen , 2002; Huk and Shadlen , 2005; Kiani et al . , 2008; Churchland et al . , 2011 ) , and it is supported in the current study by the time of the earliest significant effects of stimulus strength on LIP responses . We chose an ending point that equated the attrition of trials from each of the averages—that is , the fraction of trials that contribute to the averages . Attrition of trials associated with the fastest RTs biases the estimate of the buildup rate toward shallower slope because the faster rise times drop out of the average . An alternative approach would be to estimate the buildup rate using identical time intervals in both regimes , but Monte Carlo simulations revealed that it would result in a larger underestimation in the high speed regime than using an approach that based the endpoint on a matched level of trial attrition ( results not shown ) . The latter approach does not eliminate the bias entirely , but it minimizes the difference in the bias of the buildup rate estimates in the two regimes . For the single neuron analysis ( Figure 5B ) , we derived estimates of the evidence-independent response for each neuron and the mean RT for the corresponding behavioral session . Because of the noise inherent in measurements from single neurons , we estimated the evidence-independent signal by averaging responses across all motion strengths and choices . For the same reasons as described above , for each neuron we fit a line to the responses in an interval beginning 200 ms after stimulus onset and ending at a point in time when 50% of the trials for highest motion strength contributed to the average , thereby ensuring inclusion of at least 50% of trials at all time points for all motion strengths . From the linear fits , we obtained the firing rate 300 ms after motion onset ( i . e . , reflecting 100 ms of processing time in LIP ) and divided this by the mean response 75 ms ( using 50 ms bin width ) before saccades to Tin . This yields an estimate of the urgency signal in units of ‘fraction of neural bound’ for each neuron . The scatter plot ( Figure 5B ) displays the relationship between this statistic and the mean RT for each of the 70 experiments . We report standard correlation coefficients and report significance based on the Fisher z-transformation . Statistical comparisons of neural measures were based on regression analyses using weighted least squares , which respect the standard error for point estimates and their correlation , when applicable . To test the effect of motion strength on firing rate during a specified epoch for each speed-accuracy regime , we fit the model: ( 1 ) ynfr=β0+β1Cwhere ynfr is the mean normalized firing rate for each motion strength during the measured epoch , C is the motion strength , and for all equations , βi are fitted regression coefficients . The null hypothesis is that motion strength does not affect firing rate ( H0: β1 = 0 ) . To test the effect of motion strength on the buildup rate of the neural responses for each speed-accuracy regime , we fit the model: ( 2 ) ybu=β0+β1Cwhere ybu is the buildup rate at each motion strength ( see above ) and C is the motion strength . The null hypothesis is that motion strength does not affect buildup rate ( H0: β1 = 0 ) . To test whether motion strength influenced the buildup rate differently between the two speed-accuracy regimes , we fit the expanded model: ( 3 ) ybu=β0+β1C+β2ISA+β3ISACwhere ybu is the buildup rate at each motion strength ( see above ) , C is the motion strength , and ISA is an indicator variable for the speed-accuracy regime ( 1 for speed , 0 for accuracy ) . The null hypothesis that the regime does not affect the offset of the buildup rate is: H0: β2 = 0 . The null hypothesis that the regime does not affect the relationship between buildup rate and motion strength is: H0: β3 = 0 . To test whether firing rate depended on RT during a particular epoch for each speed-accuracy regime , we binned trials into RT quintiles for each individual neuron . Then , we combined responses across each RT quintile for each speed-accuracy regime and fit the model: ( 4 ) ynfr=β0+β1TRTwhere ynfr is the normalized firing rate for each RT quintile and TRT is the mean RT across all neurons for trials that fall into the corresponding quintile . The null hypothesis that RT does not affect firing rate is: H0: β1 = 0 . To compare mean responses derived from particular epochs in the two regimes , we used either a t test or weighted regression to detrend the responses . For example , the difference in mean firing rate in the 200 ms epoch preceding the saccade is captured by the following equation , ( 5 ) y=β0+β1t+β2ISA+β3ISAtwhere y is the running mean normalized firing rate . The time-dependent terms achieve detrending . The null hypothesis , H0: β2 = 0 , is evaluated by an F statistic using weighted regression that incorporates the covariance of the running sample means . For both monkeys , β3 was not significantly different from zero , and removal of this interaction term had no effect on the conclusions . To make neurally-constrained fits of the behavioral data , we measured the change in neural excursion in the two speed-accuracy regimes , as follows . We fit the evidence-independent neural responses ( urgency signals derived from responses combined across motion strength and choice ) from 200 ms after motion onset ( the start of decision-related ramping activity ) to the time point beyond which fewer than 50% of the trials for the highest motion strength contributed to the averages ( same reasons as explained above ) . This fit was based on the following hyperbolic function: ( 6 ) ynfr=r0+r∞tt+t1/2where ynfr is the measured population neural response for each speed-accuracy regime . The parameter corresponds to the neural response at decision time t = 0 ( 200 ms after motion onset ) . r0+r∞ corresponds to the asymptotic value of the urgency signal . The parameter t1/2 is a rate parameter that controls how fast the urgency signal approaches its asymptote . We measured the difference between this urgency signal ( which begins at r0 at t = 0 ) and rb , the neural activity at the time of putative decision commitment ( 75 ms before the saccade [Gold and Shadlen , 2007] ) , to determine the neurally-measured excursion as a function of time for each speed-accuracy regime . Thus , the neurally-measured excursion is described by the following equation: ( 7 ) Ei ( t ) =rb , i−r0 , i−r∞ , itt+t1/2 , iwhere the subscript i indexes each of the two regimes . The pair of neural excursion functions derived from the two speed-accuracy regimes can be used to constrain the excursion in the bounded accumulation model . In the model , the excursion is determined by the height of the decision bound , so we convert the time-dependent excursion E ( t ) to a time-dependent bound B ( t ) . E ( t ) is in units of normalized neural firing rate and the model's B ( t ) is in units of the standard deviation of a Wiener process ( see Palmer et al . , 2005; Shadlen et al . , 2006 ) . To achieve the conversion , we apply an identical scale factor , α , to the urgency functions from the two regimes . The bound is thus described by ( 8 ) Bi ( t ) =αEi ( t ) where the subscript i indexes each of the two regimes . The bounded accumulation model was then fit to the choice proportions and the mean RTs from correct choices in both regimes with three free parameters that are all shared: α as described above , and k and tnd¯ as described previously . Because there is not an analytic solution for drift-diffusion with time-varying bounds , we solved for mean RTs and accuracy numerically ( i . e . , propagating the probability distributions of accumulated evidence to the absorption bounds ) . Importantly , the only differences in the model between the two regimes are the parameters in Equation 7 that are directly estimated from the neural responses . Thus , the neurally-constrained fitting procedure allows us to assess whether the measured differences in neural activity can account for the behavioral changes associated with SAT without any additional free parameters that are different between the two regimes . We also compared these fits to neurally constrained fits employing only the static or the dynamic component of E ( t ) : Bi ( t ) =αEi ( 0 ) and Bi ( t ) =α[Ei ( t ) −Ei ( 0 ) ] , respectively . These models have the same degrees of freedom as the full model ( i . e . , α , k , tnd¯ ) . Comparisons were based on the Bayes Information Criterion ( BIC ) . Models with lower BIC values are preferred over alternatives with higher BIC values .	Many actions involve a trade-off between speed and accuracy , with typing being a good example: the faster you try to type a sentence , the more mistakes you are likely to make . Mathematical models have successfully reproduced the speed-accuracy trade-off , but it is not clear how the brain represents and weighs up these two factors . Now , Hanks et al . have shown how single neurons in a region of the brain called the lateral intraparietal cortex vary their firing rate to optimize the balance between speed and accuracy . Two macaque monkeys were trained to fixate on a single dot on a screen and then move their eyes in one of two directions in response to movies of random dots on a video screen . Initially , the monkeys received a reward immediately after every correct response , whereas incorrect responses were punished with a very short time-out . Under these conditions , the optimal strategy is to respond quickly at the expense of accuracy . In a separate block of trials , the monkeys were again rewarded for correct responses , but this time their reward was delayed if they responded too quickly . The most effective strategy now is to respond accurately , but more slowly . In both the ‘high speed’ and ‘high accuracy’ conditions , the firing of neurons in lateral intraparietal cortex increased while the dots were on the screen . As soon as the firing rate reached a threshold—representing the point at which the monkey had accumulated enough evidence to make a decision about the direction of movement—the monkey moved its eyes . Previous theories had suggested that when speed was the priority , the level of activity required to trigger a decision would be lower than when accuracy was emphasized . Surprisingly , however , the threshold did not differ between the ‘high speed’ and ‘high accuracy’ conditions . Instead , neurons displayed a higher initial firing rate whenever speed was prioritized , enabling the monkey to make a decision on the basis of less evidence . This finding is consistent with human brain imaging studies that have shown increased baseline activity in decision-making circuitry when speed is prioritized over accuracy . Studying these mechanisms could help to reveal why some individuals are more impulsive decision-makers than others .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2014	A neural mechanism of speed-accuracy tradeoff in macaque area LIP
The RAS pathway is central to epidermal homeostasis , and its activation in tumors or in Rasopathies correlates with hyperproliferation . Downstream of RAS , RAF kinases are actionable targets regulating keratinocyte turnover; however , chemical RAF inhibitors paradoxically activate the pathway , promoting epidermal proliferation . We generated mice with compound epidermis-restricted BRAF/RAF1 ablation . In these animals , transient barrier defects and production of chemokines and Th2-type cytokines by keratinocytes cause a disease akin to human atopic dermatitis , characterized by IgE responses and local and systemic inflammation . Mechanistically , BRAF and RAF1 operate independently to balance MAPK signaling: BRAF promotes ERK activation , while RAF1 dims stress kinase activation . In vivo , JNK inhibition prevents disease onset , while MEK/ERK inhibition in mice lacking epidermal RAF1 phenocopies it . These results support a primary role of keratinocytes in the pathogenesis of atopic dermatitis , and the animals lacking BRAF and RAF1 in the epidermis represent a useful model for this disease . The largest organ of the body , the skin , allows exchange with the environment while shielding the organism from insults of mechanical , chemical , and infectious nature . In the skin , the epidermis acts as a mechanical barrier which prevents water loss and the entry of potentially harmful chemicals; in addition , through the interplay between keratinocytes in the epidermis and immune cells based mainly in the underlying dermis , the skin works as an immunological barrier actively defending the body from pathogens ( Pasparakis et al . , 2014; Nestle et al . , 2009 ) . Maintaining these barrier functions throughout life requires continuous regeneration of the epidermis and appropriately balanced immune responses . Dysregulation of the regenerative process can lead to a wide range of defects , from epidermal thickening to malignancies; and an imbalance in the skin’s immune reactions can lead to recurrent infections , or inflammatory/allergic diseases such as psoriasis or atopic dermatitis . The RAF/MEK/ERK signaling pathway plays an essential role in the epidermis . Downstream of the EGFR and RAS , it acts in hair follicle development and wound healing; its mild activation in Rasopathies , genetic diseases caused by activating mutations in the pathway , results in skin phenotypes ranging from thickening of palms and soles to the development of papillomas; and strong activation in keratinocytes results in tumorigenesis ( Kern et al . , 2011; Ratushny et al . , 2012 ) . In animal models , epidermis-restricted inducible activation of RAF or MEK causes massive cutaneous hyperplasia and reduced differentiation ( Khavari and Rinn , 2007 ) . Consistently , RAF1 ( also known as CRAF ) and BRAF are essential for the development and progression of RAS-induced tumors although they fulfil different roles , with RAF1 preventing keratinocyte differentiation ( Ehrenreiter et al . , 2009 ) and BRAF promoting proliferation through MEK/ERK activation ( Kern et al . , 2013 ) . The requirement for MEK/ERK in epidermal proliferation has been independently assessed ( Khavari and Rinn , 2007; Dumesic et al . , 2009; Scholl et al . , 2009; Scholl et al . , 2009 ) . In line with their prominent roles in cell proliferation and tumorigenesis in a number of tissues , the components of the EGFR/RAS/RAF/MEK/ERK pathway are attractive targets for molecule-based therapy . Many inhibitory compounds have been developed that are currently in clinical trials or have reached the clinic . Reflecting the function of the pathway in skin , cutaneous toxicities are one of the main adverse effects of these therapies; they can be severe and lead to an interruption of the therapy or to its termination ( Belum et al . , 2013; Curry et al . , 2014; Dy and Adjei , 2013 ) . These adverse effects can be roughly classified as inflammatory reactions , elicited chiefly by blocking EGFR or MEK; and proliferative events , caused by multikinase inhibitors such as sorafenib or by more specific RAF inhibitors ( vemurafenib , dabrafenib ) . Cutaneous inflammation induced by agents blocking EGFR ( erlotinib or cetuximab ) and by MEK inhibitors ( selumetinib and trametinib ) is characterized by papulopustular rash , pruritus , and suppurative folliculitis in ≥80% of the patients ( Curry et al . , 2014 ) . The effects of erlotinib and cetuximab are on-target , as recently demonstrated by two studies showing similar inflammatory phenotypes in mice with epidermis-restricted EGFR ablation ( Mascia et al . , 2013; Lichtenberger et al . , 2013 ) . Skin rashes and pruritus are also observed in patients treated with inhibitors targeting RAF . Up to half of the cutaneous reactions induced by these compounds , however , consist of anomalous epidermal proliferation events ranging from different forms of keratosis to the development of drug-induced papillomas , keratoacanthomas , and squamous cell carcinomas ( 4–30% depending on the study and the inhibitors used ) ( Anforth et al . , 2013 ) . These side effects are mechanism-based and rely on the paradoxical activation of MEK/ERK promoted by the inhibitors in cultured cells , animal models , and patients ( Samatar and Poulikakos , 2014; Holderfield et al . , 2014 ) . Consistent with this , combination treatment ( RAF plus MEK inhibitors ) reduces cutaneous toxicity ( Flaherty et al . , 2012; Mattei et al . , 2013 ) . In contrast to RAF inhibitor treatment , inducible epidermis-restricted ablation of BRAF and RAF1 causes the rapid regression of RAS-driven tumors without apparent cutaneous toxicity ( Kern et al . , 2013 ) . Here , we systematically test the effect of compound BRAF/RAF1 ablation in epidermis and show that it gives rise to a progressive disease strongly resembling human atopic dermatitis . Mechanistically , the disease relies on the combination of reduced ERK and increased stress kinase activation , leading to chemokine/cytokine overproduction and chronic , systemic inflammation . BRAF and RAF1 were ablated in the epidermis by introducing the Krt5-Cre transgene ( Tarutani et al . , 1997 ) into a Braf f/f; Raf1f/f background ( Kern et al . , 2013 ) ( Figure 1A ) . The mice ( heretofore referred to as Δ/Δep2 ) were born at Mendelian ratio but with their eyes open ( Figure 1B ) , probably as a result of the migration defects of RAF1-deficient keratinocytes ( Ehrenreiter et al . , 2005 ) . With time , however , they started to show symptoms of a progressive skin disease . These included intense itching and scratching resulting in partial alopecia and self-inflicted wounds ( Figure 1B ) . Histological examination of the non-affected areas revealed thickening of the epidermis correlated with increased proliferation and expansion of both the basal and suprabasal keratinocytes but not with keratinocyte apoptosis ( Figure 1C and Figure 1—figure supplement 1A ) . Filaggrin ( FLG ) expression was similar to that of the controls ( Figure 1—figure supplement 1A ) . We also observed a rich dermal infiltrate comprised of activated ( β1 Tryptase+ ) mast cells , granulocytes , dendritic cells and , less abundant , T cells and macrophages ( Figure 1C–D and Figure 1—figure supplement 1B ) . In line with the dermal inflammatory reaction , Δ/Δep2 epidermis was characterized by the robust expression of the keratinocyte activation marker K6 , of the cell adhesion molecule ICAM1 , and by the sporadic expression of MHC class II molecules ( Figure 1E ) , all upregulated in inflammatory conditions including atopic dermatitis ( Freedberg et al . , 2001; Fan et al . , 2003; Caughman et al . , 1992 ) . Consistent with this , Δ/Δep2 epidermal lysates enriched in K5/K10-positive keratinocytes ( Doma et al . , 2013 ) contained increased amounts of the cytokine TSLP , associated with skin allergic disorders ( Ziegler et al . , 2013 ) ( Figure 1F ) , as well as of other chemokines and cytokines , with CCL7 , IL18 , IL5 and IL13 levels significantly higher than controls ( Figure 1G ) . 10 . 7554/eLife . 14012 . 003Figure 1 . Compound deletion of BRAF and RAF1 in the epidermis leads to severe skin inflammation in adult mice . ( A ) BRAF and RAF1 are efficiently deleted in epidermal cells as shown by PCR analysis of tail tissue and immunoblotting of epidermal lysates isolated from 3 weeks old F/F2 and Δ/Δep2 ( n = 4 ) . ACTB is shown as a loading control . ( B ) Macroscopic appearance of newborn and adult F/F2 and Δ/Δep2 mice . Arrowhead highlights the open eye phenotype of Δ/Δep2 pups . ( C ) Hematoxylin/eosin ( H and E ) staining shows epidermal thickening and dermal inflammatory infiltrates in Δ/Δep2 mice . BrdU incorporation confirms hyperproliferation in the basal layer of Δ/Δep2 epidermis . The numbers in the inset represent BrdU+ cells/mm2 of epidermis ( n = 5–7 , mean ± SEM ) . Infiltrating cells: activated mast cells ( β1 Tryptase+ ) , granulocytes ( esterase+ ) , dendritic cells ( CD11c+ ) . ( D ) Quantification of the infiltrating cells: T cells ( CD4+ and CD8+ ) , macrophages ( F4/80+ ) , total mast cells ( MC , toluidine blue+ ) , granulocytes ( GR , esterase+ ) . ( E ) Increased expression of K6 , ICAM1 , and MHC II in Δ/Δep2 keratinocytes/epidermis . Representative images ( C , E ) and quantification ( D ) of 5–7 individual couples . Scale bars , 50 µm . ( F ) Inflammatory chemokines and cytokines in epidermal lysates ( n = 3–4 ) . TSLP levels were determined by immunoblotting and quantified by Image J . ACTB served as a loading control . The results were normalized by arbitrarily setting one of the F/F2 samples as 1 and plotted as mean ± SEM . Data represent mean ± SEM . p = 0 . 011 , p1 = 0 . 001 , p2 = 0 . 007 , p3 = 0 . 001 , p4 = 3 . 06E-6 , p5 = 1 . 37E-4 , p6 = 0 . 002 , p7 = 0 . 049 , p8 = 0 . 042 , p9 = 0 . 046 and p10 = 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 14012 . 00310 . 7554/eLife . 14012 . 004Figure 1—figure supplement 1 . Local inflammation in adult mice lacking BRAF and RAF1 in the epidermis . ( A ) Immunohistochemical analysis of epidermal differentiation ( K5 , K10 and FLG ( filaggrin ) ) and apoptosis ( TUNEL+ ) . ( B ) Skin sections of adult F/F2 and △/△ep2 mice showing dermal infiltration: total mast cells ( toluidine blue , TB+ ) , macrophages ( F4/80+ ) and T cells ( CD4+ and CD8+ ) . Arrows indicate positive cells . Quantification is shown in Figure 1D ( n = 5–7 ) . Scale bars , 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 14012 . 004 Epidermal ablation of BRAF and RAF1 had profound systemic effects . The mice failed to thrive ( Figure 2A ) and presented with enlarged spleens and lymph nodes . The splenomegaly could largely be attributed to increased numbers of Mac1+Gr1+ cells ( Figure 2B ) , a cell type found both in the spleen of a mouse model of FITC contact hypersensitivity/S . aureus infection and in the blood and skin infiltrates of atopic dermatitis patients ( Skabytska et al . , 2014 ) . The lymph nodes contained elevated numbers of activated T , B , and dendritic cells ( Figure 2C ) . In the blood , we observed leukocytosis and increased amounts of the chemokines CCL2 and CCL7 , as well as of GCSF; in addition , serum IgEs were elevated in Δ/Δep2 mice ( Figure 2D ) . Thus , the phenotype of adult Δ/Δep2 mice resembled a skin-specific allergic disease . 10 . 7554/eLife . 14012 . 005Figure 2 . Inflammatory response in adult Δ/Δep2 animals . ( A ) The body weight of Δ/Δep2 mice is significantly reduced compared to their littermates ( n = 5 ) . The data was analyzed by two-way analysis of variance ( ANOVA ) test . ( B ) Increased spleen/body weight ratio and increased numbers of splenic Mac1+Gr1+ cells in adult Δ/Δep2 animals ( n = 4–5 ) . ( C ) Enlarged lymph nodes and and activated T , B , and dendritic cells in adult Δ/Δep2 ( n = 4–8 ) . T cells ( CD4+ or CD8+ ) and B cells ( B220+ ) activation was determined by costaining with CD69; activated dendritic cells were identified as CD11c+ and MHC IIhi or CD80+ . ( D ) Circulating blood cells and plasma levels of chemokines and IgE antibodies in adult mice ( n = 6–8 ) . Data represent mean ± SEM . p1 = 0 . 0002 , p2 = 0 . 023 , p3 = 0 . 002 , p4 = 0 . 002 , p5 = 9 . 83E-8 , p6 = 6 . 31E-7 , p7 = 2 . 91E-7 , p8 = 0 . 004 , p9 = 0 . 001 , p10 = 0 . 008 , p11 = 0 . 011 , p12 = 0 . 046 , p13 = 0 . 050 , p14 = 3 . 65E-4 and p15 = 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 14012 . 005 To determine whether the systemic phenotype was secondary to the severe skin inflammation observed in adult animals , we examined Δ/Δep2 mice at weaning ( 3 weeks of age ) , at which stage they did not groom more than control littermates nor showed any signs of skin rash , except a mild eyelid irritation ( Figure 3—figure supplement 1A ) . Skin architecture and keratinocyte proliferation were not altered at this stage ( Figure 3—figure supplement 1A ) . In terms of dermal infiltrate , a 2-fold increase in mast cells could already be observed , but these cells were not activated ( Figure 3A and Figure 3—figure supplement 1A ) . Granulocytes and dendritic cell numbers were indistinguishable in 3 weeks old F/F2 and Δ/Δep2 littermates ( Figure 3—figure supplement 1A ) . In the epidermis , ICAM1 expression was slightly upregulated , while K6 and MHC II expression could not be detected ( Figure 3A and Figure 3—figure supplement 1B ) . TSLP , CCL2 , CCL7 , and a number of cytokines ( GMCSF , IL6 , 4 , 5 , 13 , 2 ) were already significantly elevated in the epidermal lysates , while only a trend could be observed for others ( Figure 3B and Figure 3—figure supplement 1C ) . Systemically , enlarged lymph nodes contained increased amounts of activated T , B , and dendritic cells , while the spleen was normal both in size and cell composition ( Figure 3C and Figure 3—figure supplement 1D ) . In the blood , the number of monocytes and granulocytes was increased , as were the levels of CCL2 , CCL7 , and GCSF . However , no significant difference in serum IgEs could be observed at this point ( Figure 3D ) . The increase in T , B , and dendritic cells in lymph nodes and the granulocytosis could be traced as far back as 10 days of age , at which time point GCSF was the only cytokine elevated in the serum of Δ/Δep2 mice ( Figure 3E–F ) . Thus , ablation of both Braf and Raf1 in mouse epidermis does not directly disturb skin architecture but rather affects the cross-talk between keratinocytes and the innate and adaptive immune system . 10 . 7554/eLife . 14012 . 006Figure 3 . Local and systemic inflammatory phenotype in young Δ/Δep2 animals . ( A ) Local inflammation in 3 weeks old Δ/Δep2 animals . Total mast cells ( toluidine blue staining , TB; quantified in the plot , n = 4–5 ) and ICAM1 staining . Scale bars , 25 µm . ( B ) Inflammatory chemokines and cytokines in epidermal lysates ( n = 4–5 ) . TSLP levels were determined by immunoblotting and quantified and analyzed as in Figure 1F . TUBA served as a loading control . ( C , D ) Systemic inflammatory parameters in 3 weeks old mice . ( C ) Lymph node size and composition ( n = 4 ) . ( D ) Circulating blood cells ( n = 8 ) and plasma concentration of IgE ( n = 8 ) and chemokines ( n = 9 ) . ( E , F ) Systemic inflammatory parameters in 10 days old mice . ( E ) Lymph node size and composition ( n = 4–10 ) . ( F ) Hemogram showing elevated amounts of granulocytes ( upper panel , n = 4–7 ) and plasma chemokine levels showing increased GCSF ( n = 4 ) . Data represent mean ± SEM . p1 = 0 . 016 , p2 = 0 . 041 , p3 = 0 . 013 , p4 = 0 . 040 , p5 = 0 . 013 , p6 = 0 . 032 , p7 = 0 . 026 , p8 = 0 . 007 , p9 = 0 . 029 , p10 = 0 . 048 , p11 = 0 . 015 , p12 = 0 . 018 , p13 = 0 . 033 , p14 = 0 . 036 , p15 = 3 . 00E-04 , p16 = 3 . 88E-05 , p17 = 0 . 026 , p18 = 0 . 021 , p19 = 0 . 048 , p20 = 0 . 034 , p21 = 0 . 042 , p22 = 0 . 008 , p23 = 0 . 001 , p24 = 0 . 018 , p25 = 0 . 005 , p26 = 0 . 001 , p27 = 0 . 001 , p28 = 0 . 011 and p29 = 0 . 014 . DOI: http://dx . doi . org/10 . 7554/eLife . 14012 . 00610 . 7554/eLife . 14012 . 007Figure 3—figure supplement 1 . Local and systemic response in 3 weeks old △/△ep2 animals . ( A ) Eyelid inflammation in 3 weeks old △/△ep2 animals . Skin architecture ( H&E ) , proliferation rate ( percentage of BrdU+ cells in the epidermis ) , mast cells activation ( β1 Tryptase+ ) , granulocytes ( esterase+ ) and dendritic cell ( CD11c+ ) numbers are indistinguishable in 3 weeks old F/F2 and △/△ep2 littermates . Quantifications of the proliferating cells and granulocytes are shown in the insets ( n = 3 ) . ( B ) Three weeks old △/△ep2 epidermis does not express the activation markers K6 and MHC II . Scale bars , 25 µm . ( C ) Inflammatory chemokines and cytokines in 3 weeks old epidermal lysates ( n = 5 ) . ( D ) Normal spleen size and composition in 3 weeks old △/△ep2 mice ( n = 4–6 ) . Data represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 14012 . 007 Inflammatory skin conditions often involve impaired skin barrier function . Δ/Δep2 embryos performed normally in an outside-in dye penetration assay , indicating that stratum corneum permeability was not affected ( Figure 4A ) . We next monitored the weight of E18 . 5 embryos of different genotypes for 9 hr after birth as a proxy for transepidermal water loss and thus for the barrier function of tight junctions . Δ/Δep2 embryos lost three-fold the weight of littermate controls ( Figure 4B ) . Molecularly , reduced expression of E-cadherin ( CDH1 ) , an adherens junction protein known to regulate tight junctions ( Tunggal et al . , 2005 ) , and of the tight junction protein claudin 1 ( CLDN1 ) , which is crucial for the maintenance of the inside-outside barrier ( Furuse et al . , 2002 ) and has been found downregulated in atopic dermatitis patients ( De Benedetto et al . , 2011 ) was still evident at P3 in Δ/Δep2 epidermis , while occludin expression was normal ( OCLN ) ( Figure 4C ) . The architecture of P3 Δ/Δep2 skin was indistinguishable from that of controls ( Figure 4—figure supplement 1A ) ; P3 epidermal lysates contained highly variable amounts of chemokines and cytokines , predominantly CCL2 , IL6 and IL18 . All were increased in the Δ/Δep2 epidermal lysates , albeit not significantly ( Figure 4—figure supplement 1B ) . The E-cadherin and claudin 1 downregulation was transient and was no longer detectable in 3 weeks old Δ/Δep2 mice ( Figure 4D ) . Single deletion of RAF1 ( RAF1Δ/Δep ) and BRAF ( BRAFΔ/Δep ) did not lead to increased body weight loss in E18 . 5 embryos; consistently , the expression levels of tight junction proteins were normal in P3 epidermis ( Figure 4—figure supplement 1C–D ) . 10 . 7554/eLife . 14012 . 008Figure 4 . Transient inside-outside barrier defects in Δ/Δep2 animals . ( A ) Intact outside-in barrier function ( determined by toluidine blue penetration of the stratum corneum ) in E19 . 5-day-old Δ/Δep2 embryos compared to controls ( n = 6 ) . Representative pictures; two independent experiments were performed with identical results . ( B ) Increased water loss in the Δ/Δep2 E18 . 5 embryos as demonstrated by weight analysis . Results are displayed as percentage of initial weight ( n = 41 for F/F2 and n = 11 for Δ/Δep2 ) . The data was analyzed by two-way analysis of variance ( ANOVA ) test . ( C , D ) Immunoblot analysis of CDH1 , CLDN1 and OCLN expression in epidermal lysates of 3 days old ( C , n = 3; quantification shown in the plot , performed as in Figure 1F ) or 3 weeks old Δ/Δep2 animals . TUBA and GAPDH are shown as loading controls . Data represent mean ± SEM . p1 = 0 . 0001 , p2 = 0 . 028 , p3 = 0 . 020 . DOI: http://dx . doi . org/10 . 7554/eLife . 14012 . 00810 . 7554/eLife . 14012 . 009Figure 4—figure supplement 1 . Skin architecture and inflammatory factors in 3 days old F/F2 and △/△ep2 animals . ( A ) Similar skin architecture in F/F2 and △/△ep2 pups based on H and E staining and on the analysis of differentiation markers ( K5 , K10 and involucrin ) . Scale bars , 50 µm . ( B ) Inflammatory chemokines and cytokines in epidermal lysates ( n = 3 ) . ( C ) Transepidermal water loss as determined by body weight analysis in E18 . 5 embryos lacking either BRAF ( BRAFΔ/Δep ) or RAF1 ( RAF1Δ/Δep ) in the epidermis ( n = 10 for F/F2 , n = 4 for Δ/Δep2 , n = 3 for BRAFΔ/Δep and n = 6 for RAF1Δ/Δep ) . The data was analyzed by two-way analysis of variance ( ANOVA ) test . ( D ) Expression levels of tight junction proteins in the epidermis of P3 BRAFΔ/Δep or RAF1Δ/Δep pups quantified and plotted as in Figure 1F ( n = 3 ) . TUBA is used as a loading control . Data represent mean ± SEM . p1 = 0 . 0442 . DOI: http://dx . doi . org/10 . 7554/eLife . 14012 . 009 To assess the relevance of the transient perinatal barrier defect of Δ/Δep2 mice , we deleted Braf and Raf1 in 3 weeks old mice using tamoxifen-inducible Krt5-Cre ( Indra et al . , 1999 ) . These animals , termed Δ/Δep2TX , showed conversion of both F to Δ alleles and strongly reduced expression of BRAF and RAF1 proteins in tail tissue ( Figure 5A ) . Δ/Δep2TX mice developed a milder disease than the Δ/Δep2 mice , characterized by much slower kinetics ( 8 months between ablation and overt symptoms ) , moderate keratinocyte hyperproliferation ( assessed as increased epidermal thickness ) and activation ( as determined by K6 and ICAM1 expression ) , as well as by a modest increase in activated mast cells and granulocytes in the dermis ( Figure 5B ) . At the systemic level , we observed mild splenomegaly with an increase in Mac1+Gr1+ cells ( Figure 5C ) as well as enlarged lymph nodes containing activated lymphocytes and dendritic cells ( Figure 5D and Figure 5—figure supplement 1 ) . Increased numbers of lymphocytes and granulocytes were found in the blood , while IgE levels were comparable to controls ( Figure 5E ) . Thus , circumventing the transient barrier defect of the Δ/Δep2 animals postponed and attenuated the clinical manifestation of the disease . 10 . 7554/eLife . 14012 . 010Figure 5 . Local and systemic inflammation in Δ/Δep2TX mice . ( A ) PCR analysis of tail tissue ( left ) and immunoblot analysis of epidermal lysates obtained from Δ/Δep2TX animals . ( B ) Macroscopic appearance of Δ/Δep2TX mice and histological analysis of H&E sections . Scale bars , 25 µm . Infiltrating cells: mast cells ( MC; TB+ ) , activated mast cells ( β1 Tryptase+; modest ) , granulocytes ( GR; esterase+ ) . The plot shows a quantification of the histological analysis . ( C ) Mild splenomegaly with increased numbers of Mac1+Gr1+ cells in Δ/Δep2TX animals . ( D ) Activated T cells , B cells and dendritic cells in the lymph nodes of Δ/Δep2TX animals . ( E ) Mild lymphocytosis and significantly elevated granulocyte numbers in Δ/Δep2TX mice . The right panel shows comparable IgE plasma levels in control and Δ/Δep2TX animals . Data are plotted as mean ± SEM ( n = 5; p1 = 0 . 034 , p2 = 0 . 014 , p3 = 0 . 005 , p4 = 2 . 63E-4 , p5 = 0 . 001 , p6 = 0 . 001 , p7 = 0 . 019 and p8 = 0 . 042 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14012 . 01010 . 7554/eLife . 14012 . 011Figure 5—figure supplement 1 . Representative FACS analysis of lymph node and spleen cells isolated from adult F/F2TX Δ/Δep2TX animals . The percentage of activated lymphocytes in lymph nodes was assessed by combining lineage specific markers ( CD4 and CD8 for T lymphocytes; B220 for B lymphocytes ) with the activation marker CD69 . Activated dendriditic cells were identified by staining with CD11c and MHC II antibodies . Myeloid cells in the spleen were analyzed by staining with Mac1 and Gr1 antibodies . The percentage of single and double positive cells is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 14012 . 011 To gain insight in the molecular correlates of the phenotype , we analyzed MAPK signaling in the epidermis . We observed decreased activation of ERK and increased activation of JNK , assessed by the phosphorylation of both MAPK and their respective downstream targets pRSK and pJUN , in non-lesional epidermis of adult Δ/Δep2 and Δ/Δep2TX mice; in contrast , phosphorylation of the p38 target MAPKAPK2 was not affected ( Figure 6A and B ) . A more detailed analysis of 3 weeks old animals confirmed low ERK activation and increased phosphorylation of JNK ( and , to a lower degree , p38 ) in Δ/Δep2 epidermal lysates ( Figure 6C ) . Consistent with this phosphorylation pattern , the levels of the dual specificity phosphatase DUSP1 and DUSP10 , negative regulators of the stress kinases ( Patterson et al . , 2009 ) , were low in the Δ/Δep2 lysates . BRAFΔep lysates were characterized by reduced ERK phosphorylation only , while RAF1Δep lysates showed increased phosphorylation of all three MAPK . The expression of ICAM1 as marker of inflammation was detected only in the Δ/Δep2 lysates , leading to the hypothesis that the combination of high JNK , low ERK activation is at the basis of the inflammatory phenotype of Δ/Δep2 animals ( Figure 6C ) . 10 . 7554/eLife . 14012 . 012Figure 6 . Molecular consequences of BRAF/RAF1 deletion in primary keratinocytes and epidermis . ( A , B ) Immunohistochemical analysis of pERK and pJNK , their downstream targets pRSK and pJUN , and the p38 downstream target pMAPKAPK2 in adult F/F2 and Δ/Δep2 ( A ) , and F/F2TX Δ/Δep2TX epidermis ( B ) . Scale bars , 50 µm . The plots on the right show the percentage of positive cells in the epidermis ( n = 4–5 ) . ( C ) Immunoblot analysis of MAPK signaling in 3 weeks old epidermal lysates ( n = 4 ) , quantified as in Figure 1F . ACTB is shown as a loading control . Phosphorylation is expressed as the ratio between the signals obtained obtained with the phosphospecific antibody and with the protein-specific antibody . In both cases , the data are normalized to one of the F/F2 samples , which was arbitrarily set as 1 . Data are plotted as mean ± SEM . p1 = 2 . 73E-4 , p2 = 4 . 15E-5 , p3 = 0 . 042 , p4 = 0 . 031 , p5 = 0 . 001 , p6 = 0 . 023 , p7 = 0 . 038 , p8 = 0 . 049 , p9 = 0 . 010 , p10 = 0 . 030 , p11 = 0 . 033 , p12 = 0 . 023 and p13 = 0 . 018 . DOI: http://dx . doi . org/10 . 7554/eLife . 14012 . 012 We tested this hypothesis by treating 6 weeks old RAF1Δep animals with a MEK inhibitor ( MEKi; trametinib , in clinical use; daily by gavage for 32 days ) to decrease ERK activation and determine whether this would phenocopy the Δ/Δep2 disease . MEKi efficiently reduced ERK phosphorylation and slightly increased JNK phosphorylation in both control and RAF1Δep epidermal lysates , but increased p38 phosphorylation was only observed in the RAF1Δep ( Figure 7A ) . Consistently , MEKi reduced the expression of the phosphatases DUSP1 and 10 and of CLDN1 in RAF1Δep epidermis only ( Figure 7A ) . Within a month , the RAF1Δep animals developed an inflammatory skin disease characterized by K6 and ICAM1 expression in the epidermis , by the presence of activated dermal mast cells and by an increase in TSLP ( Figure 7B ) . In control animals , inhibitor treatment did not affect , or reduced chemo- or cytokine amounts; GMCSF , IL5 , IL2 and CCL4 were reduced significantly , and a trend could be observed for IL4 , IL17 , IL27 , and CXCL1 . In contrast , a comparison between inhibitor-treated RAF1F/F and RAF1Δep lysates revealed a significant upregulation of CCL2 , GMCSF , IL18 , IL5 , IL13 , IL2 , IL17 and CCL4 ( Figure 7C and Figure 7—figure supplement 1 ) . This correlated with increased Mac1+Gr1+ cells in the spleen and activated T , B , and dendritic cells in the lymph nodes ( Figure 7D ) as well as with increased circulating IgEs and granulocytosis ( Figure 7E ) . Thus , inhibiting ERK in the RAF1Δep animals recreates the immunological environment necessary to bring about a disease very similar to that observed in Δ/Δep2 mice . 10 . 7554/eLife . 14012 . 013Figure 7 . MEK/ERK inhibition in RAF1Δep animals phenocopies the Δ/Δep2 phenotype . RAF1Δep animals were treated with a MEK inhibitor ( MEKi; trametinib , daily by gavage for 32 days ) . ( A ) Immunoblot of epidermal lysates showing the effect of MEKi on the phosphorylation and expression of the indicated proteins . GAPDH is shown as a loading control . ( B ) Macroscopic appearance ( top panels ) and histological analysis of vehicle versus MEKi-treated animals . Mast cells ( TB+ ) and activated mast cells ( β1 Tryptase+ ) are quantified in the plots on the left . Scale bars 25 μm . ( C ) Inflammatory chemokines and cytokines in epidermal lysates of MEKi treated-mice . TSLP levels were determined by immunoblotting and quantified and analyzed as in Figure 1F . ACTB served as a loading control . ( D ) Increased numbers of splenic Mac1+ Gr1+ cells and of activated T cells , B cells and dendritic cells in the lymph nodes of MEKi-treated RAF1Δep animals . ( E ) Mild monocytosis and granulocytosis in MEKi-treated RAF1Δep animals and elevated amount of plasma IgE . Data represent mean ± SEM ( n = 3; p1 = 0 . 002 , p2 = 0 . 017 , p3 = 0 . 003 , p4 = 0 . 025 , p5 = 0 . 041 , p6 = 0 . 022 , p7 = 0 . 023 , p8 = 0 . 029 , p9 = 0 . 010 , p10 = 0 . 032 , p11 = 0 . 044 , p12 = 0 . 053 , p13 = 0 . 015 , p14 = 0 . 001 , p15 = 0 . 031 , p16 = 0 . 049 , p17 = 0 . 026 , p18 = 0 . 038 , p19 = 0 . 015 , p20 = 0 . 001 , p21 = 0 . 005 , p22 = 0 . 002 , p23 = 0 . 033 , p24 = 0 . 006 , p25 = 0 . 039 , p26 = 0 . 025 , p27 = 0 . 004 , p28 = 0 . 020 , p29 = 0 . 027 and p30 = 0 . 028 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14012 . 01310 . 7554/eLife . 14012 . 014Figure 7—figure supplement 1 . Epidermal chemokine and cytokine levels in MEKi treated mice . Inflammatory chemokines and cytokines in epidermal lysates of vehicle or MEKi treated mice . Data represents mean ± SEM ( n = 3; p1 = 0 . 043 , p2 = 0 . 054 and p3 = 0 . 029 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14012 . 014 Collectively , the data above suggest that JNK activation was responsible for the inflammatory skin phenotype . To assess whether this was the case in vivo , we treated Δ/Δep2 animals with the specific peptide inhibitor D-JNKI1 . D-JNKI1 efficiently blocked JUN phosphorylation , increased ERK and RSK phosphorylation ( Figure 8A ) , and prevented the development of the disease as determined by ICAM1 and TSLP expression in the epidermis as well as by eyelid inflammation and mast cells accumulation in the dermis ( Figure 8A–B ) . Consistent with the data in Figure 3—figure supplement 1B , K6 expression could not be observed in the interfollicular epidermis in any of the experimental groups ( Figure 8—figure supplement 1A ) . Compared to untreated 3 weeks old animals , vehicle ( TAT-peptide ) alone had minor effects on the level of some chemokines and cytokines; the increase in IL17 and IL18 in TAT-peptide-treated Δ/Δep2 vs F/F2 animals became significant; the opposite was observed for CCL7 , IL13 and IL4 ( compare Figures 3B and 8C; and Figure 3—figure supplement 1 with Figure 8—figure supplement 1B ) . Treatment with D-JNKI1 reduced chemo- and cytokines in epidermal lysates of Δ/Δep2 animals to levels indistinguishable from , or lower than ( CCL2 , CCL4 , GMCSF and IL4 ) , those of F/F2 controls . Exceptions were IL6 , which was significantly reduced in Δ/Δep2 , but not in F/F2 epidermal lysates; and IL5 , which was not affected by the inhibitor in either genotype ( Figure 8C and Figure 8—figure supplement 1B ) . At the systemic level , D-JNKI1 normalized the numbers of activated B and dendritic cells in lymph nodes , while the numbers of activated T cells where reduced in both F/F2 and Δ/Δep2 lymph nodes ( Figure 8D ) . Thus , JNK activation and the resulting chemo- and cytokine accumulation in the epidermis are required for disease development . 10 . 7554/eLife . 14012 . 015Figure 8 . D-JNKI1 treatment rescues inflammation in Δ/Δep2 mice . Mice were treated with D-JNKI1 or TAT peptide ( 22 mg/kg i . p . at 10 days of age ) and analyzed after 12 days ( A ) D-JNKI1 treatment prevents disease onset in Δ/Δep2 mice . Immunoblot of epidermal lysates showing the effect of D-JNKI1 on the phosphorylation and expression of the indicated proteins , quantified as in Figure 1F . ACTB is shown as a loading control . ( B–D ) Decreased eyelid inflammation , mast cells infiltration ( B; TB+; quantified in the plot on the right ) , epidermal chemokine/cytokine levels ( C ) and activated T cells , B cells and dendritic cells in lymph nodes ( D ) in D-JNKI1-treated Δ/Δep2 mice . Scale bars , 25 µm . Data represent mean ± SEM ( n = 3–5; p1 = 0 . 026 , p2 = 0 . 042 , p3 = 0 . 022 , p4 = 0 . 048 , p5 = 0 . 044 , p6 = 0 . 020 , p7 = 0 . 025 , p8 = 0 . 018 , p9 = 0 . 016 , p10 = 0 . 014 , p11 = 0 . 023 , p12 = 0 . 011 , p13 = 0 . 039 , p14 = 0 . 049 , p15 = 0 . 015 , p16 = 0 . 003 , p17 = 1 . 70E-4 , p18 = 0 . 008 , p19 = 0 . 008 , p20 = 0 . 004 , p21 = 0 . 003 , p22 = 0 . 017 , p23 = 0 . 026 , p24 = 0 . 027 , p25 = 0 . 005 , p26 = 2 . 13E-6 , p27 = 4 . 50E-8 , p28 = 1 . 39E-5 , p29 = 0 . 001 , p30 = 0 . 001 , p31 = 0 . 001 , p32 = 0 . 023 , p33 = 2 . 35E-4 , p34 = 0 . 050 , p35 = 0 . 002 , p36 = 0 . 050 and p37 = 0 . 012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14012 . 01510 . 7554/eLife . 14012 . 016Figure 8—figure supplement 1 . K6 expression and epidermal chemokine and cytokine levels in D-JNKI1-treated mice . ( A ) K6 expression is indistinguishable in TAT or D-JNKI1 treated F/F2 and △/△ep2 littermates . Scale bars , 25 µm . ( B ) Inflammatory chemokines and cytokines in epidermal lysates of TAT or D-JNKI1-treated mice . Data represent mean ± SEM ( n = 3–5; p1 = 0 . 005 , p2 = 0 . 001 , p3 = 0 . 043 , p4 = 0 . 026 , p5 = 0 . 032 , p6 = 0 . 016 and p7 = 0 . 051 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14012 . 01610 . 7554/eLife . 14012 . 017Figure 8—figure supplement 2 . The inflammatory phenotype of Δ/Δep2 mice is not rescued by MyD88 , caspase 1/11 , or TNF knockout . Macroscopic appearance , spleen and lymph node size and circulating blood cell analysis are shown for the indicated genotypes ( A–C ) . ( A ) Representative pictures of 4 month old Δ/Δep2 , Δ/Δep2 MyD88-/- and control animals . Plots on the right represent the ratio between total splenocytes or lymph node cell numbers and body weight ( n = 3–4 ) . ( B ) Representative pictures and hemogram of 4 month old Δ/Δep2 , Δ/Δep2 caspase 1/11-/- and control animals ( n = 5–6 ) . ( C ) Representative pictures and hemogram of 4 month old Δ/Δep2 , Δ/Δep2 TNF-/- and control animals ( n = 4–5 ) . The macroscopic appearance of at least ten mice per genotype was monitored . Data represent mean ± SEM . p1 = 0 . 041 , p2 = 0 . 052 , p3 = 0 . 024 , p4 = 0 . 023 , p5 = 0 . 026 , p6 = 0 . 034 , p7 = 0 . 023 , p8 = 0 . 007 , p9 = 0 . 031 , p10 = 0 . 011 , and p11 = 0 . 034 . DOI: http://dx . doi . org/10 . 7554/eLife . 14012 . 017 To determine whether the increased JNK activation observed in the Δ/Δep2 animals was due to a specific signal , we mated them to MyD88 knockout animals ( Adachi et al . , 1998 ) , to block both TLR ( with the exception of TLR3 ) and IL1 signaling ( Janssens and Beyaert , 2002 ) . We also used TNFA knockout mice to prevent TNF signaling ( Kuprash et al . , 2005 ) , and caspase 1/11 knockout mice to ablate IL1 and IL18 production ( Smith et al . , 1997 ) . None of these knockouts altered the progression ( onset around 2 months of age ) or severity of the Δ/Δep2 skin disease ( Figure 8—figure supplement 2 ) . The Myd88 knockout , however , reduced the splenomegaly observed in the Δ/Δep2 mice . This specific phenotype is caused by an increase in Mac1+/Gr1+ splenocytes ( Figure 2B ) , and its selective rescue in the compound Δ/Δep2;MyD88 knockout animals is likely due to the crucial role of MyD88 in the generation of these cells ( Arora et al . , 2010; Delano et al . , 2007 ) . A concomitant increase in JNK activation and decrease in ERK activation is the basis of the skin disease of Δ/Δep2 animals . To test whether this molecular phenotype was cell-autonomous and to gain some insight in the molecular mechanisms underlying JNK activation , we established and analyzed primary keratinocytes cultures . Δ/Δep2 keratinocytes were “primed” for inflammation: they constitutively expressed ICAM1 ( Figure 9A ) and , upon stimulation , significantly increased amounts of chemokines ( CCL2 , CCL7 ) and cytokines ( IL6 , IL18 , TSLP , IL13 , Figure 9B; IL4 was not detected under these conditions ) . In addition , Δ/Δep2 keratinocytes showed decreased ERK activation in response to EGF and a strong increase in pJNK when co-treated with proinflammatory stimuli ( Figure 9A ) . Treatment with D-JNKI1 , a cell-penetrating , protease-resistant peptide that prevents JNK interaction with its JBD-dependent targets ( Borsello and Forloni , 2007 ) , reduced the expression of the inflammatory proteins ICAM1 , TSLP , CCL2 and CCL7 , but did not have any major effect on ERK activation and on the stress kinase phosphatases DUSP1 and DUSP10 , expressed at lower levels in Δ/Δep2 keratinocytes ( Figure 9C–D ) . 10 . 7554/eLife . 14012 . 018Figure 9 . Increased stress kinase signaling and JNK pathway-dependent cytokine and chemokine production by primary keratinocytes lacking BRAF and RAF1 . ( A ) Reduced ERK phosphorylation and increased JNK/p38 activation in primary Δ/Δep2 keratinocytes stimulated with EGF and/or TNFα and IL1β for 15 min . ( B ) Increased cytokine and chemokine production in primary Δ/Δep2 keratinocytes treated with EGF , TNFα and IL1β for 24 hr . Cytokine and chemokine production was determined by multiplex analysis , except for TSLP which was quantified by ELISA . Data represent mean ± SEM of 3–5 biological replicates . ( C–D ) Cells were pretreated with D-JNKI1 inhibitors prior to stimulation with EGF , TNFα and IL1β for 15 min ( C ) or 24 hr ( D ) . Data represent the mean ± SEM of technical replicates ( n = 3 ) . ( E–F ) Effect of shRNA-mediated Mlk3 silencing on ERK and JNK phosphorylation and ICAM1 expression ( E; stimulation with EGF , TNFα and IL1β for 15 min ) and on the expression of Ccl2 and Tslp mRNA ( F; stimulation with EGF , TNFα and IL1β for 24 hr ) by F/F2 and Δ/Δep2 keratinocytes . shRen , shRNA targeting Renilla , used as a control; sh1 and sh2 , targeting Mlk3 , binding sites nucleotide 2266–2285 and 2383–2402 , respectively . The shRNAs were encoded by lentiviral vectors coexpressing GFP . GFP immunoblots are shown to confirm similar levels of infection in all samples . Data represent mean ± SEM of 4 biological replicates . Each keratinocyte culture represents a pool of three mice . Immunoblots are representative of three independent experiments . p1 = 0 . 041 , p2 = 0 . 040 , p3 = 1 . 89E-4 , p4 = 0 . 018 , p5 = 0 . 046 , p6 = 0 . 020 , p7 = 0 . 008 , p8 = 0 . 016 , p9 = 0 . 001 , p10 = 0 . 018 , p11 = 3 . 23E-4 , p12 = 1 . 47E-4 , p13 = 0 . 007 , p14 = 0 . 03 , p15 = 0 . 035 , p16 = 0 . 023 and p17 = 0 . 046 . DOI: http://dx . doi . org/10 . 7554/eLife . 14012 . 01810 . 7554/eLife . 14012 . 019Figure 9—figure supplement 1 . Compound knockdown ( KD2 ) of BRAF and RAF1 induce the expression of inflammation markers by HaCat cells in a MLK3/JNK-dependent manner . ( A ) Reduced ERK and increased JNK/p38 activation in BRAF and RAF1 knockdown ( KD2 ) HaCat cells stimulated with EGF , TNFα and IL1β for 15 min . ( B ) D-JNKI1 reduces ICAM1 and CCL2 ( n = 4 ) expression in KD2 cells treated with TNFα . ( C ) MEKi induces ICAM1 and CCL2 ( n = 3 ) expression in RAF1KD cells treated with TNFα . In ( B–C ) , ICAM1 expression was measured after a 3 hr , CCL2 expression after a 24 hr treatment with TNFα . ( D ) Effect of MLK3 silencing on ERK and JNK phosphorylation in WT and KD2 cells stimulated as in ( A ) . MLK3 was silenced using a pool of oligonucleotides targeting the following regions: 686–704; 1489–1507; 2122–2138; and 2348–2366 . MLK3 KD cells stimulated as in ( B–C ) show a decrease in JNK activation , ICAM1 and CCL2 ( n = 7 ) expression . Immunoblots are representative of three independent experiments . qPCR data represent mean ± SEM of three independent experiments run in duplicates ( p1 = 4 . 62E-4 , p2 = 0 . 013 , p3 = 0 . 050 , p4 = 8 . 60E-8 , p5 = 0 . 050 , p6 = 0 . 001 , p7 = 0 . 001 and p8 = 0 . 012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14012 . 019 To ensure that the effects observed in the Δ/Δep2 cells and tissue could be reproduced by acute ablation , we performed knockdown experiments in the human keratinocyte cell line HaCat . Concomitant silencing of BRAF and RAF1 ( double knockdown , KD2 ) abolished basal ERK phosphorylation , decreased ERK activation and increased JNK activation by a combination of EGF and proinflammatory stimuli ( Figure 9—figure supplement 1A ) . Constitutive ICAM1 expression was not observed in KD2 cells , but they expressed higher levels of this molecule when treated with TNFα . ICAM1 expression was reduced by D-JNKI1 in KD2 cells and increased by MEKi in RAF1KD cells ( Figure 9—figure supplement 1B–C ) . CCL2 mRNA accumulation , which depends on concomitant stress kinase activation and ERK inhibition , was also elevated in a JNK-dependent manner in KD2 cells ( Figure 9—figure supplement 1B ) ; MEK inhibition strongly increased CCL2 mRNA in WT cells ( Pastore et al . , 2005 ) and even more so in RAF1KD cells ( Figure 9—figure supplement 1C ) . Mixed lineage kinase 3 ( MLK3 ) activates the JNK pathway ( Gallo and Johnson , 2002 ) and can function both as a positive regulator of ERK signaling , via kinase-independent mechanisms ( Chadee and Kyriakis , 2004; Chadee et al . , 2006 ) and as a negative regulator of ERK , via JNK-dependent mechanisms ( Shen et al . , 2003 ) . Importantly , MLK3 binds to JIP1 ( Whitmarsh et al . , 1998 ) , and signal flow from this upstream kinase would be interrupted by D-JKNI1 . We thus determined whether MLK3 was implicated in the imbalance in ERK/JNK signaling observed in Δ/Δep2 primary keratinocytes and in KD2 HaCat cells . MLK3 downregulation by two independent shRNAs strongly reduced both ERK and JNK phosphorylation in F/F2 cells treated with EGF and proinflammatory stimuli; however , in Δ/Δep2 cells , only JNK phosphorylation was reduced ( Figure 9E ) . Consistently , MLK3 downregulation reduced ICAM1 upregulation and the induction of Ccl2 and Tslp mRNA ( Figure 9F ) . Essentially the same results were obtained by downregulating MLK3 in WT and KD2 HaCat cells ( Figure 9—figure supplement 1D ) . Thus , MLK3 is responsible for JNK activation and the induction of inflammatory molecules in Δ/Δep2 keratinocytes and HaCat cells . Keratinocytes have long been known to contribute to the pathogenesis of skin inflammatory disorders , but whether they do so by reacting to stimuli produced by immune cells or by initiating the cascade leading to the disease is still a matter of debate . We describe an essential role of BRAF and RAF1 in keratinocytes in the control of allergic inflammation . Mice lacking BRAF/RAF1 in keratinocytes develop a disease clinically very similar to human atopic dermatitis ( Bieber , 2010 ) . It starts with a barrier defect accompanied by the reduced expression of crucial tight junction proteins and by a marginally increased expression of chemokines and cytokines , particularly CCL2 , IL6 and IL18 . It then progresses to a stage in which the skin appears unaffected , yet both local ( mast cell infiltration , chemokine and cytokine production ) and systemic ( T and B cell activation , increased amounts of circulating leukocytes and chemokines ) anomalies become evident while TSLP and IgE levels are only slightly increased . This is reminiscent of the early phase of atopic dermatitis in children , prior to IgE sensitization ( Bieber , 2010 ) . Unlike the majority of atopic dermatitis patients ( De Benedetto et al . , 2012 ) , however , the Δ/Δep2 animals don’t have a stratum corneum defect , which may help explain the lack of symptoms at this stage . Adult mice present with a full-fledged disease characterized by rich dermal infiltrates , increased cytokine production , clearly elevated IgEs and mast cell activation , in good correlation with the intense pruritus causing extensive scratching and self-inflicted wounds . The initial barrier defect contributes to full-fledged disease , since ablation of BRAF/RAF1 after the third week causes a less severe condition lacking IgE sensitization and mast cell activation , yet recapitulating the systemic symptoms of the disease . This could be likened to late-onset , IgE-sensitization-independent dermatitis ( Bieber , 2010 ) . Thus , BRAF and RAF1 act together to ensure the timely establishment of the inside-out barrier of the epidermis and to prevent allergic inflammation . The data support the hypothesis that the keratinocytes play a primary role in the pathogenesis of atopic dermatitis , and the Δ/Δep2 animals may be useful as a model for this disease ( Figure 10 ) . 10 . 7554/eLife . 14012 . 020Figure 10 . Molecular and physiological defects in mice lacking BRAF and RAF1 in the epidermis . Compound BRAF/RAF1 ablation in keratinocytes induces an imbalance in MAPK signaling , resulting in low ERK , high JNK activation . This causes early inside-outside barrier defects accompanied by reduced CDN1 expression ( yellow arrows ) , followed by a breakdown of the immunological barrier and local as well as systemic allergic inflammation akin to atopic dermatitis , characterized by the presence of Th2 cytokines in the epidermis . The phenotype can be prevented by inhibiting the JNK pathway in Δ/Δep2 animals and cells , and phenocopied by inhibiting the ERK pathway in Raf1Δ/Δep animals . Systemic effects ( lymph node involvement , circulating IgEs ) are separated from local effect by a dashed line . B ( B cells ) , T ( T cells ) , Mc ( Mast cells ) , D ( dendritic cells ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14012 . 020 In this respect , our experiments provide information about the minimal set of keratinocyte-derived factors required for the establishment of allergic , atopic dermatitis-like inflammation . In the phase before IgE sensitization and mast cell activation , we detected significantly higher levels of TSLP , CCL2 , CCL7 , GMCSF , IL2 , IL5 , IL6 , and IL13 in Δ/Δep2 epidermal lysates; most chemokines and cytokines in the Δ/Δep2 animals were sensitive to D-JNKI1 inhibitor treatment , which brought Δ/Δep2 and F/F2 epidermal lysates down to indistinguishable levels . Conversely , the set of cytokines ( TSLP , IL5 , IL13 , IL18 , borderline IL2 and IL17 ) observed in adult , diseased Δ/Δep2 mice was similar to that promoting symptomatic disease in RAF1Δep animals treated with trametinib , which showed an increase in TSLP , IL2 , IL5 , IL13 , as well as in IL18 . These data can be interpreted to mean that ERK signaling negatively regulates the production of allergy-promoting factors . However , it is important to note that trametinib treatment reduced IL2 and IL5 as well as GMCSF in the F/F2 epidermis , indicating that sensitivity to ERK inhibition must be seen in the context of altered cell-autonomous signaling ( increased JNK activation in RAF1Δep epidermis ) and/or tissue milieu . Along the same lines , Δ/Δep2 keratinocytes treated with inflammatory stimuli were significantly more efficient than F/F2 in producing CCL2 , CCL7 , IL6 , IL18 , TSLP and IL13 , but not IL2 or IL5 , resembling , but not fully reproducing , the spectrum observed in epidermal lysates . At this stage , it is impossible to determine whether these discrepancies might be due to the presence of cell types other than keratinocytes in the epidermal lysates or whether the culture and stimulation conditions used did not recapitulate the situation in vivo . Be that as it may , the range of chemokines and cytokines overrepresented in the Δ/Δep2 mice and in the trametinib-treated RAF1Δep mice is consistent with allergic inflammation . In this respect , strong expression of CCL2 has been reported in the basal keratinocytes of atopic dermatitis patients ( Giustizieri et al . , 2001 ) , and expression of CCL7 , which can attract basophils , eosinophils , mast cells and Th2 cells , can be induced by allergens in atopic skin ( Ying et al . , 1995 ) . TSLP , the signature cytokine of allergic disease , promotes the functions of these same cell types ( Ziegler et al . , 2013 ) , and is responsible for the extreme itching accompanying atopic dermatitis ( Wilson et al . , 2013 ) . High levels of TSLP ( Liu , 2006 ) , but also of IL5 and IL13 have been reported in the skin of atopic dermatitis patients , particularly in those with elevated IgEs ( Jeong et al . , 2003 ) . In mice , it could be shown that the expression of TSLP ( Liu , 2006 ) , IL4 , IL5 and IL13 induces atopic dermatitis as well as asthma ( Lee and Flavell , 2004 ) , and IL5 knockout mice exposed to allergens are less prone than wild-type to develop skin eosinophilia and epidermal thickening ( Spergel et al . , 1999 ) . Finally , expression of IL13 in keratinocytes is sufficient to induce a disease mimicking atopic dermatitis ( Zheng et al . , 2009 ) . Thus , these cytokines appear instrumental for the development of atopic dermatitis with IgE sensitization . IL18 , on the other hand , can stimulate both Th1 and Th2 responses ( Nakanishi et al . , 2001 ) , and appears to be causally involved in a different type of atopic dermatitis . IL18 is essential for the induction of dermatitis by the epidermis-restricted expression of caspase 1 , the enzyme that generates active IL18 from its precursor . In addition , expression of IL18 in keratinocytes induces a late-onset dermatitis associated with mastocytosis , but independent of IgE production ( Konishi et al . , 2002 ) . More recently , IL18 knockout mice were shown to be resistant to infection-associated atopic dermatitis , in a mouse model generated by perturbing the stratum corneum with detergent prior to topical application of S . aureus protein A ( Terada et al . , 2006 ) . However , stratum corneum defects could not be observed in Δ/Δep2 mice , and caspase 1 ablation does not modify the course of the Δ/Δep2 disease . Thus , IL18 is not a determining factor in the context of the Δ/Δep2 dermatitis , which rather resembles a Th2/IgE-driven disease . Ablation of BRAF and RAF1 in the epidermis has non-redundant functions converging on MAPK cascades . Decreased ERK and increased JNK activation are observed in Δ/Δep2 epidermis in vivo independently of the severity of the disease as well as in Δ/Δep2 keratinocytes and BRAF/RAF1 knockdown HaCat cells . Thus , the imbalance in MAPK signaling represents an intrinsic defect of cells and tissues lacking BRAF/RAF1 . While BRAF ablation , as in many other tissues/organs , is responsible for reduced ERK activation ( Desideri et al . , 2015 ) , RAF1 ablation correlates with a previously unnoticed increase in stress kinase activation . Consistent with this , we have observed a reduction in the expression of two DUSPs involved in the inactivation of stress kinases , which is strongest in the Δ/Δep2 epidermis . This is in line with the reported positive role of p38 in the induction of DUSP expression and of ERK in its attenuation at multiple levels ( Caunt and Keyse , 2013; Taxman et al . , 2011 ) , which would predict that decreased ERK signaling would be needed to reduce DUSP expression and stabilize JNK phosphorylation when p38 is active . Importantly , both DUSPs have been implicated in the production of cytokines by immunocompetent cells and in murine models of inflammatory and autoimmune disease ( Lang et al . , 2006 ) . We could further identify MLK3 as the kinase responsible for JNK activation as well as for the expression of inflammatory molecules in primary Δ/Δep2 keratinocytes and BRAF/RAF1 knockdown HaCat cells costimulated with EGF , TNFα and IL1β ( as a proxy for the alterations observed in the Δ/Δep2 animals ) . The lack of basal JNK activation in unstimulated cells suggests a cross-talk at the level of BRAF/RAF1 and MLK3 rather than direct MLK3 activation by BRAF/RAF1 ablation . Such a cross-talk was previously reported to occur in the context of ERK signaling , which is positively regulated by MLK3 via the stabilization of the RAF dimer ( Chadee and Kyriakis , 2004; Chadee et al . , 2006 ) . It is possible that in turn , BRAF and RAF1 would restrict the involvement of MLK3 in the JNK pathway , at the same time promoting ERK and reducing JNK activation in stimulated cells ( Figure 10 ) . While the mechanistic details remain to be fully elucidated , the relevance of the imbalance in MAPK signaling for the development of the atopic dermatitis-like disease in Δ/Δep2 animals was shown in vivo in two complementary ways: 1 ) by preventing disease onset with the specific peptide inhibitor D-JNKI1; and 2 ) by inducing disease in healthy RAF1Δep animals with a MEK inhibitor ( Figure 10 ) . The second set of experiments also clearly shows that the function of BRAF and RAF1 in the epidermis is not redundant . Finally , the combined RAF knockout and inhibitor data help explain why in patients the cutaneous toxicity of RAF inhibitors , which increase ERK activation , is mostly related to hyperproliferation ( Anforth et al . , 2013 ) , while inhibitors of ERK activation promote the onset or exacerbate the course of cutaneous inflammatory reactions involving stress kinase activation ( Curry et al . , 2014 ) . All strains were maintained on a Sv/129 background . Krt5-CreBraf f/f;Raf1f/f mice were generated for this study by mating Krt5-Cre;Braff/f:: Krt5-Cre;Raf1f/f animals . These strains , the Krt5-CreER ( TX ) ;Braf f/f;Raf1f/f mice as well as their genotyping and the tamoxifen-induced deletion of the RAF alleles have been previously described ( Kern et al . , 2013; Ehrenreiter et al . , 2005 ) . Inducible RAF deletion was performed at the age of three weeks . Krt5-CreBraf f/f;Raf1f/f were mated with MyD88 ( Adachi et al . , 1998 ) , TNF ( Kuprash et al . , 2005 ) , and caspase 1/11 ( Smith et al . , 1997 ) knockout mice , all maintained on a C57BL/6 background , to test the contribution of the respective signaling pathways to the Krt5-CreBraf f/f;Raf1f/f phenotype . In selected experiments , the MEK inhibitor GSK1120212 ( trametinib , Selleckchem , Germany ) was applied daily by gavage for 32 days ( Doma et al . , 2013 ) . The peptide inhibitor D-JNKI1 was synthesized at the Istituto di Ricerche Farmacologiche 'Mario Negri' ( Milano , Italy ) , as previously described ( Borsello et al . , 2003 ) . Ten days old animals were injected once i . p . ( 22 mg/kg ) with D-JNKI1 or TAT peptide control ( ( D-Pro1 , 2 ) retro- ( D-HIV-TAT ( aa 48-57 ) , Enzo Life Sciences , NY , USA , BML-El384-0001 ) . Animal experiments were authorized by the Austrian Ministry of Science , Research and Economy . H&E staining , TUNEL , BrdU incorporation and immunohistochemistry were carried out as described ( Ehrenreiter et al . , 2009 ) on paraffin or cryostat sections . The following antibodies were used: β1 Tryptase ( AF1937 ) and ICAM1 ( AF796 ) from R&D Systems ( Minneapolis , MN ) ; CD11c ( 14–0114 ) and MHC II ( 14–5321 ) from Affymetrix eBioscience ( Santa Clara , CA ) ; K6 ( 905701 ) , K5 ( 905501 ) , K10 ( 905401 ) and Involucrin ( 924401 ) from BioLegend ( San Diego , CA ) ; F4/80 ( AbDSerotec , UK , MCAP497 ) ; CD4 ( 550280 ) and CD8 ( 550281 ) from BD Biosciences ( San Jose , CA ) ; Filaggrin ( ab24584 ) , pRSK ( ab32413 ) and pJNK ( ab4821 ) from Abcam ( UK ) ; pJUN ( 9164 ) and pERK ( 4376 ) from Cell Signaling Technology ( Danvers , MA ) . Granulocytes were visualized using the Naphthol AS-D Chloroacetate ( specific esterase ) kit ( Sigma , 91C-1KT ) according to the manufacturer's instruction . Toluidine blue staining for mast cells was carried out as described ( Mascia et al . , 2013 ) . Histology images were acquired with a ZEISS microscope Imager M1 ( 20x/0 . 5 or 10x/0 . 3 Plan-NeoFluar objectives ) equipped with ZEISS AxioCamMRc5 . Data were analyzed with ZEISS Axiovision Release 4 . 8 . 1 software ( Carl Zeiss , Germany ) . Fluorescent images were acquired with ZEISS Axioplan2 microscope ( 40x objective , Zeiss Plan – NEOFLUAR; Num: ap . 1 . 3 ) equipped with Spot Pursuit Camera ( Visitron Systems , Germany ) and analyzed with VisiView software ( Visitron Systems ) . Peripheral blood cell counts were acquired using V-Sight ( Menarini Diagnostics , Italy ) . Spleen and lymph node cells were stained with antibodies against CD11c ( 550261 ) , CD4 ( 553051 ) , MHC II ( 557000 ) , CD69 ( 557392 ) , CD80 ( 553768 ) , CD8 ( 553032 ) , B220 ( 553090 ) , Mac1 ( 553310 ) and Gr1 ( 553128 ) all from BD Bioscience and analyzed by FACSCalibur ( BD Bioscience ) and FlowJo V10 software ( Ashland , OR ) . Cytokines and chemokines were detected in cell supernatants , serum samples and epidermal tissue lysates using the Affymetrix eBioscience bead-based multiplex immunoassay . Data were analyzed with FlowCytomix Pro2 . 4 software . GCSF ( R&D Systems , DY414 ) , TSLP ( R&D Systems , DY555 ) and IgEs ( Bethyl Laboratories , E90-115 ) in serum samples were detected by ELISA according to the manufacturer's protocol . Water loss assay and toluidine blue dye staining of embryos were carried out as described ( Tunggal et al . , 2005 ) . HaCaT cells obtained from the DKFZ and mouse keratinocytes were maintained as described ( Doma et al , 2013 ) . BRAF ( L-003460-00 ) , MLK3 ( L-003577-00 ) and RAF1 ( L-003601-00 ) were silenced using ON-TARGETplus SMARTpool siRNAs ( Thermo Fisher Scientific , Waltham , MA ) . Nontargeting pool ( D-001810-10-20 ) was used as control . In accordance with the supplier's protocol , 5×105 cells were transfected with 25 nM of the previously mentioned oligos . Cells were treated with EGF ( 2 ng/ml , R&D Systems , 2028-EG ) and/or with TNFα ( 2 . 5 ng/ml , Millipore , Billerica , MA , 654245 ) and IL1β ( 2 . 5 ng/ml , R&D Systems , 401-ML/CF ) as indicated . For cytokine/chemokine assays supernatants were collected 24 hr later . In selected experiments , cells were pretreated for 1 hr with medium containing DMSO ( for trametinib ) or TAT peptide only ( for D-JNKI1 ) or with the following inhibitors: D-JNKI1 ( 2µM ) or trametinib ( 5µM ) . Two independent shRNAs against mouse MLK3 ( shRNA1 , binding site 2266–2285 and shRNA2 , binding site 2383–2402 ) were designed as 97-bp oligomers containing a 20bp targeting sequence embedded in a shRNAmir stem , amplified and cloned into Xho and EcoRI sites of the miRE lentiviral recipient vector pRRL . SFFV . GFP . miRE . PGK . Puro ( SGEP ) ( Fellmann et al . , 2013 ) . The SGEP plasmid containing Renilla shRNA served as a control . Lentiviral vectors were transfected in 293T cells . Viral supernatants were collected after 24 and 48 hr and passed through a 0 . 45-μm filter ( Sarstedt , Germany ) . Each fresh viral supernatant was used for primary keratinocyte spinfection ( 1500 g , 30 min ) . Primary keratinocyte cultures were harvested 72 hr after first transduction . Cell and epidermal lysates prepared as previously described ( Doma et al . , 2013 ) were immunoblotted using the following primary antibodies ( 1:1000 ) : TUBA ( T9206 , ) from Sigma; pERK1/2 ( 9101 ) , ERK1/2 ( 9102 ) , JNK1/2 ( 9258 ) , pJNK 1/2 ( 9251 ) , pJUN ( 9164 ) , pMAPKAPK2 ( 3041 ) , pp38 ( 9211 ) , p38 ( 9212 ) and ICAM1 ( 4915 ) from Cell Signaling Technology; ACTB ( sc-1616 ) , 14-3-3 ( sc-1657 ) , BRAF ( sc-5284 ) , RAF1 ( sc-133 ) and MLK3 ( sc-166639 and sc-536 ) from Santa Cruz Biotechnology ( Dallas , TX ) ; CDH1 ( 610181 ) from BD Biosciences; OCLN ( ab31721 ) , pRSK ( ab32413 ) , TSLP ( ab188766 ) and DUSP10 ( ab140123 ) from Abcam; GAPDH ( ABS16 ) and DUSP1 ( 07–535 ) from Millipore; ICAM1 ( AF796 , R&D Systems ) and CLDN1 ( 374900 , Invitrogen , Karlsbad , CA ) . RNA was isolated using Nucleospin RNA II kit ( Macherey-Nagel , Germany ) . cDNA was prepared using Oligo ( dT ) 18 primer , dNTPs , and RevertAidReverse Transcriptase ( Thermo Fisher Scientific ) . qPCR was performed using Go Taq qPCR Master mix ( A6002 , Promega , Madison , WI ) . Relative expression was calculated by the ΔΔCT method using ACTB as housekeeping gene . Human CCL2 primers used: for 5`-CAGCCAGATGCAATCAATGC-3´ and rev 5´-GCACTGAGATCTTCCTATTGGTGAA-3´ , human ACTB for 5`-AGAGCTACGAGCTGCCTGAC-3´ and rev 5´-AGCACTGTGTTGGCGTACAG-3´ . Mouse Ccl2 primers used: for 5`-CCCAATGAGTAGGCTGGAGA-3´ and 5`-AAAATGGATCCACACCTTGC-3´; mouse Tslp for 5`-CGACAGCATGGTTCTTCTCA-3´ and 5`-CGATTTGCTCGAACTTAGCC-3´ , mouse ActB for 5`-CCTCTATGCCAACACAGTGC-3´ and 5`-GTACTCCTGCTTGCTGATCC-3´ . Histological samples from at least three animals per condition and genotype were analyzed by counting or measuring at least 3–5 microscopic field/section and analyzed by ImageJ software . Values are expressed as mean ( ±SEM ) . The number of biological replicates and , where applicable , technical replicates are indicated in the figure legends . p values were calculated using the two-tailed Student's t test , hetero- or homoskedastic as determined by a previous F-test of equality of variances or , when indicated , by two-way analysis of variance ( ANOVA ) test . A p value ≤ 0 . 05 is considered statistically significant .	The skin is the largest organ of the body and shields it from damage . It is also home to cells of the immune system , which protect the body from infections . To maintain its role as a barrier , the skin regenerates throughout life , constantly producing new cells to replace the old ones . If this process goes wrong and the skin regenerates too much , the skin may produce tumors . Likewise , disrupting the immune barrier can lead to autoimmune or allergic skin diseases such as psoriasis or atopic dermatitis ( also known as eczema ) . The outer layer of the skin is called the epidermis , and is made up of cells known as keratinocytes . A family of proteins called RAF plays an important role in controlling how keratinocytes behave . These proteins are part of a signaling pathway that controls the production of new cells and is disrupted in skin tumors . Therapies that inhibit RAF are effective treatments for these tumors but have side effects that can affect the skin so severely that the treatment must be interrupted . Keratinocytes also play a role in the development of allergic skin diseases . However , it is not known whether they do so by triggering the disease themselves , or by responding to stimuli produced by immune cells . Raguz et al . investigated what would happen if RAF proteins were removed from keratinocytes in the epidermis of mice . This caused the mice to develop an allergic disease similar to human atopic dermatitis . This was an unexpected effect and different from the side effects caused by drugs that inhibit RAF proteins . By analyzing the signaling pathway that RAF is part of , Raguz et al . discovered that removing RAF from the epidermis reduces the pathway’s ability to prevent excessive stress signals being sent throughout the skin . Under these conditions , the keratinocytes bring about inflammation and allergy by activating the immune cells in the skin and elsewhere . Overall , the results presented by Raguz et al . indicate that allergic dermatitis can be triggered by defects in keratinocytes rather than in immune cells . Furthermore , RAF in the epidermis appears to prevent allergic skin diseases . Future studies could use mice that lack RAF in their epidermis to further understand atopic dermatitis and to investigate the way in which drugs that target RAF can damage the skin .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "immunology", "and", "inflammation" ]	2016	Epidermal RAF prevents allergic skin disease
In Saccharomyces cerevisiae , RNA polymerase II ( Pol II ) selects transcription start sites ( TSSs ) by a unidirectional scanning process . During scanning , a preinitiation complex ( PIC ) assembled at an upstream core promoter initiates at select positions within a window ~40–120 bp downstream . Several lines of evidence indicate that Ssl2 , the yeast homolog of XPB and an essential and conserved subunit of the general transcription factor ( GTF ) TFIIH , drives scanning through its DNA-dependent ATPase activity , therefore potentially controlling both scanning rate and scanning extent ( processivity ) . To address questions of how Ssl2 functions in promoter scanning and interacts with other initiation activities , we leveraged distinct initiation-sensitive reporters to identify novel ssl2 alleles . These ssl2 alleles , many of which alter residues conserved from yeast to human , confer either upstream or downstream TSS shifts at the model promoter ADH1 and genome-wide . Specifically , tested ssl2 alleles alter TSS selection by increasing or narrowing the distribution of TSSs used at individual promoters . Genetic interactions of ssl2 alleles with other initiation factors are consistent with ssl2 allele classes functioning through increasing or decreasing scanning processivity but not necessarily scanning rate . These alleles underpin a residue interaction network that likely modulates Ssl2 activity and TFIIH function in promoter scanning . We propose that the outcome of promoter scanning is determined by two functional networks , the first being Pol II activity and factors that modulate it to determine initiation efficiency within a scanning window , and the second being Ssl2/TFIIH and factors that modulate scanning processivity to determine the width of the scanning widow . Transcription of eukaryotic protein-coding genes is carried out by RNA polymerase II ( Pol II ) in three sequential steps: initiation , elongation , and termination ( Roeder , 2019 ) . Accurate initiation requires minimally the assistance of five general transcription factors ( GTFs ) TFIIB , TFIID , TFIIE , TFIIF , and TFIIH , which together with Pol II , comprise the basal transcription machinery . At the beginning of transcription , this machinery assembles at a defined DNA region for each transcript called a promoter , melts the double-stranded DNA yielding a region of unwound DNA forming a ‘transcription bubble’ . Within this bubble a position or positions will be identified to serve as transcription start sites ( TSSs ) . Initial promoter melting appears to occur stereotypically ~20–25 nt downstream of promoter elements such as the TATA box across eukaryotes , though most promoters lack a TATA box or other strong sequence signature . During initiation , the process of TSS selection determines the identity and distribution of transcript isoforms that differ by their 5’ ends . Differences in 5’ UTR can alter transcript properties such as translation efficiency or transcript stability through differences in sequence or RNA secondary structure ( Arribere and Gilbert , 2013; Rojas-Duran and Gilbert , 2012; Malabat et al . , 2015; Sample et al . , 2019; Cuperus et al . , 2017; Akirtava and McManus , 2021; Lin et al . , 2019 ) . Furthermore , in conjunction with activators and coactivators , the efficiency of the initiation process will also establish mRNA synthesis rates . How TSS selection is governed by these factors is not well understood for the majority of eukaryotic promoters that utilize multiple TSSs . Transcription initiation by Saccharomyces cerevisiae Pol II has been the subject of extensive analysis both in vivo and in vitro , and thus provides a powerful model for system for mechanistic studies of TSS selection . TSS selection by S . cerevisiae Pol II occurs over a range of positions located ~40–120 bp downstream of the core promoter region . Numerous lines of evidence suggest that TSS selection by S . cerevisiae Pol II involves a unidirectional scanning mechanism in which the preinitiation complex ( PIC ) assembles at an upstream core promoter and interrogates consecutive downstream positions for usable TSSs ( Giardina and Lis , 1993; Kuehner and Brow , 2006; Hampsey , 2006; Fishburn et al . , 2016; Qiu et al . , 2020 ) . TFIIH is proposed to drive Pol II scanning through ATP-dependent DNA translocase activity ( Fishburn et al . , 2016; Tomko et al . , 2021; Tomko et al . , 2017; Fishburn et al . , 2015; Fazal et al . , 2015 ) . An optical-tweezer-based single molecule analysis of reconstituted S . cerevisiae PICs indicated that an ATP/dATP-induced activity within the PIC causes shortening of the distance between upstream and downstream DNA ( Fazal et al . , 2015 ) . This shortened distance approximates the distance downstream from TATA elements where TSSs are positioned in yeast ( 40–120 nt ) ( Struhl , 1987 ) and suggests downstream DNA movement and compaction during promoter scanning by the PIC . Separately , a magnetic tweezer-based single molecule assay suggested that an initial melted region of 6 nt ( a ‘bubble’ ) is the direct consequence of TFIIH’s ATPase activity ( Tomko et al . , 2017 ) . Due to inability of magnetic tweezers to detect DNA compaction in the particular setup used , how the Pol II machinery reaches downstream TSSs , whether through generation of a large bubble or translocation of a small bubble was not clear . Nevertheless , both studies agree that an ATP-dependent PIC activity for promoter opening is likely Ssl2 within TFIIH , which has been demonstrated as a DNA translocase within purified TFIIH in vitro ( Fishburn et al . , 2015 ) . In support of Ssl2/TFIIH’s role in movement of downstream DNA toward the PIC , ssl2 mutants have been identified as altering TSS selection at ADH1 and showed genetic interactions with sua7 ( TFIIB ) mutants ( Goel et al . , 2012 ) . Specifically , the identified ssl2 mutants shifted TSSs upstream at ADH1 . Polar shifts in TSSs distributions have been observed in mutants within Pol II , the GTFs TFIIB and TFIIF , and the PIC cofactor Sub1 ( Kuehner and Brow , 2006; Qiu et al . , 2020; Goel et al . , 2012; Yang and Ponticelli , 2012; Khaperskyy et al . , 2008; Pal et al . , 2005; Majovski et al . , 2005; Freire-Picos et al . , 2005; Ghazy et al . , 2004; Chen and Hampsey , 2004; Faitar et al . , 2001; Pappas and Hampsey , 2000; Wu et al . , 1999; Bangur et al . , 1999; Pardee et al . , 1998; Sun et al . , 1996; Sun and Hampsey , 1995; Pinto et al . , 1994; Berroteran et al . , 1994; Pinto et al . , 1992; Hampsey et al . , 1991; Knaus et al . , 1996 ) . In a promoter scanning initiation mechanism , altering initiation efficiency is predicted to alter TSS distributions in a polar fashion when initiation efficiency increases or decreases . We have recently observed polar ( directional ) shifts at essentially all promoters examined across the genome in yeast for tested Pol II and GTF mutants , as predicted for scanning operating universally across promoter classes ( Qiu et al . , 2020 ) . We found that hyperactive Pol II catalytic mutants shifted TSSs’ distributions upstream at promoters genome-wide , consistent with a higher probability of initiation at every TSS , and thus , initiation happening on average earlier in the scanning process . Conversely , hypoactive Pol II catalytic mutants shift TSS distributions downstream at promoters genome-wide , consistent with initiation happening later in the scanning process . Our previous data on genetic interactions between Pol II and TFIIF or TFIIB support the idea that these mutations altered initiation additively ( Jin and Kaplan , 2014 ) , consistent with their acting in the same pathway during scanning . Tested TFIIB mutants appeared to reduce initiation efficiency while tested TFIIF mutants appeared to increase initiation efficiency . Consistent with this idea , a double mutant between TFIIF and a hyperactive Pol II mutant had stronger effects on TSS shifts than either mutant alone across all promoters . Pol II mutants are proposed to control initiation efficiency because active site residues important for catalytic activity alter TSS distributions correlating with the strengths of their catalytic defects in RNA chain elongation . Initiation by promoter scanning should be dependent both on Pol II catalytic rate together and by whichever factors control the actual scanning step , that is , presumptively the rate and processivity of TFIIH as the putative scanning translocase . Therefore , to understand how promoter scanning works , it is critical to understand how TFIIH contributes and how its activity is regulated within the PIC . We have described the scanning process previously using a ‘Shooting Gallery’ analogy ( Qiu et al . , 2020; Kaplan , 2013 ) . In this model , initiation is controlled by the rate ( TFIIH’s translocase activity ) in which a ‘target’ ( TSS ) passes the ‘line of fire’ ( the Pol II active site ) along with the rate of firing ( Pol II catalytic activity ) and the size of the target ( innate sequence strength ) . Together , the cooperation and competition between these rates determines the probability a target is hit ( initiation happening ) . Alteration of enzymatic activities supporting initiation , either the Pol II active site or TFIIH translocation , should have predictable effects on overall TSS distributions when initiation proceeds by scanning . In addition to the TFIIH translocation rate , it is predicted that processivity of TFIIH DNA translocation should strongly modulate scanning , but in ways distinct from controlling innate initiating efficiency . Here , TFIIH processivity would control the probability that a TSS could be reached during a scanning foray from a core promoter , which appears to be facilitated by Ssl2’s translocase activity ( Fazal et al . , 2015 ) . Optical tweezer experiments are consistent with TFIIH within the PIC having median processivity on the order of ~90 bp , consistent with the average distance of TSSs downstream of yeast TATA boxes . However , purified holo-TFIIH from yeast has much reduced measured processivity and human TFIIH has essentially none ( Tomko et al . , 2021 ) . Given that TFIIH activity is predicted to be altered extensively by cofactor interactions in both transcription and nucleotide excision repair ( NER ) , it will be important to understand TFIIH functions within the PIC in vivo . How alterations to Ssl2/TFIIH translocase activity control TSS distributions has not been extensively investigated . To test if distinct alterations to Ssl2 function have broad effects on promoter scanning and TSS selection , we used existing and newly identified ssl2 alleles to examine their effect on TSS distributions genome-wide . Our novel alleles were identified through use of genetic reporters we have developed and found to be sensitive to different kinds of initiation defects ( Malik et al . , 2017 ) . We have found that ssl2 alleles affect TSS distributions for the majority of promoters in yeast and for all promoter classes . Furthermore , we find that ssl2 alleles alter TSS selection distinctly from how changes to the Pol II active site alter TSS selection , consistent with a distinct role for Ssl2 in promoter scanning . ssl2 alleles appear to extend or truncate scanning windows at promoters genome-wide , consistent with increase or decrease in the processivity of scanning . Scanning-window truncating alleles map throughout the Ssl2 structure , consistent with a hypothetical loss of functions ( LOF ) in Ssl2 DNA translocase enzymatic activity that lead to decreased TFIIH processivity . Conversely , scanning-window extending ssl2 alleles are much more localized within the Ssl2 N-terminus , including conserved residues within regions that connect Ssl2 helicase domains to the TFIIH component and regulator of Ssl2 activity Tfb2 . Our alleles are consistent with alteration to an Ssl2 regulatory domain resulting in modulated translocase activity or TFIIH processivity . We further test our model for initiation by promoter scanning through examination of genetic interactions of initiation-altering Pol II/GTF alleles and ssl2 alleles . The genetic interactions between Pol II/GTF and ssl2 alleles support the idea of two major networks controlling TSS selection in S . cerevisiae . One network shapes TSS distributions through affecting initiation efficiency , represented by Pol II , TFIIB , and TFIIF functions . A second network appears to alter TSS distributions through regulating TFIIH’s processivity , and includes Ssl2 , Sub1 , and potentially TFIIF . To understand how TSSs are identified by promoter scanning and the potential roles for TFIIH , we first examined previously identified ssl2 mutants ( Goel et al . , 2012; Gulyas and Donahue , 1992; Qiu et al . , 1993; Lee et al . , 1998 ) for transcription-related phenotypes that we have demonstrated are predictive of specific initiation defects ( Figure 1A ) . Two such phenotypes relate to altered initiation at the IMD2 gene , whose promoter is regulated by a TSS switch ( Figure 1A , IMD2 , imd2Δ::HIS3; Kuehner and Brow , 2008; Jenks et al . , 2008 ) . We have previously shown that tested mutants that shift TSSs upstream due to altered promoter scanning result in an inability to express a functional IMD2 transcript , causing sensitivity to the IMPDH inhibitor mycophenolic acid ( MPA ) ( Qiu et al . , 2020; Jin and Kaplan , 2014; Malik et al . , 2017; Kaplan et al . , 2012; Figure 1A , IMD2 ) . In the presence of GTP starvation induced by MPA , wild-type ( WT ) strains shift start site usage at IMD2 from TATA-proximal GTP-initiating TSSs to a downstream ATP-initiating TSS . This shift results in a functional IMD2 transcript that is required for yeast to survive MPA . Catalytically hyperactive Pol II mutants ( termed GOF for ‘gain-of-gunction’ ) do not shift TSS usage to the downstream functional IMD2 TSS but instead shift to intermediately positioned non-functional upstream sites , rendering yeast sensitive to MPA ( MPAS ) ( Malik et al . , 2017 ) . Pol II GOF mutants with MPAS phenotypes and a tfg2 MPAS mutant were additionally found to shift TSS distributions upstream at ADH1 and subsequently genome-wide ( Qiu et al . , 2020; Kaplan et al . , 2012; Eichner et al . , 2010 ) . Correlation between strength of MPA-sensitive phenotypes and quantitative upstream TSS shifts at ADH1 and genome-wide suggest that MPA sensitivity is a strong predictor for upstream-shifting TSS mutants . Conversely , we found previously that mutants shifting TSSs downstream ( reduced catalytic activity ‘LOF’ Pol II mutants ) constitutively express IMD2 in the absence of using MPA as inducer ( Kaplan et al . , 2012 ) . Pol II TSS downstream shifting phenotypes at IMD2 can be detected using a reporter allele where IMD2 is replaced by HIS3 , placing HIS3 under control of IMD2 promoter and TSS selection ( Figure 1A , imd2Δ::HIS3; Malik et al . , 2017 ) . Indeed , these same LOF Pol II mutants shift TSS distributions downstream at ADH1 and genome-wide ( Qiu et al . , 2020; Jin and Kaplan , 2014; Kaplan et al . , 2012 ) . These previous results suggest that we have phenotypes predictive of alterations to promoter scanning in both directions and can form the basis of a system to characterize mutants across the genome for effects on TSS selection by promoter scanning . We used site-directed mutagenesis and plasmid shuffling to recreate and phenotype five previously described ssl2 mutants , reasoning that this would allow us a first glimpse at the effects and potential diversity present in these classic alleles . These alleles are ssl2-rtt ( ssl2 E556K ) ( Lee et al . , 1998 ) , ssl2-DEAD ( ssl2 V490A/H491D ) , and SSL2-1 ( ssl2 W427L ) ( Gulyas and Donahue , 1992 ) , ssl2-508 ( ssl2 H508R ) ( Goel et al . , 2012 ) , and rad25-ts24 ( ssl2 V552I/E556K ) ( Qiu et al . , 1993; Figure 1B and C ) . Analysis of these five ssl2 mutants showed phenotypes consistent with altered TSS selection ( Figure 1C ) . First , ssl2-DEAD and ssl2-508 exhibited strong and weak MPAS phenotypes , respectively . These results were consistent with prior analysis of initiation at ADH1 in these mutants ( Goel et al . , 2012 ) suggesting that our genetic phenotypes using IMD2 are predictive of potentially global alterations to TSS selection . SSL2-1 exhibited a His+ phenotype , which is predictive of downstream shifts in TSS usage and consistent with its identification as a dominant mutant bypassing an inhibitory stem loop in the his4-316 mRNA ( Gulyas and Donahue , 1992 ) . We now can rationalize the original suppressor of stem loop ( SSL ) phenotype of SSL2-1 as usage of TSS downstream or within the inhibitory 5’ stem loop his4-316 sequence ( though not apparent in Gulyas and Donahue , 1992 ) . ssl2-rtt and rad25-ts24 show temperature-sensitive phenotypes as expected; however , assayed transcription-related plate phenotypes were not observed . Due to the absence of His+ or MPAS phenotypes , we predicted that ssl2-rtt or rad25-ts24 alleles would not shift TSS usage . Notably , there was no lys2-128∂ Spt- phenotype observed among these five existing ssl2 alleles , in contrast to our previous observation of an Spt- phenotype in a subset of MPAS Pol II TSS alleles ( Kaplan et al . , 2008; Braberg et al . , 2013 ) , our first indication that ssl2 alleles may alter TSS selection in a distinct fashion from Pol II mutants . To quantitatively examine TSS usage of these ssl2 mutants , we chose ADH1 initiation . ADH1 has been widely used as a model gene for TSS studies in S . cerevisiae . It contains two major TSSs that are 27 and 37 nt upstream of the start codon ( Figure 1D ) . Using primer extension , transcription products of these two TSSs appear as bands of differing mobility on denaturing polyacrylamide gels ( Figure 1—figure supplement 1A , left panel WT ) . Other TSSs’ positions show minor usage . In most studies , the two major starts are compared qualitatively , but such comparisons miss meaningful alterations that may tell us about initiation mechanisms . To establish that genetic phenotypes using IMD2 correlate with altered TSS selection elsewhere in the genome , our quantitative analysis of the ADH1 promoter divides TSSs observed by primer extension into six bins from upstream to downstream , in which bin 3 and 5 each contain the TSS for one of the major ADH1 mRNA isoforms ( Figure 1D and Figure 1—figure supplement 1A , left panel WT ) . In order to compare a mutant’s TSS distribution to that of WT , distributions are normalized , and the WT distribution is subtracted from tested mutant distributions bin by bin ( Figure 1—figure supplement 1A , middle and right panel ) . Negative or positive values indicate that the mutant has relatively lower or higher usage for TSSs in that particular bin , respectively ( Figure 1—figure supplement 1A , right panel ) . For example , the Pol II GOF allele E1103G increases relative TSS usage at upstream minor sites ( TSS bin 2 ) and decreases relative TSS usage at the downstream major site ( TSS bin 4 ) ( Figure 1—figure supplement 1A , E1103G ) . Because of the dramatic effect of E1103G on TSS usage , the change of TSS usage can be easily visually detected on a primer extension gel ( Figure 1—figure supplement 1A , left panel E1103G ) . However , less visually obvious but highly reproducible phenotypes are detected upon quantification ( Figure 1—figure supplement 1A , right panel E1103G-WT ) . As predicted from plate phenotypes observed and a previous study ( Goel et al . , 2012 ) , ssl2-DEAD and ssl2-508 showed upstream shifts in their ADH1 TSS distributions ( Figure 1C and E ) . However , we observed that these ssl2 alleles were quantitatively distinct in the amount of upstream shifting from Pol II and GTF alleles with comparable MPA sensitivities . ssl2-DEAD and ssl2-508 appeared primarily to reduce downstream TSS usage ( loss in bin 5 and gain in bin 3 ) , whereas E1103G has its largest gain in bin 2 ( compare Figure 1—figure supplement 1A E1103G , and Figure 1E ssl2-DEAD and ssl2-508 ) . Consistent with the prediction based on its imd2Δ::HIS3 phenotype , the His+ SSL2-1 mutant shifted the overall TSS distribution downstream through increased relative downstream TSS utilization ( Figure 1C and E ) . Previously , it had been concluded that SSL2-1 had no obvious effects on TSSs distribution when comparing usage of just the two major starts ( Goel et al . , 2012 ) . Two other mutants , ssl2-rtt and rad25-ts24 , had no obvious effects on TSS utilization at ADH1 , consistent with their lack of plate phenotypes ( Figure 1C and E ) . We additionally constructed and tested human disease-related XPB mutations ( Oh et al . , 2006; Cleaver et al . , 1999; Weeda et al . , 1997; Weeda et al . , 1991 ) in the yeast SSL2 system , together with mutations in the ultra-conserved arginine-glutamic acid-aspartic acid ( RED ) motif . As human disease-related alleles are many times found in conserved residues , we reasoned that some may have effects on Ssl2 biochemistry detectable in our sensitive system . We examined four human disease-related mutants that confer distinct inherited autosomal recessive disorders xeroderma pigmentosum ( XP ) , trichothiodystrophy ( TTD ) , and Cockayne syndrome ( CS ) . Of these , Q592* , which creates a C-terminally truncated Ssl2 protein ( Oh et al . , 2006 ) , confers lethality ( Figure 1—figure supplement 1B ) , and T176P ( Cleaver et al . , 1999; Weeda et al . , 1997; Weeda et al . , 1991 ) confers little if any growth defects and no MPAS or His+ phenotypes ( Figure 1F ) , consistent with unaltered TSS usage ( Figure 1G ) . In contrast , F156S ( Cleaver et al . , 1999; Weeda et al . , 1991; Vermeulen et al . , 1994 ) conferred a mild MPAS phenotype and shifted TSS distribution upstream at ADH1 ( Figure 1G ) . Mutant Y750* , which mimics a disease-related C-terminally truncated protein ( Goel et al . , 2012; Sweder and Hanawalt , 1994; Weeda et al . , 1990 ) , shows a mild to moderate level MPAS phenotype ( Figure 1F ) and shifts TSS distribution upstream at ADH1 ( Figure 1G ) . The lethal phenotypes of RED motif substitutions in ssl2 revealed their essential roles in S . cerevisiae . These results suggest that a subset of human disease alleles can alter TFIIH functions when placed in the yeast system . Our establishment of a genetic system sensitive to ssl2 mutant mediated initiation defects allowed us to obtain a broad set of alleles for the study of Ssl2 function in promoter scanning by large-scale genetic screens ( see Materials and methods ) . Our genetic screening has identified at least two phenotypically distinguishable classes of ssl2 alleles: the first class is putatively defective for the induction of the IMD2 gene , resulting in sensitivity to MPA , the second class confers constitutive expression of imd2∆::HIS3 , resulting in a His+ phenotype ( Figure 1A , Figure 2—figure supplement 1 ) . Other transcription-related or conditional phenotypes , Spt- ( Simchen et al . , 1984 ) , suppression of gal10∆56 ( GalR ) ( Kaplan et al . , 2005; Greger and Proudfoot , 1998 ) or temperature sensitivity ( Csm- , Tsm- ) , were observed in distinct subsets of alleles of the two major classes ( Figure 1A , Figure 2—figure supplement 1 ) . The Spt- phenotype reporter used in our strains , lys2-128∂ , detects activation of a TSS within a Ty1 ∂ element at the 5’ end of the LYS2 gene ( Simchen et al . , 1984 ) . Importantly , a subset of TSS-shifting alleles show the Spt- phenotype and it is useful to further classify identified ssl2 alleles . We observed that Spt- and His+ phenotypes were dominant for tested alleles ( not shown ) , suggestive of possible GOF; in contrast , there were no dominant alleles found among MPAS mutants , consistent with either recessive LOF mutations or the nature of the phenotype ( sensitivity ) or both . We then asked if TSS usage at ADH1 was altered predictably in the mutant classes as we observed for existing ssl2 mutants and previously studied Pol II and other GTF mutants . We find that the two major classes of ssl2 mutants exhibited predicted TSS shifts , with all tested MPAS alleles shifting TSS usage upstream and all His+ alleles shifting TSS usage downstream ( Figure 2—figure supplement 2 ) validating our genetic method for identifying ssl2 alleles conferring altered initiation properties . We examined our substitution mutants in the context of the structure of Ssl2 within the yeast PIC as determined by the Cramer lab ( Schilbach et al . , 2021 ) to understand how these alleles relate to PIC architecture and interactions . Substitutions causing MPAS phenotypes and upstream TSS shifts alter amino acids distributed across the protein , with a large number in conserved domains and highly conserved residues ( Figure 2A , Figure 2—figure supplement 3 ) . In contrast , mutations related to His+ phenotypes , while also generally conserved , alter amino acids clustered N-terminally , within a domain predicted ( Luo et al . , 2015; He et al . , 2016 ) and now observed to be homologous to Ssl2’s interaction partner Tfb2 ( Schilbach et al . , 2021; Greber et al . , 2019; Figure 2A ) . This domain was first fully observed in a human core TFIIH structure ( PDB 6NMI ) ( Greber et al . , 2019 ) but has now been observed in just published higher resolution PIC structures ( Schilbach et al . , 2021; Figure 2B ) . This key visualization of the conserved Ssl2 N-terminus allows placement of most ssl2 mutant residues identified as His+ or both His+ and Spt- ( Figure 2C and D ) . In addition , the Spt- phenotype , not previously observed for classical ssl2 alleles , was found exclusively within a subset of stronger downstream TSS shifting His+ mutants ( Figure 2A ) . The coincidence between His+ and Spt- phenotypes for ssl2 mutants is in contrast to what is observed for Pol II mutants . In our previous studies of Pol II mutants , the Spt- phenotype was observed in Pol II catalytic center substitutions , overlapping with MPAS to a large extent , and tightly linked with increased Pol II activity ( Kaplan et al . , 2012; Kaplan et al . , 2008; Braberg et al . , 2013 ) . Pol II alleles with both Spt- and MPAS phenotypes increased the efficiency of TSS usage resulting in upstream shifts to TSS distributions across promoters in vivo . However , in our identified ssl2 alleles , none of the Spt- mutants also conferred MPAS ( Figure 2A ) . These observations together are consistent with distinct effects on structure and function in ssl2 mutant classes . Some TSS shifting substitutions are located on the Ssl2 surface ( Figure 2B ) and a subset ( e . g . D610 ) are located proximal to DNA ( Figure 2C ) . D610A confers an upstream TSS shift . In contrast , R636C , which is also close to DNA , confers a downstream TSS shift . F498 , which is located in the groove of Ssl2 lobe 1 and facing DNA , caused an upstream TSS shift when substituted with leucine . Additionally , a small patch of residues with many TSS shifting substitutions is found on the Ssl2 lobe 1 surface ( Figure 2A ) . These substitutions are from the helicase domain 1 and shift TSSs upstream , including D522V , K523N , N528I , F529L , and E537G . In addition , substitutions I383T and K372E are in residues proximal to this small patch but shift TSS downstream . Intriguingly a number of alleles of both classes are found in the Ssl2 N-terminal domain homologous to Tfb2 ( TFB2C-like or ‘Clutch’ ) that forms interaction with Tfb2 and bridges Tfb2 with Ssl2 helicase domain 1 ( Figure 2A–D ) . As Tfb2/p52 recruits Ssl2/XPB into TFIIH ( Jawhari et al . , 2002 ) and stimulates Ssl2/XPB catalytic activity ( Coin et al . , 2007 ) , this interface is of special interest for how other factors might communicate with Ssl2/XBP function in both transcription and NER . Highlighting the uniqueness and potential plasticity of this region , we identified multiple substitutions at the conserved N230 in this domain , with N230D/S conferring MPAS and an upstream TSS shifts , while N230I conferred both His+ and Spt- phenotypes and a downstream TSS shift ( Figure 2A–D ) . These results suggest that altered Ssl2 DNA or intradomain interactions alter Ssl2 function in TSS selection in distinct ways , likely through distinct effects on Ssl2 biochemical activity , discussed below and in Discussion . We highlight two alleles as representative of distinct ssl2 allele classes in comparison to examples of the two Pol II allele classes: ssl2 N230D , which is MPA sensitive and appears to reduce ability to use downstream TSSs at ADH1 , and ssl2 N230I , which shows both His+ and Spt- phenotypes and shifts TSSs downstream relative to WT at ADH1 ( Figure 2E and F rpb1 alleles , 2G and H ssl2 alleles ) . As we have shown previously , rpb1 mutants also fall into two major classes regarding transcription phenotypes and TSS shifts . As a comparison , rpb1 E1103G is MPAS and shifts TSS usage upstream while rpb1 H1085Y is His+ and shifts TSS usage downstream ( Malik et al . , 2017; Kaplan et al . , 2012 ) . MPA sensitivities for both rpb1 E1103G and ssl2 N230D correlate with an upstream TSS shift at ADH1 , similarly to our observation for existing ssl2 MPAS alleles , ssl2-DEAD and ssl2-508 . However , all ssl2 MPAS alleles examined appear to shift TSS distribution upstream by limiting or truncating TSS usage at downstream sites and not by activating lowly used upstream sites as rpb1 E1103G does ( and as known TFIIF alleles do ) ( Fishburn et al . , 2015; Figure 2F and H ) . This behavior suggests that rpb1 and ssl2 alleles may alter TSS usage by distinct mechanisms . To gain an insight into the impact of TFIIH’s activity on TSSs genome-wide , we have examined 5′ ends of RNA transcripts for these two classes of ssl2 allele in S . cerevisiae by performing TSS sequencing ( TSS-seq ) ( Qiu et al . , 2020; Vvedenskaya et al . , 2015; Figure 3A ) . In total , six ssl2 alleles were analyzed along with a WT control , including three His+ alleles ( L225P , N230I , and R636C ) that shift TSSs downstream at ADH1 , and three MPAS alleles ( N230D , D522V , and Y750* ) that shift TSSs upstream at ADH1 . Furthermore , we performed TSS-seq on previously analyzed Pol II WT , E1103G , and H1085Y for direct comparison purposes using our updated protocol ( Materials and methods ) . The positions and counts of the 5′ ends of uniquely mapped reads were extracted to evaluate correlation and assess the reproducibility between biological replicates ( Figure 3B , Figure 3—figure supplement 1A ) . We previously found that clustering of correlation coefficients among libraries could distinguish Pol II mutants into GOF and LOF groups ( Qiu et al . , 2020 ) . Here , we also found that mutant phenotypic classes were distinguished by this clustering with ssl2 and Pol II alleles separated suggesting effects observed at individual promoters are predictive of effects observed across the genome ( Figure 3C , Figure 3—figure supplement 1B ) . ssl2 alleles shift TSS distribution genome-wide . As in Qiu et al . , 2020 , we focused on initiation within a cohort of 5979 promoters for a large number of mRNA genes and non-coding RNAs . These promoters are separated into Taf1-enriched or -depleted subclasses as a proxy for the two primary promoter classes in yeast ( Rhee and Pugh , 2012 ) . We took a simple approach to examine how mutants affect observed TSS distributions using a few key metrics , for example , the position in a promoter window containing the median read in the distribution as a measure of where the TSS distribution is in a particular promoter window ( Figure 4A ) . We observed decreased TSS signal downstream of the WT median TSS signal in ssl2 N230D and other ssl2 MPAS alleles ( Figure 4B , Figure 4—figure supplement 1 ) . In contrast , substantially increased TSS usage was observed at downstream sites in ssl2 N230I and other ssl2 His+ alleles ( Figure 4B , Figure 4—figure supplement 1 ) . To quantify the change of TSS usage , we measured the TSS shift between WT and mutant strains by subtracting the WT median TSS position within each promoter window from the mutant median TSS position ( Figure 4C ) . For each mutant , we show the median TSS shifts across promoters in both heatmap and boxplot format ( Figure 4D–E , Figure 4—figure supplement 2A ) . As predicted from our model , ssl2 MPAS alleles shift the median TSS positions upstream at most promoters , showing a similar profile to Pol II GOF mutant ( Figure 4D , Figure 4—figure supplement 2A ) . Also as predicted from our model , ssl2 His+ alleles shift median TSS positions downstream at the majority of promoters and show a similar profile to a Pol II LOF mutant ( Qiu et al . , 2020; Figure 4D , Figure 4—figure supplement 2A ) . Additionally , mutants were clustered into upstream and downstream classes based on shifts in promoter median TSS position ( Figure 4D , Figure 4—figure supplement 2A ) . Principal component analysis ( PCA ) of TSS shifts distinguishes ssl2 and Pol II mutants into four major classes , namely Pol II LOF ( downstream shifting ) and GOF ( upstream shifting ) , and ssl2 upstream and downstream shifting mutants ( Figure 4—figure supplement 2B ) . We observed that the magnitude of TSS shift was consistent with the strengths of putative TSS shift-dependent growth phenotypes in ssl2 upstream shifting mutants ( Figure 4E ) . For example , alleles of N230D , D522V , and Y750* , from the left to right , show a gradient of MPAS phenotypes in genetic tests ( Figure 2A ) , while also showing a gradient in TSS shift magnitudes across promoters ( Figure 4E ) . Notably , the extents of TSS shifts in ssl2 alleles are less than for Pol II activity mutants , indicating a more dramatic effect of Pol II’s catalytic activity on TSS distributions ( Figure 4E ) . These results are consistent with phenotypes of mutants being predictive of global TSS defects among mutants for a particular gene but not necessarily between genes , which we discuss later as indicative of different mechanisms for Pol II and Ssl2 effects on TSS selection . To evaluate the effects of ssl2 and Pol II alleles on scanning distance , the width between positions of the 10th and 90th percentiles of the TSS signal at each promoter window was determined ( TSS ‘spread’ ) ( Figure 5A ) . We observed obvious narrowing of TSS spreads in ssl2 upstream shifting mutants and widening of TSS spreads in ssl2 downstream shifting mutants ( Figure 5B–D ) . The profiles of the TSS spread difference between mutant and WT at each individual promoter ( Figure 5A ) also differentiates mutants into clear shift classes ( Figure 5C; Figure 5—figure supplement 1A , B ) . Changes in TSS spread for ssl2 alleles are distinct from how Pol II mutants alter TSS spread ( Figure 5D ) . Both classes of ssl2 mutant show large bias in direction of shift in spread , relative to WT across a number assessments ( Figure 5—figure supplement 2D-F ) . Consistently , MPAS ssl2 alleles ( those that shift TSSs upstream ) showed narrowing in TSS spread at the majority of promoters while His+ ssl2 alleles ( those that shift TSSs downstream ) showed widening in TSS spread at the majority of promoters . These results extend the idea that while classes of initiation allele may be easily identified for Pol II or SSL2 using the same genetic phenotypes , their effects on TSSs at individual promoters are quantitatively and likely qualitatively distinct . To explain the observed quantitative differences between Pol II initiation efficiency alleles and ssl2 alleles , we hypothesize that ssl2 alleles that narrow TSS spreads ( ssl2 N230D and similar alleles ) , resulting in upstream shifts in TSS distributions , are defective in scanning processivity due to decreased Ssl2/TFIIH translocase activity . In contrast , ssl2 alleles that show increased TSS spreads and usage at downstream sites ( N230I and similar alleles ) behave as increased scanning processivity ( GOF ) alleles , due to an increase in Ssl2/TFIIH activity . Alternatively , ssl2 N230I might instead be a scanning rate GOF allele that decreases initiation efficiency across TSSs by decreasing the exposure time of each TSS within Pol II active site . As a consequence , there would be , hypothetically , increased TSS usage at downstream sites due to increased Pol II flux reaching those positions , similar to Pol II LOF efficiency alleles . To probe mechanisms of Ssl2 function , we designed ssl2 genetic interaction studies for which we have specific predictions based on their possible roles ( Figure 6—figure supplement 1 ) . These studies emerge from our prior observations of genetic interactions from three angles between initiation factors such as Pol II activity mutants and alleles of GTFs ( Jin and Kaplan , 2014 ) . First , we have examined general effects on growth between classes of mutant , which can manifest as synthetic sickness or lethality , suppression , or lack of effects on growth . Second , we have examined suppression or enhancement of transcription-related phenotypes in double mutants , allowing detection of additive or epistatic interactions based on putative transcription defects at the genetic reporter loci . Finally , we have quantitatively examined effects of double mutants on TSS distributions at ADH1 , allowing additive or epistatic interactions to be observed at the individual promoter level . In our previous studies , we found that Pol II activity mutants and alleles of GTFs ( sua7/TFIIB and tfg2/TFIIF ) showed additivity/suppression for transcription-dependent phenotypes as well as additivity/suppression for alterations to TSS distributions at ADH1 . These studies suggested that these GTF alleles were functioning in same pathway as Pol II catalytic mutants , namely controlling the efficiency of initiation across individual TSSs . In order to understand how ssl2 alleles interact with other initiation factors and how Ssl2 functions within the network of activities that are essential for TSS selection , we generated double mutants among a collection of ssl2 alleles , Pol II activity-altering alleles , sua7-1 , tfg2∆146–180 , and sub1∆ . To streamline display a large number of genetic interactions and growth phenotypes , we have converted general growth on plates and growth on phenotype-specific media to qualitative scores ( Figure 6D–F , Figure 6—figure supplement 2 ) and these are shown as heatmaps ( Figures 6 and 7 ) with primary data shown in Figure 6—figure supplement 3 , Figure 7—figure supplement 1 , and Figure 7—figure supplement 2 . Our detailed studies are summarized as follows ( for detailed discussion , see version 1 of the pre-print of this work Zhao et al . , 2021 ) . In contrast to the suppressive/additive interactions that were broadly observed between Pol II and TFIIB/TFIIF alleles ( Jin and Kaplan , 2014 ) , we observe primarily epistatic effects between Pol II and ssl2 alleles . First , the broad synthetic lethality or enhancement/additivity of transcriptional phenotypes through combining Pol II and TFIIB/TFIIF alleles that shift TSS distributions in the same direction were not observed between Pol II and ssl2 alleles ( Figure 6 ) . Examples of epistatic interactions , where double mutants between Pol II and ssl2 alleles have phenotypes of either the Pol II single mutant or the ssl2 single mutant , were found in a number of cases . Each case supports a model where ssl2 alleles are functioning through scanning processivity and not initiation efficiency directly . This epistasis is best reflected by nearly complete absence of synthetic lethality between ssl2 downstream shifting alleles and Pol II downstream shifting alleles ( Figure 6A–C ) , in contrast to interactions between all other classes of downstream shifting allele ( e . g . Pol II , sua7/TFIIB , sub1∆; Jin and Kaplan , 2014 ) . Epistasis for both transcription phenotypes and ADH1 TSS shifts was also observed between Pol II upstream shifting alleles and both classes of ssl2 allele , meaning that double mutants had phenotypes of Pol II single mutants ( Figure 6A–K , Figure 6—figure supplement 3 ) . For each of these cases , results support a model where if initiation is efficient enough or early enough in a scanning window , that is , due to increased Pol II initiation activity , then increase in scanning processivity ( e . g . ssl2 N230I ) loses ability to alter TSS distributions , while a decrease in scanning processivity is buffered against due to high enough gain in transcription efficiency in tested Pol II alleles . Similarly , both classes of ssl2 alleles appeared epistatic or non-additive with Pol II downstream shifting alleles ( Figure 6A–C , Figure 6—figure supplement 3 ) , also consistent with determination of scanning window by ssl2 activity to be upstream of ability of Pol II mutants to alter TSS distributions through altered initiation efficiency . Interactions between ssl2 alleles and other GTFs or sub1∆ reveal complexities that are of special note as they suggest non-obvious roles/interactions between these factors and Ssl2 function ( Figure 7 , Figure 7—figure supplements 1 and 2 ) . We anticipated that sua7-1 and tfg2∆146–180 , encoding mutant forms of TFIIB and TFIIF respectively , would behave strictly as TSS efficiency alleles due to their additive behavior with Pol II alleles ( Jin and Kaplan , 2014 ) , and therefore would similarly show epistatic effects with ssl2 alleles . Notably , lethal phenotypes were observed between individual ssl2 alleles and sua7-1 or tfg2∆146–180 alleles for combinations between single mutants that alter TSS distributions in the same direction , distinct from their interactions with Pol II alleles ( Figure 7A ) . We suggest two possibilities for this observation: first , sua7-1 and tfg2∆146–180 could confer additional defects causing increased sensitivity to ssl2 defects , for example , altered PIC integrity; second , sua7-1 and tfg2∆146–180 might be sensitized to increased Ssl2 processivity ( for sua7-1 ) or decreased Ssl2 processivity ( for tfg2∆146–180 ) in addition to their altered TSS efficiency effects ( see Discussion ) . When we combined alleles of sua7-1 or tfg2∆146–180 with ssl2 alleles that shift TSS distributions in opposite directions , interactions were complex but significant epistasis was observed . Consistently , double mutants shifted ADH1 TSS distributions to similar extent as the tfg2∆146–180 single mutant ( Figure 7C , Figure 7—figure supplement 1 ) as predicted for an increase in initiation efficiency buffering against effects of increase in scanning processivity . Sub1 , a conserved factor ( yeast homolog of mammalian PC4 ) was previously found to facilitate Pol II transcription in a variety of ways ( Garavís and Calvo , 2017; Calvo , 2018 ) , to be recruited to the PIC ( Sikorski et al . , 2011 ) , and to alter accessibility of promoter single-stranded DNA , consistent with initiation functions ( Lada et al . , 2015 ) . sub1∆ has extensive genetic interactions with initiation factors and itself causes TSSs to shift downstream ( Wu et al . , 1999; Knaus et al . , 1996; Braberg et al . , 2013; Koyama et al . , 2008 ) , though its actual role in initiation is unknown . We previously found sub1∆ to confer a His+ phenotype for the imd2Δ::HIS3 initiation reporter ( Malik et al . , 2017 ) and furthermore found that Pol II GOF alleles appeared epistatic to sub1∆ , leading to the proposal that sub1∆ effects in initiation were distinct from TFIIB or TFIIF alleles ( Jin and Kaplan , 2014 ) . Because we have observed similar epistatic interactions between ssl2 and Pol II alleles , we considered that Sub1 might also be behaving as a scanning processivity factor . Therefore , we predicted the possibility of additive effects between two types of processivity alleles , namely ssl2 upstream and downstream shifting alleles and sub1Δ , if they are acting independently . First , no strong genetic interactions ( lethality ) were observed between ssl2 and sub1Δ alleles , save for one specific case ( Figure 7A , Figure 7—figure supplement 2A ) . Second , the majority of sub1∆ interactions with ssl2 alleles appear to be additive when examining TSS distributions at ADH1 as predicted for factors are acting on processivity independently . Third , and notably , we identified allele-specific interactions between sub1∆ and specific ssl2 alleles within classes of ssl2 allele that until these experiments have not been distinguishable . For example , most upstream shifting ssl2 alleles were additive with sub1∆ for TSS distributions at ADH1 , resulting in mutual suppression of TSS distribution shifts ( Figure 7D ) . In contrast , sub1∆ was epistatic to ssl2 Y750* , suggesting that a putative block to processivity due to C-terminal truncation of Ssl2 can be relieved by sub1∆ , and potentially may be due to altered Sub1 function in ssl2 Y750* . Finally , allele specificity of ssl2 F498L was revealed by these genetic experiments . This TSS upstream shifting ssl2 allele was unexpectedly synthetic lethal with both sua7-1 and sub1∆ suggesting heretofore undetected phenotypic differences from other alleles of the same class . Results of genetic interaction studies are consistent with two distinct networks controlling TSS selection by scanning ( Figure 8 ) . Additive/suppressive interactions were observed within networks while specific classes of mutants showed epistatic interactions between networks . One network impinges on Pol II catalysis and initiation efficiency , and genetic analyses suggest that the Pol II active site collaborates with activities of TFIIB and TFIIH in this process , consistent with experiments indicating effects of TFIIB and TFIIF on Pol II catalytic activity ( e . g . , Khaperskyy et al . , 2008; Cabart et al . , 2014; Sainsbury et al . , 2013 ) . The other , we propose , impinges on scanning processivity through TFIIH with the participation of Sub1 . Our genetic interactions also uncover functional connections between TFIIB and TFIIF and Ssl2 that are distinct from Pol II active site mutants . These results support predictions of altered PIC function for TFIIB and TFIIF mutants beyond phosphodiester bond formation and will be interesting to test in biophysical experiments . Extensive epistasis observed between networks ( Figure 8 ) supports predictions for how efficiency and processivity should interact during initiation by promoter scanning ( Figure 9 , see Discussion ) . We found previously that polar shifts of TSS distribution in Pol II catalytic activity mutants were accompanied with alteration in PIC localization as detected by ChIP-exo ( Qiu et al . , 2020 ) . Shift in PIC components upon alteration to Pol II catalytic activity suggested that extent of scanning might be coupled to Pol II initiation , or that alteration to Pol II initiation kinetics affects observed distributions of GTFs . We performed these same ChIP-exo experiments on Sua7 and Ssl2 for two ssl2 alleles , N230D and N230I ( Figure 9—figure supplement 1A , B ) . Both shifted PIC localization genome-wide with the same polarity as they shift TSS distributions . We note that TAP-tagging Ssl2 confers slight phenotypes on its own and slight enhancement of ssl2 N230D and slight suppression of ssl2 N230I ( Figure 9—figure supplement 2 ) . However , each tagged mutant was compared to the tagged WT and the results are robust and distinct for each mutant . The extent of ChIP-exo shifts were as strong or stronger than Pol II mutant shifts although Pol II mutants have stronger effects on TSS distributions ( Qiu et al . , 2020 ) . These results could be consistent with scanning by TFIIH on DNA uncoupled from the Pol II initiation decision , that is , ssl2 mutants extend PIC scanning to downstream positions even though initiation has occurred . Such a result would be consistent with the similar behavior for DNA compaction in optical tweezer analysis of initiation wherein dATP-supported reactions ( presumptive TFIIH scanning-driven DNA translocation through Ssl2 use of dATP ) and NTP-supported initiation reactions ( TFIIH translocation and Pol II initiation allowed ) have similar behavior ( Fazal et al . , 2015 ) . The extent or mechanism of uncoupling between promoter scanning upon productive initiation is unknown and represents an open question in initiation mechanisms . We have previously described how Pol II determines the efficiency of a TSS in a ‘Shooting Gallery’ model , where the rate at which a TSS ( conceived of as a target ) passes the active site , the rate of firing ( catalytic activity ) , and the size of the target ( innate sequence strength ) together contribute to the probability a target is hit ( initiation happens ) ( Figure 9A; Qiu et al . , 2020; Kaplan , 2013 ) . Alteration of enzymatic activities supporting initiation , either the Pol II active site or TFIIH translocation , will have predictable effects on individual TSS usage and the overall TSS distributions when initiation proceeds by scanning . In Pol II mutants with altered catalytic activity that is known to affect transcription efficiency , we observed polar changes to TSS distributions ( Qiu et al . , 2020 ) . Distributions will also necessarily be shaped by an additional factor: Pol II flux . Pol II flux describes the relative number of polymerases encountering a given start site , which has a higher value at upstream TSSs and a lower value at downstream TSSs , resulting in reduced apparent usage at downstream position distinct from their inherent efficiencies ( Figure 9B ) . Additionally , the potential upstream and downstream constraints for defining the scanning window will have effects on TSS distributions . Studies suggest that very upstream TSSs close to the presumed location of PIC assembly show reduced transcription initiation ( Faitar et al . , 2001 ) . The physical basis for defining the upstream boundary of the scanning window has not yet been determined . An obvious constraint is the minimum space required for PIC assembly . Moreover , we hypothesize that downstream constraints for defining the scanning window could be TFIIH’s processivity , the +1 nucleosome , or both . Previous single molecule studies suggested that TFIIH drives downstream scanning distances similar in length to the distribution of TSSs at yeast promoters ( Fazal et al . , 2015 ) . We propose that TSS distribution of a promoter is established by the cooperation of Pol II’s catalytic activity and TFIIH’s processivity for reaching and activating TSSs at promoter sites . When Pol II has increased catalytic activity , for example , in Pol II catalytic activity GOF alleles , upstream TSSs will increase in efficiency ( Figure 9C and D , Pol II GOF ) . In this allele class , usage of downstream TSSs also will decrease due to reduction in Pol II flux reaching downstream sites due to prior initiation . Conversely , when Pol II has decreased catalytic activity , TSSs at upstream sites will be less efficiently used , more slowly reducing Pol II flux ( Figure 9C and D , Pol II LOF ) . Inability to initiate earlier in scanning will result in increased TSS usage at downstream sites and a flattening and spreading of the TSS distribution ( as demonstrated by an efficiency curve with decreased slope ) . We hypothesize that alleles with increased processivity ( processivity GOF allele ) will expand the scanning window by allowing the PIC to scan further downstream while attempting initiation , increasing the probability that downstream TSSs are reached during any individual scanning event ( Figure 9E , processivity GOF ) . As a consequence , processivity GOF alleles increase the potential for scanning downstream but only if Pol II flux ( Pol II molecules still scanning ) persists to reach those sites . In contrast , a processivity LOF allele would limit the Pol II machinery’s access to downstream TSSs sites by reducing the scanning window ( Figure 9E , processivity LOF ) . Consequently , there would be an upstream shift in TSS distribution compared to WT , without the activation of additional upstream TSSs . Prior biochemical and more recent structural analyses indicate that TFIIH is a fascinating complex with numerous contacts suggested or predicted to modulate or control TFIIH enzymatic subunits’ activities ( Nogales and Greber , 2019 ) . For example , Ssl2/XPB must be activated during transcription initiation to allow promoter opening . Recent structures suggest that interactions with both Mediator and TFIID may position parts of TFIIH for different functions in initiation ( Abdella et al . , 2021; Chen et al . , 2021; Rengachari et al . , 2021 ) . Genetic and biochemical studies also suggest that TFIIH may itself impose a block to initiation that is then relieved by TFIIH activity through Ssl2/XPB ( Alekseev et al . , 2017; Lin et al . , 2005 ) . Both TFIIH ATPase subunits , Rad3/XPD and Ssl2/XPB , must also be regulated for TFIIH’s function in NER with Rad3/XPD held inactive during transcription and released for NER ( reviewed in Greber et al . , 2019; Nogales and Greber , 2019 ) . XPB mutations in patients that result in XP are straightforwardly interpreted as conferring NER defects , however transcriptional phenotypes may also be present depending on mutation ( Oh et al . , 2006; Cleaver et al . , 1999; Weeda et al . , 1997 ) . Mutations in XPB that cause TTD localize to conserved residues in the XPB N-terminus where we have identified a number of mutations . TTD mutations in XPB have been interpreted as reducing the amount of TFIIH in the cell through potential destabilization , while one appears to impact folding and activity of XPB ( Greber et al . , 2019 ) . Our identification of putative LOF and GOF mutations in this domain in S . cerevisiae underscores the idea that observed conservation in this region may control key inputs to Ssl2/XPB activity . The substitutions we have identified are largely in residues conserved from yeast to humans ( Figure 2—figure supplement 3 ) , and we suggest that these residues detect potential paths for allosteric regulation of Ssl2/XPB . Only a subset of our alleles confer UV sensitivity , suggestive of NER defects ( van Eeuwen et al . , 2021 ) . These are C-terminal and this suggests that our alleles uniquely alter Ssl2 modulation in transcription or that transcriptional functions of Ssl2 are sensitized to defects that do not appreciably lead to UV sensitivity . DNA translocases are the engines for chromatin remodeling and much regulation of chromatin remodelers relates to coupling of ATP hydrolysis to translocation potential ( Clapier et al . , 2017; Clapier et al . , 2016 ) . Mutations in remodelers that increase or decrease coupling have strong effects on remodeling . We posit that it is likely that a number of our alleles will act through altered coupling of ATPase activity and translocation , with the end result being increase or decrease in scanning processivity . Biochemical studies will reveal specific aspects of TFIIH activity that are altered by these substitutions . Our putative GOF alleles are concentrated in the TFB2C-like N-terminal domain and DRD of Ssl2 . The TFB2C domain has been implicated as a target of Tfb3/Mat1 in restricting Ssl2/XPB activity in holoTFIIH ( Luo et al . , 2015; Nogales and Greber , 2019; Greber et al . , 2017 ) . However , upon assembly into the PIC , Tfb3/Mat1 releases the N-terminus of Ssl2 ( Abdella et al . , 2021; Schilbach et al . , 2017 ) . This Ssl2 domain is also targeted by Tfb2/p52 ( Schilbach et al . , 2021; He et al . , 2016; Greber et al . , 2019; Schilbach et al . , 2017 ) , which has long been implicated in modulating Ssl2/XPB activity in addition to assembling it into TFIIH . We have identified one His+ allele at the Ssl2-Tfb2 interface , though most are in internal interfaces at the putative nexus of Ssl2 NTD , DRD , and HD1/ATPase lobe 1 ( Figure 2 ) . Key open questions relate to how the PIC communicates to Ssl2/XPB to engage and open promoter DNA , what the mechanistic basis for any imposition of initiation block by Ssl2/XBP is , what the basis of its subsequent relief is , and how translocation is terminated upon or subsequent to productive initiation , that is , are the processes coupled in anyway ? Aibara , Schilbach et al . have recently imaged the human PIC captured in two states , suggestive of pre- and post-translocation intermediates of TFIIH ( Aibara et al . , 2021 ) . These structures show loss of contact between TFIIH through the MAT1 RING domain and the Pol II stalk/TFIIE in the proposed post-translocated state . This raises an attractive model for uncoupling TFIIH translocase from Pol II after a single translocation step , functionally limiting initiation to the small window of exposed TSSs within reach of the Pol II active site . These structures were from a minimal PIC lacking Mediator and TFIID and therefore it remains to be determined if this translocase is in fact uncoupled , as other potential TFIIH/PIC contacts remain . Other events during initiation may also propagate changes to the PIC , such as lengthening of the nascent RNA to potentially clash with TFIIB and potential subsequent reorganization of the PIC by this event , or due to Pol II CTD phosphorylation . In many organisms , nucleosome-depleted regions promote initiation bidirectionally and these regions are flanked by positioned nucleosomes ( Vo Ngoc et al . , 2017; Duttke et al . , 2015; Chen et al . , 2016; Kaplan , 2016 ) . Nucleosomes would potentially act as competitors for double-stranded DNA being translocated by TFIIH . Transcription activity drives histone dynamics at promoters , consistent with TFIIH translocation proposed to function akin to a chromatin remodeler ( Tramantano et al . , 2016 ) . How the +1 nucleosome might feed back on scanning or regulate TFIIH translocation is an open question . In the absence of the remodeler RSC’s function in yeast , nucleosomes move upstream into normally nucleosome depleted regions and inhibit or narrow TSS usage at a number of promoters ( Klein-Brill et al . , 2019 ) . These results are consistent with nucleosomes competing with TFIIH for promoter DNA . However , these results are conditional on RSC depletion , and it is not clear if promoter nucleosome remodeling under normal conditions obviates the ability for +1 nucleosomes at active promoters to provide a block to scanning or initiation on the edges of their positions . How Pol II specifies multiple TSSs at individual promoters across eukaryotes has long been an open question . The observation of downstream-located TSSs relative to where DNA melting occurs in S . cerevisiae led to the original proposal of a scanning mechanism for TSS identification ( Giardina and Lis , 1993 ) . In contrast to how Pol II finds TSSs in S . cerevisiae for all promoters , in other eukaryotes including other fungi , a scanning process is not required for promoters with defined architecture specified by a TATA element ( Lu and Lin , 2021; Breathnach and Chambon , 1981 ) . These promoters use highly focused TSSs immediately and precisely downstream of the DNA melting sites ~30 bp downstream of the TATA element +1 position . For example , Lu et al . have pinpointed the split between scanning from TATA-promoters and non-scanning species within the Saccharomycetes ( Lu and Lin , 2021 ) . However , most eukaryotic promoters are TATA-less and use multiple TSSs ( Vo Ngoc et al . , 2017; Li et al . , 2015Kawaji , 2014; Saito et al . , 2013; Nepal et al . , 2013; Chen et al . , 2013; Yamashita et al . , 2011; Hoskins et al . , 2011; Kawaji et al . , 2006; Carninci et al . , 2005; Kadonaga , 2012; Haberle and Stark , 2018 ) . Therefore , whether or not scanning is also a mechanism in higher eukaryotic promoters , or minimally for a subset of eukaryotic promoters or within a specified window , is still an unanswered question as there have been no formal tests of this mechanism . It has been suggested , however , that each individual TSS is recognized as an individual promoter due to sequence signatures apparent in comparison of thousands of TSSs in humans ( Luse et al . , 2020 ) . Very recent results of cryo-EM studies on human PICs , especially in structures visualizing TFIID , indicate that promoter classes may assemble PIC components in distinct fashion within a single organism ( Chen et al . , 2021 ) , yet these assembly pathways similarly position an upstream TSS proximal to the Pol II active site , consistent with proposals that human promoters could contain information for assembling PICs individually ( Luse et al . , 2020 ) . Recent results suggest that there may be plasticity in TSS selection from individual PICs upon mutation of Inr sites in mouse , potentially supporting a type of scanning in mammals ( Chou et al . , 2021 ) . That diverse initiation mechanisms are supported by highly conserved factors suggests that we may yet to find additional unexpected plasticity in initiation across evolution in eukaryotes . Yeast strains are derived from a GAL+ of S288C ( FY2 ) ( Winston et al . , 1995 ) . Yeast strains used in this study are listed in Supplementary file 1 . Bacterial strains and plasmids used in this study are listed in Supplementary file 2 . Yeast media used in this study were made as previously described ( Jin and Kaplan , 2014; Malik et al . , 2017; Kaplan et al . , 2012; Amberg et al . , 2005 ) . Briefly , YP medium is made of yeast extract ( 1% w/v; BD ) and peptone ( 2% w/v; BD ) . Solid YP medium contained bacto agar ( 2% w/v; BD ) , adenine ( 0 . 15 mM , Sigma-Aldrich ) , and tryptophan ( 0 . 4 mM , Sigma-Aldrich ) . YPD medium uses YP medium components supplemented with dextrose ( 2% w/v , VWR ) . YPRaf medium uses YP medium components supplemented with raffinose ( 2% w/v , Amresco ) and antimycin A ( 1 mg/ml; Sigma-Aldrich ) . YPRafGal medium uses YP medium components and supplemented with raffinose ( 2% w/v ) , galactose ( 1% w/v; Amresco ) , and antimycin A ( 1 mg/ml; Sigma-Aldrich ) . Minimal media ( SC- ) was made with a slightly modified ‘Hopkins mix’ ( 0 . 2 % most amino acids w/v ) , and supplemented with Yeast Nitrogen Base containing ammonium sulfate ( without amino acids , BD ) , bacto agar ( 2% w/v; BD ) , and dextrose ( 2% w/v , VWR ) . The original ‘Hopkins mix’ and the slight modification were as previously described ( Kaplan et al . , 2012; Amberg et al . , 2005 ) . All amino acids are from Sigma-Aldrich . SC-Leu +5-FOA ( 5-fluoroorotic acid ) is minimal medium of SC-Leu supplemented with ( 5-FOA ) , ( 1 µg/ml , Gold Biotechnology ) . SC-Leu+ MPA media is minimal media of SC-Leu supplemented with MPA ( 20 µg/ml , Sigma-Aldrich , from a 10 mg/ml stock in ethanol ) . SC-His +3AT is minimal medium of SC-His supplemented with 3-aminotriazole ( 0 . 5 mM , Sigma-Aldrich ) . Yeast phenotyping assays were performed by spotting 10-fold serial dilutions of saturated YPD-liquid yeast cultures on various solid media , as previously described ( Kaplan et al . , 2012 ) . Yeast cells on various media were cultured at 30°C except for temperature sensitivity phenotypes , which were at 16°C ( YPD 16 ) and at 37°C ( YPD 37 ) . Yeast growth on specific media was recorded by taking pictures every 24 hr after an initial 16 hr of growth , from day 2 ( 40 hr ) to day 7 for all media except for YPRaf/Gal ( pictures to day 9 ) . Growth phenotypes on specific media were scored on days when WT yeast reached mature colony sizes , as follows: YPD on day 2 ( 40 hr after spotting ) ; SC-Leu , SC-His , and SC-Trp on day 3 ( 64 hr ) ; YPRaf on day 4 ( 88 hr ) ; SC-Lys and SC-Leu+MPA on day 5 ( 112 hr ) ; and YPRaf/Gal on day 7 ( 160 hr ) . To illustrate the strength and distribution of mutants on the two-dimensional structure of Ssl2 ( Figure 2 ) , growth phenotypes are converted to a numerical score , using the scale 0–6 to indicate the level of growth , where 0 indicates no growth and 6 indicates full growth . The level of growth is positively correlated with the strength of phenotypes for SC-His , SC-Lys , and YPRaf/Gal medium , so the ‘growth score’ is directly used as ‘phenotyping score’ for making a heatmap . For other media , the level of growth is negatively correlated with the strength of phenotype , thus growth score 0–6 is inversely converted to the phenotyping strength score 6–0 , with 6 growth score converted into phenotyping score 0 to indicate no phenotype , with 5 growth score converted into phenotyping score 1 to show a weak phenotype , and so on . The heatmap uses light to dark color showing weak ( phenotyping score 1 ) to strong phenotypes ( phenotyping score 6 ) , no phenotype ( phenotyping score 0 ) has no color . To detect putative usage of TSSs in yeast , a primer extension ( PE ) assay was performed as previously described ( Kaplan et al . , 2012 ) , modifying original protocol in Ranish and Hahn , 1991 . Briefly , 30 µg of total RNA isolated from yeast cells was used for each PE reaction . A primer complementary to ADH1 mRNA was end-labeled with gamma-P32 ATP and T4 PNK and annealed to total RNA . Reverse-transcription was then performed by adding M-MLV reverse-transcriptase ( Fermentas/ThermoFisher ) and RNase Inhibitor ( Fermentas/ThermoFisher ) . RNase A was added to remove RNA after reverse-transcription . Products were detected by running an 8% acrylamide gel ( 19:1 acrylamide:bisacrylamide ) ( Bio-Rad ) , 1× TBE , and 7 M urea . PE gels were visualized by phosphorimaging ( GE Healthcare or Bio-Rad ) and quantified by ImageQuant 5 . 1 ( GE ) or Image Lab software ( Bio-Rad ) . ssl2 mutant screening ssl2 mutants were created by PCR-based random mutagenesis coupled with a gap repair . Briefly , mutation of SSL2 ( ssl2* ) was accomplished by standard PCR reactions using Taq polymerase ( New England Biolabs ) . ssl2* PCR products were then transformed into yeast along with a linearized pRS315 SSL2 LEU2 plasmid with most of the WT SSL2 sequence removed by restriction digest . Leu+ transformants were selected . Homologous sequences on each end of the ssl2* PCR products and the gapped SSL2 vector allowed homologous recombination , resulting in a library of gap-repaired plasmids containing potential ssl2* alleles . Since SSL2 is essential , these yeast cells are pre-transformed with a pRSII316 SSL2 URA3 plasmid to support growth , while the genomic SSL2 was deleted to allow plasmid SSL2 alleles to exhibit phenotypes . After gap repair , cells retaining pRSII316 SSL2 URA3 plasmids were killed by replica-plating transformants to medium containing 5-FOA ( GoldBio ) . Yeast cells were then plated on YPD media for growth and replica-plated to a variety of media to screen for mutants that have transcription-related or conditional phenotypes . Plasmids from yeast mutants were recovered and transformed into Escherichia coli for amplification , followed by sequencing to identify mutations . Mutant yeast candidates were additionally mated with yeast cells that contain a WT SSL2 URA3 plasmid to create diploid strains and perform phenotyping again to determine dominance/recessivity of ssl2 mutations . Plasmid shuffling on diploid strains was performed by adding 5-FOA on the medium so that presumably only presumptive ssl2* was kept . This was followed by an additional phenotyping to determine if the mutant phenotype is plasmid linked or not . All mutants described here were verified by retransformation into a clean genetic background . Yeast cell cultures were grown in triplicates and cells were harvested at mid-log phase at a density of 1 × 107 cells/ml , as determined by cell counting . For S . cerevisiae TSS-seq , cells collected from 50 ml of S . cerevisiae culture and 5 ml of Schizosaccharomyces pombe culture were mixed and total RNA was extracted as described ( Schmitt et al . , 1990 ) . We performed cDNA library construction for TSS-seq essentially as described by Vvedenskaya et al . , 2015; steps are described as follows; 100 μg of the isolated total RNA was treated with 30 U of DNase I ( QIAGEN ) and purified using RNeasy Mini Kit ( QIAGEN ) . A Ribo-Zero Gold rRNA Removal Kit ( Illumina ) was used to deplete rRNAs from 5 μg of DNase-treated RNAs . The rRNA-depleted RNA was purified by ethanol precipitation and resuspended in 10 μl of nuclease-free water . To remove RNA transcripts carrying a 5’ monophosphate moiety ( 5’-P ) , 2 μg of rRNA-depleted RNA were treated with 1 U Terminator 5′-Phosphate-Dependent Exonuclease ( Epicentre ) in the 1× Buffer A in the presence of 40 U RNaseOUT in a 50 μl reaction at 30°C for 1 hr . Samples were extracted with acid phenol-chloroform pH 4 . 5 ( ThermoFisher Scientific ) , and RNA was recovered by ethanol precipitation and resuspended in 30 μl of nuclease-free water . Next , to remove 5′-terminal phosphates , RNA was treated with 1 . 5 U CIP ( NEB ) in 1× NEBuffer three in the presence of 40 U RNaseOUT in a 50 μl reaction at 37°C for 30 min . Samples were extracted with acid phenol-chloroform and RNA was recovered by ethanol precipitation and resuspended in 30 μl of nuclease-free water . To convert 5’-capped RNA transcripts to 5′-monophosphate RNAs ligatable to 5′ adaptor , CIP-treated RNAs were mixed with 12 . 5 U CapClip ( Cellscript ) and 40 U RNaseOUT in 1× CapClip reaction buffer in a 40 μl reaction and incubated at 37°C for 1 hr . RNAs were extracted with acid phenol-chloroform , recovered by ethanol precipitation and resuspended in 10 μl of nuclease-free water . To ligate the 5′ adapter , the CapClip-treated RNA products were combined with 1 μM 5′ adapter oligonucleotide s1086 ( 5′-GUUCAGAGUUCUACAGUCCGACGAUCNNNNNN-3′ ) , 1× T4 RNA ligase buffer , 40 U RNaseOUT , 1 mM ATP , 10% PEG 8000 and 10 U T4 RNA ligase 1 in a 30 μl reaction . The mixtures were incubated at 16°C for 16 hr and the reactions were stopped by adding 30 μl of 2× RNA loading dye . The mixtures were separated by electrophoresis on 10% 7 M urea slab gels in 1× TBE buffer and incubated with SYBR Gold nucleic acid gel stain . RNA products migrating above the 5′ adapter oligo were recovered from the gel as described ( Pinto et al . , 1994 ) , purified by ethanol precipitation and resuspended in 10 μl of nuclease-free water . To generate first strand cDNA , 5′-adaptor-ligated products were mixed with 0 . 3 μl of 100 μM s1082 oligonucleotide ( 5′-GCCTTGGCACCCGAGAATTCCANNNNNNNNN3′ N = A/T/G/C ) containing a randomized 9 nt sequence at the 3′ end , incubated at 65°C for 5 min , and cooled to 4°C . A solution containing 4 μl of 5× First-Strand buffer , 1 μl ( 40 U ) RNaseOUT , 1 μl of 10 mM dNTP mix , 1 μl of 100 mM DTT , 1 μl ( 200 U ) of SuperScript III Reverse Transcriptase and 1 . 7 μl of nuclease-free water was added to the mixture . Reactions were incubated at 25°C for 5 min , 55°C for 60 min , 70°C for 15 min , and cooled to 25°C . 10 U RNase H was added , the mixtures were incubated 20 min at 37°C and 20 μl of 2× DNA loading solution ( PippinPrep Reagent Kit , Sage Science ) were added . Nucleic acids were separated by electrophoresis on 2% agarose gel ( PippinPrep Reagent Kit , external Marker B ) to collect species of ~90 to ~ 550 nt . cDNA was recovered by ethanol precipitation and resuspended in 20 μl of nuclease-free water . To amplify cDNA , 9 μl of gel-isolated cDNA was added to the mixture containing 1× Phusion HF reaction buffer , 0 . 2 mM dNTPs , 0 . 25 μM Illumina RP1 primer ( 5′-AATGATACGGCGACCACCGAGATCTACACGTTCAGAGTTCTACAGTCCGA-3′ ) , 0 . 25 μM Illumina index primers RPI3-RPI16 ( index primers have the same sequences on 5′ and 3′ ends , but different on 6 nt sequence that serves as a barcode ( underlined ) ; RPI3: 5′ -CAAGCAGAAGACGGCATACGAGATGCCTAAGTGACTGGAGTTCCTTGGCACCCGAGAATTCCA-3′ ) , and 0 . 02 U/μl Phusion HF polymerase in 30 μl reaction . PCR was performed with an initial denaturation step of 10 s at 98°C , amplification for 12 cycles ( denaturation for 5 s at 98°C , annealing for 15 s at 62°C and extension for 15 s at 72°C ) , and a final extension for 5 min at 72°C . Amplified cDNAs were isolated by electrophoresis on 2% agarose gel ( PippinPrep Reagent Kit , external Marker B ) and products of ~180 to ~ 550 nt were collected . cDNA was recovered by ethanol precipitation and resuspended in 13 μl of nuclease-free water . Barcoded libraries were pooled and sequenced on an Illumina NextSeq platform in high output mode using custom primer s1115 ( 5′-CTACACGTTCAGAGTTCTACAGTCCGACGATC-3′ ) . Quality control on TSS-seq library FASTQ files was performed to remove reads with low quality using fastq_quality_filter in the FASTX-Toolkit ( http://hannonlab . cshl . edu/fastx_toolkit/ ) package with parameters ‘fastq_quality_filter -v -q 20 p 75’ . Cutadapt ( Martin , 2011 ) was then used to remove the 6 nt 5’ linker with parameter of ‘cutadapt -u 6’ . The resulting reads were trimmed from 3′ end to 35 nt long with parameter of ‘cutadapt -l 35 --minimum-length = 35’ . Trimmed reads were mapped to the S . cerevisiae R64-1-1 ( SacCer3 ) genome using Bowtie ( Langmead et al . , 2009 ) with allowance of no more than two mismatches with suppression of non-uniquely mapped reads ‘bowtie -p3 -v2 -m1 -q --sam --un’ , reported in sam files . Uniquely mapped reads were then extracted from sam files using SAMtools ( Li et al . , 2009 ) and output in bam format ‘samtools view -F 4 S -b’ . Bam files were then sorted and converted into bed files by SAMtools ‘samtools sort -o’ , and BEDTools ( Schwalb et al . , 2011 ) ‘bedtools bamtobed -cigar’ . Customized commands were then used on bed files to identify the genomic coordinate of the 5′ end of each uniquely mapped read ‘awk 'BEGIN{FS = OFS = "\t"} $6=="+" {$3=$2 + 1} $6=="-" {$2=$3–1} {print}'’ . BEDTools was then used to determine pileup ( TSS coverage ) across the genome with parameters of ‘bedtools genomecov -g R64 . new . genome -i -bg -strand -’ and ‘bedtools genomecov -g R64 . new . genome -i -bg -strand +’ , resulting in stranded bedGraph files . FASTQ files of individual library were directly processed or contacted by strains/mutants to generate bedGraph files for correlation analysis . For each of 5979 selected yeast promoters , TSS usage was examined within 401 nt wide window , spanning 250 nt upstream and 150 nt downstream of the previously annotated median TSS . Using customized bash and R scripts , TSS coverage from the bedGraph files of a library or a mutant were assigned into the defined windows to generate a 401 × 5979 TSS count table , with each row representing one of the 5979 promoters in the same order as the promoter annotation file , each column represents a promoter position , and the number in each cell representing 5′ ends mapping to that position . These count tables were stored in csv files . Using customized R script and the count table of a library or a mutant , an expression-spread-median file containing promoter expression , median TSS position of the promoter , TSS spread of the promoter was generated . The median TSS position was defined as the actual TSS containing the 50th percentile of the promoter window . The spread of TSS , which measures the width of the middle 80% of TSS distribution , was calculated by subtracting positions of 10th percentile and 90th percentile of TSS counts in 401 nt promoter window and adding 1 . The positions of 10th percentile and 90th percentile of TSS counts in each promoter window were also stored in this expression-spread-median file . The streamlined codes to generate bedGraph files , the prompter annotation file , and the customized scripts to generate count tables and expression-spread-median files can be found at the GitHub repository https://github . com/Kaplan-Lab-Pitt/Ssl2_scanning . TSS coverage data in library-based bedGraph files were used to examine the correlation between TSS libraries by pairwise comparison . A custom R script was used to filter bedGraph files to examine genome positions with greater than two counts in each library . Log2 transformed TSS counts at the same genomic location in two examined libraries were plotted for all TSS sites to create a heat scatter plot using the LSD R package ( Schwalb et al . , 2011 ) and the Pearson correlation coefficient was calculated . The correlation coefficients deriving from all pairwise comparisons were plotted by a web-based heatmap tool Morpheus ( https://software . broadinstitute . org/morpheus/ ) and clustered by Euclidean distance . Replicates with correlation coefficient greater than 0 . 85 and the shortest Euclidean distance to each other in the clustering analysis among all the analyzed libraries were recognized as having good sequencing reproducibility and used for downstream analysis . For each of 5979 selected promoters , TSS usage was examined within a 401 nt wide window , spanning 250 nt upstream and 150 nt downstream of the previously annotated median TSS ( Qiu et al . , 2020 ) . Using BEDTools and customized R scripts , TSS coverage from the bedGraph files of a library or mutant were assigned into the defined windows to generate a 401 × 5979 TSS count table , with each row representing one of the 5979 promoters , each column represents a promoter position , and the number in each cell representing 5’ ends mapping to that position . The count table was filtered to keep data from n = 4392 promoters with ≥100 sequence reads on average per WT library and used for downstream analyses . The filtered count table was row-normalized to get the relative TSS usage at each promoter position . TSS distribution differences were determined by subtracting normalized WT data ( concatenated from both RPB1 and SSL2 WT libraries ) from normalized mutant data and visualized using heatmaps ( Morpheus ) . Data analyses for TSS distributions were based on customized R scripts and results were plotted in GraphPad Prism 8 ( https://www . graphpad . com/scientific-software/prism/ ) unless otherwise indicated . The median TSS position was defined as the actual TSS containing the 50th percentile of the TSS distribution and was determined for each promoter . ‘TSS shift’ represents the difference in nucleotide of the median TSS position for each promoter between two libraries or mutants . The distributions of TSS shifts for n = 4392 promoters with ≥100 sequence reads on average per WT library or selected promoter classes in each library or mutant were illustrated by both heatmap ( Morpheus ) and boxplots . The spread of TSS , which measures the width of the middle 80% of TSS distribution , was calculated by subtracting positions of 10th percentile and 90th percentile of TSS reads in 401 nt promoter window and adding 1 . TSS spread of selected promoters are shown in boxplot and compared between libraries by performing one-way ANOVA . The differences of TSS spreads between WT and the mutant for selected promoter classes were presented in heatmap ( Morpheus ) and/or boxplot . ChIP-exo data processing was performed as described by Rossi et al . in Nature Communications , 2018 , and Qiu et al . in Genome Biology , 2020 . Briefly , ChIP-exo libraries were sequenced on a NextSeq 500 in paired-end mode to generate 40 ( read1 ) × 36 bp ( read2 ) reads . Reads passing Q30 quality threshold were then aligned to the sacCer3 genome using the BWA-MEM alignment algorithm ( v0 . 7 . 9a ) with default parameters ( Li , 2013 ) . After alignment , PCR duplicates were removed using Picard and SAMtools assuming unique combinations of read1 and read2 were PCR duplicates . Using ScriptManager v0 . 12 ( https://github . com/CEGRcode/scriptmanager , RRID:SCR_021797 ) , BAM files of a library were assigned into two 401 × 5979 matrices and saved in CDT files , which stores counts of 5’ position of protein binding on top and bottom strands , respectively . The same as in TSS-seq data analysis , each row of 401 × 5979 matrix representing one of 5979 promoters and each column representing a position in the 401 nt promoter window . These 401 × 5979 matrices were also saved in csv format . Matrices from the same mutant were combined into a single matrix by adding counts in library matrices at the same dimension and saved in csv files . Similar to TSS-seq analysis , the customized R script and the matrix of a library or a mutant were used to generate an expression-spread-median file containing promoter expression , median 5’ position of protein binding , the binding site of the spread of the promoter , and saved in txt files . Genomics datasets generated in the current study are available in the NCBI BioProject and SRA , under the accession numbers of PRJNA681384 and SRP295731 , respectively . The processed genomic data files are available in GEO , under the accession number of GSE182792 . The streamlined commands to generate TSS-seq bedGraph files , count tables , tables of expression , spread , and median TSS can be found at https://github . com/Kaplan-Lab-Pitt/Ssl2_scanning , ( copy archived at swh:1:rev:fdcccee50e4b6b801048c163d1ac71585958aec6 , Zhao , 2021 ) . ChIP-exo and data analysis was performed as described by Rossi et al . , 2018 and Qiu et al . , 2020 . Source data files are listed in Supplementary file 3 .	In eukaryotic organisms such as yeast , the process of converting genes into proteins begins with the transcription of DNA sequences into mRNA molecules . An enzyme called RNA Polymerase II ( Pol II ) is responsible for creating new strands of mRNA , but a variety of other so called transcription factors is also needed to kickstart the transcription process . These transcription factors are delivered to genes , where they attach to specific sequences , or promoters , which sit at the beginning of each gene . Once these transcription factors are in place , the double stranded DNA is unzipped to provide access to the DNA that will serve as the template for transcription . In budding yeast , Pol II and another specific transcription factor , known as TFIIH , work together to scan these promoter sequences to find the appropriate start sites of mRNA synthesis . However , several aspects of this process , such as how TFIIH works in promoter scanning , how far its scanning functions can extend , and how its activity is controlled , are currently poorly understood . Zhao et al . have investigated these questions in budding yeast . Using a range of genetic and genomic techniques , Zhao et al . found that certain sections of TFIIH were involved in choosing specific transcription start sites of mRNA synthesis during promoter scanning . These sections were identical in different eukaryotic organisms from yeast to humans , suggesting that these regions may be important for tuning or controlling the activity of TFIIH . Moreover , in yeast , the activity of TFIIH determines how far the scanning unit was able to move along the promoter DNA . Finally , Zhao et al . found that the initiation by promoter scanning was regulated by two distinct networks . The first network controlled how well mRNA synthesis could be initiated at individual transcription start sites; and the second network – driven by TFIIH – controlled which promoter sequences could be scanned to initiate transcription . This research provides an in-depth look into the early steps of the process of converting DNA into mRNA . The biological machinery used to initiate and control this action is highly conserved between yeast and humans , suggesting that the mechanisms for controlling the activity of these factors could be similar , even if their initiation processes may differ .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "genetics", "and", "genomics" ]	2021	Ssl2/TFIIH function in transcription start site scanning by RNA polymerase II in Saccharomyces cerevisiae
Many motile microorganisms react to environmental light cues with a variety of motility responses guiding cells towards better conditions for survival and growth . The use of spatial light modulators could help to elucidate the mechanisms of photo-movements while , at the same time , providing an efficient strategy to achieve spatial and temporal control of cell concentration . Here we demonstrate that millions of bacteria , genetically modified to swim smoothly with a light controllable speed , can be arranged into complex and reconfigurable density patterns using a digital light projector . We show that a homogeneous sea of freely swimming bacteria can be made to morph between complex shapes . We model non-local effects arising from memory in light response and show how these can be mitigated by a feedback control strategy resulting in the detailed reproduction of grayscale density images . A local reduction of speed in a crowd of walking pedestrians or urban car traffic usually results in the formation of high-density regions . Mathematically speaking , a self-propelled particle , moving by an isotropic random walk with space dependent speed , explores the available space with a probability density that is inversely proportional to the local value of the speed ( Schnitzer , 1993; Tailleur and Cates , 2009; Cates and Tailleur , 2015 ) , spending a longer time in slow regions than in fast ones . As a consequence , any mechanism that allows to substantially modulate the local propulsion speed can be used as an efficient strategy for controlling the density ( Stenhammar et al . , 2016 ) . In swimming bacteria , proteorhodopsin ( PR ) provides a particularly convenient way of achieving spatial speed control through the projection of light patterns . PR is a light-driven proton pump that uses photon energy to pump protons out of the cytoplasm , thus modulating the corresponding electrochemical gradient across the inner membrane ( Béjà et al . , 2000 ) . Since this proton motive force drives the rotation of the flagellar motor ( Gabel and Berg , 2003 ) , PR puts a ‘solar panel’ on every cell allowing to remotely control its swimming speed with light ( Walter et al . , 2007 ) . This has been recently exploited to control the rotational speeds of bio-hybrid micromachines using bacteria as micropropellers ( Vizsnyiczai et al . , 2017 ) . In light driven Janus colloids , spatially asymmetric speed profiles can also affect particle orientation giving rise to an artificial phototaxis mechanism ( Lozano et al . , 2016 ) . Arlt et al . ( 2018 ) have recently shown that , by projecting a masked illumination pattern , ‘photokinetic’ ( Wilde and Mullineaux , 2017 ) bacteria can be accumulated in dark regions and depleted from brighter ones thus forming binary patterns . Here we demonstrate that bacterial density can be controlled to form reconfigurable complex patterns . To do this we exploit a Digital Light Processing ( DLP ) projector to shape light with a megapixel resolution and with a dynamic range of 256 ( 8 bit ) light intensity levels . We experimentally investigate the limits of validity of the predicted law of inverse proportionality between local density and speed . We demonstrate that observed deviations from this law can be quantitatively described by a theoretical model that explicitly takes into account memory effects in light response . Finally we show that a model independent feedback control loop allows density shaping with high spatial resolution and gray level accuracy . Density shaping in photokinetic bacteria or colloids relies on the connection between density and local speed . The density ρ⁢ ( 𝐫 ) ∝1/v⁢ ( 𝐫 ) is always a steady-state solution of the master equation of a self-propelled particle with isotropic reorientation dynamics and spatially varying speed v⁢ ( 𝐫 ) ( Cates and Tailleur , 2015 ) . In our case , v⁢ ( 𝐫 ) =v⁢ ( I⁢ ( 𝐫 ) ) where 𝐫= ( x , y ) is the position in the two dimensional image space and I⁢ ( 𝐫 ) the light intensity at 𝐫 . This result assumes that all bacteria have the same response to light and that they instantaneously adapt to intensity variations so that the speed only depends on the local value of I . However , suspensions of swimming bacteria , are known to be characterized by motility properties that occur in broad distributions and , in principle , a broad spectrum of light responses could be found . To quantify the effect of light on the speed distributions of our strain , we monitored bacterial dynamics while projecting a chessboard pattern consisting of 12 different levels of light intensity . To do this we couple a DLP projector to a custom video-microscopy setup working in both bright-field and dark-field mode ( see Figure 1 and Materials and methods ) . Using differential dynamic microscopy ( DDM ) ( Cerbino and Trappe , 2008; Wilson et al . , 2011; Maggi et al . , 2013 ) we extract the light dependent values of the mean v and standard deviation σ of the speed distributions ( Figure 2 ( a ) ) ( see Materials and methods ) . Although the absolute value of the swimming speed depends on several factors ( strain , growth medium , etc . ) we find a non linear speed response that is consistent with what previously reported for PR expressing bacteria . In particular the speed versus intensity curve is very well fitted by a hyperbola showing saturation for large light intensity values as already reported in ( Walter et al . , 2007; Schwarz-Linek et al . , 2016; Vizsnyiczai et al . , 2017; Arlt et al . , 2018 ) . We also find that , throughout the intensity range , v and σ are directly proportional ( see Figure 2 ( b ) ) suggesting a homogeneous growth law for individual cell speeds vi⁢ ( I ) =vis⁢f⁢ ( I ) , where vis is the saturation speed for the i-th cell and f⁢ ( I ) is a dimensionless function saturating at one for large intensities . Interestingly , in this scenario , the mean speed v⁢ ( I ) will also be proportional to f⁢ ( I ) so that the probability density ρi for the generic i-th cell will be proportional to the inverse of the local mean speed:ρi⁢ ( 𝐫 ) ∝1vi⁢ ( I⁢ ( 𝐫 ) ) ∝1f⁢ ( I⁢ ( 𝐫 ) ) ∝1v⁢ ( I⁢ ( 𝐫 ) ) . This implies that , despite broad speed distributions , the local density is expected to be only determined by the mean local speed , as set by the light intensity pattern and measured by DDM . To validate this hypothesis we use a DLP projector to display a complex light pattern onto a 400 µm thick layer of cells ( Figure 1 ) that have been preliminarily exposed to a uniform bright illumination for 5 min . This time is much longer than the speed response time of bacteria ( Figure 4 ) thus ensuring that cells are initialized to swim at maximal speed . The projecting system has an optical resolution of 2 µm approximately matching the size of a single cell which represents the physical ‘pixel’ of our density images . This value sets the limit for the minimum theoretical resolution of density configurations that would be achievable if bacteria could be precisely and statically arranged in space . In practice , as we will see , the real resolution will always be larger for two main reasons: ( i ) bacteria do not respond instantly to light temporal variations thus introducing a blur in the target speed map , ( ii ) the stationary state is an ensemble of noisy patterns that constantly fluctuate because of swimming and Brownian motions of bacteria . After projecting an inverted Mona Lisa image , bacteria start to concentrate in dark regions while moving out from the more illuminated areas . After about 4 min , dark field microscopy reveals a recognizable bacterial ‘replica’ of Leonardo’s painting , where brighter areas correspond to regions of accumulated cells . Once a stationary pattern is reached , we collect a time averaged image from which we extract the bacterial density ρ* shown in Figure 3 normalized to be one when uniform ( see Materials and methods ) . We split the density image into components corresponding to the same illumination level I . We then average the density within each component and report in Figure 3 ( c ) the obtained value as a function of the inverse mean speed 1/v⁢ ( I ) obtained from the speed vs intensity curve in Figure 2 ( a ) . The high speed side of the graph can be very well fitted by a straight line with a finite intercept q=0 . 5 . This suggests that 50% of the total scattering objects in the field of view responds to light and can be spatially modulated while the remaining 50% can be attributed to cells which are non-motile or insensitive to light and also to stray light generated away from the focal plane . The ratio between the maximum and minimum modulation of the light sensitive component can be obtained as max⁢[ρ-q]/min⁢[ρ-q]=2 . A strong deviation from linearity is evident in the high density/low speed region . A violation of the ρ∝1/v law can be attributed to many factors that are not included in the simple theory discussed above . In particular , the theory assumes that speed is a local function of space which would only be the case if bacteria instantly adapt to temporal changes in light intensity . However , previous studies have evidenced that speed response is not instantaneous but it displays a relaxation pattern characterized by multiple timescales . In addition to a fast relaxation time associated to the membrane discharge ( Walter et al . , 2007 ) , ATP synthase and the dynamics of stator units of the flagellar motors introduce slower timescales in the speed relaxation dynamics ( Tipping et al . , 2013; Arlt et al . , 2018 ) . Using DDM we measured speed response to a uniform light pattern whose intensity switches between two values every 100 s . Results are reported in Figure 4 and clearly show the presence of two timescales . The short time scale is smaller than our time resolution ( 1 s ) and appears as an instantaneous jump that accounts for about a fraction β=0 . 44 of the total relaxation . A slower relaxation follows and it is well fitted by an exponential function with a time constant τm=35 s that is the same for both the rising and falling relaxations . As a result , bacteria will experience an effective speed map that is a blurred version of what we would expect for an instantaneous response V⁢ ( 𝐫 ) =v⁢ ( I⁢ ( 𝐫 ) ) . In the case of smooth swimming cells , with a two-step light response , we calculate that , for weak speed modulations around a baseline value V0 , the actual speed map can be obtained as a simple convolution ( see Materials and methods ) : ( 1 ) w⁢ ( x , y ) ≃β⁢V⁢ ( x , y ) + ( 1-β ) ⁢∫V⁢ ( x-x′ , y-y′ ) ⁢γ⁢ ( x′ , y′ ) ⁢𝑑x′⁢𝑑y′with γ⁢ ( x , y ) a convolution kernel that is analytic in Fourier space: ( 2 ) γ~⁢ ( q ) =kq⁢arctan⁡ ( qk ) where q=qx2+qy2 is the modulus of the wave vector and k−1=v0τm . Assuming that the stationary distribution will remain isotropic even in the presence of memory , the relation ρ∝1/w will then still be valid provided one uses the actual , blurred speed map w and not the original map V corresponding to instantaneous response . We can then anticipate the stationary density once we calculate the actual speed map w with the convolution formula ( 1 ) . Recalling that only a fraction α of cells is motile the estimated normalized stationary density will be ( 3 ) ρ∗ ( x , y ) =α ( w ( x , y ) −1w−1¯−1 ) +1where w−1¯ is the spatial average of the inverse actual speed w⁢ ( x , y ) . Plotting ρ* as a function of 1/V we find that this simple model quantitatively explains the density behavior in Figure 3 with the natural choice of free parameters α=0 . 5 , β=0 . 44 , and k-1=v¯⁢τm=59 µm where v¯ is the spatial average of V⁢ ( x , y ) . This result is certainly encouraging and prompts for a deeper and more systematic investigation using simpler light patterns . Our aim here , however , is to investigate the resolution limits of density shaping with photokinetic bacteria . Although memory effects could be reduced by deleting ATP-synthase genes ( Arlt et al . , 2018 ) , resulting in a suppression of the slow component in speed relaxation , response will never be truly instantaneous . In addition to memory effects , steric and/or hydrodynamic interactions will always affect density by hindering bacteria penetration and further accumulation in densely packed regions . The combined action of these effects on the stationary density map cannot be easily incorporated in a theoretical model that provides precise quantitative predictions with a priori known parameters . An alternative and more practical strategy to improve the accuracy of density shaping , is to implement a feedback control loop . We set up an automated control loop that performs one iteration every 20 s , comparing the current density map to the desired target image and updating the illumination pattern accordingly ( see Materials and methods ) . Light levels are increased in those regions with a density that exceeds the target value and decreased where density is lower . At each step the light intensity increment is simply proportional to the pixel by pixel difference of target and actual density images . We quantify the distance between the obtained density pattern and the target image by the sum of the squared pixel-by-pixel difference after a proper rescaling procedure ( see Materials and methods ) . As soon as the feedback is turned on at time t0=6 min ( Figure 5 ( a ) ) the distance from the target starts to decrease reaching a new stationary value after about 4 min . The final stationary density map shows a level of detail significantly higher than the image obtained before the feedback was turned on ( Figure 5 ( b , c ) ) . The image is stable and resolution does not deteriorate for hours . However a dark region surrounds the illuminated area so that swimming cells that exit the field of view do not come back and the number of motile cells in the illuminated area is reduced by about 50% in one hour . This issue is the main limiting factor for the lifetime of our patterns and could be solved in the future by using sample chambers that match the size of the illuminated area . As evidenced in Figure 5 ( d ) , the feedback precompensates this blurring effect due to memory by converging to a sharpened version of the initial target image . Results of comparable quality have been replicated four times in each of the two independent biological replicates . A remarkable feature of this optically controlled material lays in its intrinsic dynamic and reconfigurable nature . As a demonstration of this property we show the dynamic morphing of a bacterial layer from an Albert Einstein’s to a Charles Darwin’s portrait ( see Figure 6 and Video 1 ) . The morphing is triggered by instantly switching between the two target images on the DLP projector and exponentially converges to the second target in a characteristic time of 1 . 8 min . Although light induced density modulations can be achieved in large samples of Brownian ( passive ) colloidal systems , active materials provide a largely superior performance in terms of both required power and response time . The stationary density distribution of Brownian colloids is governed by the Boltzmann law stating that the logarithmic ratio of maximum to minimum density is given bylog⁡ ( ρmaxρmin ) =-Δ⁢UkB⁢T , where Δ⁢U is the optical energy difference between brightest and darkest regions . For a plastic or glass bead of radius a=1µm in water , this energy difference can be estimated in the Rayleigh regime as ( Ashkin et al . , 1986 ) Δ⁢U≈a3⁢I/c where I is the maximum power density and c is the speed of light . For a contrast level comparable to the ones shown above , log⁡ ( ρmax/ρmin ) ≈1 , the required power density would be I≈kB⁢T⁢c/a3≈1 W/mm2 . This value is three orders of magnitude larger than the maximum power densities used in our experiments . Our active system is also much faster than its Brownian counterprt as evidenced by comparing the timescales of pattern formation . In both the active and passive case , dynamics is governed by the interplay of diffusion and drift with characteristic timescales τdiff and τdrift . Calling ℓ=1 mm the largest length scale in the target pattern , we obtain for the active case τdrift≈ℓ/v≈200 s where we used v=5 µm/s as the typical swimming speed . A diffusion time scale can be obtained as τdiff≈ℓ2/2⁢D where D is the active translational diffusion coefficient . For wild type bacteria , moving with a run and tumble dynamics , D=v2⁢τrun where τrun≈1 s is the mean duration of a run . The corresponding characteristic diffusion time is τdiff≈ℓ2/2⁢D≈2×104 s . This timescale can be considerably reduced if , as in our case , we suppress tumbling and use a smooth swimming strain with a much larger reorientation time τrot≈20 s ( Berg , 1993 ) . In this case the diffusion timescale drops down to τdiff≈103 s leading to a faster relaxation towards the stationary state . In the Brownian case the drift timescale is given by τdrift≈μ⁢kB⁢T/ℓ≈106 s where μ=50 µm/s pN is the mobility of a 1 µm radius microsphere . The same colloidal particle has a diffusivity D=μ⁢kB⁢T=0 . 2 µm2/s giving again τdiff≈ℓ2/2⁢D≈ 106 s . Summarizing , using optical forces to achieve a comparable density modulation for Brownian particles would require a thousand times larger powers and from 103 to 104 longer times . In conclusion , we have shown that a suspension of swimming bacteria , with optically controllable speed , can provide a new class of light controllable active materials whose density can be accurately , reversibly and quickly shaped by employing a low power light projector . An alternative strategy to produce static patterns of bacteria is by ‘biofilm lithography’ where an optical template of blue light is used to induce the expression of membrane proteins that promote cell-cell and cell-substrate attachment ( Jin and Riedel-Kruse , 2018 ) . As opposed to our motility driven density shaping , these patterns are static and form over a time scale of several hours . Interestingly , the two techniques could be used in combination to achieve faster and more complex ( e . g . three dimensional ) biofilm lithography by fixing with blue light a template pattern obtained by speed modulation with green light . By further genetic engineering , bacteria could be eventually encapsulated in silicate shells ( Müller et al . , 2008 ) producing solid permanent structures for micro-mechanics or micro-optics applications . Finally , the possibility of spatial and temporal control of motile bacteria density could also lead to novel strategies for the transport and manipulation of small cargoes inside microdevices Maggi et al . ( 2018 ) . For all the experiments we used the E . coli strain HCB437 ( Wolfe et al . , 1987 ) transformed with a plasmid encoding the PR under the control of the araC-pBAD promoter ( Biobricks , BBa_K1604010 inserted in pSB1C3 plasmid backbone ) . E . coli colonies from frozen stocks are grown overnight at 33∘C on LB agar plates supplemented with kanamycin ( Kan 30 µg/mL ) and chloramphenicol ( Cam 20 µg/mL ) . A single colony is picked and statically cultivated overnight at 33∘C in 10 mL of M9 broth ( M9 salts with 0 . 2% glucose , 0 . 2% casaminoacids ) supplemented with antibiotics as before . The overnight culture is diluted 100-fold into 5 mL of the previous medium , grown at 33∘C , 200 rpm . 5 mM arabinose and 20 µM retinal are added once OD590 ≈ 0 . 2 , keeping the culture in the dark to avoid retinal degradation . Once OD590 ≈ 0 . 8 , cells are collected by centrifugation ( 1500 rcf , 5’ ) . The resulting pellet is washed twice by centrifugation ( 1500 rcf , 5’ ) with motility buffer ( MB: 0 . 1 mM EDTA , 10 mM phosphate buffer and 0 . 2% Tween20 ) . The bacterial suspension is eventually adjusted to a working OD≈590 2 . 0 . 100 µL of the prepared bacterial suspension is injected into a glass capillary ( CM Scientific - Rect . boro capillaries 0 . 40 × 4 mm ) , sealed at both sides with vacuum grease ( Sigma-Aldrich ) . The capillary is then placed under the optical microscope and observed in bright and dark field illumination focusing on the plane in the middle of the capillary . Bright field and dark field imaging were performed using a custom inverted optical microscope equipped with a 4 × magnification objective ( Nikon; NA = 0 . 13 ) and a high-sensitivity CMOS camera ( Hamamatsu Orca-Flash 2 . 8 ) ( see Figure 1 ) . Light shaping was performed using a digital light processing ( DLP ) projector ( Texas Instruments DLP Lightcrafter 4500 ) coupled to the same microscope objective used for imaging . The size of a ( squared ) DLP projector pixel imaged on the sample plane results to be 2 µm . All dark-field images have been divided by the flat field image obtained by applying a Gaussian filter on the respective homogeneous bacterial suspension ( under uniform green illumination ) as detailed in the following . Flat back-ground subtraction ( fifth percentile of the image histogram ) was only applied to the frames appearing in Figure 6 . To perform DDM we compute the quantity: ( 4 ) g⁢ ( 𝐪 , t′ , t ) =⟨\|M⁢ ( 𝐪 , t′ ) -M⁢ ( 𝐪 , t′+t ) \|2⟩where M⁢ ( 𝐪 , t ) is the spatial Fourier transform at the wave-vector 𝐪 of the bright-field image captured at time t . Assuming time-translational invariance and isotropy in the bacterial movement g⁢ ( 𝐪 , t′ , t ) depends only on the time-lag t in Equation ( 4 ) and on the modulus of the wave vector q=\|𝐪\| , that is g⁢ ( 𝐪 , t , t′ ) =g⁢ ( q , t ) . The g⁢ ( q , t ) is connected to the intermediate scattering function ( ISF ) F⁢ ( q , t ) : ( 5 ) g⁢ ( q , t ) =A⁢ ( q ) ⁢F⁢ ( q , t ) +B⁢ ( q ) where A⁢ ( q ) and B⁢ ( q ) are time-independent factors related , respectively , to the number and shape of bacteria , and to the background noise in the images . Following Wilson et al . , 2011 we use the ISF model for independent smooth swimming cells: ( 6 ) F⁢ ( q , t ) = ( 1-α ) ⁢e-q2⁢D⁢t+α⁢e-q2⁢D⁢t⁢∫0∞𝑑v′⁢P⁢ ( v′ ) ⁢sinc⁢ ( q⁢v′⁢t ) where α is the fraction of motile cells , D the Brownian diffusion coefficient and P⁢ ( v′ ) the Schultz distribution: ( 7 ) P⁢ ( v′ ) =1v′⁢ ( Z+1v⁢v′ ) Z+1⁢exp⁡ ( -Z+1v⁢v′ ) Γ⁢ ( Z+1 ) Here v is the mean speed , Γ is the Euler gamma function and Z is related to the speed standard deviation σ by the formula Z= ( v/σ ) 2−1 . The relevant parameters discussed in the main text v and σ are extracted by fitting the experimental g ( q , t ) in the q-range 0 . 45 μm-1<q<1 . 2 μm-1with Equations 5 , 6 and 7 . We assume that dark field images are proportional to the bacterial density modulated by a slowly varying envelope due to inhomogeneities in illumination . This flat field correction ρ0⁢ ( 𝐫 ) is obtained by acquiring 100 frames at 25 fps at the beginning of the experiments when the bacterial density is homogeneous . These frames are averaged and then filtered with a Gaussian kernel with a large standard deviation ( ≈100µm ) . The normalized experimental density ρ∗⁢ ( 𝐫 ) is obtained by first computing a raw density field ρ⁢ ( 𝐫 ) ( 50 frames average at 25 fps ) and then dividing by ρ0 , that is ρ∗⁢ ( 𝐫 ) =ρ⁢ ( 𝐫 ) /ρ0⁢ ( 𝐫 ) . For computing the distance between ρ∗⁢ ( 𝐫 ) and the target density ρtar⁢ ( 𝐫 ) we first rescale ρ∗⁢ ( 𝐫 ) so that the 10-th and 90-th percentile of the two image histograms coincide:ρs⁢ ( 𝐫 ) =ρ90%tar-ρ10%tarρ90%∗-ρ10%∗⁢ ( ρ∗⁢ ( 𝐫 ) -ρ10%∗ ) +ρ10%tar . The distance 𝖽𝗂𝗌𝗍 is then computed as:𝖽𝗂𝗌𝗍=[∑𝐫[ρs⁢ ( 𝐫 ) -ρtar⁢ ( 𝐫 ) ]2]12where the sum runs over the image pixels 𝐫 . A smooth swimming cell with a two-step response to light intensity and traveling in the x direction over a light imposed speed profile V⁢ ( x ) , will have an actual speed at time t given by: ( 8 ) v⁢ ( t ) =β⁢V⁢ ( x⁢ ( t ) ) + ( 1-β ) ⁢∫0∞V⁢ ( x⁢ ( t-t′ ) ) ⁢e-t′/τm⁢𝑑t′where x⁢ ( t ) is the position of the cell at time t . For a weakly varying speed profile V⁢ ( x ) =V0+δ⁢V⁢ ( x ) we can transform the above time convolution into a space convolution by assuming x⁢ ( t ) ≈x⁢ ( 0 ) +V0⁢t: ( 9 ) v⁢ ( x ) ≃β⁢V⁢ ( x ) + ( 1-β ) ⁢k2⁢∫-∞∞V⁢ ( x-s ) ⁢e-k⁢\|s\|⁢𝑑swhere we have also assumed that an equal number of bacteria will be traveling in the opposite direction . Assuming isotropy and using the fact that , in the weak modulation limit , density is homogeneous at the zero order , we can easily generalize to the three dimensional case and write: ( 10 ) v⁢ ( 𝐫 ) ≃β⁢V⁢ ( 𝐫 ) + ( 1-β ) ⁢k4⁢π⁢∫V⁢ ( 𝐫-s⁢u^ ) ⁢e-k⁢\|s\|⁢𝑑s⁢𝑑Ω=β⁢V⁢ ( 𝐫 ) + ( 1-β ) ⁢∫V⁢ ( 𝐫-𝐫′ ) ⁢Γ⁢ ( 𝐫′ ) ⁢d3⁢r′where Γ⁢ ( 𝐫 ) = ( k/4⁢π⁢r2 ) ⁢e-k⁢r is a 3D convolution kernel . Our projecting light system is such that light patterns do not vary significantly across the entire sample depth so that we can assume V⁢ ( x , y , z ) =V⁢ ( x , y ) and express the effective speed distribution as a 2D convolution ( 11 ) v⁢ ( x , y ) ≃β⁢V⁢ ( x , y ) + ( 1-β ) ⁢∫V⁢ ( x-x′ , y-y′ ) ⁢γ⁢ ( x′ , y′ ) ⁢𝑑x′⁢𝑑y′with γ⁢ ( x , y ) =∫-∞∞Γ⁢ ( x , y , z ) ⁢𝑑z . Although the convolution kernel γ is not analytic in real space , its Fourier transform is analytic making convolutions easy to calculate numerically: ( 12 ) γ~⁢ ( q ) =kq⁢arctan⁡ ( qk ) where q=qx2+qy2 . The ( n+1 ) -th illumination pattern at the pixel 𝐫 is updated as follows:In+1⁢ ( 𝐫 ) =In⁢ ( 𝐫 ) +P⁢Δ⁢ρ⁢ ( 𝐫 ) , where P>0 is a proportional control parameter and Δ⁢ρ is the difference between the scaled density and the target density:Δ⁢ρ⁢ ( 𝐫 ) =ρs⁢ ( 𝐫 ) -ρtar⁢ ( 𝐫 ) If , for example , the scaled density is larger than the target at 𝐫 the projected power density at that pixel will be increased . That will increase the speed of bacteria and consequently reduce their local density .	Many bacteria can move in response to environmental signals . This helps guide them towards better conditions for growth and survival . Escherichia coli is a bacterium that can swim quickly through liquid , using tiny propeller-like structures that rotate many times per second . These ‘propellers’ are powered by a cellular motor , called the flagellar motor , which similar to an electric motor , requires an energy source to drive movement . Proteorhodopsin , a protein originally isolated from free-swimming micro-organisms in the ocean , is an alternative energy source that helps bacteria move . The protein is located close to the surface of the cell , where it acts like a solar panel and captures energy from light . In cells powered by proteorhodopsin , the intensity of light from their environment determines their swimming speed: brighter light means faster movement , and less light , slower movement . Proteorhodopsin is now also a useful tool in the laboratory . For example , genetically engineering bacteria to produce proteorhodopsin provides a way to control their movement remotely , using a light source . Swimming bacteria , much like cars in city traffic , are known to accumulate in areas where their speed decreases . By controlling swimming speed with proteorhodopsin , researchers can manipulate the local density of bacteria simply by projecting different patterns of light . To study the factors influencing this phenomenon , Frangipane et al . used genetically modified E . coli that could respond to light via proteorhodopsin to make layers of cells that could then have light patterns projected onto them . The results showed that the bacteria responded slowly to these stimuli , which was the main factor limiting the resolution of the final pattern they formed . A simple feedback mechanism , which compared the pattern formed by the cells to the desired image and updated the projected light accordingly , was enough to solve this problem . This way , the layers of E . coli could be turned into a near-perfect copy of the original image . This work allows us to control the movement of large populations of bacteria more precisely than ever before . This could be extremely valuable for building the next generation of microscopic devices . For example , bacteria could be made to surround a larger object such as a machine part or a drug carrier , and then used as living propellers to transport it where it is needed .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "physics", "of", "living", "systems" ]	2018	Dynamic density shaping of photokinetic E. coli
The absence of ‘shovel-ready’ anti-coronavirus drugs during vaccine development has exceedingly worsened the SARS-CoV-2 pandemic . Furthermore , new vaccine-resistant variants and coronavirus outbreaks may occur in the near future , and we must be ready to face this possibility . However , efficient antiviral drugs are still lacking to this day , due to our poor understanding of the mode of incorporation and mechanism of action of nucleotides analogs that target the coronavirus polymerase to impair its essential activity . Here , we characterize the impact of remdesivir ( RDV , the only FDA-approved anti-coronavirus drug ) and other nucleotide analogs ( NAs ) on RNA synthesis by the coronavirus polymerase using a high-throughput , single-molecule , magnetic-tweezers platform . We reveal that the location of the modification in the ribose or in the base dictates the catalytic pathway ( s ) used for its incorporation . We show that RDV incorporation does not terminate viral RNA synthesis , but leads the polymerase into backtrack as far as 30 nt , which may appear as termination in traditional ensemble assays . SARS-CoV-2 is able to evade the endogenously synthesized product of the viperin antiviral protein , ddhCTP , though the polymerase incorporates this NA well . This experimental paradigm is essential to the discovery and development of therapeutics targeting viral polymerases . SARS-CoV-2 has infected hundreds of million humans worldwide , causing millions of deaths , with numbers still on the rise . We are currently living through the third coronavirus outbreak in less than 20 years , and we are desperately in need of broad-spectrum antiviral drugs that are capable of targeting this emerging family of human pathogens . To this end , nucleotide analogs ( NAs ) represent a powerful approach , as they target the functionally and structurally conserved coronavirus polymerase , and their insertion in the viral RNA induces either premature termination or a lethal increase in mutations . The coronavirus polymerase is composed of the nsp12 RNA-dependent RNA polymerase ( RdRp ) , and the nsp7 and nsp8 co-factors , with a stoichiometry of 1:1:2 ( Kirchdoerfer and Ward , 2019; Hillen et al . , 2020; Gao et al . , 2020; Wang et al . , 2020 ) . This polymerase is thought to associate with several additional viral proteins , including the nsp13 , a 5′-to-3′ RNA helicase ( Chen et al . , 2020; Yan et al . , 2020 ) , and the nsp14 , a 3′-to-5′ exoribonuclease ( Agostini et al . , 2018; Bouvet et al . , 2012; Ferron et al . , 2018; Ogando et al . , 2019; Subissi et al . , 2014; Eckerle et al . , 2007 ) . The latter proofreads the terminus of the nascent RNA following synthesis by the polymerase and associated factors ( Robson et al . , 2020 ) , a unique feature of coronaviruses relative to all other families of RNA viruses . Proofreading likely contributes to the stability of the unusually large , ~30 kb , coronavirus genome . In addition , proofreading may elevate the tolerance of coronaviruses to certain NAs ( e . g . , ribavirin; Smith et al . , 2013 ) and therefore should be considered in the development of potent NAs . In other words , nsp14 adds another selection pressure on NAs: not only they must be efficiently incorporated by nsp12 , they must also evade detection and excision by nsp14 . Remdesivir ( RDV ) is a recently discovered NA that showed efficacy against Ebola infection ( Siegel et al . , 2017 ) and has been successfully repurposed for the treatment of SARS-CoV-2 infection ( Agostini et al . , 2018; Gordon et al . , 2020a; Gordon et al . , 2020b; Pruijssers and Denison , 2019; Jockusch et al . , 2020; Chien et al . , 2020 ) . The success of RDV relies on its efficient incorporation by the polymerase ( Gordon et al . , 2020b; Dangerfield et al . , 2020 ) and probable evasion of the proofreading machinery ( Agostini et al . , 2018 ) . Understanding how RDV achieves these two tasks , will help to guide the rational design of more efficacious NAs for the current and future outbreaks . To this end , it is essential to build a comprehensive model describing the selection and incorporation mechanisms that control the utilization of NAs by the coronavirus polymerase and to define the determinants of the base and ribose responsible for selectivity and potency . We have therefore compared several analogs of the same natural nucleotide to determine how the nature of the modifications changes selection/mechanism of action . Magnetic tweezers permit the dynamics of an elongating polymerase/polymerase complex to be monitored in real time and the impact of NAs to be monitored in the presence of all four natural nucleotides in their physiological concentration ranges . Here , we present a magnetic tweezers assay to provide insights into the mechanism and efficacy of current and underexplored NAs on the coronavirus polymerase . To enable the observation of rare events , such as nucleotide mismatch and NA incorporation , even in the presence of saturating NTP concentration , we have developed a single-molecule , high-throughput , magnetic tweezers assay to monitor SARS-CoV-2 RNA synthesis activity at near single base resolution ( Dulin et al . , 2017 ) . A SARS-CoV-2 polymerase formed of nsp12 , nsp7 , and nsp8 ( Figure 1—figure supplement 1A ) assembles and initiates RNA synthesis at the 3′-end of the magnetic bead-attached handle , and converts the 1043 nt long single-stranded ( ss ) RNA template into a double-stranded ( ds ) RNA in the presence of NTPs and at constant force , that is , 35 pN if not mentioned otherwise ( Figure 1—figure supplement 1B–D; see Materials and methods ) . The conversion from ssRNA to dsRNA displaces the magnetic bead along the vertical axis and directly informs on the number of incorporated nucleotides ( Figure 1A , see Materials and methods; Dulin et al . , 2015a ) . During each experiment , hundreds of magnetic beads are followed in parallel ( Figure 1—figure supplement 1E ) , yielding dozens of traces of SARS-CoV-2 polymerase activity per experiment ( Figure 1B , Figure 1—figure supplement 2A ) . As previously observed for other viral RdRps ( Dulin et al . , 2017; Dulin et al . , 2015a; Seifert et al . , 2020 ) , the traces reveal substantial , heterogeneous dynamics , with bursts of activity interrupted by pauses of duration varying from ~0 . 5 s to ~60 s in Figure 1B . This dynamic is intrinsic to the polymerase elongation kinetics and does not result from viral proteins exchange ( Bera et al . , 2021 ) . To extract the elongation kinetics of the SARS-CoV-2 polymerase , we scanned the traces with non-overlapping 10-nt windows to measure the duration of time required to complete the 10 successive nucleotide-incorporation events . Each duration of time has been coined a dwell time , which is the kinetic signature of the rate-limiting event of the 10 nt addition , that is , the 10 nucleotide addition cycles themselves , or a pause ( Dulin et al . , 2017; Dulin et al . , 2015a; Seifert et al . , 2020; see Materials and methods ) . We fitted the distribution of dwell times using the stochastic-pausing model that describes well the kinetics of nucleotide addition of the coronavirus polymerase ( Bera et al . , 2021 ) and other viral RdRps ( Dulin et al . , 2017; Dulin et al . , 2015a; Seifert et al . , 2020; Dulin et al . , 2015b; Figure 1C; see Materials and methods ) . This model is composed of four distributions: a pause-free nucleotide addition rate , Pause 1 , Pause 2 , and the backtrack pauses ( Figure 1C; see Materials and methods ) , and fit parameters values extracted from triplicate experiments fall within statistical errors ( Figure 1—figure supplement 2B , C ) . Statistics and all parameter values extracted from the analysis are reported in Supplementary file 1 and Supplementary file 2 . SARS-CoV-2 polymerase elongation kinetics is well described by a robust model where the nucleotide addition rate is the kinetic signature of the nucleotide addition burst ( NAB ) pathway , from which the RdRp stochastically and rarely switches into the slow nucleotide addition ( SNA ) pathway , and even more rarely into the very slow nucleotide addition ( VSNA ) pathway , the latter being consistent in rate and probability with mismatch incorporation ( Figure 1D; Bera et al . , 2021 ) Pause 1 and Pause 2 are respectively the kinetic signatures of the SNA and VSNA pathways ( Figure 1D ) , while the long-lived pauses relate to a catalytically incompetent polymerase backtrack state , that is , the polymerase diffuses backward on the template strand leading the product strand 3′-end to unwind and exit via the NTP channel without cleavage ( see Materials and methods Bera et al . , 2021; Malone et al . , 2021 ) . Increasing the temperature from 25°C to 37°C , SARS-CoV-2 polymerase reveals a strong temperature dependence , which translates into a twofold decrease in the median replication time ( Figure 1E ) , while not affecting the RNA synthesis product length ( Figure 1F ) . Analyzing the dwell time distribution at 25°C and 37°C ( Figure 1G ) , we extracted a ~2 . 6-fold enhancement in nucleotide addition rate , from ( 65 . 6±0 . 5 ) nt . s−1 to ( 169 . 0±3 . 8 ) nt . s−1 , making the SARS-CoV-2 polymerase the fastest RNA polymerase characterized to date ( Figure 1H; Dangerfield et al . , 2020; Shannon et al . , 2020a ) . Pause 1 and Pause 2 exit rates also increased by ~3-fold ( Figure 1H ) , whereas their respective probabilities increased by twofold and fivefold ( Figure 1—figure supplement 2D ) . The latter results are rather surprising , as poliovirus and human rhinovirus C RdRps showed only an exit rate increase with no change in probability ( Seifert et al . , 2020 ) . Next , we investigated how the elongation kinetics and the product length of SARS-CoV-2 polymerase were affected by two adenosine analogs , 3′-dATP and RDV-TP ( Figure 2—figure supplement 1 ) . 3′-dATP is an obligatory terminator of RNA chain elongation for viral RdRp ( Gohara et al . , 2004 ) . RDV-TP has also been suggested to cause chain termination but only several cycles of nucleotide addition after its incorporation . If this is the true mechanism of action , then the experimental outcome of the presence of any of these two analogs should be indistinguishable in our assay . In the presence of 500 µM NTP and 500 µM 3′-dATP , the ability of the SARS-CoV-2 polymerase to reach the end of the template ( 1043 nt ) was compromised ( Figure 2A vs . Figure 1B ) . Indeed , increasing 3′-dATP concentration up to 2000 µM , only reduced the mean product length of SARS-CoV-2 polymerase by ~1 . 7-fold , from ( 940±13 ) nt to ( 566±33 ) nt ( mean±standard deviation ) ( Figure 2B ) , while not affecting the kinetics of RNA synthesis ( Figure 2C , Figure 2—figure supplement 2A–C ) . We derived a model to determine the effective incorporation rate γ , that is , the average number of nucleotide addition cycles before terminator incorporation at equimolar concentration of competing natural nucleotide ( in the presence of all NTPs ) ( see Materials and methods , Figure 2—figure supplement 3 ) . This model fits very well to the mean product length as a function of 3′-dATP:ATP stoichiometry ( Figure 2B ) , for example , γ3′−dATP , 500μMATP= ( 780±64 ) nt , meaning that the polymerase incorporates on average 780 nt before incorporating one 3′-dATP and terminating RNA synthesis ( see Materials and methods ) . A subsaturating concentration of NTP increases the probability to enter both the SNA and the VSNA pathways ( Figure 1D; Bera et al . , 2021 ) , that is , Pause 1 and Pause 2 probability , and would increase the effective incorporation rate of 3′-dATP , providing it is incorporated via any of the two SNA states . By decreasing ATP concentration from 500 µM to 50 µM , we indeed observed an increase in Pause 1 and Pause 2 probabilities by more than twofold and threefold , from 0 . 060±0 . 002 to 0 . 149±0 . 005 and from 0 . 0033±0 . 0009 to 0 . 0115±0 . 0026 , respectively ( Figure 2—figure supplement 2A–F ) . Adding 500 µM of 3′-dATP significantly shortened the traces in comparison to the 500 µM ATP condition ( Figure 2A , D ) . However , the effective incorporation rate of 3′-dATP was identical at both concentrations of ATP , that is , γ3′−dATP , 50μMATP= ( 777±50 ) nt ( Figure 2E ) , which indicates that 3′-dATP incorporation is only driven by stoichiometry , despite the significant increase in the SNA ( Pause 1 ) and VSNA ( Pause 2 ) pathways probabilities . Therefore , we conclude that 3′-dATP utilizes the NAB pathway for incorporation ( Figure 1D ) . Of note , the decrease in the median replication time is due to the shortening of the product length from early termination ( Figure 2F ) . Replicating the experiment at a 3′-dATP:ATP stoichiometry of 6 but now at 25 pN showed no significant differences in final product length in comparison to the data acquired at 35 pN ( Figure 2E , Figure 2—figure supplement 4A ) . RDV-TP is an adenine analog with a 1′-cyano modification that has recently been shown to outcompete ATP for incorporation ( Gordon et al . , 2020b; Dangerfield et al . , 2020; Figure 2—figure supplement 1 ) , while exhibiting a low cytotoxicity ( Pruijssers et al . , 2020 ) . RDV-TP has been proposed to induce delayed chain termination at i+3 ( i being RDV incorporation position ) during RNA synthesis by the core polymerase ( Gordon et al . , 2020a; Gordon et al . , 2020b ) . Adding 100 µM RDV-TP in a reaction buffer containing 500 µM NTPs showed a dramatic increase in the pause density and duration , but most of the traces reached the end of the template ( Figure 3A ) . We indeed observed a final product length largely unaffected at all concentrations of RDV-TP ( Figure 3B ) , while the median time for RNA synthesis increased by more than tenfold ( Figure 3C ) , for RDV-TP concentrations increasing up to 300 µM . Therefore , the RDV-TP mechanism of action is not termination . We then investigated the origin of the pause induced by RDV-TP incorporation using our stochastic-pausing model ( Figure 3D , Figure 3—figure supplement 1B ) . While the nucleotide addition rate is unaffected by RDV-TP , all pauses are significantly impacted . The exit rates of Pause 1 and Pause 2 decreased by fourfold and tenfold ( Figure 3E ) , while their probabilities increased by twofold and fourfold , respectively ( Figure 3F ) . Most notably , the backtrack pause probability increased by 28-fold , from 0 . 0005±0 . 0001 to 0 . 0142±0 . 0015 , when increasing RDV-TP concentration up to 300 µM . The backtrack pause probability increase was such that it most likely affected the probability and the exit rates of Pause 1 and Pause 2 above 50 µM RDV-TP ( Figure 3F ) . As expected , the almost identical SARS-CoV-1 polymerase ( Kirchdoerfer and Ward , 2019 ) demonstrated a similar kinetic signature to RDV-TP incorporation ( Figure 3—figure supplement 2A–F , Supplementary file 1 ) , though to a lesser extent , for example , the backtrack probability increased by ~9-fold when raising RDV-TP concentration up to 300 µM versus 28-fold for SARS-CoV-2 polymerase . To verify whether the applied tension modifies the incorporation kinetics of RDV-TP by the SARS-CoV-2 polymerase , we replicated the experiment using 500 µM NTP and 100 µM RDV-TP at 25 pN , that is , a 10 pN lower force ( Figure 2—figure supplement 4B ) . We did not observe any significant difference between the two experiments at 35 pN and 25 pN ( Figure 2—figure supplement 4C , D ) , indicating that tension does not play a significant role in RDV-TP incorporation . Using our recently developed ultra-stable magnetic tweezers ( Bera et al . , 2021 ) , we wanted to directly monitor polymerase backtrack induced by RDV-TP incorporation . To this end , we performed an experiment with 10 µM RDV-TP and 50 µM ATP , keeping all other NTPs at 500 µM ( Figure 3GH , Figure 3—figure supplement 1A ) . We hypothesized that lowering the concentration of ATP , the natural competitor of RDV , would increase the incorporation yield of RDV and therefore polymerase backtrack probability . A close observation of the longest-lived pauses clearly demonstrates polymerase backtrack , as deep as ~30 nt ( Figure 3GH , Figure 3—figure supplement 1A ) , demonstrating that RDV incorporation induces polymerase backtrack , which leads to long-lived pauses . To verify whether the incorporation of RDV-TP is stoichiometric , we further analyzed the experiment performed at 50 µM ATP , 500 µM all other NTPs , and 10 µM RDV-TP , at 25°C and 35 pN ( Figure 3—figure supplement 3A ) ( coined low ATP and RDV-TP concentrations ) , that is , the same stoichiometry as 500 µM all NTPs and 100 µM RDV-TP ( coined high ATP and RDV-TP concentrations ) . In absence of RDV-TP , the decrease in ATP concentration from 500 µM to 50 µM increased dramatically Pause 1 and Pause 2 probability by 2 . 5-fold and 3 . 5-fold , respectively , while the backtrack pause remained unchanged . The large increase in both Pause 1 and Pause 2 probabilities further disentangled the distribution of these pauses from the backtrack pause , and we therefore did not expect a strong crossover of the latter on the former ( as observed at 500 µM all NTPs ) . We noticed an average product length of ( 633±30 ) nt at low ATP and RDV-TP concentrations , that is , a ~30% shorter than for any other conditions presented in Figure 3—figure supplement 3B , even though we acquired data for a much longer duration than at high ATP and RDV-TP concentrations , that is , 11 , 000 s vs . 1 , 600 s , respectively . Interestingly , this result resembles what was observed at a 3′-dATP:ATP stoichiometry of ~3 ( Figure 2B ) , indicating that RDV-TP induces what resembles termination at low ATP concentration . We also observed a ~2 . 3-fold longer median replication time than at high ATP and RDV-TP concentrations ( Figure 3—figure supplement 3C ) , an increase largely underestimated as a large fraction of the traces never reached the end of the template during the measurement ( Figure 3—figure supplement 3B ) . Applying the stochastic-pausing model to the dwell time distribution of the low ATP and RDV-TP concentrations data ( Figure 3—figure supplement 3D ) , we found the nucleotide addition rate unchanged , while Pause 1 and Pause 2 exit rates were lower than in absence of RDV-TP , that is , by 1 . 4-fold and 2 . 3-fold , respectively ( Figure 3—figure supplement 3E ) . At low ATP concentration , the probabilities of Pause 1 and Pause 2 were largely unaffected by the presence of RDV-TP , similarly to what was observed at 37°C ( Figure 3—figure supplement 3F ) . Most remarkably , the backtrack pause probability increased dramatically at low ATP and RDV-TP concentrations , even more so than at high ATP and RDV-TP concentrations , that is , 43-fold versus 18-fold , respectively ( Figure 3—figure supplement 3F ) . The main effect of RDV-TP is to increase the backtrack pause probability . In our previous study of the impact of T-1106-TP on poliovirus RdRp , we showed that T-1106 incorporation induces long-lived backtrack pauses that appear as termination in ensemble assays ( Dulin et al . , 2017 ) . Interestingly , lowering ATP concentration increases the potency of RDV-TP by dramatically increasing the backtrack pause probability . However , we know such a pause is catalytically incompetent ( Bera et al . , 2021 ) , and therefore the increase of the backtrack probability is an illustration of the effect of RDV-TP incorporation at low nucleotide concentration: the increased energy barrier induced by the steric clash of RDV-TP with the nsp12 serine-861 reduces dramatically the likelihood of a successful forward translocation of the polymerase ( Gordon et al . , 2020b; Kokic et al . , 2021 ) . This likelihood is even further reduced at low NTP concentration , which dramatically increases the probability of polymerase backtrack ( Dulin et al . , 2015c ) . We previously observed that increasing the temperature helped to further disentangle the distributions of the different pauses ( Seifert et al . , 2020 ) . We therefore performed an experiment at 37°C in the presence of 100 µM RDV-TP and 500 µM all NTPs ( Figure 3—figure supplement 4A ) . The nucleotide addition rate significantly increased with temperature , while this increase was not affected by the presence of RDV-TP ( Figure 1H , Figure 3—figure supplement 4B ) . On the one hand , Pause 1 and Pause 2 exit rates significantly decreased by threefold and ninefold , respectively , when the reaction was performed with RDV-TP ( Figure 3—figure supplement 4B ) . On the other hand , Pause 1 and Pause 2 probabilities were unaffected by the presence of RDV-TP ( Figure 3—figure supplement 4C ) , supporting the notion that the increase in probability in the experiments performed at 25°C was the consequence of the polymerase backtrack pause distribution biasing Pause 1 and Pause 2 distributions ( Dulin et al . , 2017 ) . The backtrack pause probability still increased by more than sevenfold , that is , from 0 . 0003±0 . 0001 to 0 . 0022±0 . 0007 . The lesser increase in the backtrack pause probability at 37°C ( 28-fold at 25°C ) is consistent with a model where RDV-MP represents a barrier to translocation , which crossing would be facilitated by increasing the thermal energy . If RDV-TP incorporation resulted in a pause of similar exit rates as Pause 1 and Pause 2 , but not mechanistically related to them , we would expect an increase in the probabilities of both pauses . However , in conditions where Pause 1 and Pause 2 distribution were clearly distinguishable from the backtrack pause distribution when having RDV-TP in the reaction buffer , that is , at 37°C and at low ATP concentration , we did not observe an increase in Pause 1 and Pause 2 probabilities . Therefore , we suggest that RDV-TP is incorporated by the SNA and VSNA pathways ( Bera et al . , 2021 ) , leading to polymerase backtrack when failing at overcoming the increased energy barrier resulting from the clash of RDV-MP with serine-861 . Pyrazine-carboxamides represent a promising family of antiviral NAs , of which the best-known member is Favipiravir ( T-705 ) , recently approved to treat influenza virus infection ( Furuta et al . , 2009 ) , and considered against SARS-CoV-2 . We studied here another member of this family , T-1106 triphosphate ( T-1106-TP ) , which is chemically more stable than T-705 , while presenting similar antiviral properties ( Dulin et al . , 2017; Shannon et al . , 2020b ) . T-1106-TP competes for incorporation against ATP and GTP in a sequence-dependent manner ( Dulin et al . , 2017; Shannon et al . , 2020b ) . Adding 500 µM of T-1106-TP in a reaction buffer containing 500 µM NTPs significantly increased the number and duration of pauses observed in SARS-CoV-2 RNA synthesis activity traces ( Figure 4A ) , leading to a 2 . 6-fold increase in median replication time ( Figure 4B ) . For comparison , 50 µM of RDV-TP induced a median replication time as 500 µM T-1106-TP , at the same concentration of competing NTP , suggesting that RDV-TP is better incorporated than T-1106-TP . The final product length was not affected by T-1106-TP , consistent with the T-1106-TP mechanism of action not being terminated ( Figure 4C; Dulin et al . , 2017 ) . Performing an experiment using the ultra-stable magnetic tweezers assay ( Bera et al . , 2021 ) with 500 µM T-1106 and 500 µM all NTPs at 35 pN force , the SARS-CoV-2 polymerase activity traces showed pauses with either a shallow backtrack ( Figure 4—figure supplement 1A ) , that is , ≤10 nt , or no significant backtrack at all ( Figure 4—figure supplement 1B ) . Investigating how increasing T-1106-TP concentration affects SARS-CoV-2 RNA synthesis kinetics ( Figure 4D , Figure 4—figure supplement 1C ) , we found that only the Pause 2 exit rate was affected , decreasing by tenfold ( Figure 4E ) . Pause 1 and Pause 2 probabilities remained constant , while the backtrack pauses increased by almost fivefold , though remaining in the low probability range , that is , ~0 . 002 ( Figure 4F ) . Repeating the experiment at 500 µM NTPs and 300 µM T-1106-TP at 25 pN , we found no difference in comparison to the data acquired at 35 pN ( Figure 2—figure supplement 4E–G ) . Here again , the tension has no significant effect . Our results suggest an incorporation of T-1106-TP only via the VSNA pathway ( Figure 1D ) , which explains its reduced promiscuity relative to RDV-TP , and is less likely than RDV-TP to induce polymerase backtrack upon incorporation . These observations contrast with our previous findings with poliovirus RdRp ( Dulin et al . , 2017 ) , where T-1106 incorporation induced deep polymerase backtrack . Therefore , the same NA may have a different mechanism of action on different RdRps . Next , we compared two uridine analog chain terminators , that is , Sofosbuvir and 3′-dUTP . Sofosbuvir presents a fluoro group at the 2′ α-position and a methyl group at the 2′ β-position , and is a non-obligatory chain terminator . Despite its low incorporation rate ( Villalba et al . , 2020 ) , Sofosbuvir has a proven antiviral effect against hepatitis C virus ( HCV ) and is an FDA-approved drug to treat HCV infection ( Kayali and Schmidt , 2014; Sofia et al . , 2010 ) . It is incorporated by SARS-CoV-2 polymerase ( Gordon et al . , 2020b; Chien et al . , 2020 ) , but has no efficacy in infected cells ( Xie et al . , 2020a ) . 3′-dUTP lacks a hydroxyl group in 3′ position , and is therefore an obligatory chain terminator . The presence of 500 µM Sofosbuvir-TP with 500 µM NTP did not affect RNA synthesis by the SARS-CoV-2 polymerase ( Figure 5A ) , while early termination events appeared in the presence of 500 µM 3′-dUTP ( Figure 5B ) . Supporting this visual observation , the mean RNA product length of the SARS-CoV-2 polymerase was unaffected by the presence of Sofosbuvir-TP ( Figure 5C ) . Raising the 3′-dUTP:UTP stoichiometry to 4 reduced the mean product length by almost fivefold , resulting in an effective incorporation rate γ3′−dUTP , 500μMUTP= ( 151±6 ) nt ( Figure 5D ) . For both NAs , the replication time was unaffected ( Figure 5—figure supplement 1A and Figure 5—figure supplement 2A ) , as well as SARS-CoV-2 RNA synthesis kinetics ( Figure 5—figure supplement 1B–D and Figure 5—figure supplement 2B–D ) . Reducing the concentration of UTP down to 50 µM while keeping the other NTPs at 500 µM , Sofosbuvir-TP caused few early termination events when increased to 500 µM ( Figure 5E ) . Replacing Sofosbuvir-TP by 3′-dUTP , we observed a much stronger effect , as no activity traces reached the end of the template at 3′-dUTP:UTP stoichiometry of 10 ( Figure 5F ) . The analysis showed a limited impact of Sofosbuvir-TP on the mean product length , with a minimum of ( 563±32 ) nt at a stoichiometry of 20 ( Figure 5G ) . 3′-dUTP was much more effectively incorporated , shortening the mean product length down to ( 67±3 ) nt at the same stoichiometry ( Figure 5H ) . Their respective effective incorporation rate at 50 µM UTP reflected these observations , that is , γsofosbuvir , 50μMUTP= ( 3908±467 ) nt and γ3′−dUTP , 50μMUTP= ( 241±9 ) nt . In other words , SARS-CoV-2 polymerase incorporates on average 3908 nt and 241 nt before incorporating either a single Sofosbuvir-TP or a single 3′-dUTP , respectively . The kinetics of RNA synthesis was unaffected by the presence of either 3′-dUTP or Sofosbuvir-TP , while their median replication time decreased at high stoichiometry , a direct consequence of the shortening of the RNA synthesis product ( Figure 5—figure supplement 1E–H and Figure 5—figure supplement 2E–H , respectively ) . Repeating the experiments for a Sofosbuvir-TP:UTP stoichiometry of 6 now at 25 pN tension ( Figure 2—figure supplement 4A ) , we did not see a significant difference in comparison with the data at 35 pN ( Figure 5—figure supplement 1E–H ) , therefore the applied tension has no influence in the incorporation of Sofosbuvir-TP . As for 3′-dATP , our data suggest that stoichiometry against the competing NTP regulates Sofosbuvir-TP and 3′-dUTP incorporation , which therefore support that these analogs utilize the NAB state pathway for incorporation ( Figure 1D ) . Our data provide further support to the poor incorporation of Sofosbuvir by SARS-CoV-2 ( Gordon et al . , 2020b; Xie et al . , 2020a ) and the low selectivity of the SARS-CoV-2 polymerase against 3′-dUTP . 3ʹ-Deoxy-3′ , 4ʹ-didehydro-CTP ( ddhCTP ) is a recently discovered natural antiviral NA produced in mammalian cells by the viperin-catalyzed conversion of CTP to ddhCTP using a radical-based mechanism ( Gizzi et al . , 2018 ) . While ddhCTP has been shown to efficiently terminate flavivirus replication both in vitro and in cells , its antiviral activity against SARS-CoV-2 polymerase remains unknown . The addition of 500 µM ddhCTP to a reaction buffer containing 500 µM NTP induces early termination events in the SARS-CoV-2 polymerase activity traces ( Figure 6A ) . Similar amount of 3ʹ-dCTP instead of ddhCTP resulted in a larger fraction of traces showing early termination events ( Figure 6B ) . The average RNA product length of the SARS-CoV-2 polymerase decreased by 1 . 4-fold when raising the ddhCTP:CTP stoichiometry to 4 ( Figure 6C ) , while it decreased by 2 . 7-fold at similar stoichiometry against CTP ( Figure 6D ) . We measured a respective effective incorporation rate γddhCTP , 500μMCTP= ( 1221±130 ) nt and γ3′−dCTP , 500μMCTP= ( 338±18 ) nt ( Figure 6C , D ) . For both NAs , the replication time ( Figure 6—figure supplement 1A and Figure 6—figure supplement 2A ) and the RNA synthesis kinetics ( Figure 6—figure supplement 1B–D and Figure 6—figure supplement 2B-D ) were largely unaffected . Reducing the concentration of CTP down to 50 µM and keeping the other NTPs at 500 µM , both ddhCTP and 3ʹ-dCTP showed a significant reduction in length of the activity traces ( Figure 6E , F ) . Analyzing the average product length , we extracted the respective effective incorporation rates at 50 µM CTP , that is , γddhCTP , 50μMCTP= ( 1360±71 ) nt and γ3′−dCTP , 50μMCTP= ( 457±21 ) nt ( Figure 6G , H ) . These values are similar as what was measured at 500 µM CTP , and confirm the better incorporation of 3ʹ-dCTP over ddhCTP . The kinetics of RNA synthesis was unaffected by the presence of ddhCTP or 3ʹ-dCTP , while their median replication time decreased at high stoichiometry , as a result of the shortening of the RNA synthesis product ( Figure 6—figure supplement 1F–H and Figure 6—figure supplement 2F-H , respectively ) . We also did not observe any impact of the applied tension for ddhCTP incorporation ( Figure 2—figure supplement 4A ) . Here again , stoichiometry against their competing natural nucleotide CTP directly dictates the incorporation of 3ʹ-dCTP and ddhCTP , further supporting the utilization of the NAB pathway for their incorporation ( Figure 1D ) . Though not as high as 3′-dCTP , the effective incorporation rate of ddhCTP should be sufficient to demonstrate a certain efficacy against viral replication in cells . Indeed , ddhCTP is a chain terminator , therefore a single incorporation is sufficient to end RNA synthesis . To verify whether ddhCTP inhibits replication in cells , we infected Huh7-hACE2 cells with SARS-CoV-2 , treated these cells with different concentrations of RDV , Sofosbuvir and ddhC , and report on the level of infection by immunofluorescence against SARS-CoV-2 N protein ( see Materials and methods , Figure 6—figure supplement 3A , B ) . While RDV showed a clear antiviral effect with an EC50 of 0 . 007 µM ( Figure 6—figure supplement 3B ) , ddhC and Sofosbuvir did not show any impact on SARS-CoV-2 replication in cells . This result suggests that SARS-CoV-2 is able to evade the antiviral properties of the endogenously synthesized antiviral NA ddhC . We hypothesized that the 3′–5′ exonuclease activity of nsp14 protects SARS-CoV-2 replication by excising ddhCMP from the nascent RNA . To test this hypothesis , we made a SARS-CoV-2 strain , which includes the amino acid substitutions D90A and E92A that remove the exoribonuclease activity of nsp14 ( Figure 6—figure supplement 4A ) . This SARS-CoV-2 mutant was unable to replicate in cells ( Figure 6—figure supplement 4B–F ) , confirming a recent report ( Ogando et al . , 2020 ) . Therefore , the role of nsp14 in the removal of ddhCMP from the SARS-CoV-2 genome could not be verified experimentally . Future experiments will be designed to address this question . We present here the first characterization of the mechanism of action of antiviral NAs against SARS-CoV-2 polymerase at the single-molecule level . We show that SARS-CoV-2 polymerase is the fastest RNA studied polymerase to date , elongating up to ∼170nt . s−1 at 37 °C ( Figure 1H ) . With our assay , we monitored the incorporation and determined the mechanism of action of several NAs , that is , 3′-dATP , 3′-dUTP , 3′-dCTP , Sofosbuvir-TP , ddhCTP , T-1106-TP , and RDV-TP , and resume their properties in Table 1 . The present study demonstrates that NA selection and incorporation are not force-dependent ( Figure 2—figure supplement 4 ) , which further validates the utilization of high-throughput magnetic tweezers to study NA mechanism of action . This result is in agreement with our recent study on SARS-CoV-2 polymerase mechanochemistry , where we showed that entry probability in NAB , SNA , and VSNA was not force-dependent , and that force mainly affected the kinetics of a large conformational subsequent to chemistry , that is , after nucleotide selection and incorporation . Our study shows that RDV-TP is not a delayed chain terminator at physiological concentration of all NTPs , but instead induces pauses in the polymerase elongation kinetics that are easily overcome at saturating NTP concentration ( Figure 3 ) . Since our preprint was published on BioRxiv in August 2020 , our finding has been corroborated by two recent studies ( Kokic et al . , 2021; Bravo et al . , 2021 ) . Similarly , T-1106-TP incorporation does not induce termination , but pauses in the polymerase elongation kinetics ( Figure 4 ) . However , RDV-TP affects both Pause 1 and Pause 2 exit rates , while T-1106-TP affects only the latter . We showed here that these two NAs do not affect the probability to enter Pause 1 and Pause 2 , suggesting that they preferably bind to the polymerase active site after it entered the SNA ( Pause 1 ) or VSNA ( Pause 2 ) pathway . Indeed , if the pauses induced by either RDV-TP or T-1106-TP incorporation were mechanistically unrelated to Pause 1 and Pause 2 , the total number of pauses would cumulate and the probability of pausing would dramatically increase , which we do not observe . We therefore suggest that RDV-TP can be incorporated by both SNA and VSNA pathways , while T-1106-TP is only incorporated by the latter . Finally , Pause 1 and Pause 2 respectively account for ~6% and ~0 . 3% of all the nucleotide addition events at a saturating concentration of NTP . This defines an upper limit for RDV-TP and T-1106-TP relative incorporation , and explains why RDV-TP is incorporated much better than Favipiravir ( Xie et al . , 2020a ) . Two recent ensemble kinetic studies investigating the mechanism of action of RDV-TP on SARS-CoV-2 elongation kinetics have recently been published . In the first one , the experiments were performed at submicromolar concentration of NTPs , and showed that RDV-TP is incorporated threefold better than ATP in such conditions ( Gordon et al . , 2020b ) . In the second one , the authors also claimed that RDV-TP was better incorporated than ATP ( Dangerfield et al . , 2020 ) , while using higher concentration of NTPs than in the first study . Both of these studies agree with our results: RDV is better incorporated by the coronavirus polymerase elongation kinetics at low concentration of natural nucleotides . Indeed , in such conditions , the probabilities of the pathways by which RDV-TP is incorporated , that is , SNA and VSNA , increase significantly ( Bera et al . , 2021 ) . In addition , we showed that RDV-TP incorporation remains noticeable at concentration as low as 20 µM , even when competing with 500 µM ATP . Being able to monitor RDV-TP incorporation at the single-molecule level in competition with saturating concentration of NTP—including ATP— , while the SARS-CoV-2 polymerase was elongating a ~1 kb long RNA product further completes the understanding of RDV mechanism of action . Our assay revealed that RDV-TP incorporation leads the coronavirus polymerase into backtrack as deep as ~30 nt ( Figure 3GH ) . This result demonstrates that the barrier induced by the clash of RDV-MP ( Kokic et al . , 2021 ) with the serine-861 of nsp12 is sufficiently strong to elicit polymerase backtrack , leading the polymerase into a pause long enough to be mistaken for a termination event in ensemble assays . We anticipate that RDV efficacy is further amplified when the polymerase is elongating through template secondary structures , which stimulates polymerase backtrack ( Bera et al . , 2021 ) . Lower ATP concentration would also decrease the probability to overcome the barrier when an uracil is encoded ~3 nt downstream the incorporated RDV-MP , increasing the backtrack pause probability , as observed here . Interestingly , RDV has a strong efficacy against SARS-CoV-2 in infected cells ( Figure 6—figure supplement 4A , B ) , which indicates that the 3′–5′ exonuclease nsp14 does not remove efficiently RDV-MP from the nucleic acid chain . Our results suggest that polymerase backtrack is therefore not an intermediate of product strand proofreading , which corroborates a preceding study showing that nsp14 poorly excise single-stranded RNA ( Ferron et al . , 2018; Liu et al . , 2021 ) . Concerning obligatory terminators , the effective incorporation rate we measured showed that 3′-dATP ( Figure 2 ) , 3′-dUTP ( Figure 5 ) , 3′-dCTP ( Figure 6 ) , and—to a lesser extent—ddhCTP ( Figure 6 ) are well incorporated by the SARS-CoV-2 polymerase , while Sofosbuvir-TP is strongly outcompeted by UTP ( Figure 5 ) . Though well incorporated , 3′-dNTP is cytotoxic , and is therefore not used as antiviral drugs ( Arnold et al . , 2012 ) . Interestingly , the effective incorporation rate of all these terminators is only affected by the stoichiometry of their respective competing natural nucleotide , and not their absolute concentration ( unlike RDV-TP ) , suggesting an incorporation via the NAB pathway ( Figure 1D ) . Indeed , we showed that NA incorporated via either the SNA or the VSNA pathway , for example , RDV-TP , would be more likely to be added in the RNA chain at low substrate concentration , independently of the stoichiometry . A steady-state kinetic study showed that NAs modified at the 2′ and 3′ positions are strongly discriminated against by their competing natural nucleotide ( Gordon et al . , 2020b ) . Such selectivity is an issue for purine-based analogs , which must compete with high concentrations of ATP and GTP in the cell . In contrast , pyrimidine-based analogs , for example , derivatives of CTP , will only need to compete with intracellular CTP pools on the order of 100 µM ( Traut , 1994 ) . These features make ddhCTP a particularly attractive antiviral NA . Furthermore , under certain conditions , the interferon α-induced viperin converts up to 30% of the cellular pool of CTP into ddhCTP , further increasing the ddhCTP:CTP stoichiometry in a direction favoring even greater potency ( Gizzi et al . , 2018 ) . However , we could not show any efficacy of ddhC in SARS-CoV-2 infected cells ( Figure 6—figure supplement 4 ) , suggesting that SARS-CoV-2 has developed ways to counter this cellular defense mechanism . Future studies will investigate whether the exonuclease nsp14 is capable of removing ddhCMP and is therefore responsible for protecting the virus against endogenously produced antiviral NAs . High-throughput , real-time magnetic tweezers present numerous advantages to study RdRp elongation dynamics , such as monitoring polymerase position with high spatiotemporal resolution while elongating kilobases long templates in the presence of saturating concentration of competing natural nucleotides , and therefore provide complementary information to discontinuous assays to understand the selectivity and/or mechanism of action of NAs . Such an assay will also reveal how adding functional capacity to the core polymerase , for example , RNA helicase activity and proofreading , modulate RdRp elongation dynamics and response to antiviral therapeutics . This protocol was described in Chien et al . , 2020 . SARS-CoV-2 nsp12: The SARS-CoV-2 nsp12 gene was codon optimized and cloned into pFastBac with C-terminal additions of a TEV site and strep tag ( Genscript ) . The pFastBac plasmid and DH10Bac Escherichia coli ( Life Technologies ) were used to create recombinant bacmids . The bacmid was transfected into Sf9 cells ( Expression Systems ) with Cellfectin II ( Life Technologies ) to generate recombinant baculovirus . The baculovirus was amplified through two passages in Sf9 cells , and then used to infect 1 L of Sf21 cells ( Expression Systems ) and incubated for 48 hr at 27°C . Cells were harvested by centrifugation , resuspended in wash buffer ( 25 mM HEPES pH 7 . 4 , 300 mM NaCl , 1 mM MgCl2 , and 5 mM DTT ) with 143 μl of BioLock per liter of culture . Cells were lysed via microfluidization ( Microfluidics ) . Lysates were cleared by centrifugation and filtration . The protein was purified using Strep Tactin superflow agarose ( IBA ) . Strep Tactin eluted protein was further purified by size exclusion chromatography using a Superdex 200 Increase 10/300 column ( GE Life Sciences ) in 25 mM HEPES , 300 mM NaCl , 100 μM MgCl2 , 2 mM TCEP , at pH 7 . 4 . Pure protein was concentrated by ultrafiltration prior to flash freezing in liquid nitrogen . SARS-CoV-2 nsp7 and nsp8: The SARS-CoV-2 nsp7 and nsp8 genes were codon optimized and cloned into pET46 ( Novagen ) with an N-terminal 6× histidine tag , an enterokinase site , and a TEV protease site . Rosetta2 pLys E . coli cells ( Novagen ) were used for bacterial expression . After induction with isopropyl β-D-1-thiogalactopyranoside ( IPTG ) , cultures were grown at 16°C for 16 hr . Cells were harvested by centrifugation and pellets were resuspended in wash buffer ( 10 mM Tris pH 8 . 0 , 300 mM NaCl , 30 mM imidazole , and 2 mM DTT ) . Cells were lysed via microfluidization and lysates were cleared by centrifugation and filtration . Proteins were purified using Ni-NTA agarose beads and eluted with wash buffer containing 300 mM imidazole . Eluted nsp12 , nsp7 , and ns8 were digested with 1% w/w TEV protease during overnight room temperature dialysis ( 10 mM Tris pH 8 . 0 , 300 mM NaCl , and 2 mM DTT ) . Digested proteins were passed back over Ni-NTA to remove undigested protein before concentrating the proteins by ultrafiltration . Nsp7 and nsp8 proteins were further purified by size exclusion chromatography using a Superdex 200 Increase 10/300 column ( GE Life Sciences ) . Purified proteins were concentrated by ultrafiltration prior to flash freezing with liquid nitrogen . This protocol was described in Shannon et al . , 2020b . All SARS-CoV proteins used in this study were expressed in E . coli , under the control of T5 promoters . Cofactors nsp7L8 and nsp8 alone were expressed from pQE30 vectors with C-terminal and N-terminal hexa-histidine tags , respectively . TEV cleavage site sequences were included for His-tag removal following expression . The nsp7L8 fusion protein was generated by inserting a GSGSGS linker between nsp7- and nsp8-coding sequences . Cofactors were expressed in NEB Express C2523 ( New England Biolabs ) cells carrying the pRare2LacI ( Novagen ) plasmid in the presence of Ampicillin ( 100 µM/ml ) and Chloramphenicol ( 17 µg/ml ) . Protein expression was induced with 100 µM IPTG once the OD600=0 . 5–0 . 6 , and expressed overnight at 17°C . Protein was purified first through affinity chromatography with HisPur Cobalt resin ( Thermo Fisher Scientific ) , with a lysis buffer containing 50 mM Tris-HCl pH 8 , 300 mM NaCl , 10 mM Imidazole , supplemented with 20 mM MgSO4 , 0 . 25 mg/ml Lysozyme , 10 µg/ml DNase , 1 mM PMSF , with lysis buffer supplemented with 250 mM imidazole . Eluted protein was concentrated and dialyzed overnight in the presence of histidine labeled TEV protease ( 1:10 w/w ratio to TEV:protein ) for removal of the protein tag . Cleaved protein was purified through a second cobalt column and protein was purified through size exclusion chromatography ( GE , Superdex S200 ) in gel filtration buffer ( 25 mM HEPES pH 8 , 150 mM NaCl , 5 mM MgCl2 , and 5 mM TCEP ) . Concentrated aliquots of protein were flash-frozen in liquid nitrogen and stored at −80°C . A synthetic , codon-optimized SARS-CoV nsp12 gene ( DNA 2 . 0 ) bearing C-terminal 8His-tag preceded by a TEV protease cleavage site was expressed from a pJ404 vector ( DNA 2 . 0 ) in E . coli strain BL21/pG-Tf2 ( Takara ) . Cells were grown at 37°C in the presence of Ampicillin and Chloramphenicol until OD600 reached 2 . Cultures were induced with 250 µM IPTG and protein expressed at 17°C overnight . Purification was performed as above in lysis buffer supplemented with 1% CHAPS . Two additional wash steps were performed prior to elution , with buffer supplemented with 20 mM imidazole and 50 mM arginine for the first and second washes respectively . Polymerase was eluted using lysis buffer with 500 mM imidazole and concentrated protein was purified through gel filtration chromatography ( GE , Superdex S200 ) in the same buffer as for nsp7L8 . Collected fractions were concentrated and supplemented with 50% glycerol final concentration and stored at −20°C . All experiments involving live SARS-CoV-2 were carried out under biosafety level 3 ( BSL-3 ) containment by personnel wearing the appropriate PPE , including powered air-purifying respirators with Tyvek suits , aprons , booties , and double gloves . Huh7 cells were purchased from Glow Biologics ( GBTC-099H ) and tested negative for mycoplasma . These cells expressed human ACE2 ( huh7-hACE2 ) after transduction by lentiviral particles derived with pWPI-IRES-Puro-Ak-ACE2 ( a gift from Sonja Best; Addgene plasmid # 154985 ) . SARS-CoV-2 , isolate USA-WA1/2020 ( NR-52281 ) , was obtained through BEI Resources and propagated once on VERO E6 cells before it was used for this study . Huh7-hACE2 cells in 96-well plates ( Corning ) were infected with SARS-CoV-2 ( USA-WA1/2020 isolate ) at MOI of 0 . 05 in Dulbecco's modified Eagle's medium ( DMEM ) supplemented with 1% fetal bovine serum ( FBS ) . Before 1 . 5 hr viral inoculation , the tested compounds were added to the wells in triplicate . The infection proceeded for 24 hr without the removal of the viruses or the compounds . The cells were then fixed with 4% paraformaldehyde , permeabilized with 0 . 1% Triton-100 , blocked with DMEM containing 10% FBS , and stained with a rabbit monoclonal antibody against SARS-CoV-2 NP ( GeneTex , GTX635679 ) and an Alexa Fluor 488-conjugated goat anti-mouse secondary antibody ( Thermo Fisher Scientific ) . Hoechst 33342 was added in the final step to counterstain the nuclei . Fluorescence images of approximately 10 , 000 cells were acquired per well with a 10× objective in a Cytation 5 ( BioTek ) . The total number of cells , as indicated by the nuclei staining , and the fraction of the infected cells , as indicated by the NP staining , were quantified with the cellular analysis module of the Gen5 software ( BioTek ) . SARS-CoV-2 WT and nsp14 exoribonuclease knockout viruses were prepared using a SARS-CoV-2 infectious clone ( Xie et al . , 2020b ) . Briefly , viral RNA was obtained by in vitro RNA transcription , and 40 μg RNA transcripts and 20 μg N gene RNA were co-electroporated into 8×106 Vero E6 cells using Gene Pulser XCell electroporation system ( Bio-Rad , Hercules , CA ) at a setting of 270 V and 950 μF with a single pulse . The electroporated cells were seeded to a T75 flask and immediately transfer to BSL-3 facility . Viral production was confirmed by RT-PCR . The supernatants of electroporated cells were harvested and centrifuged at 1000×g for 10 min to remove cell debris . 250 μl supernatant was added and mixed thoroughly with 1 ml of TRIzol LS reagent ( Thermo Fisher Scientific ) . RNA was extracted according to the manufacturer’s instructions and resuspended in 20 μl of nuclease-free water . RT-PCR was performed using the SuperScript IV One-Step RT-PCR Kit ( Thermo Fisher Scientific ) . Virus was determined by plaque assay . Approximately 1 . 2×106 Vero E6 cells were seeded to each well of a six-well plate . The viruses were tenfold serially diluted with 2% FBS DMEM medium and 200 μl of virus dilution was transferred to each well of the six-well plate . After the incubation for 1 hr at 37°C , 2 ml of overlay medium containing 2% FBS DMEM medium and 1% sea-plaque agarose ( Lonza , Walkersville , MD ) , was added to the infected cells per well . After a 2-day incubation , another 2 ml of overlay medium with neutral red ( final concentration 0 . 01% ) was added onto the first overlay . After 12 hr incubation , the plates were sealed with Breath-Easy sealing membrane ( Sigma-Aldrich , St . Louis , MO ) and plaques were counted . SARS-CoV-2 transient luciferase replicon assay was performed as previously described ( Xia et al . , 2020 ) . WT and mutant replicon RNA , and N gene mRNA were obtained through T7 in vitro transcription , and 40 μg RNA transcripts and 20 μg N gene RNA were co-electroporated into 8×106 Huh-7 cells ( ATCC , tested negative on mycoplasma ) using Gene Pulser XCell electroporation system ( Bio-Rad ) at a setting of 270 V and 950 μF with a single pulse . After 10 min recovery , electroporated cells were seeded to 24-well plates , and harvested at indicated timepoints . Luciferase signal was measured using Renilla luciferase assay system ( Promega ) and read by Cytation 5 ( BioTek ) according to the manufacturer’s protocols . The fabrication of the RNA hairpin has been described in detail in Papini et al . , 2019 . The RNA hairpin is made of a 499 bp dsRNA stem terminated by a 20 nt loop that is assembled from three ssRNA annealed together , and two handles , one of 856 bp at the 5′-end and one 822 bp at the 3′-end . The handles include either a 343 nt digoxygenin-labeled ssRNA or a 443 nt biotin-labeled ssRNA . Upon applied force above ~21 pN , the hairpin opens and frees a 1043 nt ssRNA template for SARS-CoV-2 replication . To obtain the different parts of the RNA construct , template DNA fragments were amplified via PCR , purified ( Monarch PCR and DNA Cleanup Kit ) and in vitro transcribed ( NEB HiScribe T7 High Yield RNA Synthesis Kit ) . Transcripts were then treated with Antarctic Phosphatase and T4 Polynucleotide Kinase . RNAs were purified using the RNA Clean and Concentrator-25 kit ( Zymo Research ) . Individual RNA fragments were annealed and ligated with T4 RNA ligase 2 ( NEB ) to assemble the RNA hairpin . The template contains 250 U ( 24% ) , 253 A ( 24% ) , 273 C ( 26% ) , and 267 G ( 26% ) . The high-throughput magnetic tweezers used in this study have been described in detail elsewhere ( Ostrofet et al . , 2018 ) . Shortly , a pair of vertically aligned permanent magnets ( 5 mm cubes , SuperMagnete , Switzerland ) separated by a 1 mm gap are positioned above a flow cell ( see paragraph below ) that is mounted on a custom-built inverted microscope . The vertical position and rotation of the magnets are controlled by two linear motors , M-126-PD1 and C-150 ( Physik Instrumente PI , GmbH and Co . KG , Karlsruhe , Germany ) , respectively . The field of view is illuminated through the magnets gap by a collimated LED-light source , and is imaged onto a large chip CMOS camera ( Dalsa Falcon2 FA-80–12 M1H , Stemmer Imaging , Germany ) using a 50× oil immersion objective ( CFI Plan Achro 50 XH , NA 0 . 9 , Nikon , Germany ) and an achromatic doublet tube lens of 200 mm focal length and 50 mm diameter ( Qioptic , Germany ) . To control the temperature , we used a system described in detail in Seifert et al . , 2020 . Shortly , a flexible resistive foil heater with an integrated 10 MΩ thermistor ( HT10K , Thorlabs ) is wrapped around the microscope objective and further insulated by several layers of Kapton tape ( KAP22-075 , Thorlabs ) . The heating foil is connected to a PID temperature controller ( TC200 PID controller , Thorlabs ) to adjust the temperature within ∼0 . 1° . The fabrication procedure for flow cells has been described in detail in Ostrofet et al . , 2018 . To summarize , we sandwiched a double layer of Parafilm by two #1 coverslips , the top one having one hole at each end serving as inlet and outlet , the bottom one being coated with a 0 . 01% m/V nitrocellulose in amyl acetate solution . The flow cell is mounted into a custom-built holder and rinsed with ~1 ml of 1× phosphate-buffered saline ( PBS ) . 3 µm diameter polystyrene reference beads are attached to the bottom coverslip surface by incubating 100 µl of a 1:1000 dilution in PBS of ( LB30 , Sigma Aldrich , stock conc . : 1 . 828*1011 particles per milliliter ) for ~3 min . The tethering of the magnetic beads by the RNA hairpin construct relies on a digoxygenin/anti-digoxygenin and biotin-streptavidin attachment at the coverslip surface and the magnetic bead , respectively . Therefore , following a thorough rinsing of the flow cell with PBS , 50 µl of anti-digoxigenin ( 50 µg/ml in PBS ) is incubated for 30 min . The flow cell was flushed with 1 ml of 10 mM Tris , 1 mM EDTA pH 8 . 0 , 750 mM NaCl , 2 mM sodium azide buffer to remove excess of anti-digoxigenin followed by rinsing with another 0 . 5 ml of 1× TE buffer ( 10 mM Tris , 1 mM EDTA pH 8 . 0 supplemented with 150 mM NaCl , and 2 mM sodium azide ) . The surface is then passivated by incubating bovine serum albumin ( BSA , New England Biolabs , 10 mg/ml in PBS and 50% glycerol ) for 30 min , and rinsed with 1× TE buffer . 20 µl of streptavidin-coated Dynal Dynabeads M-270 streptavidin-coated magnetic beads ( Thermo Fisher Scientific ) was mixed with ~0 . 1 ng of RNA hairpin ( total volume 40 µl ) ( see Materials and methods ) and incubated for ~5 min before rinsing with ~2 ml of 1× TE buffer to remove any unbound RNA and the magnetic beads in excess . RNA tethers were sorted for functional hairpins by looking for the characteristic jump in extension length due to the sudden opening of the hairpin during a force ramp experiment ( Figure 1—figure supplement 1C; Papini et al . , 2019 ) . The flow cell was subsequently rinsed with 0 . 5 ml reaction buffer ( 50 mM HEPES pH 7 . 9 , 10 mM DTT , 2 µM EDTA , and 5 mM MgCl2 ) . After starting the data acquisition at a force that would keep the hairpin open , 100 µl of reaction buffer containing either 0 . 6 µM of nsp12 , 1 . 8 µM of nsp7 and nsp8 for SARS-CoV-2 experiments or 0 . 1 µM of nsp12 , 1 µM of nsp7 and nsp8 for SARS-CoV-1 experiments , the indicated concentration of NTPs and of NAs ( if required ) were flushed in the flow cell to start the reaction . Sofosbuvir-TP and T-1106-TP were purchased from Jena Bioscience ( Jena , Germany ) and 3′-dATP was purchased from TriLink Biotechnologies ( San Diego , CA ) . The experiments were conducted at a constant force as indicated for a duration of 20–40 min . The camera frame rate was fixed at either 58 Hz or 200 Hz , for reaction temperature set at either 25°C or 37°C , respectively . A custom written Labview routine ( Cnossen et al . , 2014 ) controlled the data acquisition and the ( x- , y- , z- ) positions analysis/tracking of both the magnetic and reference beads in real time . Mechanical drift correction was performed by subtracting the reference bead position to the magnetic bead position . The replication activity of SARS-CoV-2 core polymerase converts the tether from ssRNA to dsRNA , which concomitantly decreases the end-to-end extension of the tether . The change in extension measured in micron was subsequently converted into replicated nucleotides NR using the following equation: ( 1 ) NR ( F ) =N∙LssF-Lmeas ( F ) LssF-Lds ( F ) where Lmeas ( F ) , LssF and Lds ( F ) are the measured extension during the experiment , the extension of an ssRNA and of a dsRNA construct , respectively , experiencing a force F , and N the number of nucleotides of the ssRNA template ( Dulin et al . , 2015a ) . The traces were then filtered using a Kaiser-Bessel low-pass filter with a cutoff frequency at 2 Hz . We removed the rare slow outliers traces from data sets ( Figure 1—figure supplement 2A ) . As previously described in Dulin et al . , 2015a , a dwell time analysis was performed by scanning the filtered traces with non-overlapping windows of 10 nt to measure the time ( coined throughout the manuscript dwell time ) for SARS-CoV-2 polymerase to incorporate ten successive nucleotides . The dwell times of all the traces for a given experimental condition were assembled and further analyzed using a maximum likelihood estimation ( MLE ) fitting routine to extract the parameters from the stochastic-pausing model . To extract the product length of the replication complex , only the traces where the beginning and the end could clearly be distinguished and for which the tether did not rupture for ten minutes following the last observed replication activity were considered . We represented the mean product length , as well as one standard deviation of the mean from 1000 bootstraps as error bars . The model is described in detail in Dulin et al . , 2017; Dulin et al . , 2015a; Seifert et al . , 2020 . There are many kinetic models that are consistent with the empirical dwell time distributions we observe , and we here work under the assumption that the probability of pausing is low enough that there is only one rate-limiting pause in each dwell time window . This assumption washes out most details of the kinetic scheme that connects pauses and nucleotide addition , but allows us to determine the general form of the dwell time distribution without specifying how the pauses are connected to the nucleotide addition pathwaypdw ( t ) ∝pnaΓ ( t;Ndw , 1kna ) +Q ( t ) ( ∑n=1Nsppnkne−knt+abt2 ( 1+t/1s ) 3/2 ) In the above expression , the gamma function in the first term contributes the portion pna of dwell times that originate in the RdRp crossing the dwell time window of size Ndw base pairs without pausing; the second term is a sum of contributions originating in pause-dominated transitions , each contributing a fraction pn of dwell times; the third term captures the asymptotic power-law decay ( amplitude abt ) of the probability of dwell times dominated by a backtrack . The backtracked asymptotic term needs to be regularized for times shorter than the diffusive backtrack step . We have introduced a regularization at 1 s , but the precise timescale does not matter , as long as it is set within the region where the exponential pauses dominate over the backtrack . From left to right , each term of Equation 1 is dominating the distribution for successively longer dwell times . A cutoff factor Qt for short times is introduced to account for the fact that the dwell time window includes Ndw nucleotide-addition steps , ( 3 ) Qt=tkna/NdwNdw-11+tkna/NdwNdw-1 The fit results dependence on these cutoffs is negligible as long as they are introduced in regions where the corresponding term is sub-dominant . Here , the cut is placed under the center of the elongation peak , guaranteeing that it is placed where pausing is sub-dominant . The normalized version of Equation 1 is the dwell time distribution fit to the experimentally collected dwell-times tii by minimizing the likelihood function ( Cowan , 1998 ) . ( 4 ) L=-∑iln⁡pdw ( ti ) with respect to rates and probabilistic weights . The fractions pn represent the probability that a particular rate kn dominates the dwell time . We want to relate this to the probability Pn that a specific exit rate dominates within a 1-nt transcription window . Assuming we have labeled the pauses so that kn−1>kn , we can relate the probability of having rate n dominating in Ndw steps to the probability of having it dominate in one step through ( 5 ) pn=∑m=0nPmNdw-∑m=0n-1PmNdw , p0=pna=PnaNdw=P0Ndw The first term in Equation 3 represents the probability of having no pauses longer than the nth pause in the dwell time window , and the second term represents the probability of having no pauses longer than the ( n-1 ) th pause . The difference between the two terms is the probability that the nth pause will dominate . This can be inverted to yield a relation between the single-step probabilities ( Pn ) and the dwell time window probabilities ( pn ) ( 6 ) Pn=∑m=0npm1/Ndw-∑m=0n-1pm1/Ndw , P0=p01/Ndw This relationship has been used throughout the manuscript to relate our fits over a dwell time window to the single-step probabilities . The above stochastic-pausing model was fit to the dwell time distributions using a custom Python 3 . 7 routine . Shortly , we implemented a combination of simulated annealing and bound constrained minimization to find the parameters that minimize Equation 2 . We calculated the statistical error on the parameters by applying the MLE fitting procedure on 100 bootstraps of the original data set ( Press et al . , 1992 ) , and reported the standard deviation for each fitting parameters . Starting with an empty active site ( E ) , we assume that there is direct binding competition between the natural nucleotide ( N ) and the NA terminator ( T , simply coined terminator ) that result in either the former bound ( Nb ) or the latter bound ( Tb ) to the active site . From these states there can be any number of intermediate states before the base is either added to the chain with probability PcatT/N , or unbinds from the pocket with probability 1−PcatT/N see Figure 2—figure supplement 3 . The effective incorporation rate is the attempt rate times the probability of success , ( 7 ) kincT/N=[T/N]KonT/NPcatT/Nand the relative probability that next incorporated base is a terminator or natural nucleotide is given by the relative effective addition rates . ( 8 ) pTpN=kincTkincN=[T][N]KonTKonNPcatTPcatN , pT+pN=1 . This can be rewritten as , pN=yy+x , pT=xy+x , x=[T][N] , y=KonNKonTPcatNPcatT In the above , x is the relative stoichiometry between T and N , while y is the relative effective incorporation rates of N and T at equimolar conditions . On an infinite construct , polymerization will proceed until the first T is incorporated , after which it terminates . At termination , the product has incorporated n−1 Ns , and finally one T , with probability . ( 9 ) P ( n ) = ( pN ) n−1pT= ( 1−pT ) n−1pT The average number of Ns and Ts incorporated on an infinite construct is therefore . ( 10 ) n∞=∑n=1∞n ( pN ) n−1pT=1/pT If the construct only allows for the addition of N Ns and Ts , the average number of Ns and Ts in the product will instead be , ( 11 ) nN=∑n=1Nn ( pN ) n−1pT+∑n=N+1∞N ( pN ) n−1pT=1− ( pN ) NpT=n∞ ( 1− ( pN ) N ) For a genome of length L , with the relative abundance q of templating bases for N and T , we thus expect there to be at most N=qL Ns and Ts incorporated at termination . At termination the product then has the average length . ( 12 ) lL=nqLq=1− ( pN ) qLqpT=l∞ ( 1− ( pN ) qL ) , l∞=1qpT Though the constructs are 1043 nucleotides long , this length is not always reached even when there are no terminators in the buffer . The average product length is about 10% shorter than the full construct length . To account for this reduction in maximal average product length , we simply fix L to be the mean product length reached without terminator in the buffer , and fit out γ from a least-square fit , weighted with the inverse experimental variance .	To multiply and spread from cell to cell , the virus responsible for COVID-19 ( also known as SARS-CoV-2 ) must first replicate its genetic information . This process involves a ‘polymerase’ protein complex making a faithful copy by assembling a precise sequence of building blocks , or nucleotides . The only drug approved against SARS-CoV-2 by the US Food and Drug Administration ( FDA ) , remdesivir , consists of a nucleotide analog , a molecule whose structure is similar to the actual building blocks needed for replication . If the polymerase recognizes and integrates these analogs into the growing genetic sequence , the replication mechanism is disrupted , and the virus cannot multiply . Most approaches to study this process seem to indicate that remdesivir works by stopping the polymerase and terminating replication altogether . Yet , exactly how remdesivir and other analogs impair the synthesis of new copies of the virus remains uncertain . To explore this question , Seifert , Bera et al . employed an approach called magnetic tweezers which uses a magnetic field to manipulate micro-particles with great precision . Unlike other methods , this technique allows analogs to be integrated under conditions similar to those found in cells , and to be examined at the level of a single molecule . The results show that contrary to previous assumptions , remdesivir does not terminate replication; instead , it causes the polymerase to pause and backtrack ( which may appear as termination in other techniques ) . The same approach was then applied to other nucleotide analogs , some of which were also found to target the SARS-CoV-2 polymerase . However , these analogs are incorporated differently to remdesivir and with less efficiency . They also obstruct the polymerase in distinct ways . Taken together , the results by Seifert , Bera et al . suggest that magnetic tweezers can be a powerful approach to reveal how analogs interfere with replication . This information could be used to improve currently available analogs as well as develop new antiviral drugs that are more effective against SARS-CoV-2 . This knowledge will be key at a time when treatments against COVID-19 are still lacking , and may be needed to protect against new variants and future outbreaks .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "physics", "of", "living", "systems", "microbiology", "and", "infectious", "disease" ]	2021	Inhibition of SARS-CoV-2 polymerase by nucleotide analogs from a single-molecule perspective
Recent studies have implicated the subthalamic nucleus ( STN ) in decisions that involve inhibiting movements . Many of the decisions that we make in our daily lives , however , do not involve any motor actions . We studied non-motor decision making by recording intraoperative STN and prefrontal cortex ( PFC ) electrophysiology as participants perform a novel task that required them to decide whether to encode items into working memory . During all encoding trials , beta band ( 15–30 Hz ) activity decreased in the STN and PFC , and this decrease was progressively enhanced as more items were stored into working memory . Crucially , the STN and lateral PFC beta decrease was significantly attenuated during the trials in which participants were instructed not to encode the presented stimulus . These changes were associated with increase lateral PFC-STN coherence and altered STN neuronal spiking . Our results shed light on why states of altered basal ganglia activity disrupt both motor function and cognition . The basal ganglia have traditionally been implicated in facilitating or inhibiting motor movements ( Albin et al . , 1989; DeLong , 1990 ) . As such , studies tying the basal ganglia to decision making have focused on decisions in which participants either make or withhold a motor command ( Zavala et al . , 2015 ) . Yet , many of the choices that we make in our daily lives involve decisions that do not explicitly require an immediate motor action . For instance , when meeting someone at a dinner party , one may decide that their name is worth trying to remember . Whether and how the basal ganglia may also participate in such non-motor decisions remains poorly understood . Given their widespread connections to cortical regions , we hypothesized that the circuits and structures that enable the basal ganglia to mediate motor decisions may also be used in an analogous manner to participate in non-motor decisions . We were specifically interested in the non-motor decision to attend to and encode items into working memory . Our ability to dynamically control access to working memory is critical as only a small portion of the large number of stimuli we are exposed to each day are relevant for our behavioral goals . Most studies of working memory have implicated several regions in the prefrontal cortex ( PFC ) in regulating whether and how memories are successfully encoded ( McNab and Klingberg , 2008; Spitzer et al . , 2010; Voytek and Knight , 2010; Heinrichs-Graham and Wilson , 2015; Wiesman et al . , 2016 ) . We hypothesized , however , that gating access to working memory , like gating movement , also involves the basal ganglia . Several lines of evidence suggest this possibility . The decision to execute or withhold a movement involves a race between competing basal ganglia pathways that facilitate or inhibit motor commands ( Schmidt et al . , 2013 ) . At the center of this race lies the subthalamic nucleus ( STN ) , a structure within the basal ganglia previously implicated in motor decisions ( Kühn et al . , 2004; Frank et al . , 2007; Cavanagh et al . , 2011; Zaghloul et al . , 2012; Alegre et al . , 2013; Zavala et al . , 2014 ) . The ability to gate movement may also grant the basal ganglia , and the STN , the ability to regulate non-motor cognitive processes such as memory . In addition , the basal ganglia exhibit extensive connections with regions of the PFC that have been implicated in working memory ( Alexander et al . , 1986; Nakano et al . , 2000; Chudasama et al . , 2003 ) , and imaging studies reveal increased basal ganglia activity during tasks that involve working memory ( Chang et al . , 2007; McNab and Klingberg , 2008; Murty et al . , 2011; Chatham et al . , 2014 ) . Empiric support for the basal ganglia’s role in gating memory also arises from studies of patients with Parkinson’s disease who have disruptions of normal basal ganglia circuitry ( Hammond et al . , 2007 ) . These patients often experience working memory deficits that correlate with the degree of motor and executive function impairment ( Owen et al . , 1997; Lewis et al . , 2003; Moustafa et al . , 2008; Chiaravalloti et al . , 2014; Trujillo et al . , 2015; Wiesman et al . , 2016 ) . Moreover , patients receiving deep brain stimulation ( DBS ) , in which high-frequency electrical stimulation is delivered directly to the STN , may experience changes in working memory performance , although studies examining whether DBS improves or worsens working memory have not been conclusive ( Hälbig et al . , 2004; Hershey et al . , 2008; Mayer et al . , 2016; Ventre-Dominey et al . , 2014; Mollion et al . , 2011 ) . Collectively , these studies suggest a link between the basal ganglia and non-motor decisions such as those used to regulate and access working memory . Despite such empiric support , however , few studies have directly examined neural activity of the human basal ganglia , and specifically the STN , during a working memory task . We address this here by examining local field potential and neuronal spiking activity in the human STN in patients undergoing DBS surgery for Parkinson’s disease as they participate in a task requiring them to make non-motor decisions to selectively attend to and encode items into working memory . Because of the known involvement of the PFC in working memory , and because of the known connections between the PFC and the basal ganglia , we also simultaneously capture intracranial EEG ( iEEG ) activity from the PFC in order to ask how these structures communicate during this process of making non-motor decisions . As such , our work builds upon previous studies establishing the role of the human STN in motor response inhibition and decision making ( Zavala et al . , 2015 ) to demonstrate the STN’s involvement in non-motor decisions used to dynamically modulate working memory . Eighteen participants ( 16 males; 58 . 4 ± 1 . 64 ( mean ± SEM ) years old ) undergoing DBS surgery for Parkinson’s disease performed a novel working memory task in which we sequentially presented eight single-digit numbers ( Figure 1A; see Materials and methods ) . We instructed the participants to attend to and memorize only numbers presented within a target shape ( square or octagon ) , which was pseudo-randomly chosen and presented before each block of eight numbers . During each block , four of the numbers appeared within the target shape ( target trials ) . The target numbers were randomly interleaved with four numbers appearing within a distractor shape ( distractor trials ) . Correct performance in each block required the participant to vocalize only the four target numbers at the end of each block . Participants vocalized the correct response on 71 . 2 ± 2 . 6% of the blocks they were shown . In the remaining 28 . 8 ± 2 . 6% blocks , the participants made one of two types of errors: they could fail to vocalize all four target numbers ( 96 . 4 ± 1 . 2% of error blocks ) or they could erroneously include a distractor number in the vocalized sequence ( 63 ± 4 . 5% of error blocks ) . On 62 . 1 ± 4 . 4% of the error blocks , they made both types of errors . We simultaneously captured micro- and macroelectrode recordings from the subthalamic nucleus ( STN ) , and intracranial EEG recordings from subdural electrodes temporarily placed over the anterior and lateral prefrontal cortex ( PFC ) as participants performed the task ( Figure 1B and Figure 1—figure supplement 1 ) . We first investigated overall changes in power during all trials in each of three recorded brain regions - the anterior PFC , the lateral PFC , and the STN - by computing the average power across all trials during each experimental session and across all electrodes in each brain region ( Figure 2A , B ) . Across sessions , all three brain regions demonstrated significant decreases in beta band ( 15–30 Hz ) power across all trials ( p < 0 . 005 , permutation test; see Materials and methods ) . We also found significant increases in theta band ( 2–8 Hz ) power in the anterior and lateral PFC during the task ( p < 0 . 005 for anterior PFC and lateral PFC , permutation test , Figure 3—figure supplement 1A ) . These changes were not specific to either left- or right-sided brain regions , as we found no significant differences in spectral power between experimental sessions captured in the separate hemispheres ( left anterior PFC vs right anterior PFC p=0 . 37 , left lateral PFC vs right lateral PFC p=0 . 55 , left STN vs right STN p=0 . 60 , permutation test; see Materials and methods ) . Although we observed these decreases in beta oscillatory power in both the anterior and lateral PFCs by averaging the spectral power across all electrode contacts , we were interested in whether the observed changes represented a focal process or a diffuse cortical phenomenon . We therefore examined changes in spectral power within each electrode contact . In 19 out of the 24 sessions with lateral PFC electrodes , at least one electrode contact exhibited a significant decrease in beta power during the task ( p < 0 . 05 , permutation test; see Materials and methods ) . This effect was not present in all electrode contacts , however . Indeed , within these 19 sessions , we found that only 2 . 6 ± 0 . 29 of seven lateral PFC electrode contacts ( 36 . 8 ± 4 . 1% ) showed a significant decrease in beta power relative to baseline ( p < 0 . 05 , permutation test; Figure 2—figure supplement 1 ) . In contrast , 14 out of 29 sessions with anterior PFC electrodes demonstrated at least one electrode contact exhibiting a significant decrease in beta power during the task ( p < 0 . 05 , permutation test ) . Within these sessions , we found that 2 . 1 ± 0 . 31 of five electrode contacts ( 42 . 9 ± 6 . 2% ) showed the change in beta power relative to baseline ( p < 0 . 05 , permutation test ) . Although we did not have access to intra-operative computed tomography ( CT ) , we estimated all anterior and lateral PFC electrode locations on each participant’s brain ( see Supplementary Information ) to determine whether the observed effects were localized to a particular brain region . We found that despite the fact that the anterior PFC , the lateral PFC , and the STN all showed a similar drop in beta power during the task , the beta power within individual participants was focal in nature , although the location of these focal drops was not consistent across participants ( Figure 2—figure supplement 1 ) . We were interested in understanding how the observed changes in beta oscillatory power evolved over the course of successive trials in each individual block . Although each individual trial exhibited a relative transient decrease in beta power , overall beta power exhibited a systematic decrease from the first through the last trial of each block in all three brain regions ( Figure 2C ) . The observed decreases across trials were not related to movement as participants were instructed to withhold all movements until receiving an instruction to vocalize their response one second after the final number was cleared from the screen . We quantified this overall decrease by examining beta oscillatory power averaged across the entire trial ( 0 to 1000 ms following stimulus onset ) . A repeated-measures ANOVA revealed a significant effect of trial number on the changes in average beta oscillatory power in all three brain regions ( anterior PFC F=4 . 88 , d⁢f=7 , p < 0 . 001; lateral PFC F=6 . 46 , d⁢f=7 , p < 0 . 001; STN F=4 . 07 , d⁢f=7 , p < 0 . 001 ) . We also performed a linear regression of beta power against trial number and found a significant relation across all experimental sessions in all three brain regions ( anterior PFC ρ=-0 . 31±0 . 1 , t⁢ ( 27 ) =3 . 4 , p < 0 . 002; lateral ρ=-0 . 31±0 . 1 , t⁢ ( 23 ) =3 . 1 , p < 0 . 005; STN ρ=-0 . 35±0 . 1 , t⁢ ( 28 ) =3 . 35 , p < 0 . 002; best fit line in Figure 2C ) . See Supplementary Data and Figure 3—figure supplement 3 for analysis of progressive decrease in beta oscillatory power separately for target and distractor trials . Our data suggest that participation in this non-motor task evokes decreases in beta oscillatory power in both the STN and PFC following the presentation of each number . The observed decreases in beta oscillatory power during each trial are similar to decreases previously observed with motor movements ( Kühn et al . , 2004; Ray et al . , 2012; Alegre et al . , 2013 ) . However , we were specifically interested in whether these changes in beta oscillatory power depend on whether the presented stimulus was a target or distractor item . We therefore investigated whether there were significant differences in oscillatory power between the target and distractor conditions . We computed the average power separately across all target and distractor trials during each experimental session . Across sessions , all three brain regions exhibited initial decreases in beta power following the presentation of both target and distractor . However , during the distractor trials , both the STN and lateral PFC demonstrated an early termination of the beta band decrease . This resulted in significantly higher beta band power relative to the target trials ( p = 0 . 020 for STN and p = 0 . 025 for lateral PFC , permutation test; Figure 3; see Materials and methods; for changes in theta band power , see Figure 3—figure supplement 1B ) . As with the changes in overall beta power during the task , we were also interested in whether the observed differences in beta power between target and distractor trials were specific to individual electrode contacts . We found that 13 out of the 24 experimental sessions with lateral PFC electrodes demonstrated a significant difference between target and distractor trials in at least one contact ( p < 0 . 05 , permutation test ) . Indeed , the observed effect averaged across the entire brain region was only mediated by a subset of electrodes . Within these 13 sessions , only 2 . 1 ± 0 . 2 of seven electrode contacts ( 29 . 7 ± 3 . 4% ) showed a significant difference in beta power between target and distractor trials ( p < 0 . 05 , permutation test; Figure 3—figure supplement 2 ) . Conversely , in the anterior PFC , only 6 out of 29 sessions demonstrated at least one contact exhibiting a significant difference in average beta power between target and distractor trials ( p < 0 . 05 , permutation test ) . Although our data demonstrate that target and distractor stimuli involved differences in beta oscillatory power during each trial , we were interested in whether these differences were directly related to the behavioral performance on each trial . As such , we next examined the relatively few error trials in which participants failed to encode the target number . If participants failed to encode these target stimuli because they treated them as distractors , then physiologically the observed beta decreases should be more similar to the distractor rather than the target trials . In the lateral PFC electrodes , we found that the successfully encoded target trials exhibited significantly greater decreases in beta power than those target trials that participants failed to encode ( p = 0 . 04 , permutation test; Figure 3—figure supplement 4 ) . Indeed , the changes in beta power during the target trials that were not encoded did not differ from the distractor trials , suggesting that the failure to encode those target trials were related to the same early rebounds in beta power that were present when participants intentionally chose not to encode the distractor numbers . We did not , however , find similar differences between the error trials and the successfully encoded target trials in the anterior PFC ( p = 0 . 17 , permutation test ) or the STN ( p = 0 . 20 , permutation test , Figure 3—figure supplement 4 ) . We also did not find significant differences between the erroneously encoded distractor trials and the correct distractor trials in any brain region ( p > 0 . 05 , permutation test , data not shown ) . As the lateral PFC and the STN both demonstrated significant differences in oscillatory power between target and distractor trials , we were next interested in asking whether these conditions also exhibited significant differences in connectivity between the two regions . Within each participant , we first identified individual lateral and anterior PFC contacts exhibiting significant differences in power between target and distractor trials that mediated the overall changes in these brain regions ( Figure 3—figure supplement 2 ) . We focused our analysis of connectivity only on the lateral PFC electrodes that exhibited a significant difference in power within participants ( p < 0 . 05 , permutation test; 12 experimental sessions with simultaneously LFP recordings within the borders of the STN; see Materials and methods; an insufficient number of individual anterior PFC electrodes exhibited significant differences in power within participants , consistent with the lack of an overall effect of trial type in the anterior PFC ) . We examined connectivity by comparing the spectral coherence between the lateral PFC and STN to the coherence observed during a baseline period ( see Materials and methods ) . As with the analysis of power , we first calculated the average coherence difference between target and distractor trials for the relevant STN-lateral PFC pairs within a session before testing for a significant trial-type related difference across all sessions . Distractor trials exhibited significantly higher coherence in the beta band between the lateral PFC and the STN than target trials ( p = 0 . 015 , permutation procedure; Figure 4 ) , suggesting that the elevated levels of beta power in both brain regions during distractor trials were accompanied by elevated levels of beta band coherence . We subsequently tested whether the higher levels of coherence were due to increased synchrony of phase or increased correlation of power ( Cohen , 2014 ) . The increased coherence we observed during distractor trials was primarily driven by higher levels of phase synchrony ( p = 0 . 025 , permutation test; Figure 4—figure supplement 1 ) . Distractor trials also had higher levels of correlation of beta power , but this difference did not survive multiple comparison testing . Finally , we extracted spiking activity from microelectrode recordings in the STN while participants were engaged in the task . We identified spiking activity from 48 neuronal firing clusters across 21 STN recording sessions ( 35 . 9 ± 3 . 7 spikes/s ( mean ± SEM ) per cluster ) . On average , across all neurons , we found a significant decrease in overall spiking activity during each trial compared to baseline ( p = 0 . 02 , permutation test; see Materials and methods; Figure 5A ) . However , the responses of individual neuronal clusters were heterogenous . Some individual clusters demonstrated decreases in spiking activity during each trial , whereas others demonstrated increases ( Figure 5B ) . Within individual recordings , we found that 47 . 9% ( 23 out of 48 clusters ) of the recorded clusters demonstrated a significant change in firing rate ( p < 0 . 05 , permutation test ) during the task ( across all trials ) when compared to baseline ( Figure 5C ) . Of these neuronal clusters that showed a change in firing , most ( 65% , 15 out of 23 clusters ) exhibited significant decreases in firing during the task , but some ( 35% , 8 out of 23 clusters ) exhibited significant increases ( average responses of both populations in Figure 5D ) . Relative to the dorsal and ventral borders of the STN , there were no significant differences in depth between the upward ( mean depth 70 . 4 ± 7 . 8%; see Materials and methods ) and the downward firing clusters ( mean depth 56 . 2 ± 5 . 4%; p = 0 . 14 , unpaired t-test ) . We separately examined the neuronal clusters that individually demonstrated decreases in spiking activity during the task , and found that the average response of these clusters across an entire block exhibited similar transient dynamics during each trial to those observed in beta oscillatory power ( Figure 5E ) . However , unlike the changes in beta oscillatory power , we did not find that there was a progressive change in overall spiking activity with subsequent trials across each block ( ρ=0 . 13±0 . 20 , t⁢ ( 14 ) =0 . 61 , p = 0 . 55 ) . Nevertheless , we examined the relationship between individual spike events and the ongoing LFP oscillations recorded from the STN macroelectrodes , and found that these neuronal clusters were significantly entrained to ongoing beta oscillations during the task ( p = 0 . 02 , permutation test; Figure 5F; see Materials and methods ) . Notably , the neuronal clusters that individually demonstrated increases in spiking activity during the task did not exhibit a significant interaction between LFP phase and spiking events ( p = 0 . 16 , permutation test; Figure 5F ) . The observed changes in STN spiking activity suggest that indeed the STN modulates its spiking output during this non-motor task . However , as with the changes in beta oscillatory power , we were also interested in whether the spiking responses of these neurons were differentially modulated by target and distractor trials . Individual recordings accounting for approximately 20% of the recorded neuronal clusters exhibited significant differences in spiking activity between target and distractor trials ( p < 0 . 05 , permutation test; Figure 5G ) . However , we found no significant overall differences in spiking activity between target and distractor trials across either decreasing ( p = 0 . 34 , permutation test; Figure 5H ) or increasing neurons ( p = 0 . 36 , permutation test ) . Moreover , although neuronal clusters exhibiting decreased spiking activity during the task were locked to the LFP beta oscillation , we found no significant differences in beta band spike phase locking between the target and distractor trials ( p > 0 . 05 , permutation test ) . This was the case when we examined the full 1000 ms window of all trials , as well as when we used smaller time windows ( 250 ms ) centered at various points within the task ( data not shown ) . Our data demonstrate that the human STN modulates both oscillatory and spiking activity as participants make decisions regarding whether to attend to and remember individual items . Our data therefore build upon previous empiric findings suggesting that the basal ganglia may play a role in regulating working memory , and provide direct evidence that the STN exhibits changes in activity during a non-motor decision . Moreover , by simultaneously recording activity from the lateral PFC , our data also demonstrate changes in oscillatory coherence between the PFC and STN , thereby tying cortical regions known to be involved in working memory with the STN . In the novel working memory task used here , participants make internal decisions to attend to and encode numbers if they appear within a target shape , or to ignore them if they appear within a distractor shape . Importantly , both choices involve no immediate motor movements , enabling us to directly ask whether the STN participates in such non-motor decisions . As when inhibiting motor movements ( Zavala et al . , 2015 ) , participants should optimally perform this task by immediately recognizing the indicator shape on each trial , and in the case of distractors , either ignore the presented item completely or prevent any processes that actively encode it . Although our data cannot distinguish these possibilities , the observed differences in neural activity suggest that participants are able to successfully make an internal distinction between the two trial types . Of note , this strategy would spare working memory capacity from storing distracting items that are unnecessary for correct completion of the block . It is possible , however , that the participants encode every number presented in each trial , along with the shape within which each number was presented . In this latter scenario , the participants would make a distinction between the target and distractor items only at the time of retrieval when vocalizing their response . Although we do not explicitly test their memory for the distractor items , it is less likely that participants are adopting this latter more challenging approach given how well they performed this task , and given the differences in neural activity observed at the time of presentation between target and distractor items . The most robust changes we observed during our task in both the STN and in the lateral PFC were related to beta oscillatory activity . Changes in STN beta oscillatory power have been extensively studied in the context of movement , decreasing during the execution of a movement , and increasing when a movement is inhibited . The relation between beta oscillatory activity and movement is most evident during stop inhibition tasks when participants are asked to specifically begin or stop a motor movement ( Kühn et al . , 2004; Ray et al . , 2012; Alegre et al . , 2013 ) . In the task described here , participants make a similar decision , albeit without moving . Remarkably , both the STN and the lateral PFC exhibit similar decreases in beta oscillatory power during this non-motor decision , with significantly earlier termination of these decreases when the presented item should be ignored . The absence of movement here therefore allows us to specifically examine changes in STN activity during non-motor decisions and show that these changes are similar to those seen during movement inhibition . However , this also makes interpreting the timing of when these changes in beta power occur more difficult since there is no explicit behavioral measure of when each individual is internally processing the numbers and distractor signals , or when they are deciding to encode or ignore each stimulus . Nevertheless , we find that the timing of beta power differences between trial types is similar between our task and the tasks that required an inhibition of an actual movement , suggesting that the changes observed here are analogous to the previously identified stop signal ( Kühn et al . , 2004; Ray et al . , 2012; Alegre et al . , 2013 ) . Moreover , in stop inhibition tasks , differences in beta power can still alter the behavior of the failed stop trials ( in which a subject failed to properly inhibit an action ) even when occurring after the erroneous response has already begun . These differences in beta power persist after that behavior has been initiated , and in the case of movement inhibition , this manifests itself as a suppression of the effort and duration of an erroneous response ( Ruiz et al . , 2011; Cohen and van Gaal , 2014 ) . In the case of our current task then , increased STN beta activity may prevent the encoding of an item into working memory , or it may impair the maintenance of that item in memory . Although rebounds in beta power have also been observed following the execution of a movement , because these rebounds occur much later in the task than the timing observed here ( Alegre et al . , 2013 ) , we think our data are more consistent with an internal stop signal rather than a post-action rebound in beta power . Given the relative timing of the observed changes in beta power and the behavioral and physiological evidence that participants are able to successfully distinguish trial types , our data are therefore consistent with previous studies suggesting that beta band activity may in fact represent the brain’s mechanism for globally suppressing all actions . The concept of global suppression has been well described in the context of movement , but more recently has been extended to include even non-motor processes such as working memory ( Wessel et al . , 2016a ) . Increases in STN beta oscillatory power have been observed when participants actively refrain from updating working memory ( Oswal et al . , 2012 ) , and following the presentation of a surprise auditory stimulus that impaired the maintenance of items in working memory ( Wessel et al . , 2016b ) . Moreover , inhibiting movements can result in an unintentional inhibition of memory ( Chiu and Egner , 2015 ) , suggesting that the suppression of actions related to movement and to memory may be linked . Indeed , the suggestion that beta band oscillatory activity could serve as a global suppression signal for both motor and non-motor functions throughout the brain finds support both in our observation that the STN and the lateral PFC exhibit similar decreases and rebounds in beta oscillatory power during this non-motor decision , and in previous studies of the PFC . In particular , beta oscillations in the lateral PFC decrease during the encoding phase of working memory ( Spitzer et al . , 2010; Brookes et al . , 2011; Heinrichs-Graham and Wilson , 2015; Kornblith et al . , 2016; Lundqvist et al . , 2016; Wiesman et al . , 2016 ) , and exhibit similar changes as found in the STN when movements are inhibited ( Swann et al . , 2009 , 2012 ) . In patients with Parkinson’s disease , therefore , excessive beta oscillatory activity may not only impact movement , but could also play a role in the observed deficits in working memory due to the global nature of this activity ( Owen et al . , 1997; Lewis et al . , 2003; Chiaravalloti et al . , 2014; Trujillo et al . , 2015; Wiesman et al . , 2016 ) . These effects may be related to the loss of dopaminergic innervation to the lateral PFC that result in increased beta oscillations ( Puig and Miller , 2012; Puig et al . , 2014; Puig and Miller , 2015 ) . Importantly , here we demonstrate that the changes in beta oscillatory power are accompanied by significant changes in beta oscillatory coherence between the lateral PFC and the STN . Although the lack of intraoperative CT imaging limited our ability to determine precisely what areas of the lateral PFC were interacting with the STN , our data provide direct evidence linking these structures and their respective roles as participants make these non-motor decisions related to memory . Interestingly , the progressive decreases in overall beta oscillatory power over successive trials observed here suggest a link with working memory load . Previous studies have demonstrated progressive increases in gamma band power in the PFC and hippocampus with working memory load ( van Vugt et al . , 2010; Roux et al . , 2012 ) , and in alpha band power in the parietal lobe ( Tuladhar et al . , 2007 ) . These progressive changes are not consistent , however , as progressive decreases in oscillatory activity with working memory load have also been observed across several cortical regions ( Meltzer et al . , 2007 ) , similar to our data . Of note , we observed such progressive decreases separately for both target and distractor trials , with no progression in the differences in beta oscillatory power between them . It is also possible , therefore , that these progressive decreases are not related to memory load , and instead emerge as each trial brings the participant closer to the forthcoming motor movement at the end of each list . Conversely , if these progressive decreases are indeed linked with memory load , then participants may be retaining some memory of the distractor items even though , during individual trials , their neural data suggest that they treat these items differently . Alternatively , the progressive decreases could be related only to increasing memory load of the target items , in which case the distractor trials may not add to the decreases , but also may not reverse them . Ultimately , the STN exerts its effects on the rest of the basal ganglia through its spiking activity . Our data demonstrate that the firing rates of a significant fraction of individual neurons within the STN were also modulated during this working memory task that did not involve an explicit motor movement . Overall firing rates averaged across all neurons decreased during each trial , consistent with the suggestion that decreases in overall STN spiking activity are facilitatory . However , just as movement elicits variability in the spiking responses of individual neurons ( Zavala et al . , 2017 ) , the decision to attend to and encode items was accompanied by heterogeneous spiking responses . Indeed , the observed overall decreases in spiking activity was mediated by only a subset of neurons . Interestingly , these neurons also exhibited significant phase locking to the ongoing beta oscillations , suggesting a link between spiking activity in the STN and LFP oscillations . Despite the overall decreases in spiking activity , however , only a subset of neurons demonstrated a significant difference in spiking activity between target and distractor trials . Moreover , we did not find evidence that there was a significant difference in spike phase locking between the two conditions . Hence , while our oscillatory data would implicate the basal ganglia at large in such non-motor decisions that distinguish target and distractor items , the spiking data would suggest that only a subset of STN neurons are involved in this process , and that these neurons do not alter the extent to which they are locked to the ongoing beta oscillations accordingly . Finally , our data demonstrate that the changes in cortical beta oscillatory power were also accompanied by changes in cortical theta oscillatory power . Unlike beta oscillations , the precise roles played by theta oscillations in the cortex and STN are less clear . The elevated levels of cortical theta oscillatory power during the distractor trials are consistent with the role cortical theta oscillations may play during conflict related response inhibition ( Cavanagh et al . , 2011 , 2012 ) . Given our previous work demonstrating increases in STN theta power and cortico-STN theta coherence during conflict ( Zavala et al . , 2014 , 2016b , 2016a ) , however , we were surprised to find no trial-type related differences in theta power in the STN during this task . One possibility , however , is that STN theta increases are specific to conflict induced slowing of response times rather than complete response inhibition . Indeed , previous studies have demonstrated that motor response inhibition as tested using Go-NoGo paradigms fails to elicit differences in theta oscillatory activity despite showing robust changes in the beta band power similar to those reported here ( Kühn et al . , 2004; Ray et al . , 2012; Alegre et al . , 2013 ) . Together , our data demonstrate that many of the circuit mechanisms of the basal ganglia that have been previously described in the context of movement are used in an analogous fashion during non-motor cognitive decisions . Although our data were collected from patients with Parkinson’s disease in an intraoperative environment , we primarily contrast beta band activity when participants were making decisions between different trial types , suggesting that these differences may exist or even be enhanced ( Oswal et al . , 2012 ) in the healthy brain . Here , we were specifically interested in the non-motor decision to attend to and encode items into working memory . Our data raise the possibility , however , that similar neural mechanisms and interactions between the basal ganglia and the cortex could be used for any non-motor decision that involves proceeding with or aborting a cognitive process . We captured intraoperative recordings in 18 participants undergoing deep brain stimulation ( DBS ) surgery of the subthalamic nucleus ( STN ) for Parkinson’s disease . The study was conducted in accordance with an NIH IRB approved protocol , and all participants gave their written informed consent to take part in the study . Participants received no financial compensation for their participation . Parkinson’s medications were stopped on the night before surgery ( 12 hr preoperatively ) . We captured recordings while participants were alert , at rest , and in an OFF state while in the operating room . As per routine DBS surgery , we used microelectrode recordings to identify the STN based on firing rate and pattern . We simultaneously advanced three electrodes , separately spaced 2 mm apart , during each recording session ( placed along a central , 2 mm lateral , and 2 mm anterior trajectory; Figure 1C ) . Each targeting electrode consisted of a microelectrode contact and a macroelectrode contact positioned 3 mm dorsal to the microelectrode tip . Macroelectrode were within the STN if the corresponding microelectrode contact was greater than 3 mm ventral to the dorsal border of the STN ( identified by increased spiking activity and background noise relative to the more dorsal zona incerta and thalamus ) . We restricted all analyses only to signals captured from electrode contacts positioned within the STN . Raw signals were sampled at 1 . 5024 and 24 . 0345 kHz from macro and microelectrode contacts , respectively , and stored using a MicroGuide Pro data acquisition system ( Alpha Omega Co . , Alpharetta , GA ) . The raw electrophysiology data and relevant code are available on Dryad ( Zavala et al . , 2018 ) . During the operative procedure , we acquired simultaneous intracranial EEG ( iEEG ) recordings from two subdural strip electrodes temporarily placed through the DBS burrhole PMT Corporation , Chanhassen , MN; Figure 1C ) . We placed a six-contact anterior prefrontal cortext ( PFC ) strip electrode consisting of a single row of six platinum contacts ( 2 . 3 mm exposed diameter with 1 cm inter-contact spacing ) in a direct anterior direction from the burr hole . We also placed an eight-contact lateral PFC strip electrode in a direction that was angled approximately 60 degrees lateral to the direction of the anterior PFC strip electrodes . We confirmed contact localization using intraoperative X-ray ( Figure 1—figure supplement 1 ) . The subdural strip electrodes were removed after completion of the behavioral task on each side . Participants performed a novel working memory task ( Figure 1A ) in the intra-operative environment . The task involved 30 blocks . During each block , we sequentially presented participants with eight randomly chosen single-digit numbers , and instructed them to attend to and encode four of the numbers ( targets ) and ignore the remaining four numbers ( distractors ) . Each number was presented within either a square or an octagon . At the beginning of each block , we pseudo-randomly chose and displayed in the center of the screen either an image of the square or the octagon to indicate the target shape for the upcoming block . The participants indicated that they understood the target shape by pressing a large handheld button . After the button press , we displayed a blank screen for 1000 ± 100 ms . We then sequentially displayed each number for 500 ms , followed by a blank screen for 500 ± 100 ms . In this manner , each block consisted of four target trials and four distractor trials that were randomly interleaved . No numbers were repeated . During the presentation of the numbers , we instructed the participants not to make any responses and to explicitly limit all movements if possible . Following the presentation of the final number in each block , a blank screen was presented on the screen for one second . We then prompted the participants to vocally retrieve the four target numbers by displaying the words 'Say the numbers’ on the screen . We manually recorded their responses by typing it into the computer . If the participant correctly vocalized all four target numbers ( in any order ) , and said no other numbers , we presented a green smiley face and the word ’CORRECT’ in the center of the screen . If the participant did not correctly vocalize the four target numbers , we presented a red sad face and the word ’WRONG . ’ In the rare case when the participant vocalized only the four distractor numbers ( and none of the target numbers ) , we presented a yellow neutral face and the phrase , ’Wrong Symbol . ’ In these cases , the participant had performed the task correctly , but had encoded and retrieved numbers presented within the wrong shape . Feedback in this case was intended to encourage participants to pay more attention to the target shape presented at the beginning of each block . For subsequent analysis , we included all trials from the correct blocks and from blocks in which the participant attended to the wrong shape because in both cases the participant had attended to just one shape and ignored the other . Most participants completed 30 blocks of 8 trials each ( four target and four distractor trials ) during each experimental session . Despite the challenging setting , most participants had were able to successfully remember all four numbers . During three sessions , however , the participant failed to correctly encode and retrieve the target numbers during several of the initial blocks but was willing to continue performing the task . In these three cases , we added additional blocks at the end of the experimental session in order to increase the total number of correct blocks included in the analysis . Participants correctly completed 22 . 8 ± 0 . 5 blocks , corresponding to 91 . 0 ± 2 . 1 target trials per participant ( and an equal number of distractor trials ) , that were included in our main analysis . During incorrect blocks , participants made two types of errors . They either failed to vocalize a target number , or they erroneously vocalized a distractor number . For our analysis of error-related activity , we only included sessions in which participants made these errors in at least five blocks . Errors in which participants failed to encode target items met this inclusion criteria in 24 sessions , resulting in 14 . 8 ± 2 . 8 error trials ( meaning trials in which the target number being displayed was not subsequently remembered ) per session . Errors in which participants erroneously vocalized distractor items met this inclusion criteria in 18 sessions , resulting in 8 . 6 ± 1 . 9 error trials ( meaning trials in which the distractor number being displayed was subsequently recalled ) per session . On the day before surgery , participants practiced a complete 30-block session of the task in order to familiarize themselves with the task . Prior to the practice session , we explained the three possible types of feedback the participants could potentially receive at the end of each block and emphasized the importance of not moving during the time period when the numbers were displayed . During the operative procedure , most participants performed one session while we captured recordings from the left STN and a second session while we recorded from the right STN . Four of the 18 participants did not complete a second session on the opposite side because of fatigue . Thus , there were 32 total intraoperative STN recordings included in the analysis . In two of these 32 experimental sessions , the LFP macro electrodes were not located within the STN . Furthermore , in three of the 32 experimental sessions , we did not implant any iEEG strip electrodes , and in five additional sessions we only implanted the anterior PFC strip electrodes without the additional lateral PFC strip electrodes . We collected the data during an 18-month period between Nov 2014 and April 2016 . Every patient who was willing and able to participate in the task was included in the analysis . We performed all analyses using custom MATLAB scripts ( Mathworks , Natick , MA; Raw data and relevant analysis code are available on Dryad [Zavala et al . , 2018] ) . We extracted local field potential ( LFP ) activity from each macroelectrode and iEEG activity from each subdural contact . We bandpass filtered both signals between 1 and 500 Hz , notch filtering at 60 Hz , and downsampled the data to 1 kHz . We referenced macroelectrode signals to the common average of all simultaneously recorded macroelectrode contacts to yield three referenced monopolar LFP channels per session . We referenced the iEEG signals by subtracting the signals of adjacent electrodes . We henceforth refer to these bipolar channels as electrode contacts . Prior to any subsequent analysis , we manually discarded all trials exhibiting a clear artifact in the LFP or iEEG trace . In order to obtain magnitude and instantaneous phase information in the frequency domain , we convolved the LFP signals captured from the STN and iEEG signals captured from the subdural contacts from each trial with complex valued Morlet wavelets ( wave number 6 ) . We used 47 logarithmically spaced ( 8 scales/octave ) wavelets between 2 and 107 Hz and convolved each wavelet with 1500 ms of LFP data from each trial ( 250 ms before to 1250 ms following number presentation ) . We used a 1000 ms buffer on both sides of the clipped data to eliminate edge effects . We squared the magnitude of the continuous-time wavelet transform to generate a continuous measure of instantaneous power for each frequency . We determined the percentage change in power ( normalized power ) for each channel and frequency by comparing the continuous measure of power to the mean power recorded from that channel during a baseline period . We defined the baseline period as the 500 ms preceding the presentation of each number . For each STN recording , we averaged the normalized power from macroelectrodes that were within the STN . Because we did not have access to intraoperative computed tomography ( CT ) imaging , we were unable to use patient specific landmarks to accurately localize individual iEEG contacts . Thus , we defined the activity for each cortical brain region as the averaged normalized power from all bipolar channels recorded from that strip . This procedure therefore resulted in a single spectrogram that we broadly assigned to each of the two cortical brain regions ( anterior and lateral PFC ) during each trial . To assess overall changes in spectral power regardless of trial type during the task compared to baseline , we performed a random-effects statistical analysis in each brain region across experimental sessions . We treated each experimental session as an independent event since in the cases when a participant completed two experimental sessions , they did so while we recorded from the left and right hemispheres separately . We first calculated the trial-averaged normalized power for each region , yielding an average normalized power spectrogram at each time point and frequency for each session . Our null hypothesis was that across experimental sessions , the average normalized power at each time-frequency point would be zero ( no change from baseline ) . We tested this hypothesis using a non-parametric permutation procedure in which the across trial average for each experimental session is the unit of observation ( Maris and Oostenveld , 2007 ) . In each region , we computed the true mean difference in spectral power between task and baseline in each session for every time point and frequency . This generates a distribution of spectrograms ( one for each session ) and a resulting average spectrogram across experimental sessions . We then generated a surrogate distribution of differences by permuting the data 200 times to generate an empiric distribution of possible mean spectrograms that are all equally probable under the null hypothesis . Each of the 200 surrogate mean spectrograms was generated by randomly either subtracting the baseline from the task data or subtracting the task data from the baseline in each session , and then recomputing the mean difference in the spectrogram across sessions . For every time-frequency point , we then compared the true mean across-session power to the mean and standard deviation of the corresponding point in the empiric distribution to generate a p-value . This p-value represents the likelihood that the true mean change from baseline for each time-frequency point represents a departure from the null hypothesis . However , this p-value for each time-frequency point does not take into account the multiple comparisons that are made across all time points and frequencies . To correct for multiple comparisons across all time points and frequencies , we used a cluster correction method based on exceedance mass testing ( Maris and Oostenveld , 2007 ) . This method assumes that a true effect at any time-frequency point is likely to be observed across multiple time points and frequencies . We defined time-frequency clusters by thresholding the across-session p-values derived from the statistical analysis described above . Any contiguous time-freqency points with a p-value less than 0 . 05 were included in each cluster . For each identified cluster , we defined a cluster statistic to be the sum of the z-scores , derived from the p-value using a normal cumulative distribution function , for all time-frequency points within that cluster . We calculated clusters using the true data , and for each of the 200 permutations . We used the maximum cluster statistic of each permutation to create an empiric distribution for significance testing . We determined whether a true cluster test statistic was significant by comparing it to the empiric distribution of maximum cluster test statistics . In this manner , significant clusters can arise from large changes in power that extend over a small number of frequencies or over a small time period , or from smaller differences that involve a larger number of time-frequency points . We considered cluster test statistics with p < 0 . 05 to be significant and corrected for multiple comparisons . We also tested whether there were any differences in response to the task that depended on the hemisphere ( left or right ) from where the data were recorded . For all of the participants that performed the task twice , we calculated the mean difference between the response on the left and right hemispheres . To test whether the difference we observed was significantly different from zero , we compared the true mean difference to the distribution of 200 surrogate differences calculated by randomly permuting the hemisphere labels ( left or right ) of each experimental session’s across-trial average prior to averaging across sessions . We tested for significance and corrected for multiple comparisons using the same cluster based procedure described above . We also tested for significant changes in power within two specific frequency bands of interest ( theta , 2–8 Hz; beta , 15–30 Hz ) based on prior evidence suggesting these frequency bands are involved in conflict and response inhibition ( Cavanagh et al . , 2011 , 2012; Cohen and Cavanagh , 2011; Brittain et al . , 2012; Zavala et al . , 2013; 2014 , 2017; Kühn et al . , 2004; Alegre et al . , 2013 ) . We used the same permutation procedure described above , but first averaged spectral power across the frequency band of interest prior to calculating any true or surrogate means . In this case , clusters were based on contiguous time points exhibiting significant differences across sessions . To identify individual electrode contacts within individual experimental sessions that demonstrated a significant change in power , we used the same statistical analysis and permutation procedure described above . In this case , however , the unit of observation for our statistical test was the individual trial . Hence , for each electrode contact , we calculated the difference in spectral power between each trial and baseline , and computed the mean across all trials . We then compared that true mean spectral power to a distribution of 200 surrogate mean spectrograms . Each of the 200 surrogates was generated by randomly either subtracting the baseline from spectral power during the trial or subtracting the spectral power during the trial from the baseline in each trial , and then recomputing the mean difference in the spectrogram across trials . To assess differences in spectral power between conditions ( target and distractor ) across experimental sessions , we used an identical permutation procedure . In this case , however , we calculated the average normalized spectral power separately for target and distractor trials in each experimental session . We then calculated the average true mean difference between target and distractor conditions across sessions . We again performed a random-effects statistical analysis , with our null hypothesis in this case being that across sessions , there was no difference in normalized power between target and distractor trials . Thus , to test whether the difference we observed was significantly different from zero , we compared the true mean difference to the distribution of 200 surrogate differences calculated by randomly permuting the condition labels ( target or distractor ) of each experimental session’s across-trial average prior to averaging across sessions . We tested for significance and corrected for multiple comparisons using the same cluster based procedure described above . We used a similar procedure to compare the correct target trials to the incorrect target trials as well as the correct distractor trials to the incorrect distractor trials . To determine whether any individual electrode contacts exhibited a significant difference in spectral power between target and distractor trials , we performed the same statistical analysis and permutation procedure . In this case , for each electrode contact , we compared the mean spectral power difference between target and distractor trials to a surrogate distribution generated by permuting the trial labels ( target or distractor ) within that individual session . We identified 13 of the 24 recordings sessions that had lateral electrodes that showed a significantly higher power during the distractor trials for at least one contact . Twelve of these recording sessions had simultaneous LFP recordings within the borders of the STN . For the anterior PFC electrodes , 6 of the 29 recording sessions contained electrodes with a significant difference in power , but only 4 of these sessions had simultaneous LFP recordings within the borders of the STN . For this reason , we focused our subsequent analyses only on the 12 lateral PFC electrode recordings that had significant power differences and simultaneous LFP recordings within the borders of the STN . To estimate the time-varying coherence between the STN and the lateral PFC , we calculated the cross-spectrum between every pair of contacts at each time-frequency point by multiplying the complex value extracted from the LFP signal at each time-frequency by the conjugate of the complex value extracted from the iEEG signal . We similarly calculated each signal’s auto-spectrum at every time and frequency . To generate a continuous-time estimate of coherence , we used the absolute value of the average cross-spectrum divided by the product of each signal’s average autospectra over 250 ms sliding windows ( step size 1 ms ) ( Lachaux et al . , 2002 ) . This results in a time-varying estimate of coherence for every time-frequency point , for every trial , and for every pair of electrodes . We determined the percentage change in coherence ( normalized coherence ) for each channel pair and time-frequency point by comparing the continuous measure of coherence to the mean coherence recorded from that channel pair during a baseline period ( 500 ms preceding each number ) . In order to compare differences in normalized coherence between target and distractor trials across participants , we used the same across-session permutation procedure described above . Because changes in coherence can be due to increased phase synchrony or increased correlations of power ( Cohen , 2014 ) , we separately examined these two measures to further characterize the relation between the STN and the lateral PFC . We calculated phase synchrony using an identical method to our measure of spectral coherence . In this case , however , we extracted the angle of the cross-spectrum at each time point , which represents the phase difference between the LFP and iEEG signal , rather than magnitude . To compare differences in phase synchrony between target and distractor trials across sessions , we used the same permutation procedure described above . To calculate the correlations in beta power between the STN and the lateral PFC , we first smoothed the time-frequency power data across time by convolving each trial’s power time series with a 250 ms window separately for each frequency . At each time point , we calculated the Spearman correlation coefficient between the STN LFP power and the lateral PFC iEEG power across trials , resulting in a time-by-frequency spectrogram of correlations for each lateral PFC-STN electrode pair . In order to compare differences in power correlations between target and distractor trials across sessions , we used the same permutation procedure described above . We extracted spiking activity by bandpass filtering microelectrode recordings between 0 . 3 and 3 kHz and resampling the filtered signals at 24 kHz . Using a spike-sorting software package ( Plexon Offline Sorter , Inc . , Dallas , TX ) , we identified spike waveforms by manually setting a negative or positive voltage threshold depending on the direction of the voltage deflection . Given the difficulty of isolating single-units in the STN ( Weinberger et al . , 2006; Sharott et al . , 2014 ) , it is possible that some of the units we recorded reflected the activity from more than one neuron . We will therefore subsequently refer to each individual microelectrode recording as a neuronal cluster . We defined the location of each cluster as a percentage reflecting the relative depth between the dorsal and ventral border of the STN as defined by the microelectrode recordings . We extracted 1500 ms of spiking data from each trial for each microelectrode . We excluded all trials with an average firing rate greater or less than 10 standard deviations from the average firing rate across all trials . We then calculated the continuous-time firing rates for each recording by smoothing the spike train from each trial ( 1 ms bins ) with a Gaussian kernel ( standard deviation 50 ms ) . To generate a normalized firing rate , we compared continuous time firing rates for each trial to the mean and standard deviation of the firing rates during the 500 ms baseline period and then averaged across trials . To determine whether there were any overall changes in firing activity across all sessions , we used the same permutation procedure as described above , comparing the true mean continuous firing rate across all trials ( both target and distractor trials ) in each session to baseline and treating the session as the unit of observation . Similarly , we determined whether each individual cluster exhibited a significant change in firing during the task using the same permutation procedure , in this case comparing the differences in firing rate between trials and baseline across all trials within a given cluster and treating the trial as the unit of observation . We subsequently calculated for each time point the percentage of all recordings that showed a significant difference from baseline across all trials . Finally , to determine whether a given cluster exhibited a significant difference between the target and distractor conditions , we calculated the mean difference in firing at every time point between target and distractor trials . We then used the same within-session permutation procedure described above to test for significance . We subsequently calculated for each time point the percentage of all recordings that showed a significant difference between target and distractor trials . In order to analyze the relationship between neuronal firing and ongoing LFP phase oscillations , we calculated spike-phase interactions for each neuronal cluster ( Zavala et al . , 2017 ) . Briefly , we tabulated the instantaneous phase of the macroelectrode LFP signal during all spiking events captured on the associated microelectrode during each trial . We then calculated the normalized mean vector length , ( r ) , across all trials during the first 1000 ms following each trials stimulus onset . We normalized the resulting spike-phase locking values by permuting phase information across trials . Whereas in the true case , spike times from a given trial were assigned an instantaneous phase from that same time point in the same trial , in each permuted case we assigned to each spike time an instantaneous phase from the same time point in a different trial drawn at random from the pool of other trials . We permuted phase information 200 times resulting in a distribution of 200 surrogate r values for each cluster . We compared the true r value with the mean and standard deviation of the distribution of permuted r values to generate a normalized spike-phase locking value for each neuronal cluster ( R ) . To determine if individual frequencies exhibited significantly non-zero R values across all neuronal clusters , we used the same permutation procedure described above to correct for multiple comparisons . We first calculated the mean normalized spike-phase locking value , R , across all neuronal clusters . We then compared this true mean value to a distribution of 200 surrogate mean R values . Each of the 200 surrogates was generated by randomly either subtracting zero from a neuronal cluster’s R value or subtracting the cluster’s R value from the from zero in each cluser . We then recomputed the mean R value across clusters for each surrogate , and in each frequency , assigned a p value based on the comparison between the true mean R and the surrogate distribution . To correct for multiple comparisons across frequencies , we used the same cluster correction method based on exceedance mass testing ( Maris and Oostenveld , 2007 ) . In this case , clusters of significant frequencies were based on contiguous frequency points exhibiting significantly non-zero R values across neuronal clusters . Because only 10 of the 15 downward firing clusters had simultaneously recorded macroelectrodes within the borders of the STN , only these 10 clusters were included in the analysis . All 8 upward firing clusters had simulataneously recorded LFP electrodes within the STN , so all 8 clusters were used . We subsequently repeated the analysis separately for the target and distractor trials . The trial scrambling method used to generate the R values allowed us to normalize spike-phase locking values observed for a given trial type by the probability of observing spike-phase locking by chance given the distribution of phases and spike times observed during that trial type . In order to compare differences in spike-phase-locking between target and distractor trials across clusters , we used the same permutation procedure described above .	Should I run to catch the bus , or wait for the next one ? Deciding whether or not to move may be a common part of our life , but many decisions we make every day do not involve any physical activity , such as deciding whether something is important or not . But does the brain distinguish between decisions involving movement and those without ? Previous research has shown that a cluster of neurons called the basal ganglia and a region within these clusters called the subthalamic nucleus play a critical role in both movement and decision-making that involves activity . The basal ganglia are widely connected to other regions in the brain such as the prefrontal cortex , which is important for decision making and forming associated memories . Therefore , scientists think that the basal ganglia may also play a role in making decisions that do not involve movement , such as deciding whether to pay attention and form relevant memories . Indeed , in patients with Parkinson’s disease – a condition that damages parts of the brain and affects their movement – the degree of physical impairment correlates with their memory deficits . To test this , Zavala et al . recorded the activity of the subthalamic nucleus and the prefrontal cortex of patients with Parkinson's disease while they performed a task that required them to decide whether to remember something or not . The recordings were taken as the patients underwent ‘deep brain stimulation surgery’ , which is used to treat the symptoms of Parkinson’s . The results showed that brain activity in both regions decreased during the task . This is what normally happens when people make movements , but in this case , these decreases occurred even when individuals were not moving and instead just formed memories . In addition , individuals were presented with specific items to remember and others to forget . When they were presented with items to forget , the activity rebounded back to its original levels . Similar activity patterns have also been observed when individuals decided to stop a movement . This confirms that the subthalamic nucleus plays a role in decision-making and shows that this area is involved in decisions , even when they do not involve a movement . A full understanding of the purpose of the subthalamic nucleus and basal ganglia will help us understand why changes in the activity of basal ganglia leads to the memory and movement deficits seen in Parkinson’s disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Human subthalamic nucleus activity during non-motor decision making
We have reconstituted a eukaryotic leading/lagging strand replisome comprising 31 distinct polypeptides . This study identifies a process unprecedented in bacterial replisomes . While bacteria and phage simply recruit polymerases to the fork , we find that suppression mechanisms are used to position the distinct eukaryotic polymerases on their respective strands . Hence , Pol ε is active with CMG on the leading strand , but it is unable to function on the lagging strand , even when Pol δ is not present . Conversely , Pol δ-PCNA is the only enzyme capable of extending Okazaki fragments in the presence of Pols ε and α . We have shown earlier that Pol δ-PCNA is suppressed on the leading strand with CMG ( Georgescu et al . , 2014 ) . We propose that CMG , the 11-subunit helicase , is responsible for one or both of these suppression mechanisms that spatially control polymerase occupancy at the fork . Composition of the eukaryotic replisome and the function of its various proteins is an area of active investigation . Cellular studies reveal that eukaryotes use two different DNA polymerases for the leading and lagging strands , Pols ε and δ , respectively ( Lee et al . , 1989; Weinberg and Kelly , 1989; Tsurimoto et al . , 1990; Waga and Stillman , 1998; Benkovic et al . , 2001; Pursell et al . , 2007; Kunkel and Burgers , 2008; Nick McElhinny et al . , 2008; Stillman , 2008 ) . Priming is performed by Pol α , a 4-subunit enzyme that contains an RNA primase and DNA polymerase activity and makes short RNA-DNA hybrid primers of 25–35 nucleotides ( Kaguni and Lehman , 1988; Waga and Stillman , 1998; Benkovic et al . , 2001; Stillman , 2008 ) . The 11-subunit CMG complex consisting of the Mcm2-7 ‘motor’ , a GINS heterotetramer , and one Cdc45 subunit ( Moyer et al . , 2006; Ilves et al . , 2010; Costa et al . , 2011; Costa et al . , 2014 ) provides the helicase activity . Numerous other proteins travel with eukaryotic replication forks and have no bacterial homolog or known function . In addition , many replication fork-associated proteins undergo modifications in response to the cell cycle or DNA damage . While in vitro synthesis of the leading strand replisome has been accomplished with the purified CMG complex from budding yeast ( Georgescu et al . , 2014 ) , the discontinuous lagging strand is a much more difficult process and the number of proteins required for lagging strand synthesis is currently unknown . Indeed , epitope tagging of CMG subunits , followed by cell extract pull-outs and mass spectrometry , has identified a large ‘replisome progression complex’ ( RPC ) that contains CMG along with several other factors , some of which are essential for cell viability ( Gambus et al . , 2006; Gambus et al . , 2009 ) . Thus , RPCs contain CMG along with Mcm10 , Ctf4 , Pol α , Mrc1 , Csm3 , Tof1 , FACT , and Topo I . Furthermore , there is evidence that nucleosomes may be involved in determining the size of eukaryotic lagging strand fragments ( Smith and Whitehouse , 2012 ) . We demonstrate here a 31 protein system , requiring the three replicative Pols α , ε , and δ , which performs both leading and lagging strand synthesis and generates Okazaki fragments of a size similar to those observed in cells . Study of the eukaryotic replisome identifies a new process that has no precedent in bacterial systems . Bacteria use simple recruitment processes to attract and hold polymerases to the fork . These are typically mediated by polymerase interactions with other proteins at the replication fork , such as the helicase and sliding clamps ( Benkovic et al . , 2001 ) . However , we find that in addition to recruitment processes that attract polymerases to the fork , eukaryotes use suppressive mechanisms , which prevent polymerase action on one strand or the other . Thus , while Pol ε extends the leading strand , its activity is suppressed on the lagging strand , even in the absence of Pol δ . In opposite fashion , we demonstrate here that Pol δ is active on the lagging strand in the presence of Pol ε and Pol α . This activity stands in contrast to the suppression of Pol δ on the leading strand shown in our earlier report ( Georgescu et al . , 2014 ) . We also find that Pol α functions with CMG on both the leading and lagging strands . However , Pol α lacks a proofreading exonuclease and thus has lower fidelity than Pols ε and δ . Interestingly , Pol α extension activity is suppressed by Pol ε on the lagging strand , even though Pol ε is inactive on the lagging strand . Likewise , Pol δ suppresses Pol α on the leading strand despite its inefficient extension of this strand . Thus , multiple suppression reactions exist that prevent the activity of a polymerase positioned on the ‘wrong’ strand , and the only active solution is the asymmetric Pol ε/δ leading–lagging strand replisome with use of Pol α to prime the strands . Pol α is the eukaryotic primase and thus is required for lagging strand studies . To obtain the 4-subunit Pol α , we reconstituted it by expressing the Pol1 polymerase subunit of Pol α in yeast and Pol12 and the primase subunits , Pri1 and Pri2 , in Escherichia coli . A mixture of cells containing the 4 subunits were lysed , and Pol α was purified as an intact 4-subunit holoenzyme ( Figure 1—figure supplement 1 ) . We then studied the behavior of Pol α with the 11-subunit CMG helicase and RPA on the forked DNA substrate . We initially assumed the lagging strand–specific Pol α would not function on the leading strand with CMG , especially since Pol δ-PCNA did not function well on the leading strand with CMG in our earlier study ( Georgescu et al . , 2014 ) . To test Pol α function with CMG , we primed the leading strand with an oligonucleotide and examined Pol α DNA synthesis with CMG using 32P-dCTP ( see scheme in Figure 1B ) . In contrast to our expectations , Pol α was highly active with CMG and completely extended the leading strand ( Figure 1B , lanes 4–6 ) . Dropout reactions demonstrated that CMG is absolutely required and therefore Pol α cannot perform this action alone ( Figure 1—figure supplement 2 ) . Models of priming at bidirectional origins in the SV40 and bacterial systems indicate that leading strand primers are formed on the lagging strand of one fork and then extended from the leading strand of the opposite fork ( Waga and Stillman , 1998; Méndez and Stillman , 2003; Kaguni , 2011; O'Donnell et al . , 2013 ) . The current study uses a unidirectional linear fork and thus has no opposite fork from which to prime the leading strand . Therefore , we did not expect to observe leading strand synthesis using an unprimed leading strand fork . However , Pol α was fully active on unprimed DNA and thus the primase within Pol α is capable of priming the leading strand directly , while the polymerase subunit of Pol α is able to extend it ( Figure 1B , lanes 1–3 ) . We support and expand this observation again later in Figure 2 . 10 . 7554/eLife . 04988 . 006Figure 2 . Pol α requires CMG for priming activity during unwinding of forked DNA . ( A ) Scheme of assays comparing Pol α activity using either CMG helicase or the strand displacing ϕ29 polymerase . ( B ) Autoradiograph of DNA products using either 32P-dCTP ( leading ) or 32P-dGTP ( lagging ) . Use of a DNA-primed leading strand fork ( PF ) or an unprimed fork ( UF ) is indicated in the figure . Pol α was present at 10 nM , and reactions were for 20 min . Lanes 1 and 2 represent control reactions of ϕ29 polymerase alone . DOI: http://dx . doi . org/10 . 7554/eLife . 04988 . 006 A titration of Pol α into reactions with CMG shows an increasing rate of leading strand elongation as the Pol α concentration is raised , indicating that DNA synthesis by Pol α is distributive ( autoradiograph illustrated in Figure 1C left panel and quantified on the right panel ) . The leading strand is continuously extended to full-length product; consequently , once Pol α has switched to the DNA elongation mode , it does not appear to switch back to the priming mode . This behavior has not previously been observed at a moving fork with CMG . Moreover , there are reports to suggest that the eukaryotic fork is discontinuous on both strands ( reviewed in [Langston and O'Donnell , 2006] ) . The results of Figure 1 , however , indicate that the leading strand of the eukaryotic replication fork is synthesized continuously under our assay conditions . To determine whether this minimal replication system is competent to prime and extend the lagging strand , we utilized 32P-dGTP to specifically label lagging strand products . We assembled CMG on the forked DNA , then titrated Pol α into the reaction in the presence of RPA , dNTPs , and rNTPs . We expected to observe the small hybrid RNA/DNA 25–35 nucleotide primers known to be generated by Pol α ( Kaguni and Lehman , 1988; Waga and Stillman , 1998; Benkovic et al . , 2001; Stillman , 2008 ) . However , we were surprised to observe sizeable Okazaki fragments ( autoradiograph illustrated in Figure 1D left panel and quantified on the right panel ) . At 10–20 nM Pol α , a concentration consistent with intracellular concentrations of Pol α subunits ( Ghaemmaghami et al . , 2003; Kulak et al . , 2014 ) , Okazaki fragments were in the range of 200 bp , similar to their length in vivo ( Smith and Whitehouse , 2012 ) . The fact that the Okazaki fragment size decreases with increasing Pol α concentration indicates that priming is stochastic , occurring with lower frequency as the Pol α concentration is decreased ( Figure 1D ) . These results demonstrate that small Okazaki fragments are intrinsic to Pol α action with CMG in the absence of nucleosomes . To determine if CMG is needed for Pol α priming function , we used ϕ29 DNA Pol , an efficient strand displacing enzyme ( Blanco et al . , 1989 ) , to perform leading strand synthesis and generate a lagging single strand ( ss ) without CMG and asked whether Pol α primase could prime the lagging strand ssDNA template and produce Okazaki fragments with RPA present ( illustrated in Figure 2A ) . The control shows ϕ29 Pol efficiently extends a primer through the 3 . 2 kb duplex ( Figure 2B , lanes 1 and 5 ) and thus produces an ssDNA lagging strand . However , the lagging strand ssDNA did not serve as a template to generate Okazaki fragments when Pol α and RPA were present ( Figure 2B , lane 6 ) . In contrast , lagging strand synthesis by Pol α with CMG supports robust Okazaki fragment production by Pol α ( lane 4 ) . The requirement of CMG for Pol α primase activity suggests that Pol α forms a transient but specific interaction with CMG for priming action , since use of a heterologous enzyme to unwind DNA does not support Pol α priming activity . Although both Ctf4 and the essential Mcm10 proteins are known to bind Pol α ( Gambus et al . , 2009; Warren et al . , 2009 ) , these results demonstrate that neither Ctf4 nor Mcm10 are required for Pol α priming function with CMG in vitro . To support this observation , we examined our protein preparations for trace contaminating Ctf4 and/or Mcm10 by mass spectrometry , but no Ctf4 or Mcm10 was detected . Consistent with results of Figure 1B , Pol α also primes the leading strand directly and priming is stimulated by the presence of CMG ( compare lanes 3 and 7 ) . The results suggest Pol α binds CMG in the absence of Ctf4 or Mcm10 and therefore we tried to pull down a complex of Pol α with CMG but did not obtain a positive result . Thus the interaction , if it exists , may be weak , consistent with weak helicase–primase interactions in phage T4 and E . coli replication systems that can be deduced by activity assays but elude direct detection methods ( Benkovic et al . , 2001 ) . To determine if Okazaki fragments are distributed over the length of the DNA template , we used 32P-dGTP to label Okazaki fragments and then analyzed the distribution of radioactivity across the DNA by restriction enzyme analysis in a native gel ( Figure 3 ) . The analysis using CMG and Pol α shows that all the restriction fragments are radioactive and therefore Okazaki fragments are synthesized along the entire length of the DNA ( lanes 4–9 ) . To create size markers , ϕ29 Pol was used to extend the leading strand using 32P-dCTP , followed by restriction digestion ( lanes 10–12 ) . The analysis also confirms that Pol α cannot perform priming and extension in the absence of CMG ( lanes 1–3 ) and that ϕ29 Pol cannot perform lagging strand synthesis ( lanes 13–15 ) . 10 . 7554/eLife . 04988 . 007Figure 3 . Okazaki Fragments are produced along the entire DNA . ( A ) Restriction enzyme map of the 3 . 2 kb substrate for Psi I and Ear I . ( B ) Lagging strand reactions were performed as detailed in ‘Materials and methods’ using an unprimed forked DNA , CMG , RPA , and either Pol α ( lanes 4–6 ) or Pol α and Pol ε ( lanes 7–9 ) , then were either untreated ( lanes 4 , 7 ) , treated with Psi I ( lanes 5 , 8 ) , or treated with Ear I ( lanes 6 , 9 ) . A control leading strand reaction using only ϕ29 Pol is shown in lanes 10–12 . Pol α without CMG ( lanes 1–3 ) and ϕ29 alone ( lanes 13–15 ) gave no lagging strand products . The ( * ) mark incomplete digestion products . The reaction products were analyzed on a native 2% agarose gel . DOI: http://dx . doi . org/10 . 7554/eLife . 04988 . 007 Pol ε is the leading strand enzyme and presumably takes over the leading strand from Pol α after this distributive enzyme dissociates from DNA . To determine if the Pol α/ε switch occurs as expected , we preloaded CMG on the forked DNA , then added increasing amounts of Pol α either with or without Pol ε , and stopped the reactions after 20 min . If Pol ε takes over the leading strand from Pol α , full-length products will be observed sooner in reactions that contain Pol ε because in the presence of CMG , this enzyme synthesizes DNA faster than Pol α . The results show full-length product in all the lanes containing Pol ε with Pol α ( autoradiograph in Figure 4A left , compare lanes 1–4 with 5–8; the quantification on the right panel is based on the autoradiograph analysis shown in Figure 4—figure supplement 1A ) . These observations indicate that after Pol α primes the DNA , Pol ε takes over and rapidly extends the leading strand . The reactions lack RFC/PCNA , but the presence of RFC/PCNA does not alter the outcome ( Figure 4—figure supplement 2 ) . At sufficiently high concentrations , RFC/PCNA competes with Pol α and suppress its extension activity with CMG , as previously reported without CMG ( Mossi et al . , 2000 ) ( Figure 4B and quantification analysis shown in Figure 4—figure supplement 1B ) . The results show that RFC/PCNA also inhibits Pol α polymerase extension of primers on the lagging strand . 10 . 7554/eLife . 04988 . 008Figure 4 . Pol ε switches with Pol α on the leading strand but is not active on the lagging strand . ( A ) Left: scheme of the assay . Middle: titration of Pol α into leading strand reactions in the absence of Pol ε ( lanes 1–4 ) or in the presence of 20 nM Pol ε ( lanes 5–8 ) . The reactions were 20 min and contained unprimed DNA fork template . Right: histogram illustrating total DNA synthesis obtained from Typhoon laser scan analysis in Figure 4—figure supplement 1A ( the error bars represent Standard Fit Errors obtained from the Gaussian fit analysis ) . ( B ) Left: scheme of the assay . Titration of RFC-PCNA into a primed leading strand assay containing 10 nM Pol α with or without 10 nM Pol ε . Right: RFC-PCNA inhibits Pol α ( lanes 1–5 ) probably by competing for the 3′ terminus as illustrated . When present , Pol ε rapidly extends the leading strand and is not inhibited by RFC-PCNA ( lanes 6–10 ) . Replication reactions were performed in the presence of 32P-dCTP for 15 min . Right: histogram illustrating total DNA synthesis obtained from Typhoon laser scan analysis in Figure 4—figure supplement 1B ( the error bars represent Standard Fit Errors obtained from the Gaussian fit analysis ) . ( C ) Left: leading and lagging strand synthesis is monitored in the same reaction plus or minus Pol ε . Each reaction was divided to separately monitor either the leading ( 32P-dCTP ) or lagging ( 32P-dGTP ) strand . Pol ε is absent in the reaction of lanes 1 and 2 , and Pol ε is present in the reaction of lanes 3 and 4 . Right: histogram illustrating total DNA synthesis obtained from Typhoon laser scan analysis in Figure 4—figure supplement 1C ( the error bars represent Standard Fit Errors obtained from the Gaussian fit analysis ) . Lane analysis of the autoradiographs from panels A , B , and C are shown in Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04988 . 00810 . 7554/eLife . 04988 . 009Figure 4—figure supplement 1 . Analysis of replication products in Pol α and Pol α−ε titrations illustrated in Figure 4A–C . The autoradiographs shown in Figure 4 were analyzed by Typhoon laser scanner , and the lane profiles were normalized to the corresponding molecular weight at each pixel , as previously described ( Kurth et al . , 2013 ) . This corrects for the fact that longer products incorporate more radiolabel ( α32P-dCTP or α32P-dGTP ) than shorter products . Each line trace was fit to a single or multiple Gaussian functions , shown as a thin dashed line in each scan . For ease of understanding , each of the panels A , B , and C correspond to the autoradiographs displayed in Figure 4A–4C , respectively ( a cutout of the gels in Figure 4 are inserted , for ease of identification ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04988 . 00910 . 7554/eLife . 04988 . 010Figure 4—figure supplement 2 . Pol ε excludes Pol α from the leading strand by taking over the primer whether RFC and PCNA are present or not . ( A ) Autoradiogram of CMG mediated Pol α extension of the leading strand on the unprimed 3 . 2 kb forked DNA , in the presence or absence of Pol ε , as indicated in the figure . Reactions were performed for the indicated times and otherwise performed as described in ‘Materials and methods’ . The gel lanes were analyzed by Typhoon laser scanner ( as in Figures 1 , 4 ) , and the plot on the right side quantitates the progression of DNA product length at the times indicated in the gel . The numbers in the plots represent the rate of the Pol α ( blue ) or Pol α-ε ( red ) obtained from the linear fit of the data points . ( B ) Autoradiogram of reactions performed as described in panel A , except 20 nM each of RFC and PCNA were present in all reactions . The plot on the right quantitates the data as described for panel A . DOI: http://dx . doi . org/10 . 7554/eLife . 04988 . 01010 . 7554/eLife . 04988 . 011Figure 4—figure supplement 3 . CMG and Pol ε form a stable CMGE complex . ( A ) Scheme of the bead assay to detect protein interactions . CMG-Pol ε complex was bound to Strep-Tactin magnetic beads via a strep-Pol ε , then washed and eluted with biotin; products were analyzed by 8% SDS-PAGE . Mixtures of proteins were as described in ‘Materials and methods’ and included 80 pmol strep-Pol ε and 120 pmol CMG . ( B ) SDS-Coomassie Blue stained PAGE of eluted protein complexes . The left lane , labeled ‘BMW markers’ represent the Broad Molecular Weight markers ( Bio-Rad ) ; lanes 1 and 2 show the protein preps used in the analysis . The location of each subunit is indicated . Cdc45 and Dpb2 co-migrate . The ‘*’ marks an impurity present in the strep-Pol ε prep . Lane 3 illustrates the biotin elution of the CMGE complex . Lane 4 represents the negative control in which the same amounts of CMG was mixed , in the absence of strep-Pol ε , then incubated with the beads , washed and eluted with biotin . The clear spaces between the last two lanes indicate these two lanes were taken from the same gel but were not adjacent to one another . Lanes 5 and 6 are a higher contrast of the lanes 3 and 4 to more clearly show the Sld5 and Psf1 subunits . DOI: http://dx . doi . org/10 . 7554/eLife . 04988 . 011 The ability of Pol ε to suppress Pol α function with CMG might be facilitated by direct interaction of Pol ε with CMG , thus holding Pol ε at the leading strand primer terminus and preventing Pol α from binding . The Dpb2 subunit of Pol ε is known to bind the Psf1 subunit of GINS ( Sengupta et al . , 2013 ) , and we have previously used a glycerol gradient to demonstrate that intact Pol ε can bind CMG helicase , forming a 1:1 CMG-Pol ε ( CMGE ) complex ( Langston et al . , 2014 ) . In Figure 4—figure supplement 3 , we document that the CMGE complex can also be reconstituted in a bead-based protein binding assay . In these experiments , we purified Pol ε containing an N-terminal StrepTag , incubated it with CMG , and then added Strep-Tactin magnetic beads . Biotin-specific elution of CMG in complex with StrepTag-Pol ε demonstrates that CMG is retained by Pol ε , forming the 15-protein ‘CMGE’ leading strand complex . No CMG was eluted from the column in the absence of Pol ε . Considering that Pol ε takes over the primed template from Pol α on the leading strand , one may presume that Pol ε will also take over from Pol α on the lagging strand . In Figure 4C , we separately monitored leading and lagging strand synthesis in Pol α/CMG reactions in the absence or presence of Pol ε . The first two lanes are Pol α/CMG on the leading ( lane 1 ) and lagging ( lane 2 ) strands . At the concentration used , Pol α extended the leading strand DNA to about half of full length ( lane 1 ) , and the signal for Okazaki fragment synthesis was quite strong ( lane 2 ) . When Pol ε was added , it took over the leading strand and extended it full length ( Figure 4C , lane 3 ) ; however , lagging strand Okazaki fragments were inhibited , a result we did not expect ( lane 4 ) . We also tested a higher concentration of Pol ε to see if more Pol ε was required to efficiently extend the Okazaki fragments , but lagging strand synthesis was not enhanced ( Figure 4C , lane 6 ) ( for quantification see Figure 4C right panel and Figure 4—figure supplement 1C ) . This result is contrary to conventional wisdom because one would normally expect more synthesis upon addition of more polymerase . This is especially true considering that Pol ε enhances the leading strand in the very same reaction in which it inhibits the lagging strand ( i . e . , the only difference is the radioisotope ) . The results of Figure 4C indicate that Pol ε activity is suppressed on the lagging strand . Moreover , Pol ε suppresses Pol α , suggesting that Pol ε gains access to lagging strand primed sites but is inactive on them . Considering that Pol ε extends DNA on RPA-coated primed ssDNA in the absence of CMG ( Wold , 1997; Garg and Burgers , 2005; Georgescu et al . , 2014 ) , it is possible that CMG controls the activity of Pol ε on the lagging strand . We further supported this observation in Figure 5A by titrating Pol ε into Pol α priming reactions and monitoring lagging strand synthesis . As Pol ε is titrated into the reaction , Okazaki fragment synthesis is inhibited , confirming that Pol ε suppresses Pol α polymerase but is unable to extend lagging strand primers ( see quantification in Figure 5A , right panel ) . Although Pol ε inhibits Pol α on the lagging strand , a combination of Pol ε and RFC/PCNA inhibits Pol α more than either Pol ε or RFC/PCNA alone ( Figure 5—figure supplement 1A , lanes 1–2 and 4–7 vs lanes 8–11; quantification shown in Figure 5—figure supplement 1B ) . Inhibition of Pol α synthesis by RFC/PCNA is also consistent with results in the SV40 system ( Tsurimoto and Stillman , 1991 ) and model studies using primed ssDNA ( Figure 5—figure supplement 2 ) ( Mossi et al . , 2000 ) . Hence , RFC/PCNA contributes to the observed inhibition of Pol α synthesis in addition to Pol ε . 10 . 7554/eLife . 04988 . 012Figure 5 . Pol δ functions on the ‘Pol ε suppressed’ lagging strand . ( A ) Titration of Pol ε into lagging strand reactions containing Pol α/CMG results in inhibition of the lagging strand in the absence of RFC-PCNA . Similar reactions containing RFC-PCNA give even more inhibition on the lagging strand ( Figure 5—figure supplement 1 ) . Reactions were for 20 min . ( B ) Pol δ is titrated into lagging strand reactions containing Pol α , Pol ε , RFC-PCNA , and CMG under conditions in which Pol ε and RFC-PCNA inhibit lagging strand synthesis . Lagging strand reactions ( 32P-dGTP ) contain 10 nM each Pol α and Pol ε ( lanes 1–4 ) , or 20 nM each Pol α and Pol ε ( lanes 5–8 ) ; RFC-PCNA are at 20 nM each . CMG concentration was 24 nM in all reactions . Lanes 9 and 10 are controls with no CMG but contain 20 nM each of Pol α , Pol ε , RFC , PCNA , and either no Pol δ ( lane 9 ) or 20 nM Pol δ ( lane 10 ) . Reactions were for 20 min . The plots on the right of panels A and B represent quantifications of lagging strand replication reactions ( using α−32P-dGTP ) as described in the legend of Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 04988 . 01210 . 7554/eLife . 04988 . 013Figure 5—figure supplement 1 . Pol ε and RFC/PCNA inhibit Pol α DNA polymerase on the lagging strand , but Pol ε cannot extend primed sites with or without RFC/PCNA . ( A ) Autoradiograph of lagging strand replication reactions using α32P-dGTP . CMG is pre-incubated and loaded on unprimed forked DNA , as described in ‘Materials and methods’ , then either 10 nM Pol α ( lane 1 ) , 10 nM each Pol α and Pol ε ( lane 2 ) or only 10 nM Pol ε ( lane 3 ) are added along with RPA , dNTPs , and rNTPs; RFC-PCNA are absent . The Pol ε/CMG control shows no product synthesis on the lagging strand , as expected . The analysis of DNA products in lanes 1 and 2 show a 2 . 1 fold decrease in total DNA synthesis for the Pol α/Pol ε reaction relative to Pol α alone . Next , RFC-PCNA was titrated into reactions containing 10 nM Pol α with no Pol ε ( lanes 4–7 ) , or 10 nM each of Pol α and Pol ε ( lanes 8–11 ) . Reactions were 20 min . ( B ) Histogram illustrating total DNA synthesis obtained from Typhoon laser scan analysis of the autoradiograph in panel A ( the error bars represent Standard Fit Errors obtained from the Gaussian fit analysis ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04988 . 01310 . 7554/eLife . 04988 . 014Figure 5—figure supplement 2 . RFC and PCNA inhibit Pol α DNA polymerase activity on ssDNA model templates . Top: scheme of the assay . Primer extension assays utilized RPA-coated singly primed 5 . 4 kb φX174 ssDNA . RFC-PCNA is titrated into singly primed ssDNA replication assays containing 10 nM Pol α as described in ‘Materials and methods’ . Concentrations of RFC , PCNA , and reaction times are indicated in the figure . DOI: http://dx . doi . org/10 . 7554/eLife . 04988 . 014 Our earlier study demonstrated that Pol δ-PCNA was slow and distributive on the leading strand in the presence of CMG and at 50 mM potassium glutamate ( Georgescu et al . , 2014 ) . However , Pol δ-PCNA is highly processive during synthesis of the 5 . 4 kb primed RPA-coated ϕX174 ssDNA under the 50 mM potassium glutamate conditions used in the 3 kb replisome assays ( Langston and O'Donnell , 2008 ) . Pol δ-PCNA has been shown to be much less processive in reactions containing 125 mM added NaOAc ( Chilkova et al . , 2007 ) . A high ionic strength is known to decrease processivity of enzymes , as the highly processive bacterial Pol III-β clamp replicase is decreased significantly at 100 mM NaCl ( Griep and McHenry , 1989 ) . Potassium glutamate is the physiological osmolyte in E . coli ( Richey et al . , 1987 ) , but to the authors' knowledge , neither the intracellular ionic strength nor the major intracellular osmolytes are known for budding yeast . We do not know why Pol δ-PCNA is not very active on the leading strand with CMG . However , we have shown in this report that RFC/PCNA competes with Pol α on the leading strand ( Figure 4B , lanes 1–5 ) , indicating that PCNA is in fact loaded onto the leading strand . We demonstrate in Figure 6A that Pol δ added to the reaction exacerbates inhibition of Pol α in the presence of RFC/PCNA , while both Pol α and Pol ε are active in the presence of CMG and the amount of RFC/PCNA used ( see quantifications in Figure 6B , C ) . Hence , Pol δ-PCNA does not function with CMG and severely competes and limits the function of Pol α with CMG . The addition of Pol δ to reactions containing Pols α and ε shows no significant effect on the leading strand reactions ( Figure 6—figure supplement 1 ) . The result of this three polymerase reaction is consistent with and anticipated from our earlier study of leading synthesis using primed forks ( i . e . , no Pol α ) in which we demonstrate that Pol δ is inefficient and distributive on the leading strand and that Pol ε suppresses Pol δ by taking over the 3′ primed site in the CMG-dependent leading strand reaction ( Georgescu et al . , 2014 ) . 10 . 7554/eLife . 04988 . 015Figure 6 . Polymerases switch with Pol α on the leading strand . ( A ) Alkaline agarose gel following the time course of leading strand extension using the indicated DNA polymerases . Reactions were assembled on unprimed forked DNA in presence of 24 nM CMG for 10 min before adding 15 nM Pol α ( lanes 1–5 ) , 15 nM each Pol α and Pol ε ( lanes 6–10 ) , and 15 nM each Pol α and Pol δ ( lanes 11–15 ) ; all reactions contained 6 nM RFC and 20 nM PCNA . Reactions were initiated upon adding RPA and nucleotides as described in ‘Materials and methods’ . The rates of Pol α reactions are high in this experiment because the amount of Pol α used here promotes relatively rapid fork progression as documented in Figure 1 and Figure 1—figure supplement 1 . Still , the addition of Pol ε gives slightly faster forks due to the intrinsically faster rate of CMG-Pol ε over the rate of the distributive Pol α with CMG . ( B ) Autoradiograph quantification as described in the legend to Figure 4 . ( C ) The analysis of DNA products at the end-point reaction ( 25 min ) reveals a 1 . 8 fold increase in total DNA synthesis for the Pol α/Pol ε reaction relative to Pol α alone ( lane 10 vs lane 5 ) ; the same comparison of total DNA synthesis in Pol α vs the Pol α/Pol δ reaction reveals a 3 . 2 fold decrease in total DNA synthesis ( lane 15 vs 5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04988 . 01510 . 7554/eLife . 04988 . 016Figure 6—figure supplement 1 . Pol δ does not inhibit the leading strand replication activity of Pol ε when all three polymerases are present . ( A ) Autoradiographs of leading strand replication reactions using α32P-dCTP ( left ) or 32P-5′ labeled primed fork ( right ) . CMG is pre-incubated and loaded on DNA , as described in ‘Materials and methods’ , then the different polymerases are added into the reaction at the specified concentrations along with RFC-PCNA , RPA , dNTPs , and rNTPs . There is a slight end-labeling product observed in the 32P-dCTP experiment ( 20′ control time points ) , and therefore , we repeated these reactions using the 32P-5′ labeled primed fork . At high CMG concentration and in the presence of all three polymerases , there is a very efficient replication of the forked DNA template . a , b and c mark the location of 32P-primers that are not extended , location of 32P-primers extended up to the fork junction , and location of a minority of 32P-primer extension by strand displacement activity of Pol δ in 20 min control reactions , respectively . ( B ) Pol ε functions on the ‘Pol δ suppressed’ leading strand . Left: scheme of the reaction . Right: Pol ε is titrated into leading strand reactions containing Pol α , Pol δ , RFC-PCNA , and CMG , and under these conditions , the Pol δ and RFC-PCNA inhibit leading strand synthesis by Pol α . The leading strand reactions contain 10 nM RFC , 20 nM PCNA , and the specified amounts of polymerases and CMG . DOI: http://dx . doi . org/10 . 7554/eLife . 04988 . 016 Next , we tested whether Pol δ-PCNA could function on the lagging strand with CMG in the presence of Pol ε , Pol α , RFC , PCNA , and RPA , conditions in which Pol ε extends the leading strand with CMG but suppresses Okazaki fragment extension by Pol α . In Figure 5B , Pol δ was titrated into lagging strand reactions , and surprisingly , Pol δ was capable of extending Okazaki fragments under these circumstances . In fact , it is the only polymerase that we observed to perform this function . Hence , CMG does not prevent Pol δ function on the lagging strand , even while Pol δ-PCNA activity is suppressed on the leading strand . In order to compare total leading and lagging strand synthesis rates , we setup standard replication reactions by loading CMG onto the nucleotide-biased forked DNA , followed by the addition of all three DNA polymerases in the presence of RFC and PCNA; the reactions were divided , and replication was initiated by the addition of either 32P-dCTP or 32P-dGTP along with ATP , RPA , dNTPs , and rNTPS ( see ‘Materials and methods’ for experimental details ) . Quantification of leading and lagging strand synthesis using either 32P-dCTP ( leading ) or 32P-dGTP ( lagging ) in the three polymerase replisome system shows similar amounts of leading and lagging strand synthesis ( Figure 7 ) . We tested the rate of dNTP incorporation using primed as well as unprimed forked DNA templates . While leading and lagging strand rates are similar using either DNA forked substrate , there is 20–25% more synthesis using primed DNA forks relative to unprimed DNA forks . We presume this difference reflects the additional time required for Pol α to prime the leading strand on the unprimed fork DNA , relative to use of forks that are pre-primed with a DNA oligonucleotide ( i . e . , primed forks ) . 10 . 7554/eLife . 04988 . 017Figure 7 . The leading and lagging strands are replicated at similar rates . ( A ) Time course of leading–lagging strand replication reactions with all three polymerases at 10 nM each using either a pre-primed fork ( Left panel ) or unprimed fork ( Right panel ) . Experiments were performed in triplicate , using either 32Pα-dCTP or 32Pα-dGTP for leading–lagging replication reactions , respectively ( for experimental details , see ‘Materials and methods’ section ) . The numbers shown represent the rate of incorporation ( fmol dNTPs/s ) and the SE obtained from the linear fit . ( B ) Left: comparative histogram depicting total DNA synthesis at the 20 min time point; Right: comparative histogram of the rate of incorporation of leading–lagging strand replication reactions on pre-primed and unprimed forked DNA substrates . DOI: http://dx . doi . org/10 . 7554/eLife . 04988 . 017 The current report reveals that suppression reactions specify correct polymerase placement at the fork and that when a polymerase occupies the ‘wrong strand’ it is excluded from functioning . This stands in contrast to current views in which the mechanism of polymerase placement is thought to be via recruitment ( i . e . , each polymerase binds a particular protein on each strand ) . Indeed , polymerase recruitment by binding clamps and the helicase underlies attraction of polymerases to replication forks of bacteria , its phages , and the SV40 virus . Recruitment is also partly responsible for polymerase placement in eukaryotes . For example , Pol ε binds directly to CMG helicase , stabilizing it on the leading strand ( Langston et al . , 2014 ) . However , the current study reveals the importance of suppression of polymerase action to the specific use of Pol ε and Pol δ on the leading and lagging strands , respectively . Thus , Pol ε function is suppressed on the lagging strand , even when Pol δ is not present . Likewise , our earlier study , confirmed here , showed that Pol δ-PCNA function is suppressed on the leading strand , even in the absence of Pol ε ( Georgescu et al . , 2014 ) . We were surprised to find that Pol α polymerase activity is highly functional with CMG on both leading and lagging strands in the absence of other polymerases . Pol α lacks the high fidelity of Pols ε and δ and does not provide bulk leading or lagging strand synthesis in cells , and thus processes must exist that suppress the polymerase activity of Pol α . In fact , we find many ways that Pol α polymerase is suppressed . One mechanism is by Pol ε positioning on the lagging strand . Interestingly , Pol ε is also suppressed on the lagging strand; perhaps , Pol ε is suppressed from function on the lagging strand by the relative orientation in which CMG holds Pol ε . On the leading strand , Pol ε simply prevents Pol α polymerase extension by switching with it and becoming processive with CMG . This can be likened to the switch of Pol δ with Pol α that prevents leading strand synthesis by Pol α in the SV40 system ( Tsurimoto and Stillman , 1991 ) . Pol α polymerase activity is also inhibited by RFC/PCNA as shown in this report and an earlier study ( Mossi et al . , 2000 ) . The exclusion processes that underlie eukaryotic fork leading/lagging strand function are summarized in Figure 8 . Pol α primes both leading and lagging strands ( diagram A ) . Pol ε prevents further polymerase activity of Pol α by trading places with it , probably by waiting for the distributive Pol α to dissociate from DNA ( diagram B ) ; Pol ε then takes over the leading strand . Diagram C illustrates that Pol δ/RFC/PCNA also switch with Pol α but that Pol δ-PCNA cannot function in the presence of CMG on the leading strand . However , Pol δ-PCNA is uniquely capable of extending Okazaki fragments in the complete system ( diagram D ) . Hence , the three-polymerase-CMG replisome is the product of polymerase suppression reactions that enable a unique asymmetric arrangement of DNA polymerases to advance the replication fork . 10 . 7554/eLife . 04988 . 018Figure 8 . Exclusion reactions specify polymerase action at the eukaryotic fork . ( A ) Pol α interacts with CMG to prime the leading and lagging strands . Pol α can extend DNA on both strands with CMG in vitro , but it lacks high fidelity and does not replicate bulk DNA in vivo . ( B ) Pol ε can switch with Pol α on both strands , in the presence or absence of RFC/PCNA , but Pol ε is not active on the lagging strand . ( C ) Pol δ/RFC/PCNA can switch with Pol α on both strands , but Pol δ is inactive with CMG on the leading strand . ( D ) Presence of all three polymerases , Pols α , δ , and ε , provides active leading/lagging strand synthesis . DOI: http://dx . doi . org/10 . 7554/eLife . 04988 . 018 Yeast Pol δ is highly processive with PCNA on primed ssDNA under the conditions of this report ( Langston and O'Donnell , 2008 ) , yet Pol δ-PCNA is not effective in extending the leading strand with CMG helicase . The distributive behavior of Pol δ-PCNA with CMG suggests that CMG exerts a negative effect upon Pol δ , causing it to frequently dissociate from PCNA/DNA . A possible explanation for how CMG may cause distributive behavior of Pol δ with PCNA/DNA lies in our earlier finding that Pol δ self-ejects from PCNA upon completing replication of a template ( Langston and O'Donnell , 2008 ) . This property was originally noted for the highly processive E . coli Pol III replicase ( O'Donnell , 1987; Studwell et al . , 1989 ) . Pol III remains firmly attached to the β clamp until reaching the last nucleotide ( Stukenberg et al . , 1994 ) . This also holds true for T4 polymerase upon colliding with a hairpin ( Hacker and Alberts , 1994 ) . Self-ejection of polymerase is thought to be useful for recycling among numerous lagging strand primers ( Benkovic et al . , 2001 ) . In eukaryotes , however , the self-ejection process may be used to promote polymerase asymmetry at the fork . A straightforward mechanism could be that CMG occludes the leading strand ssDNA , tricking Pol δ-PCNA to eject as if it were at the end of an Okazaki fragment . If CMG were to trigger the Pol δ−PCNA self-ejection process , it would result in distributive behavior of Pol δ-PCNA , suppressing its activity on the leading strand . Another possible mechanism could be that the 3′ terminus of the leading strand is sequestered and held in a posture that Pol δ cannot freely access . The suppression of Pol ε on the lagging strand could possibly result from the geometry of CMG-Pol ε ( CMGE ) complex , directing the single Pol ε molecule to the leading strand . Suppression of Pol α by Pol ε on the lagging strand could perhaps be explained by competition of these polymerases for CMG , where Pol ε is suppressed from extending the primer . In this connection , the Pol2 gene encoding the catalytic polymerase actually contains two polymerase structures: the active polymerase/exonuclease on the N-terminal half and B-family polymerase in the C-terminal half of Pol2 that is presumed to be inactive ( Tahirov et al . , 2009 ) . One possible mechanism by which Pol ε may suppress synthesis on the lagging strand could be that CMG positions Pol ε such that the ‘inactive polymerase’ binds lagging strand primers , preventing their elongation by Pol α and Pol ε . Interestingly , genetic studies have shown that the C-terminal ‘inactive polymerase’ region of Pol ε is required for cell viability , while the N-terminal region containing the active polymerase is dispensable ( Dua et al . , 1999; Kesti et al . , 1999 ) . The binding of Pol ε to CMG involves Dpb2 , and possibly the Dpb3 , Dpb4 subunits of Pol ε , and regions of Pol2 as well . Expression and purification of a mutant Pol ε that lacks the active N-terminal half of Pol2 ( ΔN-Pol ε ) yields a 4-subunit complex , while the active polymerase half of Pol ε is monomeric ( Hogg et al . , 2014 ) . Hence , the ΔN-Pol ε probably connects to CMG similar to wild-type Pol ε and may carry out a vital function at the fork . Alternatively , the C-terminal portion of Pol2 could be important to an origin activation step prior to replisome formation . Although direct evidence is lacking , Pol δ is presumed to function on the leading strand in the ΔN-Pol ε genetic background . However , the finding in this report that Pol α functions on the leading strand with CMG while Pol δ-PCNA is suppressed leaves open the possibility that Pol α could participate as the leading strand polymerase in ΔN-Pol ε mutants . Numerous proteins travel with the replication fork , and it is not possible to know a priori how many are needed for functional leading and lagging strand replication . This is the first study to reconstitute leading/lagging strand replication with three pure polymerases in a eukaryotic system . It reveals that many of the proteins that move with forks are not required to recapitulate leading/lagging strand synthesis in vitro . For example , Ctf4/AND-1 is essential in most cells ( not in budding yeast ) , and yeast Ctf4 helps recruit Pol α bind to RPCs ( Gambus et al . , 2009 ) , yet the current study shows that Pol α functionally interacts with CMG without Ctf4 . Hence , the complete role of Ctf4 remains unknown . Likewise , the essential Mcm10 protein is known to bind Pol α ( Warren et al . , 2009 ) , but this study shows that Mcm10 is not required for fork function in vitro . The same can be said for a myriad of proteins that form the RPC complex , including Mrc1 , Csm3 , Tof1 , Topo I , Mcm10 , Ctf4 , and FACT complex ( Gambus et al . , 2006; Gambus et al . , 2009 ) . It is important to note , however , that in vitro fork progression is 4–5 ntds/s , 5–10 times slower than in vivo measurements , and thus one or more of these proteins may be required to increase the rate of fork movement . The current report also reveals that Pol α can prime the leading strand directly . To our knowledge , before this report , all proposed models of leading strand initiation in bacteria and SV40 show priming on the lagging strand of a bidirectional origin , which becomes the leading strand of one fork , rather than directly priming the leading strands ( Waga and Stillman , 1998; Méndez and Stillman , 2003; Kaguni , 2011 ) . We show here that this is not necessary . Pol α is robust in initiating the leading strand directly and targets CMG to do so . Cellular studies indicate that nucleosomes are involved in determining Okazaki fragment size ( Smith and Whitehouse , 2012 ) , but data in this report show that short Okazaki fragments of the size observed in vivo do not require nucleosomes . However , nucleosomes may still be needed for greater precision in Okazaki fragment length . We also demonstrate that Pol ε forms a stable and isolable complex with CMG , which stands in contrast to the inability to isolate Pol ε in RPCs as defined by a complex that stays associated during two successive affinity columns ( Gambus et al . , 2006 ) . Presumably , the time involved in preparing RPC through two columns is sufficiently long for the CMGE complex to dissociate . However , some Pol ε can be recovered when isolated in a single-step pull down with epitope tagged CMG from cell extracts ( Sengupta et al . , 2013 ) . Reconstitution of cellular replisomes in vitro should provide a framework to explore the effects of other proteins that move with forks and of post-translational modifications that control eukaryotic forks in response to the cell cycle and DNA checkpoint mechanisms . Reconstituted systems should also enable detailed study of factors that maintain the epigenetic state of a cell during replication . Radioactive nucleotides were from Perkin Elmer . Unlabeled nucleotides were from GE Healthcare . DNA modification enzymes and ϕ29 DNA polymerase were from New England Biolabs . DNA oligonucleotides were from Integrated DNA Technologies . Protein concentrations were determined using the Bio-Rad Bradford Protein stain and bovine serum albumin as a standard . Buffer A is 20 mM Tris-HCl , pH 7 . 5 , 5 mM DTT , 0 . 1 mM EDTA , and 4% glycerol . Buffer B is the same as buffer A except 20 mM Tris-acetate , pH 7 . 5 was used in place of 20 nM Tris-HCl . Buffer C is 25 mM Tris-Cl pH 7 . 9 , 10% glycerol , 1 mM DTT , 1 mM MgCl2 , 5 mM imidazole , 20 mM KOAc , and 350 mM KCl . Buffer D is 25 mM Tris-OAc pH 7 . 6 , 40 mM K-OAc , 40 mM K glutamate , 2 mM Mg-OAc2 , 1 mM DTT , 20% glycerol , and 0 . 25 mM EDTA . Stop buffer is 1% SDS , 40 mm EDTA . Buffer H is 20 mM Hepes pH 7 . 5 , 10% glycerol , 1 mM EDTA , 2 mM DTT , 350 mM KCl , 1 mM ATP , and 4 mM MgCl2 . Proteins were purified as described: RPA ( Henricksen et al . , 1994 ) , E . coli SSB ( Georgescu et al . , 2011 ) , PCNA ( Yao et al . , 2003 ) , RFC containing full-length RFC1 ( Finkelstein et al . , 2003 ) . Pol ε was expressed in yeast as described ( Georgescu et al . , 2014 ) , in which a 3XFLAG tag was placed on the N-terminus of Pol2 in pRS425/GAL and transformed into yeast along with pJL6-expressing genes encoding Dpb2 , Dpb3 , and Dpb4 ( Chilkova et al . , 2003 ) . All buffers were degassed before use to prevent oxidation of Fe-S centers . Pol δ was expressed and purified as described ( Georgescu et al . , 2014 ) . CMG was purified as described ( Georgescu et al . , 2014 ) . Ctf4 with a N-terminal 3XFLAG tag was expressed from a Gal1/10 promoter and purified from yeast using an anti-FLAG column , similar to methods described for Pol ε ( Georgescu et al . , 2014 ) . Pol α was prepared by integrating the gene encoding Pol α with a C-terminal 3XFLAG tag into the yeast pRS402 vector under control of the Gal1/10 promoter , then integrated into strain OYO1 ( ade2-1 ura3-1 his3-11 , 15 trp1-1 leu2-3 , 112 can1-100 bar1Δ MATa , pep4::KANMX6 ) , a strain constructed from W303 ( a gift from Alan Tackett and Brian Chait , Rockefeller University ) ( Georgescu et al . , 2014 ) . The Pol12 , Pri1 , and Pri2 subunits were cloned into E . coli vectors pRSFDuet , pCDFDuet , and pACYCDuet/RIL , respectively ( Novagen Inc . , Madison , WI ) . Pol12 was transformed into E . coli BL21 ( DE3 ) codon plus RIL ( Agilent , Santa Clara , CA ) , then induced with IPTG for 8 hr at 15°C . Pri1 and Pri2 were co-expressed in E . coli BL21 ( DE3 ) cells by IPTG induction for 8 hr at 15°C . A 12 l culture of induced yeast cells for Pol1 and 1 l of each induced E . coli cultures for Pol12 and Pri1 and Pri2 were co-crushed in a cryogenic mill as described for CMG ( Georgescu et al . , 2014 ) . Frozen crushed cells were thawed at 12°C and the re-suspended with 20 ml of 250 mM Hepes pH 7 . 4 , 1 . 25 M potassium glutamate , 5 mM EDTA . The lysate was spun at 42 , 000×g for 1 hr at 4°C . Anti-Flag agarose ( 1 . 2 ml ) was added to the supernatant ( 80 ml ) and incubated with slow rotation for 1 . 5 hr at 12°C . Beads were collected by centrifugation at 1500×g , washed twice with 5 ml 50 mM Hepes pH 7 . 4 , 250 mM potassium glutamate , 1 mM EDTA , then loaded into a gravity column . The column was washed with 15 ml 50 mM Hepes pH 7 . 4 , 250 mM potassium glutamate , 1 mM EDTA , 10% glycerol , and Pol α was eluted with 5 ml 50 mM Hepes pH 7 . 4 , 250 mM potassium glutamate , 1 mM EDTA , 10% glycerol containing 20 μg/ml Flag peptide . The eluent was diluted to a conductivity equal to 150 mM NaCl using 25 mM Hepes pH 7 . 4 , 1 mM EDTA , and 10% glycerol , then applied to a 1 ml Heparin agarose column . Pol α was eluted with a gradient of 100 mM to 1 M NaCl in 25 mM Hepes pH 7 . 4 , 1 mM EDTA , 10% glycerol . Peak fractions were pooled , aliquoted , and stored at −80°C . Typical yield of Pol α was about 3 mg . To make the linear fork DNA substrate , a 3 . 2 kb sequence of DNA was designed such that one strand lacked dC residues and thus the other strand lacked dG . The G-rich strand was examined to eliminate runs of four G residues . The resulting 3260 bp sequence was synthesized by Biomatik ( Wilmington , Delaware ) . The first four and last two bases represent overhangs generated by BsaI and BtsC1 , respectively . Both of these enzymes cut outside of their recognition sequences , and their recognition sequences are excluded from the template . CGGTATTCTAACCACATTAATCTACACCTCTCACACACTACTCATATCATCTTCCAAAACCCACCTTTAAAAAACCCTTTATCCACACTCATCACCATTTCACCAACCTTTTTCTTAATTCTACACAAATCCAATTAACCTATCTCCAATTTTAACTCCATCACCTCTTATATTAACCCACCTACTACTCCAACAATACCCTATCAAATCTACTTCTATCTCAAAACTATCACCTACTCCTTCCATCATAATCCACTCTTATCAATTAAACAATTATCCTTCTTTCCCACCATCACTCACCATCTTTTCTTAACACCCTTAACATTTCCTTTTATAAAACACTTCCAATCCTATTTTCTCACTATCCCACCCACCATAAAAACTATCTCACCCTAACTCAACCCTTTCCCTCTCACCAACACTCCTTTATCACACACACTTTACCACACAAATCCCTCCATCATACACCTTTACCCTCAAATCCTAAACCACCTAACTATTCCACACAATATATCTACAAAAATTTACTTTTCCACATCTCCAACCCTTCCAACACCTTAATCCCAAACCTTAACTAACCTCTCTTTAATACTTCCTCCCATTCCCACCACATACACCAAATTATCTTCAACTCAAAAACCTAAACTCTCCTTTTTATTCCCTATAAAAACTCTTAACCCTCCAATATACAACTCTAACTAACTTCATTATCAACCAATCTTCCTCTACTTCCTCATCTTATAATTTATCCATTCAAAATAACCTACCTACCACCTCTCCTCTTCACTTCCTACCCTAAAATCACCACCTTATCCCTAATTTACCTCTTTCAACTTTCCTTAACCCAACTTCTCAATCCTACTTCACTTACTTCTTATAAAACCATCATTATCACACTACACATTACTCTATCTATCCAATCATCACCTTCTACAATCCAAACTATCCCACTACCCTCTCATTCTACCTTTTCATCTATCTCAAACTATCCACCAATCCATACCTCAAACTTTAACCACCCACTCCAAATCTACAACATAAAATTAACCTATCACATTCCAAACTAATCACCCTAACCCTAACACCCTTTTATCCTCACCAAAATTACCATTTTCCTCTTTACTCATAAACAAACATTCTCACCCATTTATAAAACACTTAATACCCACTTAATTCACTTCCTTTTTCTACCTCACCATCATCAACTCCTAATATCAACAACCCAAAATCACCACCTATATCCTCATCTCCTATATAAAAAACTTCACATCTCAACCTCAAACCACCTATTCCACTTAATCCCAATCAACCTATCAACTCTACAACCTACTCTTCAAATACATCTCCTATCACTTTCCCACCTCCTTCAATCAATTATAACTTTATCACCTAAACATTCTAAAATCTCCTATCCACTACATCACATAAACCTAAACCTACTACCAACCTACCATTATCCTTATCAAACATCATCAATTCCATCTTTCTTTATACCCTCTCCATATCTCTCTCATTAAAAAAACCAAAATCTAACAACTTCTTATTTCTAATCAAAAAAACAATCAACCTAACTCATAAAATCTTCACCTTAAAATTCCTTTACATTTAACAAATCCACTCTCCCTATCTTTTTCATATCAAACTTATCTAAACCACTATCCTCATTTATCCTAATACTCCATATACAACACCAAATTTCTTATATCTCATCTAAAATCCTCCACCAATATAAACTCCTCTTTACCATTTCCACTCAACACACCAAATCTTATTATTCCATCAAATCTAATCCATTACCATCATCAACCCTAATAAACCTACTTCCCAACTTTTATCTCTCCACTACCACACCAAAAATTAACCTCCTCTAAAAACTATCATTCCCTTTACCTCTTCCACATTCCACCTATAACTCCTCATCTTAAAACCAATCAACACCAAACAACTATCTCACCATATTTCCTCTCCAAACCAACAAATTAACAATCCTACCCACTCCAACCTCCACATTACTAATAACTAAACTTACCTTACCTACCACACCCTATCAACCATATTTAAAAAATTACTTATTCACTAACTAAAACATCACCCACAAACTTAAATCATCACCTCCTCTTTTCCACCTTATTCACCAACCCAATCTATCTATCTCACCTATACCTTTCCCTAATATCTTTTACTAACCCTATAAATACCACAATTCTAAAAAACCCATACTTATCTCACACATCACTTTATACTTCACTCTTAAAATACCCTCCAATATATATTACAACCCAAAAATATCTCCCTTCTATCTCCTACACACAAATTATACCACTTTTAAACCACTCCTCATCTCTAACCCAACCCTCTACAATTCCATACATCTCTACTATCAACATCACTCCTTCTTTTCCACTCTTCTCTCCACATCTTTATTAAACATCTCCTCCTCATTTTCACATAACTATTTACTAAATAAATTTACCTAAACTACATTTATTAAAACCCTACAACATACTCCTTATTCTCCTACCTACCATTCTCTAATCTCTTTACATTCTACTACTTTCCTACCTACTATCTCAATAAATTCACTTTCCTTTCACCACACACCAACACACCTTCCTCCAAATTCTTTATATCTCCTTCTCCTAACCAAATTCCTCACTAATAACATCTTACCTCCCTACCTTTTTCCTTCTACCCTCCACCATTTCCCAACCTCATACTCAATAATCAATTTACCCTCCCACAACATAACTCTATTAACACCCATTTTCTATCCATCAACTTCCTATTACTTAAATTATCTTTTAAACCAATAAAACTCCACCTCAACCACCCACCTCTCTCTTTCAATCCAATTTCAATCTTTCCAACCATTCTATCTACCCTAAACTATTAACTATCTATCTCACCCACAATCCCTCCTACAATTCACAACAACATTCCACCACTTCACTTTATCTTCACCTCCAAACTTATTCCTTCCCATTATCACCTTCTCCAAAACCCACAAACATCTAACTCTCATCTCTCAACACTTTCTACCCATTCTCTCTAACAAAATTCCCTTACTCTTTATTCACAAAACCACTAAAATCACCACAACAACCCAACAAAAACAAAATCCTCACTTACCTATACTCAATAAATCCTTCAACTCATTATTCTATTTCTAACCCTAAATCAAAACTCCCATATCTACCATTCTTTCCACTCAATTAAATCCCACCAACCCTTATTTTCCTCCAATAACTTAACC . The synthetic 3 . 2 kb DNA was cut from the plasmid using BsaI-HF and BtsCI and purified from low-melting agarose gel . 35 pmol of the 3 . 2 kb linear DNA template was ligated to a fivefold excess of forked junction on the BsaI end and to a 10-fold molar excess of a short blocking duplex on the BtsCI end . To make the forked junction , 1050 pmol 1T oligonucleotide was annealed to 175 pmol 160mer oligonucleotide , as described ( Georgescu et al . , 2014 ) . The 160mer is the leading strand template and contains a 3′ biotin and three 3′ terminal thiophosphate linkages , to protect against nucleases . The synthetic fork contains a 5′-phosphorylated-ACCG overhang in the duplex region that ligates to the BsaI end of the 3 . 2 kb nucleotide-biased duplex . To protect the BtsCI end of the 3 . 2 kb duplex from excision and fill-in by the proofreading exonucleases of Pols δ and ε , a short protecting duplex was ligated to the BtsCI end of the forked junction . The blocking duplex was formed upon annealing 350 pmol 5′-tggttagtatagcaagtagagg-3′ and 2100 pmol tctacttgctatactaaccat3′-′3-dT . After ligation , this results in a 3 . 2 kb duplex with two 5′ terminal nucleotides at one end , to resist digestion by 3′–5′ exonuclease inherent in the DNA polymerases . Both the blocking duplex and the synthetic forked junction were present in the ligation reaction . Ligation was for 18 hr at 15°C with T4 DNA ligase . Excess non-ligated oligonucleotides were removed by gel filtration over a 20 ml bed volume of Sepharose 4B ( GE Healthcare , Piscataway , NJ ) equilibrated in 10 mM Tris-acetate pH 7 . 5 , 1 mM EDTA , and 100 mM sodium acetate , pH 7 . 44 . Peak fractions containing the nucleotide-biased linear forked DNA were pooled and stored at −20°C . When the substrate was primed , a DNA oligonucleotide was annealed to the leading strand template , DNA 37mer ( C2 ) as described ( Georgescu et al . , 2014 ) . Replication assays contained 1 . 5 nM linear forked DNA , 24 nM CMG , 400 nM RPA ( unless noted otherwise ) , 20 nM PCNA ( unless noted otherwise ) , 6 nM RFC ( unless noted otherwise ) , and Pol α , Pol ε , and Pol δ as indicated in the figure legends , in 25 mM Tris-acetate pH 7 . 5 , 10 mM Mg-acetate , 50 mM potassium glutamate , 5 mM DTT , 0 . 1 mM EDTA , 40 μg/ml BSA , 0 . 1 mM AMP–PNP , 5 mM ATP , 200 μM each rCTP , rUTP , rGTP , 60 μM of each unlabeled dNTP , and 20 μM of the labeled dNTP . Reactions were staged as follows . CMG was added first and pre-incubated with the DNA and 0 . 1 mM AMP-PNP for 10 min at 30°C , then the noted polymerases together with the RFC and PCNA ( where indicated ) were added , along with dATP , dCTP for an additional 2 min . Replication was then initiated upon the addition of a solution containing the RPA , ATP , dTTP , and dGTP . It is important to note that for leading strand replication reactions , we used 20 μΜ dCTP and 10 μCi 32P-dCTP , while for lagging strand replications , we used 20 μM dGTP and 10 μCi 32P-dGTP . Exceptions to this protocol are noted in the figure legends . Timed aliquots were removed and quenched upon adding SDS and EDTA to final concentrations of 0 . 5% and 20 mM , respectively . Quenched reactions were analyzed in 0 . 7% or 2% alkaline agarose gels and imaged in a Typhoon 9400 PhosphorImager ( GE/Molecular Dynamics , Berkeley , CA ) . In order to compare total DNA synthesis rates on leading and lagging strand , we performed standard replication reactions containing all three DNA polymerases at 10 nM , CMG ( 30 nM ) , RFC ( 10 nM ) , PCNA ( 20 nM ) , and RPA ( 400 nM ) in presence of 1 mM ATP , rNTPs ( C , G , U at 100 μM ) , and 30 μM dNTPs; reactions were divided , and either 32P-dCTP ( leading ) or 32P-dGTP ( lagging ) was added . Timed reactions were stopped with an equal volume of 2× STOP solution ( 40 mM EDTA and 1% SDS ) and spotted on DE81 filter papers , then analyzed using a Perkin Elmer Liquid Scintillation Analyzer ( Perkin Elmer , Tri-Carb 2910 TR ) . Separately , we performed control replication reactions using ϕX174 ssDNA coated with RPA , containing a known amount of Gs and Cs , confirm that 32P-dCTP and 32P-dGTP are equally incorporated by each of the DNA polymerases in our experimental conditions . Three replication reactions were performed as described above with the following differences . The first reaction used 1 . 5 nM unprimed forked DNA and 10 nM Pol α with 20 μM dGTP as well as 10 μCi 32P-dGTP to label the lagging strand; the second reaction utilized 1 . 5 nM unprimed forked DNA and 10 nM each Pol ε and Pol α along with 20 μM dGTP and 10 μCi 32P-dGTP to label the lagging strand , and the third reaction contained DNA-primed forked DNA and 1 U of ϕ29 DNA polymerase along with 20 μM dCTP and 10 μCi 32P-dCTP to label the leading strand . Each reaction was quenched upon heating to 65°C for 10 min to inactivate the CMG and polymerases . Then , reactions were divided into three tubes , one was untreated , the second was treated with EarI , and the third was treated with PsiI , adjusting the reaction buffer for each enzyme with the commercial provided buffer . Reactions were analyzed in a 2% native agarose gel followed by autoradiography in a Typhoon 9400 PhosphorImager ( GE/Molecular Dynamics ) . Reactions contained 1 . 5 nM ϕX174 circular ssDNA ( as circles ) primed with a DNA 30-mer and pre-incubated for 10 min with 420 nM RPA ( as heterotrimer ) in 20 mM Tris-Cl ( pH 7 . 5 ) , 50 mM potassium glutamate , 5 mM DTT , 0 . 1 mM EDTA , 40 μg/ml BSA , 8 mM MgOAc , 0 . 5 mm ATP , 5% glycerol , and 60 μM , each of dGTP and dATP . Reactions also contained the indicated amounts of RFC , PCNA , and Pol α and were pre-incubated for 5 min at 30°C . DNA synthesis was initiated by adding 15 μl of 60 μM dCTP , 20 μM dTTP , 15 μCi of ( α-32P ) dTTP and incubated at 30°C . At the times indicated , 25-μl aliquots were removed and quenched by addition of an equal volume of 1% SDS/40 mm EDTA . Products were analyzed in 0 . 7% alkaline agarose gels . Gels were dried , exposed to PhosphorImager screens , and imaged using a Typhoon 9400 PhosphorImager ( GE/Molecular Dynamics ) . Proteins were premixed in the following amounts: 80 pmol StrepTag-Pol ε , 120 pmol CMG , and when present , 200 pmol Ctf4 ( as trimer ) . Proteins were brought to a final volume of 200 μl in 100 mM sodium phosphate , 150 mM NaCl pH 8 . 0 , and incubated with 50 μl ( as a 10% slurry ) Strep-Tactin magnetic beads ( Qiagen , Valencia , CA ) for 1 hr at 4°C with end-over-end mixing . Beads were collected in a magnetic separator and washed twice in 200 μl 100 mM sodium phosphate , 150 mM NaCl pH 8 . 0 , then eluted in 75 μl 10 mM biotin in 100 mM sodium phosphate , 150 mM NaCl pH 8 . 0 for 20 min on ice . Samples were analyzed in 8% SDS-PAGE followed by staining with Coomassie Blue Denville stain .	Cells must replicate their DNA before they divide so that the newly formed cells can each receive a copy of the same genetic material . DNA replication requires complex molecular machinery called a replisome , which comprises multiple proteins , enzymes , and other molecules . First , an enzyme called a helicase starts to unwind the double-stranded DNA into two single strands . This process continues while other enzymes , called polymerases , use the exposed single strands as templates to make complementary new strands of DNA . One of these new strands is built continuously and called the ‘leading strand’ . The other newly forming strand—the ‘lagging strand’—is made in the opposite direction , as a series of short fragments that are later joined together . The replisomes in bacterial cells have been well studied , but many researchers are investigating the composition of the replisome in animals , plants , and fungi ( collectively called eukaryotes ) . Now , Georgescu et al . have essentially rebuilt a eukaryotic replisome from 31 different proteins in a test tube and confirmed that it can make both leading and lagging DNA strands—just like in a normal cell . Further experiments revealed that the polymerase enzyme that operates on the leading strand cannot work on the lagging strand and vice versa . This exclusivity is unique to eukaryotic DNA replication , as bacterial polymerases can use either DNA strand as a template . Georgescu et al . then found that the eukaryotic polymerases are actively prevented from copying the ‘wrong’ strand of DNA and suggest that the helicase enzyme that unwinds the DNA might be behind this activity . Important future studies must now address how the replisome deals with obstacles created by certain DNA-binding proteins and damaged DNA and how it interfaces with the molecules that control cell division and DNA repair .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology" ]	2015	Reconstitution of a eukaryotic replisome reveals suppression mechanisms that define leading/lagging strand operation
Schistosomes are parasitic flatworms infecting hundreds of millions of people . These parasites alternate between asexual reproduction in molluscan hosts and sexual reproduction in mammalian hosts; short-lived , water-borne stages infect each host . Thriving in such disparate environments requires remarkable developmental plasticity , manifested by five body plans deployed throughout the parasite’s life cycle . Stem cells in Schistosoma mansoni provide a potential source for such plasticity; however , the relationship between stem cells from different life-cycle stages remains unclear , as does the origin of the germline , required for sexual reproduction . Here , we show that subsets of larvally derived stem cells are likely sources of adult stem cells and the germline . We also identify a novel gene that serves as the earliest marker for the schistosome germline , which emerges inside the mammalian host and is ultimately responsible for disease pathology . This work reveals the stem cell heterogeneity driving the propagation of the schistosome life cycle . Flatworms include more than 44 , 000 parasitic species that form one of the largest groups of metazoan endoparasites ( Loker and Hofkin , 2015 ) . Their life cycles typically involve asexually and sexually reproducing stages , each with its own distinct body plan and strategy to enhance transmission between multiple hosts ( Clark , 1974; Pearce and MacDonald , 2002; Viney and Cable , 2011 ) . Although the life cycles of these parasites were established more than a century ago , they have only recently been studied in cellular and molecular terms ( Matthews , 2011 ) . Since many parasitic flatworms are pathogenic , their life cycles are also the routes for disease transmission ( Hoffmann et al . , 2014 ) . Therefore , a deeper understanding of these life cycles is significant from both basic science and medical perspectives , as blocking transmission is an effective approach to fighting parasitic diseases . Focusing on the cells that may drive such parasitic life cycles , we study Schistosoma , a parasitic flatworm infecting over 250 million people , which causes the major neglected tropical disease , schistosomiasis ( Hoffmann et al . , 2014 ) . Schistosomes are transmitted through snail intermediate and human definitive hosts . Their life cycle begins with the parasite egg excreted from the mammalian host into water , releasing a free-swimming miracidium larva . The miracidium penetrates a snail host and transforms into a mother sporocyst that undergoes asexual clonal expansion to produce many daughter sporocysts that leave the mother and colonize other snail tissues . These daughters either self-renew to produce more daughters or enter embryogenesis to produce infective cercariae ( Cheng and Bier , 1972; Jourdane and Theron , 1980; Whitfield and Evans , 1983; Taft et al . , 2009 ) . This cloning process is repeated , allowing massive numbers of cercariae to be produced from a single miracidium . Mature cercariae emerge from the snail into water , then burrow through the skin of a mammalian host to become schistosomula . This transition initiates the sexual portion of the life cycle . Schistosomula then migrate to species-specific niches in the host vasculature and develop into juvenile worms ( Basch , 1991; Wilson , 2009 ) . Juveniles remodel their tissues extensively to build a functional digestive system , and after they begin feeding on host blood , undergo massive growth and develop sexual reproductive organs de novo ( Clegg , 1965 ) . Male and female worms pair to produce fertilized eggs , which are excreted to continue the life cycle . A long-standing hypothesis proposes that a lineage of totipotent stem cells , called ‘germinal cells’ , persists throughout the schistosome life cycle and drives reproduction ( Cort et al . , 1954; Clark , 1974; Whitfield and Evans , 1983 ) . Histological and ultrastructural studies defined these cells in miracidia and sporocysts by their stem cell-like morphology and rapid proliferation ( Schutte , 1974; Pan , 1980 ) . Recently , we showed that these germinal cells indeed drive proliferation within developing sporocysts and share some molecular signatures with stem cells from diverse organisms ( Wang et al . , 2013 ) . In contrast , the cellular source of the schistosome germline , which underlies sexual reproduction in the mammalian host , remains an open question . Furthermore , because somatic stem cells were only recently identified in adult schistosomes ( Collins et al . , 2013 ) , the relationships between germinal cells , germ cells , and somatic stem cells are unclear . To clarify these relationships , we transcriptionally profiled stem cells from Schistosoma mansoni asexual ( sporocyst ) and sexual ( juvenile ) stages at both population and single-cell levels . We identified four transcriptionally distinct populations and validated this heterogeneity by in situ hybridization . By characterizing the behavior of these stem cells at major developmental transitions , we found that larvally derived stem cells serve as the source for the parasite’s adult stem cells . We also identified a novel gene that is activated during development inside the mammalian host and serves as the earliest marker for the schistosome germline . This work reveals the stem cell heterogeneity underlying the development and propagation of these important parasites . Each miracidium carries 10–20 germinal cells ( Pan , 1980; Cort et al . , 1954; Wang et al . , 2013 ) , which expand massively and differentiate to produce many daughter sporocysts ( Figure 1A , and Figure 1—figure supplement 1 ) . Our recent work has shown that germinal cells exhibit heterogeneity within this population ( Wang et al . , 2013 ) , revealed by the distinct proliferation kinetics and expression of a schistosome homolog of nanos ( Wang and Lehmann , 1991 ) , a conserved regulator of germ cell development ( Juliano et al . , 2010; Wang et al . , 2007 ) also expressed in the schistosome adult stem cells ( Collins et al . , 2013 ) . To characterize this heterogeneity further , we isolated and transcriptionally profiled these stem cells from in vitro-transformed mother sporocysts ( Figure 1B ) . Principal component analysis ( PCA ) of single-cell transcriptomes revealed three major cell classes ( Figure 1C ) . We designated these classes based upon their respective markers: κ-cells ( kappa indicates klf+nanos-2+ ) ; φ-cells ( phi indicates fgfrA , B+ ) ; and δ-cells ( delta indicates double-positive for nanos-2 and fgfrA , B ) . The difference between κ and δ/φ-cells extends along PC1 , and contributes to ~30% of the total variance among cells , whereas the difference between δ and φ-cells is secondary , delineated by PC2 and contributing ~10% of the total variance . For example , nanos-2 exhibits almost equal loadings on both PCs , negative on PC1 , positive on PC2 , consistent with its expression in both κ and δ-cells . Based on projections along the first two PCs ( Treutlein et al . , 2014 ) , we identified additional genes that contribute to the distinctions between classes: a schistosome p53 homolog and a zinc finger protein ( zfp-1 ) expressed abundantly in δ-cells and at lower levels in φ-cells; and a hes family transcription factor ( hesl ) expressed specifically in φ-cells ( Figure 1D and E , Supplementary file 1 ) . We validated these transcriptomic findings by fluorescent in situ hybridization ( FISH ) on in vitro-cultured mother sporocysts ( Figure 1—figure supplement 2 ) . Unfortunately , the κ class-specific marker klf was expressed at very low levels ( Figure 1E ) , beneath the detection limits of our current FISH protocol . In addition to these class-defining genes , the divergence of the three cell classes is manifested by hundreds of other genes that exhibit various levels of statistically significant differences between classes ( Figure 1—figure supplement 3 ) . However , these genes comprise only a small fraction of transcripts detected in these cells ( N = 6 , 661 ) , and most of them are not enriched in stem cells compared to differentiated cells . Notably , very few transcripts are specific to individual cell classes , with φ-cells showing the fewest specific markers . These observations confirm that sporocyst stem cells , regardless of the subpopulation to which they belong , share a common transcriptomic profile . Examining fgfrA and nanos-2 enabled us to distinguish all three cell classes in situ: φ-cells express fgfrA , κ-cells express nanos-2 , and δ-cells express both . Thus , we followed these cells throughout intramolluscan development by monitoring fgfrA and nanos-2 expression . After the first week of infection , asexually produced embryos– identified as compact , spherical cell clusters ( Schutte , 1974 ) and from which daughter sporocysts will arise– begin to develop ( Figure 2A ) . φ-cells were distributed beneath the parasite’s outer layer and excluded from daughter embryos . δ-cells were found in large clusters within embryos . κ-cells clustered with δ-cells in embryos and were found in extraembryonic tissues as singlets or doublets , suggested to be the source of developing embryos in previous histological studies ( Schutte , 1974 ) . Two weeks post-infection , mother sporocysts contain many mature daughter sporocysts that are ready to leave the mother and migrate elsewhere in the snail . At this stage κ-cells comprised the vast majority ( >85% ) of stem cells in these daughters ( Figure 2B ) ; fewer δ-cells were observed and φ-cells were mostly excluded . However , one week later , when post-migratory daughters colonized new regions of host tissue ( Figure 2C ) , all three classes reappeared as intermingled populations , consistent with κ-cells generating the other stem cell types . Intramolluscan development culminates with the production of infectious cercariae . In early cercarial embryos ( dashed circle in Figure 2D–2E ) , φ-cells were found concentrated both anteriorly and posteriorly , where the mouth and tail bud form , respectively . Additionally , two clusters of κ-cells were observed posterior to the penetration glands ( Figure 2D–2E ) , at the site of the germinal cell cluster , considered gonadal primordia based on histological and ultrastructural studies ( Cheng and Bier , 1972; Dorsey et al . , 2002 ) . In the mature cercarial body ( Figure 2F–2G ) , the κ-cell pair posterior to the glands expands into two clusters that contain multiple cells each . In parallel , five δ-cells were detected in a regular pattern around the penetration glands , with one at the midline and two pairs laterally ( Figure 2E ) , whereas φ-cells are absent at this stage . Since the cercarial body ( but not the tail ) penetrates the mammalian host , only δ and κ-cells , but not φ-cells , may be passed to the intramammalian ( sexual ) stage . After emerging from the snail into water , cercariae burrow through mammalian host skin and their bodies transform into the next life-cycle stage , the schistosomula . At this stage , the parasites do not grow for several weeks , until they reach the hepatic portal vein . Thus , the extent of proliferation in the initial days after infection has been unclear ( Clegg , 1965 ) , with mitotic cells only detected 4 days post-infection ( Clegg and Smithers , 1972 ) . Furthermore , because the adult stem cells have only been identified recently ( Collins et al . , 2013 ) , their developmental origin has yet to be investigated . The identification of δ and κ-cells in cercariae provides a potential source of new multipotent cells upon entry into the mammalian host . We mimicked this transition by exposing cercariae to ex vivo mouse tail-skin biopsies and collecting transformed schistosomula on the other side of the skin ( Clegg and Smithers , 1972; Protasio et al . , 2013 ) . Following skin penetration , we assayed proliferation in schistosomula via EdU labeling ( Figure 3A ) . Between 22 and 36 hr post-transformation , we observed five EdU+ cells around the penetration glands , anterior to the ventral sucker ( Figure 3B ) , and confirmed that they were δ-cells ( fgfrA+nanos-2+ ) ( Figure 3C ) . During the next 12 hr , these cells completed mitosis , indicated by the appearance of five EdU+ doublets ( Figure 3B ) . Thereafter , the number of EdU+ nuclei steadily increased ( Figure 3D ) , but proliferation was restricted anteriorly to the ventral sucker , until one week later , when two clusters of ~2–3 EdU+ cells appeared in the ‘germinal cell cluster’ region posterior to the ventral sucker , where κ-cells are found ( Figure 3B , Days 9–10 ) . EdU+ cells were not detected in irradiated worms ( Figure 3E ) , consistent with previous reports that irradiation leads to developmental defects and reduced pathogenicity ( Wilson , 2009 ) . These results suggest that a small , fixed number of κ and δ-cells are transmitted to the mammalian host . Because these cells appear to be the only dividing cells in schistosomula , they are likely the source of the recently identified stem cells in adult schistosomes ( Collins et al . , 2013; 2016 ) . Intramammalian growth initiates after schistosomula migrate into the portal vein , around 2 weeks post-infection ( Clegg , 1965; Basch , 1981 ) . To characterize proliferation driving juvenile growth , we harvested EdU-labeled parasites 3 weeks post-infection . Worms displayed a range of sizes based upon differences in arrival time , enabling a developmental time course to be reconstructed from a static time point . EdU labeling revealed a posterior growth zone ( PGZ ) that extended as the parasites grew ( Figure 4A ) . In more mature juveniles , all cells in primordial testes and ovaries also incorporated EdU ( Figure 4B ) . To compare proliferating cells from juveniles to sporocyst stem cells , we transcriptionally profiled isolated cells undergoing division ( S and G2/M phase ) from juveniles and sporocysts ( Figure 4—figure supplement 1 ) . Five-hundred and seventy-three genes were commonly enriched in both populations , including previously identified schistosome stem cell markers ( e . g . nanos-2 , ago2-1 , and fgfrA ) , and cell cycle-associated transcripts ( h2a , cyclin B , and PCNA ) ( Wang et al . , 2013; Collins et al . , 2013 ) . Taken together , proliferating cells from juveniles resemble the sporocyst stem cells both morphologically and transcriptionally ( Figure 4—figure supplement 1 ) , suggesting that these cells represent juvenile stem cells , which support both somatic growth and germline development in the mammalian host . Are κ and δ-cells maintained during juvenile development ? Since κ and δ-cells can be distinguished by a small set of transcripts , we measured expression of 87 genes , including all identified cell class-specific factors , across single juvenile proliferating cells by multiplex qPCR ( Figure 4—figure supplement 2 , and Supplementary file 2 ) ( van Wolfswinkel et al . , 2014 ) . The assayed gene set contained the class-specific factors identified from sporocyst stem cells and the most highly enriched genes in juvenile vs . sporocyst stem cells . Hierarchical clustering identified two major cell classes ( Figure 4C ) . δ’-cells are similar to δ-cells , but express abundantly both δ and φ-cell markers ( including nanos-2 , fgfrA , p53 , zfp-1 , and hesl ) , indicating that these cells are the likely source of the adult somatic stem cells ( Collins et al . , 2013; 2016 ) . The other class abundantly expresses a novel schistosome-specific factor , eledh ( eled ) , which is undetectable in sporocysts but is among the most abundant transcripts in juvenile stem cells ( Supplementary file 3 ) . We designated this class ε ( epsilon indicates eled+ ) , which displays lower expression of nanos-2 and fgfrA , and similar to κ-cells , lacks expression of p53 , zfp-1 , and hesl ( Figure 4C , and Figure 4—figure supplement 3 ) . Based on these similarities in gene expression , we propose that ε-cells are likely derivatives of κ-cells ( see Discussion for more details ) . eled is a single-copy gene ( Smp_041540 ) that was previously annotated as a nuclear hormone receptor , dhr4 ( Protasio et al . , 2012; Berriman et al . , 2009 ) ; however , our analysis suggests that this gene does not encode a hormone receptor ( Figure 4—figure supplement 4 ) . As described below , this gene antagonizes nanos ( Greek for ‘dwarf’ ) ; therefore , we named Smp_041540 eledh ( Sindarin for ‘elf’ ) , based on the antagonistic relationship between dwarves and elves in Tolkien’s world . Homologs of eled are found across Schistosoma species , whereas planarian and tapeworm genomes appear to lack them . Figure 4D shows the predicted architecture of eled , which has a transmembrane domain at the N-terminus ( probability ~100% ) , a putative transmembrane domain at the C-terminus ( probability ~35% ) , and an extracellular domain in between ( probability ~85% ) . The extracellular domain contains 19% serine and 17% threonine , presenting the highest serine/threonine fraction in known proteins . In juveniles eled expression was detected in primordial testes , ovaries , and vitellaria , as well as in a gradient increasing toward the PGZ , which lacks reproductive organs in males ( Figure 5A ) . This pattern is distinct from that of the proliferation marker , h2a , which labels all stem cells and is more evenly distributed . Quantitatively , overlay of eled and h2a revealed that all gonadal eled+ cells ( Figure 5B ) and ~80% of somatic eled+ cells ( 966/1269 , from two male juveniles ) were also h2a+ . Conversely , all germ cells within gonadal primordia were eled+ , but only ~50% of the somatic h2a+ cells in the PGZ ( 1679/3576 ) and ~10% of the somatic h2a+ cells outside the PGZ ( 769/5366 ) expressed eled . In mature adults , eled expression was limited to reproductive organs ( Figure 5C ) . Thus , somatic ε-cells may be a transitory source of new tissue underlying massive juvenile growth in the PGZ . Expression of eled in gonadal primordia led us to examine earlier stages of germ cell development . We found that eled expression precedes nanos-1 , which is germline specific both in juveniles ( Figure 6—figure supplement 1 ) and adults ( Iyer et al . , 2016; Wang and Collins , 2016 ) : only a subset of eled+ cells in gonadal primordia co-express nanos-1 , and the number of eled+nanos-1+ cells increases over the course of development ( Figure 6A , B ) . Quantification reveals a sharp transition as worm length exceeds ~400 µm: before the transition , none of the gonadal eled+ cells is nanos-1+; after the transition , most if not all of the gonadal eled+ cells become nanos-1+ ( Figure 6C ) . These results suggest that germ cells may be derived from ε-cells early in juvenile development , and eled is the earliest germline marker yet identified in schistosomes . To characterize functional interactions between eled and nanos , we knocked down gene function using RNA interference ( RNAi ) ( Mann et al . , 2010; Collins et al . , 2013; Wang et al . , 2013 ) . For these experiments , juveniles were soaked in double-stranded RNA ( dsRNA ) continuously in vitro for 2 weeks . We focused on male juveniles because female development was retarded under in vitro culture . To assess gene expression changes after RNAi , we performed whole-mount in situ hybridization ( WISH ) ; we used confocal microscopy to examine testis structure . eled RNAi resulted in upregulation of nanos-2 in the PGZ ( Figure 7A ) , where the somatic ε- cells are located . Knockdown of nanos-1 or nanos-2 resulted in degenerated testes and loss of differentiated germ cells; however , the remaining gonads maintained eled expression ( Figure 7A and B ) . By contrast , knockdown of eled resulted in premature accumulation of sperm in juvenile testes ( Figure 7B ) . Thus , eled appears to inhibit , whereas nanos genes are required for , germ cell differentiation . Together , these results suggest that eled antagonizes the two schistosome nanos homologs: in the soma , it suppresses nanos-2 expression in ε-cells in the PGZ; in the germline , it inhibits germ cell differentiation . Throughout their life cycles , parasitic flatworms undergo dramatic morphological changes as they switch from free-living , infectious stages to endoparasitic forms residing in different hosts; the cellular basis of this developmental plasticity is largely unknown . Here , we have used single-cell transcriptional profiling to characterize schistosome stem cells during intramolluscan and intramammalian development . We found that these stem cells are a heterogeneous population , consisting of four major classes , distinguished by several distinct markers . This analysis enabled cellular and molecular studies to trace the origin of the adult stem cells back to a handful of larvally derived cells packaged into the infectious stage; it also provided evidence for the origin of the schistosome germline . During intramolluscan asexual development we find three classes of stem cells that can be distinguished by expression of either or both nanos-2 and fgfr: κ ( klf+nanos-2+ ) ; φ ( fgfrA+ ) ; and δ ( nanos-2+fgfrA+ ) cells . Our data are consistent with the proposed lineage depicted in Figure 8 . We posit that κ cells serve as ‘embryonic’ stem cells: in addition to nanos-2 they express a klf homolog and are found isolated in developing mother sporocysts . This observation is consistent with classic histological studies suggesting that such isolated germinal cells serve as the source of developing embryos ( Schutte , 1974; Pan , 1965 ) . These studies reported that the scattered stem cells may persist in mother sporocysts for months and continuously undergo asymmetric divisions , with one daughter cell forming an embryo while the other remains undifferentiated ( Schutte , 1974 ) . We found that κ-cells are dramatically enriched in mature , pre-migratory daughter sporocysts , consistent with the idea that these cells generate the other stem cell types observed in post-migratory daughter sporocysts . Activation of several genes associated with somatic stem cell function in adults ( e . g . fgfrA/B; p53; zfp-1 ) leads to the specification of δ-cells from κ-cells; we propose that δ-cells serve to generate somatic tissues . We suggest that downregulation of nanos-2 and activation of hesl in δ-cells leads to the formation of φ-cells . The restricted distribution of φ-cells in transitory larval structures , including the sporocyst epidermis ( tegument ) and the cercarial tail ( Figure 2A , E ) , is consistent with the observations that φ-cells are not transmitted between life-cycle stages and are absent during the sexual stage . Our previous work identified rapidly cycling stem cells that do not express nanos-2 ( in contrast to slower-cycling cells that do ) ( Wang et al . , 2013 ) ; thus , φ-cells appear to be a lineage-committed , transit-amplifying population that produces temporary larval tissues . The distinction between κ , δ , and φ-cells is also consistent with early histological studies that observed germinal cells in three characteristic tissue locations within mother sporocysts: either scattered among daughter embryos , clustered inside of embryos , or situated close to mother sporocyst walls ( Schutte , 1974; Pan , 1965 ) . The result of larval development in the snail is the production of cercariae , free-swimming infectious forms that penetrate mammalian skin to continue the life cycle . We observed five δ-cells , localized in a stereotypical pattern in the cercarial body . Using mouse skin explants to mimic the transformation that occurs after parasites enter their mammalian host , we found that these five cells proliferate during the first 24 hr after transformation . Given the similarities between the transcriptional profiles of δ-cells and the somatic stem cells found in juvenile parasites ( δ’-cells that also express hesl ) , as well as their early proliferation upon penetrating host skin , we suggest that δ-cells serve as the source of the somatic stem cells identified in adult schistosomes . Because there is no growth during the first 2 weeks post-infection ( Clegg , 1965; Wilson , 2009 ) , these early-proliferating cells may contribute to the tissue remodeling ( e . g . of the tegument or digestive system ) required for the parasite’s subsequent growth and maturation . Cercariae also contain two clusters of κ-cells , in the germinal cell cluster that is proposed to be the gonadal primordia ( Schutte , 1974; Dorsey et al . , 2002 ) . These cells do not proliferate in the initial days after transformation; instead , we detect proliferation in the posterior of the schistosomula 9–10 days after transformation in vitro . Based on the antagonistic interaction between eled and nanos , we suggest that κ-cells downregulate nanos-2 and activate expression of eled . eled is specific to the intramammalian stage of the life cycle and defines ε-cells . The absence of δ/δ’-cell markers , fgfr , zfp-1 , p53 , and hesl , in both κ and ε-cells favors the model that κ/ε-cells form a separate lineage from δ/δ’-cells . Expression of eled is dynamic: during juvenile development , eled is expressed in gonadal primordia and in a subset of proliferating cells in the posterior growth zone . In mature adults , eled expression is maintained in the reproductive system but is no longer detected somatically . Thus , ε-cells can either give rise to germline or to a transient somatic population that appears to drive the massive posterior growth exhibited by these parasites . During maturation , gonadal ε-cells subsequently activate germline-specific nanos-1 . We suggest that the activation of nanos-1 commits ε-cells to germline fate . How eled+ cells choose to generate germline ( nanos-1+ ) vs . posterior somatic cells ( nanos-1- ) will be an important avenue for future studies . Functional analysis of eled was limited to juvenile male parasites harvested after 3 weeks of growth in the mouse , due to technical constraints imposed by the parasite’s life cycle . Under these experimental conditions , eled ( RNAi ) males exhibited premature production of sperm; thus , eled appears to act as a brake on germ cell differentiation . Because it is also the earliest germline marker yet identified and could play a role in germ cell specification , new experimental techniques will have to be developed to analyze gene function during the first 3 weeks of development in the mammalian host , when the germline is formed . Similarly , rigorous experimental testing of the proposed lineage relationships between the stem cell classes ( Figure 8 ) will require introduction of lineage-tracing techniques in these parasites . Previous work noted similarities between schistosome stem cells and the neoblasts that drive regeneration in free-living planarians ( Wang et al . , 2013; Collins et al . , 2013 ) . The heterogeneity we detected in schistosome stem cells is also reminiscent of that observed in the planarian neoblasts ( van Wolfswinkel et al . , 2014 ) , and we observed a striking overlap in a group of genes co-regulated between stem cell classes from both organisms . The genes fgfr , zfp-1 , and p53 were defined as markers of an epidermally committed population in planarians ( van Wolfswinkel et al . , 2014 ) . These genes are abundantly expressed in δ/δ’-cells , the major somatic population in schistosomes . Recently , schistosome adult stem cells were shown to have a strong differentiation bias toward the tegumental lineage ( Collins et al . , 2016 ) ; the necessity of zfp-1 family genes for proper tegumental fate or , more generally , for proper differentiation , further linked these parasites to their free-living ancestors ( Wendt et al . , 2018 ) . Furthermore , in the planarian embryo , blastomeres produce temporary embryonic tissues but also give rise to neoblasts by downregulating a set of embryo-specific genes and upregulating genes associated with adult development ( e . g . , zfp-1 and p53 ) ( Davies et al . , 2017 ) . Our data suggest that a similar transition may also occur between κ-cells and δ/δ’-cells . Beyond planarians and schistosomes , stem cells have also been described in other parasitic flatworms , including tapeworms ( Hoffmann et al . , 2014; Koziol and Brehm , 2015; McCusker et al . , 2016 ) . We expect that single-cell approaches applied to additional parasitic flatworms will provide a broader overview of the role of stem cell heterogeneity in driving such complex life cycles . This study presents an important step toward understanding the fundamental mechanisms driving the propagation and long-term survival of schistosomes . As the causative agents of a neglected tropical disease impacting hundreds of millions of people , these parasites present a major threat to global health . Schistosome infection is currently treated with a single drug , praziquantel , which is used in mass drug administration programs ( Hoffmann et al . , 2014; Sokolow et al . , 2015 ) . With concerns about resistant strains emerging ( Hoffmann et al . , 2014 ) , it becomes increasingly important to understand the fundamental mechanisms driving the propagation and long-term survival of these parasites . Characterizing the roles of the stem cell populations defined here in schistosome transmission , reproductive development , and survival may ultimately lead to novel approaches to reduce the burden imposed by these parasites ( Matthews , 2011; Valentim et al . , 2013; Hoffmann et al . , 2014 ) . In vitro-transformed mother sporocysts were obtained as detailed previously ( Wang et al . , 2013; Mann et al . , 2010; Ivanchenko et al . , 1999 ) . Briefly , S . mansoni ( strain NMRI ) eggs were purified from livers harvested from schistosome-infected mice ( Swiss Webster , female , ~7 weeks post-infection ) . Free-swimming miracidia were hatched from eggs in artificial pond water and transformed in vitro to mother sporocysts by exchanging pond water with sporocyst culture medium supplemented with 1X Antibiotic-Antimycotic ( Gibco ) and 20 µg/mL gentamycin ( Gemini ) at 37°C in 5% CO2/5% O2 for 48 hr . S . mansoni cercariae were shed from infected Biomphalaria glabrata snails about 5–8 weeks post-infection by exposing snails to bright light at 26°C for 1–2 hr . Schistosomula were transformed from cercariae using skin transformation ( Clegg and Smithers , 1972; Protasio et al . , 2013 ) , in which cercariae were placed on ex vivo mouse skin biopsies ( Swiss-Webster , Taconic ) overlaid on Basch medium 169 ( Basch , 1981 ) to collect parasites that passed through the skin . Juvenile and adult worms were obtained from infected mice ( Swiss Webster NR-21963 , ~3 or 6–7 weeks post-infection , respectively ) by hepatic portal vein perfusion using 37°C DMEM . Worms were cultured at 37°C/5% CO2 in Basch Medium 169 supplemented with 1X Antibiotic-Antimycotic . In adherence to the Animal Welfare Act and the Public Health Service Policy on Humane Care and Use of Laboratory Animals , all experiments with and care of mice were performed in accordance with protocols approved by the Institutional Animal Care and Use Committees ( IACUC ) of: Stanford University ( protocol approval number 30366 ) ; University of Illinois at Urbana-Champaign ( protocol approval number 15134 ) ; University of Wisconsin–Madison ( protocol approval number M005569 ) . Artificial pond water: 0 . 125 mg/L FeCl3•6H2O , 32 . 25 mg/L CaCl2•2H2O , 25 mg/L MgSO4•7H2O , 42 . 5 mg/L KH2PO4 , 1 . 875 mg/L ( NH4 ) 2SO4 , pH 7 . 2 . Medium F ( Mann et al . , 2010; Ivanchenko et al . , 1999 ) : 1X BME vitamins , 1X BME amino acids , 6 mg/L serine , 2 . 9 mg/L proline , 2 . 4 mg/L L-alanine , 2 . 8 mg/L aspartic acid , 4 . 7 mg/L glutamic acid , 2 . 4 mg/L glycine , 2 . 4 mg/L β-alanine , 40 mg/L malic acid , 30 mg/L ketoglutaric acid , 10 mg/L succinic acid , 5 mg/L fumaric acid , 10 mg/L citric acid , 70 mg/L Na2HPO4 , 0 . 53 g/L CaCl2•2H2O , 0 . 15 g/L KCl , 0 . 45 g/L MgSO4•7H2O , 1 . 5 g/L NaCl , 4 . 5 g/L galactose , 1 g/L glucose , 25 mM HEPES , pH 7 . Sporocyst culture medium is modified from Ivanchenko et al . ( 1999 ) : 10% heat inactivated FBS , 23 . 5% Medium F , 23 . 5% DMEM/F12 , 10% Schneider’s Drosophila Medium , 2 g/L lactalbumin hydrolysate , 0 . 6 g/L galactose , 55 µM 2-mercaptoethanol , 0 . 005% Chemically defined lipid ( Invitrogen ) . Modified Basch 169 medium ( Mann et al . , 2010 ) : 1 g/L lactalbumin hydrolysate , 1 g/L glucose , 8 mg/L insulin , 1 μM serotonin , 1 μM hydrocortisone , 0 . 5 μM hypoxanthine , 0 . 2 μM triiodothyronine , 0 . 5X MEM vitamins , 5% Schneider’s Drosophila Medium , 10% heat inactivated FBS , 10 mM HEPES , pH 7 . To ablate dividing cells , cercariae or juveniles were exposed to either 200 Gy of γ-irradiation on a Gammacell-220 Excel with a Co60 source ( Nordion ) or 250 Gy of X-ray irradiation on a CellRad Faxitron source and cultured for 48 hr for juveniles and 72 hr for schistosomula . For EdU labeling , schistosomula were pulsed with 2–5 µM EdU . Juveniles were pulsed with 10 µM EdU overnight . EdU incorporation was detected by click reaction with 20–50 µM Alexa Fluor azide conjugates ( Invitrogen ) as described previously ( Wang et al . , 2013; Collins et al . , 2013 ) . Experiments were confirmed on three biological replicates , each containing a cohort of juvenile worms ( N =~50 ) collected from a separate infection . For schistosomula , two to three independent replicates were performed , each batch with ~10–15 worms for each condition . Confirming the specificity of EdU labeling , EdU+ cells were not detected in irradiated schistosomula or juveniles ( Figure 3D , and Figure 4—figure supplement 1C ) . For infected snails , animals were fed with ‘gel food’ containing 60 mg/mL spiruline and 1 mg/mL low-melting agarose in artificial pond water supplemented with 40 µM EdU ( Invitrogen ) . Following the EdU pulse , the snails were washed , cultured at 26°C for 1–2 d , and fixed as described in the previous section . RNAi was performed using previously described protocols ( Wang et al . , 2013; Collins et al . , 2013 ) . Clones were generated using oligonucleotide primers listed in Supplementary file 3 . For RNAi , juveniles were soaked in ~20 µg/mL dsRNA for 2 weeks , with media containing dsRNA refreshed daily . We noticed that worm density is critical to achieve efficient knockdown . We used 20–30 juveniles per mL of medium . Each RNAi was repeated on at least three biological replicates ( each replicate is from a separate infection ) . Every biological replicate contained two technical replicates ( each replicate is one well of a 24-well plate and contains 20–30 worms ) . In rare situations , wells with juveniles showing significantly lower activities at the end of RNAi treatments were excluded from downstream analysis , as this deterioration in overall physiology is likely caused by poor culture conditions . To assess gene expression changes after RNAi , we performed WISH as signal development can be quenched while still in the linear range . The development was performed in parallel in control and RNAi animals and stopped simultaneously . Imaging was performed with identical illumination and exposure settings . RNA FISH experiments were performed as detailed previously ( Wang et al . , 2013; Collins et al . , 2013 ) with modifications specific to each life-cycle stage . To observe various intramolluscan stages , schistosome-infected B . glabrata snails ( 10 , 15 , 25 , or 30 days post-infection , dpi ) were relaxed in sodium pentobarbital solution ( 0 . 5 mg/mL ) for 6 hr , killed in hot water ( 70–90°C ) for 30 s , deshelled , and fixed in 4% formaldehyde in artificial pond water supplemented with 0 . 2% Triton X-100% and 1% NP-40 for 24 hr at 4°C . The snail tissue was then bleached in a formamide bleaching solution ( 0 . 5% formamide , 0 . 5% SSC , and 1 . 2% H2O2 ) for 90 min , equilibrated in 30% sucrose/PBSTx ( PBS with 0 . 3% Triton X-100 ) overnight , embedded in TBS tissue freezing medium , and cryosectioned at 30 µm thickness . Dried cryosections of snail tissues were then rehydrated in PBSTx on gelatin-coated slides , permeabilized by 2 µg/mL proteinase K for 5 min , and post-fixed for 10 min . Schistosomula were killed with ice-cold 1% HCl for 30–60 s before fixation . Fixed , dehydrated in vitro-transformed mother sporocysts were rehydrated , permeabilized by 2 µg/mL proteinase K ( proK , Invitrogen ) for 5 min , and post-fixed for 10 min in 4% formaldehyde in PBSTx . Schistosomula were bleached in the formamide bleaching solution for 10 min , and permeabilized by 5 µg/mL proteinase K for 10 min . Juveniles were killed in 6 M MgCl2 for 30 s-5 min , fixed for 4 hr , dehydrated in methanol , incubated in 3% H2O2 in methanol for 30 min , then rehydrated , permeabilized by 10 µg/mL proteinase K for 10 min , and post-fixed . Adults were permeabilized by 5 µg/mL proteinase K for 45 min . The hybridization step was carried out at 52°C overnight as described previously ( Wang et al . , 2013; Collins et al . , 2013 ) . Following washes , samples were blocked with 5% heat-inactivated horse serum and 0 . 5% Roche Western Blocking Reagent in PBSTx , and then incubated with antibody peroxidase conjugates at 4°C overnight . Detection was performed using tyramide signal amplification ( TSA ) with lab-made reagents ( King and Newmark , 2013 ) . For double FISH , the first peroxidase reaction was quenched for 30 min in 0 . 1% sodium azide solution before detection of the second gene . Imaging was performed in scale solution ( 30% glycerol , 0 . 1% Triton X-100 , 4 M urea in PBS supplemented with 2 mg/mL sodium ascorbate ) . Clones used for riboprobe and dsRNA synthesis were generated as described previously ( Wang et al . , 2013; Collins et al . , 2013 ) , with oligonucleotide primers listed in Supplementary file 3 . WISH follows the same procedure of FISH , except that the detection was carried out using antibody phosphatase conjugates for chromogenic development with NBT/BCIP ( Sigma ) . All FISH/WISH experiments were repeated on at least three biological replicates , each from a separate infection . All expression patterns throughout developmental stages were confirmed on multiple animals , specifically , ~150 in vitro-transformed mother sporocysts , ~5 mother and daughter sporocysts in snails , ~20 cercarial embryos , ~10 schistosomula , ~10 juveniles , and ~5 adults per biological replicate . For intramolluscan stages , as parasites are thicker than 30 µm ( the cryosection thickness ) , we used parasite surface and penetration glands to determine the orientation and the position of the sections . To assign cell classes using FISH signals , confocal stacks were obtained from a laser-scanning microscope using over-sampled resolutions recommended by Imaris ( Bitplane ) . The stacks were resampled to give isotropic voxels , and subjected to Gaussian filtering and background subtraction . Center of labeled cell bodies was segmented channel-by-channel with Imaris using parameters empirically determined to minimize the need for manual curation . Overlapping cells from two channels were merged , and the assignment of cell classes for each individual cell was based on the ratio of integrated intensity within 10 µm ( the cell diameter ) around the respective determined centers between two channels . This analysis provides quantification of co-localization to support our observation of anatomical distributions of different cell classes . We developed a fluorescence-activated cell sorting ( FACS ) strategy to isolate proliferative stem cells . We used DyeCycle Violet ( DCV ) to label live cells proportionally to their DNA content and sorted replicating cells at either S or G2/M phase ( Hayashi et al . , 2006 ) . Schistosomes are covered by a syncytial outer layer impenetrable to typical digestive enzymes used for cell dissociation ( Hahnel et al . , 2013 ) . To overcome this barrier , we briefly treated the parasites with detergents , followed by trypsin to dissociate tissues into cell suspensions . This method dramatically improved the yield of dissociation and reduced the duration of enzymatic digestion to maximize cell viability . Specifically , in vitro-transformed mother sporocysts were permeabilized in PBS containing 0 . 1% Triton X-100% and 0 . 1% NP-40 for 20 s , and washed thoroughly to remove the surfactants . The permeabilized sporocysts were dissociated in 0 . 125% trypsin in HBSS for 10 min and triturated with a 1 mL pipette for 10 min . Cell suspensions were passed through a 100 μm nylon mesh ( Falcon Cell Strainer ) and centrifuged at 150 g for 5 min . Cell pellets were gently resuspended , passed through a 30 μm nylon mesh , and stained with Vybrant DyeCycle Violet ( 5 µM , Invitrogen ) , TOTO-3 ( 0 . 2 µM , Invitrogen ) , and calcein AM ( 0 . 1 µg/mL , Invitrogen ) in sporocyst culture medium for 30–45 min . Dissociation of juveniles was performed similarly but with the following modifications: juveniles were permeabilized for 30 s , dissociated in 0 . 25% trypsin for 20 min , and triturated with serially narrowed flamed-tip glass . Dissociated cells were analyzed on an LSR II flow cytometer or sorted using a FACAria II flow sorter ( BD Biosciences ) , with dead cells excluded based on TOTO-3 fluorescence . We confirmed that the FACS signature of the proliferative cells disappeared as early as 2 days after worms received high doses of X-ray irradiation ( Figure 4—figure supplement 1C ) . All flow sort profiles were confirmed on at least three biological replicates . 250 , 000 stem cells from either sporocysts or juveniles were sorted directly into lysis buffer ( Qiagen ) supplemented with 0 . 6% 2-mercaptoethanol ( Sigma ) , and total RNA was purified using Qiagen RNeasy mini kit . After DNase treatment and poly ( A ) selection , stranded RNA-seq libraries were prepared using TruSeq Stranded RNA Sample Prep kit ( Illumina ) , pooled in equimolar concentrations , and sequenced on a HiSeq2500 sequencer ( Illumina ) to acquire 100- or 160 bp reads with a depth of 40–100 million reads per library . To compare these transcriptomes of purified cell populations to those of whole animals , we also extracted total RNA from approximately 10 , 000 miracidia , in vitro-transformed mother sporocysts ( 48 hr post-transformation ) , or cercariae , or about 1000 juvenile worms , using the standard Trizol ( Invitrogen ) extraction method . All RNAseq data have been submitted to SRA and are available under accession number PRJNA395457 . Single stem cells from in vitro-transformed mother sporocysts were captured on a medium-sized ( 10–17 µm ) microfluidic RNA-seq chip ( Fluidigm ) using the Fluidigm C1 system . Sorted cells were resuspended at a density of 300 cells/µL , with size distribution and number density confirmed on a TC20 cell counter ( Bio-rad ) . The single-cell suspension was then mixed with Fluidigm suspension reagent at 7:3 ratio and loaded onto the chip immediately . After capture , chambers on the chip were examined quickly by phase-contrast microscopy to assess the number , size , and morphology of captured cells and by fluorescence microscopy to examine the live-dead cell stain , and only chambers containing single round-shaped live cells were included in the downstream procedures . cDNAs were prepared on the chip using SMARTer Ultra Low RNA kit for Illumina ( Clontech ) following the manufacturer’s instructions . cDNA quality was quantitated by qPCR analysis of two quality-control genes ( ago2-1 and h2a ) on an Applied Biosystems Step One Plus station using GoTaq qPCR reagents ( Promega ) . Libraries were constructed from this cDNA using Illumina Nextera XT DNA Sample Preparation kit . Library size distribution and concentration were assessed using High Sensitivity DNA analysis kit on an Agilent Bioanalyzer , as well as fluorometrically using Qubit Fluorometer ( Invitrogen ) . Libraries were then sequenced on a HiSeq 2500 to obtain 100 bp reads at a depth of 3–10 million reads per cell . Data from four biological replicates were pooled together for analysis , but two of them were later excluded from final results , as they had a high ‘dead cell’ rate . Mapping of reads to the annotated S . mansoni genome ( Protasio et al . , 2012 ) was performed using CLC Genomics Workbench 7 . 0 ( CLC Bio ) using standard settings . Expression levels were estimated based on transcripts per million ( TPM ) . Ratio of expression levels between libraries was calculated using TPM+1 , and TPM values below or equal to one were considered as being dominated by noise ( Kumar et al . , 2014 ) . In a subset of cells that passed quality control based on live-dead stains , quantitative PCR ( qPCR ) failed to detect ago2-1 and h2a , two ubiquitous stem cell markers ( Wang et al . , 2013; Collins et al . , 2013 ) . RNAseq on these cells exhibited lower overall genomic mapping rates ( <60% ) , and detected significantly fewer transcripts per cell , more intergenic reads , and strong amplification bias towards the most abundant housekeeping genes . These cells were excluded from downstream analysis . The remaining libraries were then inspected for uniformity in mapping depth throughout the ago2-1 transcript and the outlier cells were further excluded , as the schistosome stem cells ( many of them were dividing ) may be sensitive to the microfluidic capture process , leading to mRNA degradation . Although this stringent data selection step reduced the number of cells analyzed , it suppressed variations between cells that are likely attributable to technical artifacts . PCA was performed using genes expressed in at least three cells and showing variance and bimodality coefficient of expression levels across all cells greater than empirically determined thresholds . Hierarchical clustering was performed using Euclidean distance metric on expression levels standardized gene-by-gene by mean-centering and dividing by the standard deviation of expressing cells . Assignment of cell classes is based on hierarchical clustering . As TPM values are well characterized as log-normal distributions for housekeeping genes , log2 ( TPM+1 ) was used as a measure of expression level in PCA and hierarchical clustering . For single-cell analysis of juvenile stem cells , cells were sorted directly into 96-well plates that contained 5 μL 1X CellsDirect One-Step Reaction mix ( Invitrogen ) supplemented with 0 . 05 μL RNaseOUT ( Invitrogen ) in each well . Sorted cells were immediately frozen on dry ice and kept at −80°C until reverse transcription ( RT ) . After thawing plates containing sorted cells , each well was supplemented with 5 μL 1X CellsDirect One-Step Reaction mix that also contained 0 . 2 μL SuperScript III/Platinum Taq mix ( Invitrogen ) and outer primer pairs with a final concentration of 10 nM per primer . Reverse transcription was performed at 50°C for 20 min and stopped by heating the plate to 95°C for 2 min . The cDNA was then amplified for 20 cycles ( 95°C 15 s , 60°C 4 min ) before digestion of the remaining outer primers with ExoI ( 20 U/well , New England Biolabs ) at 37°C for 30 min and inactivation of the RT enzyme at 80°C for 15 min . Amplified cDNA samples were diluted 1:6 in water . We performed 10 technical replicates ( each replicate consisting of one 96-well plate ) . For quality control , 5 μL of each well was used to quantify h2a levels by qPCR . We randomly picked cells from wells that generated CT values within 4 CT around the most probable values ( ~75% of total wells ) for multiplex qPCR on the Fluidigm Biomark platform . For each reaction , 5 μL of diluted cDNA was loaded on a 96 . 96 DynamicArray IFC chip ( Fluidigm ) along with negative controls . Expression levels were assessed using inner ( nested ) primers for each gene . The primer sets are listed in Supplementary file 2 . CT values from the DynamicArray chip qPCR were determined from amplification curves with Fluidigm Real-time PCR analysis software using auto ( detectors ) thresholding and linear ( derivative ) baseline correction with a quality threshold of 0 . 65 . The limit of detection was determined as 22 CT based on negative controls; undetected genes or CT values greater than 22 were all adjusted to 22 . Expression values in log space were calculated as 22-CT . About 10% of cells showing substantially fewer numbers of genes detected were excluded from downstream analysis . To estimate technical variability , two independent sets of nested primers were designed for three genes , ago2-1 , mier ( Smp_101370 ) , and hmt ( Smp_055310 ) , expression levels of which cover the full dynamic range of the qPCR analysis . The technical noise was determined as 2–3 CT and inversely correlated with gene expression level . PCA was performed on 90 amplicons ( 87 genes with the extra three technical-variability controls ) . Subsequently , genes with the highest scores in the first two PCs were identified and used for hierarchical clustering of cells . For hierarchical clustering , expression levels were standardized gene-by-gene by mean-centering and dividing by the standard deviation of expressing cells . Fixed and bleached whole snails were rendered transparent by clearing in 50 , 75 , and 100% tetrahydrofuran ( THF ) in water , followed by dichloromethane , and hexane , successively , for 12–24 hr at each step ( Ertürk et al . , 2012 ) . Specimens were then rehydrated through 100% THF , 50% THF , and PBSTx ( PBS with 0 . 3% Triton X-100 ) , and then bleached either in 6% H2O2 in PBSTx overnight or in 0 . 5% formamide , 0 . 5X SSC , and 1 . 2% H2O2 for 90 min . Lectin stainings were performed as previously described ( Wang et al . , 2013 ) , with 12–24 hr incubation times at every step . Imaging was performed in RapiClear ( Sunjinlab ) .	Parasitic flatworms called schistosomes infect around 250 million people , causing the disease schistosomiasis . Schistosomes live complex lives , spending part of their life cycle inside snails and part of it inside mammals; short-lived , water-borne stages infect each of these hosts . To thrive in such different environments , schistosomes go through several life-cycle stages . At each stage the flatworms transition to a new body plan adapted to its new environment . Understanding how these transitions occur could help researchers devise new strategies for eliminating these parasites . Previous research suggested that stem cells help schistosomes transition to new body plans . Stem cells have the ability to transform into many different cell types , and have been found in schistosome larvae and adults . However , the relationship between the larval and adult stem cells was not clear . Wang et al . used transcriptional profiling , a technique that measures the genes currently in use in different cells , to study the stem cells in the schistosome species Schistosoma mansoni . This uncovered four types of stem cell , each of which uses a slightly different combination of genes . Examining the behaviour of these cells at different schistosome life-cycle stages revealed that certain larval stem cells produce adult stem cells . Other larval stem cells seem to be the source of the ‘germline’ cells that make gametes ( egg and sperm ) and allow the parasites to reproduce sexually . Schistosomes only produce germline cells when they are inside mammals . Wang et al . found that as juvenile flatworms develop inside mouse blood vessels , a gene called eledh becomes active in some of their stem cells . Further investigation showed that this activity is the earliest indicator that germline cells are developing and is also required for proper development of the germline . This knowledge , along with future work to characterize the roles of the stem cell populations identified by Wang et al . , could ultimately help researchers develop new ways to stop the spread of schistosomiasis .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine", "microbiology", "and", "infectious", "disease" ]	2018	Stem cell heterogeneity drives the parasitic life cycle of Schistosoma mansoni
Histone acetyl transferases ( HATs ) play distinct roles in many cellular processes and are frequently misregulated in cancers . Here , we study the regulatory potential of MYST1- ( MOF ) -containing MSL and NSL complexes in mouse embryonic stem cells ( ESCs ) and neuronal progenitors . We find that both complexes influence transcription by targeting promoters and TSS-distal enhancers . In contrast to flies , the MSL complex is not exclusively enriched on the X chromosome , yet it is crucial for mammalian X chromosome regulation as it specifically regulates Tsix , the major repressor of Xist lncRNA . MSL depletion leads to decreased Tsix expression , reduced REX1 recruitment , and consequently , enhanced accumulation of Xist and variable numbers of inactivated X chromosomes during early differentiation . The NSL complex provides additional , Tsix-independent repression of Xist by maintaining pluripotency . MSL and NSL complexes therefore act synergistically by using distinct pathways to ensure a fail-safe mechanism for the repression of X inactivation in ESCs . Histone acetyl transferases ( HATs ) are among the key architects of the cellular epigenetic landscape as the acetylation of histones is unanimously associated with transcriptionally active domains . Many HATs also have the ability to acetylate non-histone proteins extending their influence to diverse cellular pathways inside and outside of the nucleus ( reviewed in Sapountzi and Cote , 2011 ) . Based on their catalytic domains , the HATs are classified into two major families , GCN5 N-acetyl transferases ( GNATs ) and MYST HATs ( named after the founding members MOZ , Ybf2/Sas3 , Sas2 , Tip60 ) , that encompass diverse sets of protein complexes . The individual complex members enhance and modulate the enzymes' activities , guiding the versatile HATs towards specific functions . GCN5 , for example , is part of SAGA , ATAC , and SLIK complexes that are associated with distinct histone tail modifications and differential gene regulation ( reviewed in Lee and Workman , 2007; Nagy et al . , 2010 ) . In contrast , one of the well-known members of the MYST family , MOF ( also known as: KAT8 , MYST1 ) , is rather substrate-specific for lysine 16 of histone H4 ( H4K16 ) ( Akhtar and Becker , 2000 ) and its interaction partners are thought to mainly alter the specificity and extent of MOF's H4K16 acetylation ( H4K16ac ) . As part of the male-specific lethal ( MSL ) complex ( MSL1 , MSL2 , MSL3 , MOF , MLE , roX1 and roX2 lncRNAs ) in Drosophila melanogaster , MOF is recruited to the single X chromosome of male flies . The subsequent spreading of H4K16 acetylation results in transcriptional upregulation of the male X chromosome , the major means of D . melanogaster dosage compensation ( reviewed in Conrad and Akhtar , 2011 ) . In addition to the highly specialized MSL-associated role , MOF is also involved in the more universal and sex-independent regulation of housekeeping genes within the non-specific lethal ( NSL ) complex ( NSL1 , NSL2 , NSL3 , MBD-R2 , MCRS2 , MOF , WDS ) ( Mendjan et al . , 2006; Raja et al . , 2010; Feller et al . , 2012; Lam et al . , 2012 ) . MOF and most of its interaction partners are conserved in mammals , where MOF is also responsible for the majority of H4K16 acetylation ( Smith et al . , 2005; Taipale et al . , 2005 ) . MOF is essential for mammalian embryonic development and unlike the male-specific lethality in Drosophila , deletion of Mof in mice is lethal for both sexes ( Gupta et al . , 2008; Thomas et al . , 2008 ) . More specifically , mammalian MOF is critical for physiological nuclear architecture ( Thomas et al . , 2008 ) , DNA damage repair ( Gupta et al . , 2008 ) , maintenance of stem cell pluripotency ( Li et al . , 2012 ) , differentiation of T cells ( Gupta et al . , 2013 ) , and survival of post-mitotic Purkinje cells ( Kumar et al . , 2011 ) . Compared to MOF , mammalian MSL and NSL complex members are poorly understood . Nevertheless , the individual complex members appear to have important functions in vivo as mutations of the NSL complex member KANSL1 cause the core phenotype of the 17q21 . 31 microdeletion syndrome ( Koolen et al . , 2012; Zollino et al . , 2012 ) and are common amongst patients with both Down syndrome and myeloid leukemia ( Yoshida et al . , 2013 ) . Another NSL-associated protein , PHF20 has been shown to associate with methylated Lys370 and Lys382 of p53 ( Cui et al . , 2012 ) and to be required for somatic cell reprogramming ( Zhao et al . , 2013a ) . WDR5 was shown to be an essential regulator of the core transcription network in embryonic stem cells ( Ang et al . , 2011 ) . The mammalian counterpart of Drosophila MSL2 was shown to have the capacity to ubiquitylate p53 ( Kruse and Gu , 2009 ) and lysine 34 of histone 2B ( Wu et al . , 2011 ) . In the study presented here , we set out to dissect the mammalian MOF functions within the MSL and NSL complexes using genome-wide chromatin immunoprecipitation and transcriptome profiles and biochemical experiments for the core members of MSL and NSL complexes in mouse embryonic stem cells ( ESCs ) and neuronal progenitor cells ( NPCs ) . We found that the MSL and NSL members possess concurrent , as well as independent functions and that effects generally attributed to MOF are frequently accompanied by the NSL complex . The NSL complex abundantly binds to promoters of broadly expressed genes in ESCs and NPCs . These genes are predominantly downregulated upon depletion of either MOF or KANSL3 . In contrast , the MSL complex shows more restricted binding in ESCs , which expands after differentiation , particularly at NPC-specific genes . In addition to promoter-proximal binding , we discover several thousand binding sites of KANSL3 and MSL2 at promoter-distal loci with enhancer-specific epigenetic signatures . The majority of these distal regulatory sites are bound in ESCs , but not in differentiated cells , and genes that are predicted to be targeted by TSS-distal binding of MSL2 are frequently downregulated in shMsl2-treated cells . The distinct , yet synergistic actions of both complexes become very apparent at the X inactivation center ( XIC ) that encodes numerous non-coding RNAs involved in the silencing of one of the two X chromosomes in differentiating female cells . We show that the MSL but not the NSL complex directly promotes expression of Tsix , the inverse transcript and the key murine repressor of Xist during early differentiation . Depletion of MSL proteins results in attenuation of Tsix transcription , enhanced Xist RNA accumulation and ‘chaotic’ inactivation of variable numbers of X chromosomes during early differentiation . In addition to the very specific effect of MSL1/MSL2-depletion on the XIC genes , we show that MOF together with the NSL complex also influences Xist levels , but instead of affecting Tsix , MOF and KANSL3 depletion diminish key pluripotency factors involved in repressing Xist . Our study provides novel insights into the intricate interplay between MSL and NSL complexes in orchestrating gene expression . Furthermore , we demonstrate how MSLs and NSLs ensure the active state of two X chromosomes in mouse embryonic stem cells via distinct mechanisms . To examine the behavior of MSL and NSL proteins in a cell type-specific manner , we derived homogeneous populations of multipotent neuronal progenitor cells ( NPCs ) from mouse embryonic stem cells ( ESCs ) ( Conti et al . , 2005; Splinter et al . , 2011; Gendrel et al . , 2014 ) . We followed the progress of the differentiation process by monitoring cell morphology ( Figure 1A ) , as well as protein ( Figure 1B ) and transcript levels of ESC- and NPC-specific markers ( Figure 1—figure supplement 1A–C ) . To gain a better understanding of how MOF-associated complexes behave throughout the differentiation process , in parallel to cell type-specific markers , we also monitored the RNA and protein levels of MOF , MSL ( MSL1 , MSL2 ) , and NSL ( KANSL1 , KANSL3 , MCRS1 ) complex members ( Figure 1B , Figure 1—figure supplement 1A ) . Interestingly , MSL and NSL complex members showed distinct RNA and protein dynamics during the process of differentiation: KANSL1 and KANSL3 protein levels remained unchanged , whereas MSL1 , MSL2 and MOF became more abundant in NPCs accompanied by increased H4K16 acetylation ( H4K16ac ) ( Figure 1B ) . These results were confirmed using another ES cell line and its NPC derivative ( Figure 1—figure supplement 1D ) . The specificities of the antibodies were confirmed by co-immunoprecipitation assays ( Figure 1—figure supplement 2A–C ) , as well as shRNA-mediated knockdowns followed by western blot analyses ( for individual knockdowns please see below ) . 10 . 7554/eLife . 02024 . 003Figure 1 . Distinct dynamics of MOF , MSL and NSL complexes during differentiation from ESCs to NPCs . ( A ) We monitored the cell morphology during differentiation of mouse embryonic stem cells into neuronal progenitor cells ( NPC ) via embryoid body formation ( EB ) with bright field microscopy . The day of differentiation is indicated in white boxes . ( B ) Western blot analysis for ESC to NPC differentiation . Stages of differentiation together with the day of differentiation ( d0–d15 ) are indicated on top . GAPDH and histone 3 ( H3 ) were used as loading controls . For expression analysis see Figure 1—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02024 . 00310 . 7554/eLife . 02024 . 004Figure 1—figure supplement 1 . Monitoring RNA and protein levels in ESCs and NPCs . ( A ) We monitored the expression dynamics during ESC differentiation for markers of pluripotency ( Oct4 , Nanog , Rex1 , Klf4 ) , embryoid body formation ( Fgf5 ) , differentiation ( Sox2 ) , and NPC ( Nestin ) . Panels 3 and 4 contain the expression profiles for members of the MSL complex ( Msl1 , Msl2 ) , Mof , and the NSL complex ( Kansl1 , Kansl3 , Mcrs1 ) , respectively . All results are represented as relative values individually normalized to Rplp0 expression levels ( panel 2 ) on a given day and to the highest expression level of a given gene during the entire differentiation process ( highest expression level of each gene = 1 ) . The x-axes show days of differentiation . All results are expressed as means ± SD for technical replicates . For primers see Supplementary file 3C . ( B ) Bright field images illustrate the cell morphology before and after the process of differentiation . The immunofluorescence analysis indicates the specific staining for the NESTIN ( green ) in neuronal progenitors ( NPC ) ; DNA is counterstained with DAPI ( blue ) . ( C ) Expression changes for selected ESC-specific and NPC-specific markers before and after differentiation of wild-type WT26 cells using RT-PCR analysis and RNA-seq . ( D ) Western blots for proteins from two ES cell lines and their NPC derivatives . Different dilutions were loaded ( 100% , 30% , 10% ) with the order indicated on top of the blots . Anti-GAPDH was used as loading control; arrows indicate the protein of interest . DOI: http://dx . doi . org/10 . 7554/eLife . 02024 . 00410 . 7554/eLife . 02024 . 005Figure 1—figure supplement 2 . Verification of antibodies used in this study . ( A ) Immunoprecipitations from ESC nuclear extracts with antibodies specific for KANSL1 , KANSL3 or MOF , and rabbit or rat antisera . The blot was probed with indicated antibodies showing the co-immunoprecipitation of several NSL complex members . Pol II = RNA Polymerase II . ( B ) and ( C ) same as ( A ) except that immunoprecipitations were performed with antibodies specific to MSL1 ( B ) and MSL2 ( C ) . Asterisks represent the IgG signal . DOI: http://dx . doi . org/10 . 7554/eLife . 02024 . 005 To assess the distinct behaviors of the complexes in more detail , we generated genome-wide chromatin binding profiles for MSL1 , MSL2 ( MSL complex ) , KANSL3 , MCRS1 ( NSL complex ) , and MOF ( MSL and NSL ) . ChIP-seq experiments in ESCs and NPCs ( Figure 2 ) yielded large numbers of high-quality DNA sequence reads and excellent agreements between the biological replicates ( Figure 2—figure supplement 1A , Supplementary file 1A ) . Using MACS for peak calling ( Zhang et al . , 2008 ) and additional stringent filtering ( ‘Materials and methods’ ) , we scored between 1500 and 15 , 000 regions of significant enrichments for the different proteins ( Supplementary file 1B ) . 10 . 7554/eLife . 02024 . 006Figure 2 . Distinct and shared binding sites of MOF and its complexes in mouse ESCs and NPCs . We applied unsupervised clustering on the union of peaks from all ChIP-seq samples and thereby identified five distinct groups of binding for MOF , MSL and NSL proteins in ESCs and NPCs . Shown here are the input-normalized ChIP signals for each cluster of peaks including a size-matched control set of random genomic regions . The order of the regions is the same for all columns . The pie charts on the left indicate the number of regions from each cluster that overlap with gene bodies , the region 1 kb upstream of genes' TSS or intergenic regions . DOI: http://dx . doi . org/10 . 7554/eLife . 02024 . 00610 . 7554/eLife . 02024 . 007Figure 2—figure supplement 1 . ChIP-seq quality measures . ( A ) Correlation plot for all individual ChIP-seq and input samples from ESCs ( left ) and NPCs . The genome was sampled in windows of 10 kb length; the numbers of reads per bin were counted for each ChIP sample and correlated using Pearson correlation . The calculation and heatmap visualization were done with the bamCorrelate module from the deepTools suite ( Ramirez et al . , 2014 ) . ( B ) The bar chart depicts the fraction of ChIP-seq peaks for each protein that reside within each cluster shown in Figure 2 , that is approximately 30% of MSL1 peaks in ESCs locate in cluster E . Note that the absolute numbers of peaks differ between the samples ( see Supplementary file 1B for absolute peak numbers and ‘Materials and methods’ for peak calling details ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02024 . 007 To uncover patterns of co-occurrence and independent binding , we used unsupervised clustering on the input-normalized signals . This unbiased approach allowed us to determine five main groups of binding distinguished by different combinations of the proteins and cell-type-specific dynamics . As shown in Figure 2 , three large clusters of binding sites encompassed regions , where at least 1 of the investigated proteins was present both in ESCs and NPCs ( clusters A , B and C ) . The binding sites of clusters A and B predominantly overlapped with annotated transcription start sites ( TSS ) in contrast to the regions that were bound exclusively in ESCs , which tended to contain inter- and intragenic regions ( clusters D and E , Figure 2 ) . The width of the enrichments did not differ profoundly between the groups ( cluster E: 836 bp median width , cluster A: 1782 bp median width ) . We found surprisingly few regions where MOF associated primarily with MSL complex members . Instead , approximately 80% of all MOF peaks displayed strong KANSL3 and MCRS1 signals ( cluster B , see Figure 2 and Figure 2—figure supplement 1B ) , suggesting a predominant role of the NSL complex among MOF-associated complexes and a more specific role for the MSL complex at subsets of promoters and numerous intergenic and intronic regions . As the different clusters showed distinct enrichment patterns and diverse genomic localization , we set out to analyze the individual groups of binding in more detail . We first focused on the characterization of target promoters as the majority of MOF-binding was found around the TSS ( mostly clusters A and B in Figure 2 , Figure 3A ) . We identified 8947 TSSs overlapping with ChIP-seq peaks of KANSL3 and/or MCRS1 in ESCs that encompassed virtually all MOF- and MSL-bound TSSs ( Figure 3B ) . This pattern did not change substantially in NPCs where TSSs overlapping with MOF peaks almost always ( 99% ) showed significant enrichments of KANSL3 and in 35% of the cases additionally contained a peak of MSL2 ( Figure 3B , middle panel ) . Genes that were TSS-bound in ESCs tended to be bound in NPCs as well ( Figure 3B , middle panel and Figure 3—figure supplement 1A ) . We next generated RNA-seq data for ESCs and NPCs , determined genes that were expressed in both cell types ( FPKM >4 ) and found that all ChIPed proteins preferably bound to the promoters of active genes ( Figure 3C ) . Interestingly , in ESCs , genes whose TSSs were bound by members of both complexes showed higher median expression values than genes bound by only one complex ( Figure 3—figure supplement 1B ) . In contrast to the differing expression values , analysis of gene ontology ( GO ) using DAVID ( Huang da et al . , 2009 ) revealed basic housekeeping functions for both gene groups , regardless of whether they were bound by the NSL complex only or by both MOF-complexes together ( Figure 3—figure supplement 1C ) . Consistently , the promoters of all target gene groups were enriched for motifs associated with broad , non-cell-type-specific expression such as ELK1 , YY1 , CREB , and E2F ( Xie et al . , 2005; Farre et al . , 2007 ) and showed profound enrichments of CpG islands ( Figure 3—figure supplement 1D ) , which is indicative of housekeeping genes ( Landolin et al . , 2010 ) . Interestingly , when we analyzed the subset of genes that gained binding of either KANSL3 or MSL2 in NPCs , we found strong enrichments of GO terms related to embryonic development for KANSL3 targets and cell migration and neuronal development for MSL2 targets . 10 . 7554/eLife . 02024 . 008Figure 3 . Both MOF-complexes bind to the TSS of broadly expressed genes in mouse ESCs and NPCs . ( A ) Genome browser snapshots of genes targeted by MSL and NSL complexes or by the NSL complex only . Signals were sequencing-depth-normalized and from ESCs . For ChIP-qPCR-based validation of the signals see Figure 3—figure supplement 4B . ( B ) Venn diagrams of genes whose promoter regions ( TSS ± 500 bp ) overlapped with ChIP-seq peaks of NSL complex members ( KANSL3 and/or MCRS1 ) , MOF and MSL complex members ( MSL1 and/or MSL2 ) . The right-most panel depicts the overlap of genes bound by at least one factor in ESCs and NPCs . ( C ) The heatmaps display the input-normalized ChIP enrichments of MOF , MSL2 and KANSL3 around the TSS of genes that were active in ESCs as well as NPCs based on RNA-seq data that we generated for both cell types . ( D ) Summary plots of genes bound by the NSL complex in D . melanogaster for which mouse homologues were found . The input-normalized ChIP-seq signals around the TSS reveal markedly increased binding of MOF for male X-linked fly genes ( left panels ) that was not recapitulated in the mouse ( right panels; ChIP-seq signals from ESCs ) . Fly genes were scaled to 1 . 2 kb and values were extracted from published data sets , mouse genes were scaled to 30 kb . ( E ) Heatmap depicting results of RNA-seq experiments from different shRNA-treated cells . The colors correspond to log2 fold changes ( shRNA-treated cells/scrambled control ) for genes whose expression was significantly affected in all knockdown conditions . Values were ordered using hierarchical clustering . ( F ) Bar plot of gene counts for different gene classes . We determined significantly up- and downregulated genes for each knockdown condition and binned them according to their expression strength in wild-type ESCs ( high , intermediate , low ) . Then , for each gene , information about the TSS-targeting was extracted from the corresponding ChIP-seq sample . Non-target genes are neither bound at the promoter nor the gene body and were not predicted to be regulated via TSS-distal binding sites in any of the 5 ChIP-seq ESC samples . For details on the target classification see ‘Materials and methods’ . ( G ) Western blot analysis of MSL and NSL complex members and H4K16 acetylation in scrambled- , Mof- , Msl1- , and Kansl3-shRNA-treated male ESCs . Three concentrations ( 100% , 30% , 10% ) of RIPA extract were loaded per sample . Asterisks mark the position of unspecific bands; triangles indicate the protein of interest . DOI: http://dx . doi . org/10 . 7554/eLife . 02024 . 00810 . 7554/eLife . 02024 . 009Figure 3—figure supplement 1 . MSL and NSL complexes target promoters of broadly expressed genes in ESCs and NPCs . ( A ) The heatmap is related to Figure 3B as it is based on all genes that are bound by at least 1 ChIPed factor in ESCs or NPCs . The intensity of the color depicts the fraction of the 1 kb TSS-region that was covered by a binding site of MOF , MSL1 , MSL2 , KANSL3 or MCRS1 . Rows and columns were sorted using hierarchical clustering on the Euclidean distances of the overlap fractions using R . The left color bar indicates which genes are targeted in 1 or both cell types . ( B ) Distribution of expression values from RNA-seq data in ESCs and NPCs for genes targeted by MSL and NSL complex members together or by the NSL complex only . p-values were calculated using Welch t test . ( C ) Results of the GO term analysis using DAVID ( Huang da et al . , 2009 ) on genes that were bound at the TSS in ESCs by NSL complex members only or both MSL and NSL complexes . ( D ) The pie charts depict how many times annotated TSSs overlapped with a CpG island . The vast majority of genes that were bound in ESCs by MSL and NSL together or by NSL complex members alone overlapped with at least 1 CpG island ( dark and medium blue ) while approximately 2/3 of the non-target-TSS did not overlap with any CpG island ( light blue for 0 CpG islands within the queried regions ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02024 . 00910 . 7554/eLife . 02024 . 010Figure 3—figure supplement 2 . The NSL- , but not the MSL-binding mode of D . melanogaster is present in mammalian cells . ( A ) Exemplary genome browser snapshots of the X-linked fly gene CG4419 . Shown here are the sequencing-depth normalized profiles for ChIP and corresponding input samples , clearly showing a broad enrichment of MOF and MSL1 along the entire gene body in male ( m ) D . melanogaster while all other marks show sharp enrichments around the TSS ( including MSL1 and MOF in female ( F ) D . melanogaster ) which are similar to those seen for both complexes in mouse cells ( Figure 3A , D ) . ( B ) Comparison of expressed ( FPKM >4 ) mouse genes whose homologous genes are either bound or not bound by MOF and its complexes in the fly . We extracted the input-normalized ChIP-seq values for 6 kb regions around the TSS using the computeMatrix module of deepTools ( Ramirez et al . , 2014 ) . H3K4me3 signal is from a published data set , see Supplementary file 2 for the corresponding accession number . DOI: http://dx . doi . org/10 . 7554/eLife . 02024 . 01010 . 7554/eLife . 02024 . 011Figure 3—figure supplement 3 . Effects of shRNA-mediated depletion of MOF , MSL1 , MSL2 and KANSL3 . ( A ) Time course of knockdown experiments . For experimental details see ‘Materials and methods’ . Samples for RNA-sequencing and AP staining ( Figure 4—figure supplement 4 ) were extracted 4 days after puromycin selection of shRNA-treated cells . ( B ) Proliferation assay for shRNA-treated cells , starting at day 4 after puromycin selection ( Figure 3—figure supplement 3A ) . ( C ) Bar plots depicting the fractions of genes ( per chromosome ) that were significantly up- or downregulated in RNA-seq experiments from shRNA-treated cells . The left plot contains genes which were defined as TSS-targets in the respective ChIP-seq samples , the right plot contains genes that were neither classified as TSS- nor as TSS-distal targets . The labels on each bar indicate the chromosome name and the total number of genes that fulfilled the criteria for this chromosome ( significantly affected , TSS-bound or non-targeted ) . See ‘Materials and methods’ for details of the classification . DOI: http://dx . doi . org/10 . 7554/eLife . 02024 . 01110 . 7554/eLife . 02024 . 012Figure 3—figure supplement 4 . Assessment of ChIP signals around the TSSs of putative target genes as determined by ChIP-seq . ( A ) Genome Browser snapshots of several MSL/NSL ( first three from the left ) or NSL-only target genes and respective sequencing-depth-normalized ChIP-seq and input signals from ESCs . The exact genomic coordinates are indicated on top of each panel . Gene names are indicated on the bottom . ( B ) ChIP-qPCR validation for MOF ( green ) , KANSL3 ( purple ) , MSL1 ( blue ) and MSL2 ( red ) signals . Immunoprecipitated DNA was amplified by qPCR with primer sets positioned at the promoter ( P ) and end ( E ) of the coding sequence ( Supplementary file 3A ) . Results are expressed as mean ± SD of three biological replicates; cells were harvested for experiments on day 4 ( Kansl3 , Msl1 , Msl2 ) or 5 ( Mof ) of knockdown . ( C ) ChIP-qPCR for MSL1 ( blue ) , MSL2 ( red ) and KANSL3 ( purple ) in ESCs treated with sh-RNA ( scrambled or shMof ) . Signals on genes were evaluated using primers at the promoter ( P ) , and end ( E ) of the coding sequence . Results are expressed as mean ± SD of three biological replicates; cells were harvested for experiments on day 5 of Mof knockdown . DOI: http://dx . doi . org/10 . 7554/eLife . 02024 . 012 MOF has traditionally been associated with a widespread enrichment along male X-linked genes in flies that is dependent on the MSL proteins ( Figure 3D , Figure 3—figure supplement 2A ) . In our mammalian profiles , despite the presence of the MSLs , we could neither detect X-specific enrichments of MOF nor broad domains of binding along gene bodies . Furthermore , promoter-distal binding sites consisted of narrow peaks and no evidence of spreading from intronic or intergenic regions was observed ( Figures 2 , 3A , D ) . We then examined whether there was a correlation between NSL complex binding in D . melanogaster and mouse cells . Indeed , we found that mouse genes that were homologous to NSL complex targets in D . melanogaster had a high probability of being bound by the murine NSL complex as well ( Pearson's Chi squared test of independence between NSL binding in the fly and the mouse , p-value <2 . 2e−16 ) . We additionally observed that mouse genes expressed in ESCs and NPCs , whose fly homologues were NSL targets , showed stronger signals for H3K4me3 , MOF , KANSL3 , and MCRS1 ( but not for MSL1 or MSL2 ) than the mouse homologues of non-NSL-bound D . melanogaster genes ( Figure 3—figure supplement 2B; lists of NSL-bound and NSL-non-bound fly genes were from Lam et al . , 2012 ) . These findings support the notion that the function in housekeeping gene regulation by the D . melanogaster NSL complex is evolutionary conserved . To dissect the biological consequences of the gene targeting by the different MSL and NSL proteins in ESCs , we systematically depleted core members of both complexes ( MOF , KANSL3 , MSL1 , MSL2 ) ( Figure 3—figure supplement 3A ) . Interestingly , MOF- or KANSL3-depleted cells showed more severe proliferation defects than MSL1- and MSL2-depleted cells ( Figure 3—figure supplement 3B ) . We subsequently performed RNA-seq experiments from shRNA-treated cells and determined their differential expression against the scrambled control to dissect transcriptional outcomes of the depletions at a global level . We found a striking overlap between the differential expression of MSL1 and MSL2 knockdowns and a higher resemblance of MOF-dependent differential expression to that of KANSL3-depletion ( Figure 3E ) . When we specifically focused on genes that we had identified as TSS-bound in our ChIP-seq samples , we found that their transcripts tended to be downregulated in all four knockdowns in comparison to untargeted genes which showed higher fractions of upregulation . These effects were independent of the wild-type expression status of the gene or the chromosome ( Figure 3F , Figure 3—figure supplement 3C ) . Turning to the assessment of protein levels in shRNA-treated cells , we detected markedly reduced bulk H4K16 acetylation in MSL1- and MOF-depleted cells and only slight reduction upon KANSL3-depletion . This is consistent with previous reports that indicate MSL1 as the major enhancer of MOF's H4K16 acetylation ( Kadlec et al . , 2011 ) and demonstrate relaxed substrate specificity for the NSL complex ( Zhao et al . , 2013b ) . In addition , we found that MSL1-depletion affected the levels of MSL2 but not of NSL complex members while the depletion of KANSL3 moderately decreased protein levels for both complexes ( Figure 3G ) . ChIP-qPCR assays in MOF-depleted cells revealed that MSL1 and KANSL3 do not require the presence of MOF to bind to gene promoters , which is in agreement with previous observations in D . melanogaster ( Hallacli et al . , 2012; Figure 3—figure supplement 4C ) . In summary , our TSS-focused analysis shows that the localized binding of the NSL complex to the promoters of housekeeping genes appears to be a conserved feature between the mammalian and Drosophila systems . Unlike in the fly , we do not detect an MSL- and X-chromosome-specific binding mode of MOF in the mouse cell lines . Instead , both complexes narrowly bind to TSSs where their co-occurrence is associated with significantly higher median expression values than those solely bound by the NSL complex . Moreover , we found that MOF is dispensable for the TSS recruitment of its interaction partners and that depletions of the individual proteins predominantly result in the downregulation of TSS-bound genes , further supporting the fact that the promoter-binding of the MSL and the NSL complex is associated with active transcription . In addition to promoter-proximal binding , where both the MSL and NSL complex tend to ( co- ) occur constitutively in ESCs and NPCs , we identified a large proportion of binding sites where the proteins were present in a dynamic fashion , that is their binding was observed only in ESCs but not in NPCs ( Figure 2 , clusters D and E ) . In contrast to the binding mode represented by clusters A and B ( Figure 2 ) , here MSL2 , MCRS1 , and KANSL3 were predominantly enriched within introns and intergenic regions that underwent significant CpG methylation upon differentiation ( e . g . , from median 50% CpG methylation in ESCs to more than 80% in NPCs for cluster D; bisulfite sequencing data from Stadler et al . , 2011 ) . As shown in Figure 4A , CpG methylation in NPCs was particularly pronounced around the center of the regions with significant ChIP enrichments in ESCs , indicating a correlation between the loss of ChIP-seq signal for MOF , MSL1 , MSL2 , KANSL3 and MCRS1 , and DNA methylation upon differentiation . In addition , the regions of cluster D and , to a lesser extent the MSL1-rich cluster E ( Figure 2 ) , showed highly localized enrichments of DNase hypersensitivity sites ( DNase HS ) , RNA Polymerase II ( Pol II ) , p300 , methylation of histone 3 on lysine 4 ( H3K4me1 ) , and acetylation of histone 3 on lysine 27 ( H3K27ac ) in ESCs ( Figure 4A ) , which are characteristic features of enhancer regions . We thus examined whether MOF and its interaction partners were enriched on known enhancer regions , using lists of typical and super enhancers defined by binding sites of the pluripotency factors SOX2 , NANOG , and OCT4 ( Whyte et al . , 2013 ) , as well as sets of active and poised enhancers based on histone mark signatures ( Creyghton et al . , 2010 ) . 10 . 7554/eLife . 02024 . 013Figure 4 . MSL and NSL complex members are enriched at regions with enhancer marks in ESCs . ( A ) Shown here are the fractions of methylated cytosines and ChIP-seq read densities of enhancer markers for regions of ESC-specific enrichments of our proteins of interest . We downloaded the different data from public repositories ( see Supplementary file 3A for details ) and calculated the values for the regions of the ESC-specific clusters D and E and random genomic loci . Most data sets used here were from mouse ESC except one RNA Polymerase II ( Pol II ) sample from NPC . All heatmaps were sorted according to the DNase hypersensitivity values except for CpG methylation heatmaps which were sorted according to their own values . ( B ) Summary plots of input-normalized ChIP-seq signals along typical ( TE ) and super enhancers ( SE ) ( Whyte et al . , 2013 ) . Note that we show the ESC-specific TE only while on the right-hand side we show the signal for SE regions from several cell types . Enhancer regions were scaled to 30 kb ( SE ) and circa 700 bp ( TE ) . The heatmaps between the summary plots depict how much of each enhancer region overlaps with ChIP-seq peaks of MSL2 or KANSL3 . ESC = embryonic stem cells ( n = 232 ) , pro-B = progenitor B cells ( n = 396 ) , Th = T helper cells ( n = 437 ) , C2C12 = myotube cells ( n = 536 ) . ( C ) Exemplary genome browser snapshots of annotated super enhancers ( SE , pink boxes ) for three pluripotency factors displaying the sequencing-depth normalized ESC ChIP-seq signals of MSL2 , MOF and KANSL3 . See Figure 4—figure supplement 4C for additional examples . ( D ) Luciferase assays demonstrate the biological activity of regions bound by MOF-associated proteins in ESCs ( ‘in’ stands for intronic region , ‘us’ indicates that the cloned region is upstream of the gene ) . The firefly luciferase gene was cloned under a minimal promoter together with the putative enhancer region in ESCs , NPCs , and 3T3 cells . The graphs represent at least three independent experiments performed in technical triplicates; error bars represent SEM . ( E ) Bar plots depicting the fraction of significantly up- and downregulated genes per chromosome in the different shRNA-treated cells compared to shScrambled controls ( total number of significantly affected genes per sample and chromosome labels are indicated ) . All genes counted here were classified as TSS-distal target genes in the respective ChIP-seq experiments . See ‘Materials and methods’ for details of the classifications . ( F ) Western blot analyses of the pluripotency factors in scrambled- , Mof- , Kansl3- , Msl1- , and Msl2-shRNA-treated male ESCs . For additional analyses in female ESCs see Figure 6C . The respective dilution ( 100% , 30% , 10% ) of loaded RIPA extract is indicated above each panel . Asterisks mark the position of unspecific bands; triangles indicate the protein of interest . GAPDH was used as the loading control . For antibodies see ‘Materials and methods’ . DOI: http://dx . doi . org/10 . 7554/eLife . 02024 . 01310 . 7554/eLife . 02024 . 014Figure 4—figure supplement 1 . MSL2 and KANSL3 show strong enrichments at typical and super enhancers in ESCs . ( A ) Boxplots demonstrating the distribution of mean ChIP enrichments for enhancer regions defined by H3K4me1 and H3K27ac marks in ESCs ( see Creyghton et al . , 2010 for details ) that overlap with the clusters of binding defined by our ChIP-seq samples . Mean values were extracting using the UCSCtool bigWigAverageOverBed . ( B ) Summary plots for typical enhancer regions ( Whyte et al . , 2013 ) that overlapped with either MSL2 ( top ) or KANSL3 ( bottom ) peaks . Different colors indicate different ChIP-seq signals . Related to the heatmaps of Figure 4B . ( C ) Genome browser snapshots of sequencing-depth normalized ChIP-seq and input profiles for super enhancers of key pluripotency factors . DOI: http://dx . doi . org/10 . 7554/eLife . 02024 . 01410 . 7554/eLife . 02024 . 015Figure 4—figure supplement 2 . MOF is moderately enriched at non-canonical enhancers . ( A ) Summary plots of ChIP-seq values for binding sites belonging to cluster D . The regions were divided based on the presence or absence of annotated ESC enhancers ( Creyghton et al . , 2010; Whyte et al . , 2013 ) . ( B ) Heatmaps of ChIP-seq read densities of known enhancer markers for the ESC-specific binding sites of our proteins of interest ( cluster D , see Figure 2 ) and random genomic regions . The binding sites of cluster D ( excluding regions with TSSs ) were divided into two basic groups based on the presence or absence of known ESC enhancers ( Creyghton et al . , 2010; Whyte et al . , 2013 ) . The latter group was further divided into three ( arbitrarily numbered ) sub-clusters based on hierarchical clustering of the values from DNase hypersensitivity sites , p300 , H3K4me1 and our MOF sample ( in ESCs ) . Heatmaps of the ESC-enhancer-containing regions were sorted according to p300 , those of the sub-clustered regions were sorted according to the MOF signal . ( C ) Related to ( B ) , shown here are the corresponding summary plots of ChIP-seq values for cluster D binding sites that do not overlap with annotated enhancer regions ( bottom part of the heatmaps in the figure above ) . The three indicated groups are based on the hierarchical clustering that was performed on p300 , H3K4me1 and MOF values ( ‘Regions without annotated ESC enhancers’ in ( B ) ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02024 . 01510 . 7554/eLife . 02024 . 016Figure 4—figure supplement 3 . MSL2 has intergenic binding sites in DNA-hypomethylated regions that are enriched for SMAD3 binding sites . ( A ) We extracted the percentage of methylated CpGs and the input-normalized ChIP-seq values from KANSL3 and MSL2 and 5 kb surrounding the center of the regions belonging to cluster C ( Figure 2 ) and random genomic control regions . All heatmaps were sorted according to the percentages of methylated CpGs ( Stadler et al . , 2011 ) . ( B ) Motif obtained by MEME analysis on the top 200 MSL2 peaks within cluster C . ( C ) Same as for ( A ) , except that the score was the motif hit score for SMAD3 for 1 kb . See ‘Materials and methods’ for details . DOI: http://dx . doi . org/10 . 7554/eLife . 02024 . 01610 . 7554/eLife . 02024 . 017Figure 4—figure supplement 4 . Biological significance of the TSS-distal binding sites of the investigated proteins . ( A ) Genome browser snapshots of sequencing-depth normalized ChIP-seq and input profiles . Pink boxes mark the regions cloned and transfected into ESCs and NPCs for luciferase assays ( Figure 4D ) . ( B ) Genes that were significantly up- or downregulated in the respective shRNA-treatments compared to shScrambled were classified according to ChIP-seq peak overlaps ( TSS-distal , no target ) and expression strength in wild-type ESCs ( high , intermediate , low ) . See ‘Materials and methods’ for details of the classifications . ( C ) Distribution of absolute log2 fold changes ( shKansl3 or shMsl2 compared to shScrambled ) for significantly downregulated genes . Different shades of orange indicate different target classes based on ChIP-seq experiments for KANSL3 or MSL2 , respectively . p-values were calculated with Welch t test . ( D ) Alkaline phosphatase staining and morphology of ESC colonies in indicated knockdowns after 4 days growth under puromycin selection ( Figure 3—figure supplement 3A ) . MOF- and KANSL3-depleted cells demonstrate reduced alkaline phosphatase positive colonies with increased differentiation compared with MSL1- and MSL2-depleted cells and scrambled control . DOI: http://dx . doi . org/10 . 7554/eLife . 02024 . 017 Interestingly , MSL2 , KANSL3 and MCRS1 , but not MOF and MSL1 , showed profound enrichments for active and poised ESC enhancers ( Figure 4—figure supplement 1A ) as well as along the regions of super enhancers that have been described as being particularly important for maintenance of cell identity ( Whyte et al . , 2013 ) . The signals of MSL2 and KANSL3 were specific for ESC enhancers and wide-spread along super enhancer regions ( Figure 4B , C ) . We noted that enhancers overlapping with MSL2 ChIP-seq peaks tended to show lower KANSL3 enrichments and vice versa , implying that MSL2 and KANSL3 preferred different enhancer regions ( heatmaps in Figure 4B , Figure 4—figure supplement 1B ) . MOF was not enriched at super enhancers and generally , its binding to TSS-distal sites was much less pronounced than to gene promoters ( Figure 4—figure supplement 1A , 1C , Figure 2 ) . Like for TSS-specific binding , MOF was not alone ( 87% of TSS-distal MOF peak regions overlapped with either KANSL3 or MSL2 ) . Since a recent report showed H4K16 acetylation to be present at p300- and H3K27-acetylation-independent enhancer regions ( Taylor et al . , 2013 ) , we analyzed the moderate TSS-distal enrichments of MOF in more detail and observed a slight preference for TSS-distal regions that were not overlapping with previously published ESC enhancer regions ( Figure 4—figure supplement 2A ) . In fact , we detected the strongest MOF signals in regions with rather low enrichments of known enhancer marks ( see DNase HS , p300 , H3K4me1 , H3K27ac in Figure 4—figure supplement 2B and 2C ) , which suggested a preferred binding of MOF outside canonical ESC regulatory regions . In addition to ESC-specific binding of MSL2 and KANSL3 to predicted enhancers , we also identified a very distinct set of TSS-distal binding sites by MSL2 to introns and intergenic regions without enhancer-associated marks ( cluster C in Figure 2 ) . Approximately , 81% of these cluster C regions had solitary MSL2 enrichments without significant signals of any of the other ChIPed proteins . Interestingly , these MSL2 binding sites increased in number and binding strength upon differentiation to NPCs ( 829 solitary MSL2 peaks in ESCs compared to 3635 in NPCs ) . In contrast to the previously described binding sites that were characterized by the prevalence of open , active chromatin ( Figures 3 and 4 ) , here MSL2 was excluded from hypo-methylated DNA regions ( Figure 4—figure supplement 3A; note the different behavior of KANSL3 ) . When we searched the unique MSL2 binding sites for DNA motifs , we obtained a ( CAGA ) n motif ( Figure 4—figure supplement 3B ) that was previously described as a binding site for SMAD3 , a transcription factor that translates the TGF-beta receptor response into gene expression regulation ( Zawel et al . , 1998 ) . When we subsequently scanned all the binding sites for the presence of the published , original SMAD3 motif , we found a strikingly specific signal for the center of the solitary MSL2 ChIP-seq peaks ( Figure 4—figure supplement 3C ) . We conclude that MOF , MSL2 and KANSL3 specifically recognize ESC enhancers . In contrast to MSL-MOF-NSL co-occurrence at housekeeping gene promoters , we found evidence for differential and independent binding of the individual proteins to gene bodies and intergenic regions suggesting the potential for distinct tissue-specific regulatory functions of MSL2 and KANSL3 . These data reveal a newly evolved function of MSL2 and KANSL3 in mammals , which has not been observed in flies . To study the functional implications of the binding of MSL2 and KANSL3 to putative ESC enhancers , we first tested five different regions located near genes related to pluripotency and self-renewal ( Hu et al . , 2009; Young , 2011 ) . Using luciferase reporter constructs , we found strong transcriptional enhancement for all tested regions in ESCs , but not in NPCs or 3T3 cells which correlated with the presence of MSL2 and/or KANSL3 and MCRS1 in ESCs only ( Figure 4D , Figure 4—figure supplement 4A ) . We then used our RNA-seq data sets from MSL1- , MSL2- , MOF- , and KANSL3-depleted cells to assess the effects on the transcription of those genes that were not bound at promoters , but had been predicted by GREAT ( McLean et al . , 2010 ) to be regulated by TSS-distal binding sites of the respective protein . As shown in Figure 4E , we again found similar effects for KANSL3- and MOF-depleted cells compared to MSL1- and MSL2-depleted cells with the latter group showing genome-wide downregulation of predicted target genes . In fact , the numbers of TSS-distal targets of MSL1 or MSL2 that were significantly reduced in the respective shRNA-treatments were markedly larger than for genes where MSL1 or MSL2 bound to the promoter ( compare Figure 3F with Figure 4—figure supplement 4B ) . Moreover , in MSL2- , but not KANSL3-depleted cells , the effects on TSS-distally targeted genes were slightly stronger than for TSS-targets ( Figure 4—figure supplement 4C ) . While TSS-binding predominantly occurred at housekeeping genes , we noticed that the majority of enhancer regions associated with key pluripotency factors ( e . g . , SOX2 , ESRRB , MYC , REX1 , TBX3 , NANOG ) were strongly enriched for MSL2 and KANSL3 . We thus assessed the effects of the protein depletions on pluripotency factors in ESCs and found strongly reduced levels of NANOG , REX1 , and ESRRB in MOF- or KANSL3-depleted cells . Surprisingly , the pluripotency factors remained almost unaffected in cells depleted of MSL1 or MSL2 ( Figure 4F ) . These contrasting results were mirrored by decreased levels of alkaline phosphatase ( AP ) in MOF- and KANSL3- , but not in MSL1- or MSL2-depleted cells ( Figure 4—figure supplement 4D ) . These findings indicate that despite their frequent effects on TSS-distally targeted genes , MSL1 and MSL2 might not show dominant effects at genes that are bound by KANSL3 as well . Therefore , we specifically searched for regions without KANSL3 binding to identify putative MSL-specific functions . As described previously , we identified only a small subset of regions in the mouse genome where MSL complex members were enriched exclusively ( see cluster E in Figure 2 ) . Strikingly , several of these binding sites fall into a region known as the X inactivation center ( XIC ) . The XIC is the X-chromosomal region necessary and sufficient to control the inactivation of one of the two X chromosomes in females ( reviewed in Pollex and Heard , 2012 ) . The XIC site with the strongest concomitant enrichments of MSL1 , MSL2 and MOF was the major promoter ( P2 ) of Tsix and its intronic minisatellite—DXPas34 ( Figure 5A , B ) . DXPas34 is a well-characterized tandem repeat that serves as a binding platform for multiple transcription factors and contains bidirectional enhancing properties essential for the expression of Tsix , the antisense transcript of Xist ( Debrand et al . , 1999; Cohen et al . , 2007; Donohoe et al . , 2007; Navarro et al . , 2010; Gontan et al . , 2012 ) . In rodents , Tsix antisense transcription across the Xist promoter is required for regulating the levels of Xist accumulation . In turn , DXPas34 deletion impairs the recruitment of Pol II and TFIIB to the major promoter of Tsix causing its downregulation ( Vigneau et al . , 2006 ) . 10 . 7554/eLife . 02024 . 018Figure 5 . The MSL complex binds multiple loci within the X inactivation center including the Tsix DXPas34 minisatellite enhancer . ( A ) Genome browser snapshots of the mouse X inactivation center ( approximately 0 . 9 Mb ) ( upper panel ) plus enlargement of the 164 kb region between Chic1 and Jpx/Enox ( lower panel ) . The signals shown are the sequencing-depth normalized profiles for ChIP-seq from ESCs ( for corresponding profiles in NPCs see Figure 5—figure supplement 1A ) ; colored arrows indicate genes of lncRNAs . The schematic representation of the DXPas34 locus depicts the locations of the primer pairs that were used for ChIP-qPCR analyses ( Supplementary file 3B ) . ( B ) Genome browser snapshots of the DXPas34 minisatellite of sequencing-depth normalized ChIP-seq profiles in ESCs and NPCs . ( C ) ChIP-qPCR analyses of MSL1 ( blue ) , MSL2 ( red ) , MOF ( green ) , and H4K16 acetylation ( purple ) across the Tsix major promoter ( P2 ) and the DXPas34 enhancer in male ESCs treated with the indicated shRNAs . For corresponding ChIP-qPCR in female ESCs see Figure 5—figure supplement 1C . Panels in the middle show the effects of MOF depletion on the recruitment of MSL1 and MSL2 to DXPas34 and vice versa . The bottom panel shows effects of depletion of control ( dark pink ) , MOF ( light pink ) and MSL2 ( purple ) on the H4K16 acetylation signal . The labels of the x axes correspond to the arrowheads in ( A ) . Results are expressed as mean ± SD of three biological replicates; cells were harvested on day 4 ( Msl1 , Msl2 ) or 5 ( Mof ) after shRNA treatment . For primer pairs see Supplementary file 3C . DOI: http://dx . doi . org/10 . 7554/eLife . 02024 . 01810 . 7554/eLife . 02024 . 019Figure 5—figure supplement 1 . The MSL proteins bind to multiple loci within the X inactivation center ( XIC ) . ( A ) Genome browser snapshots of the mouse XIC ( top panel ) with three enlargements on Jpx , Ftx and Rnf12 genes ( lower panels ) . Red boxes with corresponding numbers mark the enlarged regions presented in the lower panels . The exact genomic coordinates are indicated on top of each panel , arrows represent genes . The signals shown are the sequencing-depth normalized ChIP-seq profiles in NPCs . ( B ) ChIP analysis of MSL1 , MSL2 and MOF across the DXPas34 minisatellite in female ESCs . The x-axis labels indicate the genomic coordinates corresponding to the arrowheads in Figure 5A . The y-axes show the percentage of ChIP recovery for MSL1 and MSL2 ( left-hand side ) and MOF ( right-hand side ) normalized to input . For all ChIP experiments , three biological replicates were used; all results are expressed as mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 02024 . 019 In addition to the DXPas34 binding site , we detected MSL peaks on the promoters , gene bodies and intronic regions of other key XIC regulators including the genes of the long non-coding ( lnc ) RNAs Xist and Jpx . Additionally , we observed peaks upstream of the Tsx gene and both at the TSS and downstream of the Rnf12 gene ( Figure 5A ) . Products of all of these genes were shown to play important roles in orchestrating the process of X inactivation ( Stavropoulos et al . , 2001; Shin et al . , 2010; Tian et al . , 2010; Anguera et al . , 2011; Chureau et al . , 2011; Gontan et al . , 2012; Sun et al . , 2013 ) . The XIC binding of MSL-MOF was specific to ESCs , as almost all enrichments were abolished upon differentiation , except for some loci upstream of Xist where traces of binding could still be detected in NPCs ( e . g . , Ftx and Jpx TSS , Figure 5—figure supplement 1A ) . We next confirmed the high ChIP-seq enrichments of MSL1 , MSL2 and MOF and assessed H4K16 acetylation on the major promoter of Tsix and along DXPas34 with ChIP-qPCR assays covering the entire region in male and female ESCs ( Figure 5C , Figure 5—figure supplement 1B ) . Interestingly , the recruitment of MOF was almost completely abolished in both MSL1- and MSL2-depleted cells , whereas the depletion of MOF had no effect on MSL1 and MSL2 binding to the Tsix major promoter and DXPas34 ( Figure 5C ) . H4K16 acetylation ChIP signals were severely reduced in both MOF- and MSL2-depleted cells . These results are in agreement with our global observations ( Figure 3G , Figure 3—figure supplement 4C ) and indicate that MSL1 and MSL2 are together necessary and sufficient for the recruitment of MOF and for the deposition of H4K16 acetylation at DXPas34 . To directly assess the functional outcome of MOF- , MSL1- , and MSL2-depletions , we studied the expression of Tsix and Xist in shRNA-treated ESCs . Unexpectedly , only MSL1- and MSL2- , but not MOF-depletion led to pronounced downregulation of Tsix both in male and female ESCs ( Figure 6A; note that in our RNA-seq data set for MSL2-depleted cells , Tsix was among the five most strongly downregulated genes ) . Downregulation of Tsix was accompanied by moderately elevated Xist RNA levels in MSL1- and MSL2-depleted ESCs whereas depletion of MOF yielded the most pronounced ( 8–15-fold ) upregulation of Xist without affecting Tsix . 10 . 7554/eLife . 02024 . 020Figure 6 . Depletion of MSL1 and MSL2 leads to downregulation of Tsix with concomitant upregulation of Xist . ( A ) Gene expression analysis for the indicated genes in male and female ESCs treated with scrambled RNA ( shScram ) or shRNA against Msl1 , Msl2 , or Mof . All results are represented as relative values normalized to expression levels in shScram ( normalized to Hprt ) and expressed as means ± SD in three biological replicates . ( B ) RNA-FISH for Huwe1 ( red ) and DXPas34 ( green ) in: scrambled control , shMsl1- , shMsl2- , and shMof-treated female ESCs . Nuclei were counterstained with DAPI ( blue ) . White arrows denote foci corresponding to Huwe1 or Tsix; dashed lines indicate nuclei borders . For additional images , phenotypes and quantifications see Figure 6—figure supplement 1A–C . For probe references see ‘Materials and methods’ . ( C ) Western blot analyses of the pluripotency factors in scrambled- , Mof- , Msl1- , and Msl2-shRNA-treated female ESCs . For corresponding expression analyses see Figure 6—figure supplement 1D , E . The respective dilution ( 100% , 30% , 10% ) of loaded RIPA extracts is shown above each panel . GAPDH was used as the loading control . For antibodies see ‘Materials and methods’ . ( D ) Western blot analyses of the transcription factors involved in regulation of the XIC in scrambled- , Mof- , Msl1- , and Msl2-shRNA-treated female ESCs . The respective dilution ( 100% , 30% , 10% ) of loaded RIPA extracts is shown above each panel . GAPDH was used as the loading control . ( E ) ChIP-qPCR analysis of REX1 ( left panel ) and YY1 ( right panel ) across the Tsix major promoter ( P2 ) and DXPas34 in male ESCs treated with the indicated shRNAs . The labels of the x axes correspond to the arrowheads in Figure 5A . For all ChIP experiments , three biological replicates were used; results are expressed as mean ± SD; cells were harvested on day 4 ( Msl2 ) or 5 ( Mof ) after shRNA treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 02024 . 02010 . 7554/eLife . 02024 . 021Figure 6—figure supplement 1 . Cells depleted of MSL1 or MSL2 , but not MOF show loss of DXPas34 foci . ( A ) RNA-FISH for Huwe1 ( red ) and DXPas34 ( green ) in shScrambled- , shMsl1- , shMsl2- , and shMof-treated female ESCs . Shown here are examples of RNA-FISH signals for multicellular colonies and loss of DXPas34 signal in MSL1- and MSL2-depleted cells . White boxes indicate cells enlarged and resented in Figure 6B . For all experiments , nuclei were counterstained with DAPI ( blue ) . ( B ) Summary of RNA-FISH for DXPas34 and Huwe1 . Red dots indicate the number of X chromosomes and green dots , DXPas34 foci ( smaller dot = reduced signal ) . Phenotypes that we observed in knockdowns are categorized into four groups containing cells with equal Huwe1/DXPas34 ratio and with DXPas34 loss . The percentages indicate how many cells per population showed the respective phenotype . ( C ) Corresponding to Figure 6B . Summary of total cell counts from RNA-FISH for ( DXPas34 ) and Huwe1 in MSL1- , MSL2- , or MOF-depleted female ESCs . ( D ) Gene expression analysis for the indicated genes in female ESCs treated with scrambled RNA ( shScram ) or shRNA against Mof , Msl1 and Msl2 . All results are represented as relative values normalized to expression levels in shScram ( normalized to Hprt ) and expressed as means ± SD in three biological replicates . ( E ) Gene expression analysis for genes of the XIC in female ESCs treated with scrambled RNA or shRNA against Msl1 , Msl2 or Mof . All results are represented as relative values normalized to expression levels in shScrambled ( normalized to Hprt ) and expressed as means ± SD for three biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 02024 . 021 To determine the effects on Tsix in individual cells , we next performed RNA-FISH with probes against DXPas34 and Huwe1 in female ESCs ( Huwe1 was used to mark X chromosomes , for probe references see ‘Materials and methods’ ) . The RNA-FISH confirmed the qPCR results as we observed global reduction and in many cases elimination of DXPas34 signals in MSL1- and MSL2- , but not in MOF-depleted cells ( Figure 6B , Figure 6—figure supplement 1A–C ) . We next wanted to understand the mechanistic differences between the Tsix-specific and the Tsix-independent effects on Xist levels that we found for depletions of MSL1/MSL2 and MOF , respectively . As pluripotency factors are additional regulators of Xist ( Navarro et al . , 2008; Nesterova et al . , 2011 ) , we assessed the consequences of the different knockdowns on the Xist-related pluripotency network in female ESCs . Like for MOF- and KANSL3-depletions in male ESCs ( Figure 4F ) , the depletion of MOF ( but not of MSL1 or MSL2 ) in female ESCs resulted in a significant decrease of transcript and protein levels of pluripotency factors that had previously been associated with Xist repression ( e . g . , NANOG and REX1; see Figure 6C , Figure 6—figure supplement 1D ) . Taken together , we detect direct binding of MSL complex members to several loci within the X inactivation center including the Tsix/Xist locus . Depletion of MSL1 or MSL2 , but not MOF led to severe downregulation of Tsix expression while depletion of MOF , MSL1 , or MSL2 resulted in elevated Xist levels . These results indicate a direct regulatory function of MSL1 and MSL2 on the DXPas34 locus and an indirect NSL-associated MOF effect on Xist expression through the pluripotency network . As loss of MSL1 and MSL2 did not affect the core pluripotency network , we set out to explore what might be the impact of MSL depletion on XIC genes ( other than Tsix and Xist ) and transcription factors involved in their regulation . As shown in Figure 6—figure supplement 1E , we observed mild effects on the expression of XIC-encoded genes involved in the regulation of X inactivation . Only depletion of MSL2 led to significant downregulation of Ftx and Jpx genes whose promoters were bound by MSL1 and/or MSL2 ( Figure 5A ) . On the other hand , depletion of MOF led to moderate upregulation of Linx lncRNA , which acts synergistically with Tsix ( Nora et al . , 2012 ) . Neither the depletion of MSL1 and MSL2 nor the depletion of MOF significantly influenced protein levels of RNF12 , YY1 , or CTCF that are known regulators of the XIC ( Figure 6D; Donohoe et al . , 2007 , 2009; Jonkers et al . , 2009; Shin et al . , 2010; Jeon and Lee , 2011 ) . Since REX1 and YY1 bind and regulate the Tsix locus ( Donohoe et al . , 2007; Gontan et al . , 2012 ) , we subsequently tested whether MSL depletion would affect the recruitment of these factors to the Tsix major promoter and DXPas34 . Indeed , the depletion of MSL2 led to significant reduction of REX1 ChIP signals across the DXPas34 locus whereas the effect on YY1-targeting was less pronounced and restricted to the Tsix major promoter ( P2 ) ( Figure 6E ) . We next assessed the consequence of MSL-dependent reduction of Tsix levels and concomitant upregulation of Xist at a cellular level using RNA-FISH for Xist upon depletion of individual MSL complex members ( for probe reference see ‘Materials and methods’ ) . Interestingly , we observed accumulating Xist lncRNA and X-chromosomal coating in a small fraction of MSL1- and MSL2-depleted female ESCs ( but not MOF-depleted cells; 4–5% of the cell population in shMsl1 and shMsl2 with comparison to 0 . 5% in scrambled control , see Figure 7A and Figure 7—figure supplement 1A–C ) . These findings suggest that the MSL1- and MSL2-dependent downregulation of Tsix is sufficient to cause occasional accumulation of Xist lncRNA in undifferentiated female ESCs . The different outcomes following MOF and MSL1/MSL2 depletion on Xist confirmed the notion that MOF and MSL1/MSL2 influence the XIC via different mechanisms . 10 . 7554/eLife . 02024 . 022Figure 7 . MSL1 and MSL2 depletion leads to enhanced and chaotic Xist accumulation in early differentiation . ( A ) RNA-FISH for Huwe1 ( red ) and Xist ( green ) in: scrambled control , shMsl1- , shMsl2- , and shMof-treated female ESCs . Nuclei were counterstained with DAPI ( blue ) . White arrows denote foci corresponding to Huwe1 or Xist; dashed lines indicate nuclei borders . For additional images , phenotypes and quantifications see Figure 7—figure supplement 1B–D . For probe references see ‘Materials and methods’ . ( B ) Expression analysis for Xist in undifferentiated , day 2 ( D2 ) and day 3 ( D3 ) differentiating female ESCs treated with scrambled RNA ( shScram ) or shRNA against Mof , Msl1 , and Msl2 . All results are represented as arbitrary units ( Xist expression in undifferentiated ESCs = 1 ) normalized to expression levels in shScram ( normalized to Hprt ) and expressed as means ± SD in three biological replicates . p-values for D2-to-D3 expression change were obtained using unpaired t test . ( C ) RNA-FISH for Huwe1 ( red ) and Xist ( green ) in: scrambled control , shMsl1- , shMsl2- , and shMof-treated differentiating female ESCs . Nuclei were counterstained with DAPI ( blue ) . RNA-FISH was performed on the sixth day of knockdown ( after 72 hr of differentiation ) . Percentages indicate number of cells with at least one Xist cloud for each of the knockdowns . For additional images of multicellular colonies see Figure 7—figure supplement 2A . ( D ) Bar plot summarizing the percentage of Xist clouds for individual knockdowns in differentiating ( DAY3 ) female ESCs for individual knockdowns . Cells were divided into three categories: cells carrying no Xist clouds ( white ) , single Xist cloud ( light green ) , or two Xist clouds ( dark green ) . For quantifications , see Figure 7—figure supplement 2B . ( E ) RNA-FISH for Xist ( green ) in: scrambled control , shMsl1- , shMsl2- , and shMof-treated differentiating ( DAY3 ) female ESCs . Here , we show examples of individual nuclei carrying different patterns of Xist accumulation . Percentages correspond to the frequency of the shown Xist pattern within the population of cells . White arrows denote Xist foci; dashed lines indicate nuclei borders . For quantifications see Figure 7—figure supplement 2B . DOI: http://dx . doi . org/10 . 7554/eLife . 02024 . 02210 . 7554/eLife . 02024 . 023Figure 7—figure supplement 1 . Depletion of MSL1 and MSL2 leads to occasional accumulation and spreading of Xist in undifferentiated ESCs . ( A ) RNA-FISH for Huwe1 ( red ) and Xist ( green ) in shScrambled- ( top left ) and shMof- ( top right ) , shMsl1- ( bottom left ) and shMsl2-treated ( bottom right ) female ESCs . Shown here are additional examples of RNA-FISH for multicellular colonies and individual cells exhibiting Xist-mediated coating ( Figure 7A ) . White boxes indicate cells enlarged in Figure 7A . White arrows denote Huwe1 and Xist foci . Dashed lines indicate nuclei borders . For all experiments , nuclei were counterstained with DAPI ( blue ) . ( B ) Summary of RNA-FISH for Xist and Huwe1 . The number of green dots indicates the number of X chromosomes within the cell while the larger dot indicates Xist accumulation . Cells were classified into three phenotypic groups with cells showing sharp , localized Xist signals ( once or twice ) or Xist ‘clouds’ . The percentages indicate how many cells per population showed the respective phenotype . ( C ) Corresponding to Figure 7A . Summary of the total cell counts from Xist and Huwe1 RNA-FISH in indicated knockdowns . DOI: http://dx . doi . org/10 . 7554/eLife . 02024 . 02310 . 7554/eLife . 02024 . 024Figure 7—figure supplement 2 . Depletion of MSL1 and MSL2 lead to enhanced Xist accumulation in differentiating ESCs . ( A ) RNA-FISH for Huwe1 ( red ) and Xist ( green ) in shScrambled- , shMsl1- , and shMsl2-treated differentiating ( DAY3 ) female ESCs . Shown here are additional examples of RNA-FISH for multicellular colonies ( Figure 7C ) . Dashed lines indicate nuclei borders . For all experiments , nuclei were counterstained with DAPI ( blue ) . ( B ) Corresponding to Figure 7C–E . Summary of the total cell counts from Xist RNA-FISH in indicated knockdowns . Percentage of cells with respective phenotype indicated in red . DOI: http://dx . doi . org/10 . 7554/eLife . 02024 . 024 Previous studies have shown that the effects of Tsix depletion on Xist accumulation and X inactivation become fully apparent after induction of differentiation ( Clerc and Avner , 1998; Debrand et al . , 1999; Lee and Lu , 1999; Luikenhuis et al . , 2001; Ohhata et al . , 2006; Sun et al . , 2006 ) . We therefore depleted MSL1 , MSL2 and MOF , and induced differentiation for 3 days by withdrawing LIF and placing the ESCs in N2B27 media . Consistent with our previous results , the induction of differentiation resulted in a stronger elevation of Xist RNA levels in MSL1- and MSL2-depleted cells in comparison to the scrambled control ( Figure 7B ) . As Tsix expression was not affected in MOF-depleted ESCs and Xist levels were already high before induction of differentiation , Xist upregulation between day 2 and 3 of differentiation was similar to the scrambled control . To monitor the effect on the X chromosome more closely , we next performed Xist RNA-FISH in MSL1- , MSL2- and MOF-depleted cells after 3 days of differentiation . All three knockdowns resulted in enhanced Xist accumulation and X-chromosomal coating ( 63 . 1% , 61 . 1% and 50 . 6% of all counted cells in shMsl1- , shMsl2- , and shMof-treated ESCs , respectively , in comparison to scrambled control with 39 . 1% of counted cells; see Figure 7C , D and Figure 7—figure supplement 2A , B ) . Interestingly , we observed that MSL1- and MSL2-depleted differentiating cells contained numerous cells with two inactive X chromosomes . The fraction of cells where both X chromosomes underwent XCI was approximately 10-fold higher in Msl1 and Msl2 knockdown compared to the scrambled control ( Figure 7E ) . These results are in agreement with previously published data from homozygous Tsix mutants that exhibit irregular , ‘chaotic’ choice for X inactivation ( Lee , 2005 ) . Taken together , our data establishes MSL1 and MSL2 among the key regulators of Tsix transcription , as the depletion of MSL proteins results in severe downregulation of Tsix transcription and enhanced accumulation of Xist during early differentiation . Our study sheds light on the interplay between MOF and its complexes in mammals . Despite the fact that the depletion of KANSL3 does not strongly reduce global H4K16 acetylation levels , we observed strikingly similar protein and transcriptome changes in KANSL3- or MOF-depleted cells ( Figure 3E–G ) . On the other hand , MSL1- and MSL2-depletion caused marked decreases of H4K16 acetylation ( Figure 3G ) . This is consistent with previous reports that established MSL proteins as the main enhancers of MOF's H4K16 acetylation activity , while the NSL complex was shown to possess broader substrate specificity and can crosstalk with histone methylases ( Cai et al . , 2010; Kadlec et al . , 2011; Zhao et al . , 2013b ) . Unexpectedly , we observed remarkably different phenotypic changes in MSL1- or MSL2-depleted cells compared to MOF- and KANSL3-depleted cells ( Figure 3E , Figure 3—figure supplement 3A ) . A striking example was the strong reduction of key pluripotency factors in KANSL3- and MOF-depleted cells that remain unaffected in MSL1- and MSL2-knockdowns ( Figure 3G , Figure 4F ) . These results support the recent finding that MOF is vital for the maintenance of pluripotency ( Li et al . , 2012 ) , but we furthermore show that this is an NSL- and not MSL-related function of MOF independent of H4K16 acetylation deposition . Taken together , our data shows that while MOF is the major acetyltransferase for lysine 16 of histone 4 ( Taipale et al . , 2005 ) , MSL-dependent H4K16 acetylation is one of several means through which MOF exerts its crucial biological functions . This notion was further supported by the finding that MOF predominantly binds to promoters of broadly expressed genes as part of the NSL complex and subsequently supports their transcription ( Figure 3A–F ) . MSL1 and MSL2 , on the other hand , bound to a relatively small subset of broadly expressed MOF-NSL-targeted genes that were significantly stronger expressed than those where MOF was exclusively present with NSL complex members ( Figure 3B , Figure 3—figure supplement 1B ) . The additive effects of the complexes on gene expression were intriguing , and whether they influence each other's activity or exert their functions separately should be studied in the future . We propose that the MSL complex fine-tunes MOF's activity and ensures precise regulation of more specific targets—after all , their presence is essential for the recruitment of MOF to NSL-independent targets ( Figure 5B ) . Our model is surprisingly similar to the picture that is emerging from Drosophila research where the NSL complex regulates housekeeping genes ( Feller et al . , 2012; Lam et al . , 2012 ) while the MSL complex fulfills a highly specialized role on the male X chromosome ( reviewed in Conrad and Akhtar , 2011 ) . In addition to insights about MOF-related functions of MSL and NSL complexes , we show for the first time additional binding of MSL and NSL proteins to TSS-distal regions with enhancer characteristics . On a global scale , MOF did not yield strong enrichments for canonical enhancers; however , both MSL2 and KANSL3 showed robust signals for TSS-distal regions in ESCs , but not in NPCs , which reflected the transcriptional activity of these regions ( Figure 2 , Figure 4A ) . This apparent MOF-independent binding of the individual proteins ( that tended to prefer different sets of enhancers; Figure 4B ) suggests that KANSL3 and MSL2 stimulate transcription even in the absence of the histone acetyltransferase . Both proteins are in principle capable of supporting transcription: the Drosophila homologue of KANSL3 can directly activate transcription in vitro ( Raja et al . , 2010 ) and human MSL2 acts as an E3 ubiquitin ligase at lysine 34 of H2B ( H2BK34ub ) ( Wu et al . , 2011 ) , which has been suggested to promote methylation of H3K4 , and thus gene expression ( Wu et al . , 2011 ) . Indeed , we observed several hundred genes that had been predicted to be regulated by TSS-distal binding sites of MSL2 or KANSL3 to be downregulated in the respective knockdowns with particularly high frequencies in MSL2-depleted cells ( Figure 4E ) . It is important to note that the subset of ESC enhancers for key pluripotency factors ( e . g . , Klf4 , Sox2 ) were bound concomitantly by KANSL3 and MSL2 and only the depletion of KANSL3 , but not of MSL1 or MSL2 diminished protein and transcript levels of these key ESC molecules ( see above ) . It is possible that KANSL3 could rescue loss of MSL2 at certain loci , but the exact mechanisms through which KANSL3 affects transcription via enhancer-binding need to be studied further . Furthermore , the pluripotency network and/or Mediator-related functions at super enhancers may be sufficient and dominant over MSL2 to maintain the expression of the pluripotency factors in the absence of MSL2 , but may well be dependent on the function of KANSL3 at these regions . When we specifically searched for regions where KANSL3 was not present together with MSL1 and MSL2 , we found that the X inactivation center ( XIC ) showed numerous signals for the MSL complex ( Figure 5 ) . The XIC , a hot-spot of regulatory lncRNAs , is an X-chromosomal region that contains the main regulators of X chromosome inactivation ( XCI ) . The proper function of XIC-located non-coding RNAs is influenced by the spatial organization of the XIC and governed by a sophisticated interplay of multiple transcription factors such as pluripotency factors ( Donohoe et al . , 2007; Navarro et al . , 2010; Deuve and Avner , 2011; Gontan et al . , 2012; Nora et al . , 2012 ) . We found that depletion of MSL1 and MSL2 severely reduced Tsix expression in male and in female ESCs , moderately increased Xist levels ( Figure 6A ) , but left pluripotency factors unaffected ( Figure 6C ) . In contrast , MOF-depleted cells showed downregulation of pluripotency factors and much higher Xist levels . Previous studies demonstrated that in undifferentiated ESCs , where pluripotency factors are highly abundant , even severe downregulation of Tsix , or Tsix-deletion has almost no effect on Xist transcription ( Morey et al . , 2001; Navarro et al . , 2005; Nesterova et al . , 2011 ) . Thus , the pronounced Xist upregulation seen in MOF-depleted cells seems to be an indirect effect due to the downregulation of pluripotency factors , while the reduction of Tsix transcripts in MSL1- and MSL2-depleted cells , where pluripotency factors remain unaffected , has milder consequences on Xist levels . Consequently , we could show that once ESCs are forced to initiate differentiation , the depletion of MOF has mild effects while MSL1- and MSL2-depleted cells , in which Tsix expression is prematurely downregulated , indeed suffer from enhanced Xist accumulation accompanied by ‘chaotic’ X inactivation ( different numbers of inactivated X chromosomes within a population of cells; Figure 7B–E ) . This is consistent with the notion that the repressive potential of Tsix on Xist accumulation and the role of Tsix and the DXPas34 locus in the process of counting and choice of XCI ( Lee , 2005; Vigneau et al . , 2006 ) becomes fully apparent during early stages of differentiation where additional repressive factors such as pluripotency factors are downregulated ( reviewed in Rougeulle and Avner , 2004 ) . We show that NSL and MSL complex members can function in concert to ensure proper regulation of gene expression , but our findings also strongly imply that members of both complexes have the capacity to act independently . In the case of the X inactivation center , we observe that the MOF-interacting proteins , despite engaging different regulatory means ( MSL1 , MSL2 through direct regulation of Tsix , and MOF-NSL through the pluripotency network ) synergize to ensure the proper expression of the X chromosomes in undifferentiated ES cells ( Figure 8 ) . Our study sets the ground for future research to dissect the intricate interactions and specific functions of MOF and its associated major regulatory proteins in more detail . All cell culture was performed in a humidified incubator at 37°C and 5% CO2 . The feeder-dependent mouse female embryonic stem cell line F1-21 . 6 was cultivated on mitomycin-C-inactivated or irradiated mouse embryonic fibroblasts ( MEFs ) . The feeder-independent mouse male ES cell line WT26 , a kind gift from the lab of Thomas Jenuwein , was cultivated on gelatin-coated dishes in ESC culture media KnockOut-DMEM ( Gibco , Carlsbad , CA ) supplemented with 1% L-glutamine ( Gibco ) , 1% penicillin/streptomycin ( Gibco ) , 1% non-essential amino acids ( Gibco ) , 1% sodium pyruvate , 1% 2-mercaptoethanol . All ESC media contained 15% FBS and 1000 U/ml ( for feeder-dependent ) or 2000 U/ml ( for feeder- ) of leukemia inhibitory factor . Male and female neuronal progenitor cell ( NPC ) lines were derived from previously mentioned ES cell lines ( see below ) . Mouse 3T3 cells ( for luciferase assays ) and human HEK293-FT cells ( for lentiviral production ) were cultivated in DMEM ( high glucose , with glutamine , Gibco ) supplemented with 10% heat-inactivated serum ( PAA Laboratories , North Dartmouth , MA ) , 1% L-glutamine , 1% penicillin/streptomycin . Mouse ESCs were differentiated into neuronal progenitor cells ( NPC ) as previously described ( Conti et al . , 2005; Splinter et al . , 2011 ) . In brief , 1 × 106 ESCs ( deprived of feeder cells ) were plated on 0 . 1% gelatin-coated dishes in N2B27 medium and cultured for 7 days with daily media changes . The cells were then dissociated from the plate using accutase ( Sigma , Germany ) and 3 × 106 cells were plated on a bacterial petri dish to induce formation of embryoid bodies in N2B27 medium supplemented with 10 ng/ml EGF and FGF2 ( Peprotech , Rocky Hill , NJ ) . After 72 hr , embryoid bodies were transferred to 0 . 1% gelatin-coated dishes to allow adhesion and expansion of NPCs from the embryoid bodies . NPC lines were maintained in N2B27 medium supplemented with EGF and FGF2 ( 10 ng/ml each ) , on 0 . 1% gelatin-coated flasks . For FISH analysis , F1-21 . 6 ESCs were grown on gelatin-coated coverslips with a MEF-inactivated monolayer for 24 hr . The Invitrogen precast gel system NuPAGE was used for SDS-PAGE . The 4–12% Bi–Tris gradient gels ( for proteins above 20 kDa ) or 12% Bis–Tris gels ( for histones and histone marks ) were loaded with samples supplemented with Roti-Load 1 sample buffer . After blotting , the membranes were blocked in 5% milk with PBS + 0 . 3% Tween-20 ( PBST ) mix for at least 1 hr at room temperature . Membranes were then incubated overnight with the primary antibody in 0 . 5% milk with PBST at 4°C . The next day , membranes were washed three times for 10 min in PBST , incubated with a suitable HRP-coupled secondary antibody for 1 hr at room temperature , washed thrice and proteins were visualized with Lumi-Light Plus Western Blotting Substrate using the Gel Doc XR+ System . For ( co ) immunoprecipitation ( IP , co-IP ) experiments , 1 ml of nuclear extract ( 0 . 5 mg/ml ) was used . IPs were performed in IP buffer ( 25 mM HEPES pH 7 . 6 , 150 mM KCl , 5 mM MgCl2 , 0 . 5% Tween20 , 0 . 2 mg/ml BSA , 1× complete protease inhibitors tablet ) . Extracts were incubated with 5 μg of the respective antibody or normal-rabbit/normal rat serum . For MSL1 15 μl of antibody serum was used . Extracts were incubated with the antibody for 2 hr , rotating at 4°C . Protein-A Sepharose beads ( GE Healthcare , United Kingdom ) , blocked with 1 mg/ml yeast tRNA and 1 mg/ ml BSA ( NEB , Ipswich , MA ) , were used for all ChIP and IP assays . Chromatin immunoprecipitation ( ChIP ) assays were performed as previously described ( Pauli , 2010 ) with minor changes . Cells were fixed in 1% molecular biology grade formaldehyde ( Sigma ) 9 min before being quenched with glycine ( 0 . 125 M final concentration ) . Cells were washed twice with ice-cold PBS and lysed on ice for 10 min with 10 ml of Farnham lysis buffer ( 5 mM PIPES pH 8 . 0 , 85 mM KCl , 0 . 5% NP-40 + Roche Protease Inhibitor Cocktail Tablet , filtered through 0 . 2 micron filter unit ) . Lysates were transferred to a Kontes dounce tissue grinder ( K885300-0015 , size B ) and dounced 15 times in order to break the cells and keep nuclei mostly intact . Crude nuclear prep was transferred to 15-ml falcon tube and nuclei pelleted by centrifugation at 2000 rpm at 4°C for 5 min . Nuclei were resuspended in RIPA lysis buffer ( 1 × PBS , 1% NP-40 , 0 . 5% sodium deoxycholate , 0 . 1% SDS + Roche Protease Inhibitor Cocktail Tablet , filtered through 0 . 2 micron filter unit ) . The nuclear extract was subjected to chromatin shearing using the Diagenode Bioruptor Plus sonicator ( at high setting for a total time of 25 min , 30 s ON , 30 s OFF ) . The sonicated mixture was centrifuged at 14 , 000 rpm at 4°C for 5 min and supernatant was collected . Chromatin was supplemented with 5 μg of primary antibody and incubated for 16 hr ( antibodies used for ChIP are listed below ) . After incubation , 50 μl of 50% slurry bead solution was added for another incubation period ( 2 hr ) , then beads were washed: four times for 15 min with RIPA lysis buffer , two times for 1 min with LiCl IP wash buffer ( 250 mM LiCl , 10 mM Tris–HCl pH 8 . 0 , 1 mM EDTA , 0 . 5% NP-40 , 0 . 5% DOC , filtered through 0 . 2 micron filter unit ) , two times for 1 min with TE buffer ( 1 mM Tris–HCl pH 8 . 0 , 1 mM EDTA , filtered through 0 . 2 micron filter unit ) . Washed beads were resuspended in 100 μl of IP elution buffer and subjected to overnight reverse cross-linking ( RNase and proteinase K digestions ) followed by DNA purification ( DNA was purified using Minelute PCR purification kit from Qiagen , Germany ) . For single IP assay 50 µl of bead solution was used . Purified ChIPed DNA was subjected to qPCR amplification ( Applied Biosystems , Carlsbad , CA ) . Input was used for normalization control . For primer pairs see Supplementary file 3 . For MSL1 antibody production , a GST-mMSL1 fusion protein ( C-terminal , residues 254–616 ) was used to immunize rabbits; the final bleed was used in experiments . Antibody specificity was verified with IP and MSL1-specific RNAi followed by Western blot analysis and ChIP assay . We used several commercial antibodies: a-KANSL1 ( PAB20355; Abnova , Taiwan ) , a-KANSL3 ( HPA035018; Sigma ) , a-MCRS1 ( 11362-1-AP; Proteintech , Chicago , IL ) , a-MOF ( A3000992A; BETHYL Montgomery , TX ) , a-MSL2 ( HPA003413; Sigma ) , a-NANOG ( A300-397A; BETHYL ) , a-OCT3/4 ( sc-5279; Santa-Cruz Dallas , TX ) , a-REX1 ( Ab28141; Abcam , England ) , a-ESRRB ( PP-H6705-00; Perseus Proteomics , Japan ) , a-KLF4 ( Ab72543; Abcam ) , a-SOX2 ( AF2018; R&D Systems , Minneapolis , MN ) , a-YY1 ( A302-779A; BETHYL ) , a-RNF12/RILM ( 16121-1-AP; Proteintech ) a-GAPDH ( A300-639A; BETHYL ) , a-NESTIN ( Ab93666; Abcam ) , a-CTCF ( Ab70303; Abcam ) , a-H3 ( Ab1791; Abcam ) , a-H4 ( Ab10158; Abcam ) , a-H4K16ac ( 07-329; Millipore , Billerica , MA ) . Enhancer candidate regions ( see below ) were cloned into the firefly luciferase plasmids ( pGL4 . 23; Promega , Witchburg , WI ) and transfected into mouse ESCs and 3T3 fibroblasts using Lipofectamine-2000 reagent and into NPCs using LTX-PLUS reagent ( Invitrogen ) . Transfections were performed according to the manufacturer's guidelines except for using a 1:6 DNA to Lipofectamine ratio . Cells were seeded 1 day prior to transfection to achieve 70–80% confluency at the time of transfection . Next , cells were fed with antibiotics-free medium ( ES medium with LIF for ESCs and OPTIMEM for NPCs and 3T3s ) at least 30 min before transfection and the medium was changed back 6–8 hr after transfection ( basal neural medium with FGF and EGF for NPCs ) . 100 ng of firefly construct with the cloned candidate region was co-transfected with 1 ng of renilla luciferase construct ( pRL-TK of Promega ) per 96-well and harvested for luciferase assay after 24 hr . Cells were harvested for luciferase assay 24 hr after transfection . The Dual Luciferase Kit ( Promega ) was used according to the manufacturer's protocol but with reduced substrate volumes of LARII and Stop&Glo reagents ( 50 µl per well of a 96-well plate with 10 µl cell lysate ) . Luminescence was measured by using Mithras plate reader ( Berthold , Germany ) . The transfection efficiency was normalized by firefly counts divided by the renilla counts . The fold enhancement value was calculated by an additional normalization to minimal promoter alone activities in each experiment ( the graphs represent at least three independent experiments that were performed in technical triplicates each with error bars representing standard error of the mean ) . The following enhancer candidate regions were amplified from mouse genomic DNA by PCR and cloned into BamHI-SalI sites ( downstream of luciferase gene ) of firefly luciferase plasmid pGL4 . 23 ( Promega ) : Intron of Esrrb ( chr12:87 , 842 , 537-87 , 843 , 719 ) with primers introducing BamHI and XhoI sites: ATAGGATCCGAAGTAATTGTCTATTGTATCAG ( forward ) , TATCTCGAGAAGAAGAAAGACTGTGTTCAACTCC ( reverse ) . Upstream of Lefty ( chr1: 182854617-182855516 ) with primers introducing BamHI and SalI sites: ATAGGATCCCTTGCGGGGGATATGAGGC ( forward ) , TATGTCGACCTGGGCCTTTCTAAGGC ( reverse ) . Upstream of Trim28 ( Kap1 ) ( chr18: 34309039-34310140 ) with primer introducing BamHI and SalI sites: ATAGGATCCGAGGACTATTTGAAGGATCTATT ( forward ) , TATGTCGACCTCACTCCCCAACCTCCATTTC ( reverse ) . Upstream of Apc ( chr18: 34309039-34310140 ) with primers introducing BamHI and SalI sites: ATAGGATCCCTGAGCAATGCTCTTCCTCACAAGC ( forward ) , TATGTCGACTTATACTCCAAATAGAATTGTCTG ( reverse ) . Intron of Tbx3 ( chr5: 120129690- 120130617 ) with primers introducing BamHI and SalI sites: ATAGGATCCATAAATAAATAAATAAATATCTGATTG ( forward ) , TATGTCGACCGCGAGTCTGGCGATGCCTTGTC ( reverse ) . cDNA was synthesized from 500 ng–1 µg of total RNA ( extracted from circa 1 million cells using Rneasy kit , Qiagen ) with random hexamers using SuperScript-III First Strand Synthesis kit ( Invitrogen ) . The qPCRs were carried in a total reaction volume of 25 µl containing 0 . 5–1 µl of cDNA , 0 . 4 µmol of forward and reverse primer mix and 50% 2 × SYBR Green PCR Master Mix ( Roche ) . Gene expression was normalized to multiple controls ( RplP0 or Hprt ) , using the 7500 software V2 . 0 . 4 for analysis ( Applied Biosystems ) . For primer pairs used for expression profiling see Supplementary file 3C . shRNA constructs were either obtained from Sigma in pLKO . 1 or designed using Genscript and cloned ( please see below for details ) . For cloning , forward and reverse complimentary DNA oligonucleotides ( Eurofins MWG Operon , Germany ) designed to produce AgeI ( 5′ ) and EcoRI ( 3′ ) overhangs were annealed at a final concentration of 2 µM in NEBuffer . The pLKO . 1-puro plasmid was digested with AgeI and EcoRI , ligated to the annealed oligonucleotides , and transformed into HB101 competent cells ( Promega ) . Plasmid DNA was purified using the QIAprep Spin Miniprep kit ( Qiagen ) , and the sequence was validated . For production of lentiviral particles , 70% confluent HEK293FT cells in a 10-cm tissue culture plate were co-transfected with 3 . 33 µg lentiviral construct , 2 . 5 µg psPAX2 packaging plasmid and 1 µg pMD2 . G envelope plasmid using Lipofectamine-2000 reagent ( Invitrogen ) . To transduce ESCs , either concentrated or diluted lentiviral particles were used . For concentrated lentivirus , transfections were scaled up and OPTIMEM ( Invitrogen ) added to the HEK293FT cells following transfection and the lentiviral supernatant collected at 48 and 72 hr post-infection . This was then concentrated using Amicon Ultra-15 centrifugal filter units ( Millipore ) and added to ESC media supplemented with LIF and 10 µg/ml polybrene ( Millipore ) . For diluted lentivirus , ESC media without LIF was added to the HEK293FT cells and the lentiviral supernatant was collected after 48 hr , filtered through 0 . 22 µm filters ( Whatmann ) , and added 1:1 with fresh ESC media supplemented with LIF and polybrene to the ESCs . ESCs were then subjected to selection with 1 . 0 µg/ml puromycin , passaged once , and harvested on day 3 , 4 , 5 or 6 of knockdown depending on the experiment ( the numbers of days are indicated in the corresponding results section ) . The following shRNA sequences were used for the knockdowns: CCGGCCTAAGCACTCTCCCATTAAACTCGAGTTTAATGGGAGAGTGCTTAGGTTTTTG ( shMsl1 , SIGMA , TRCN0000241378 ) , CCGGCCCAGTCTCTTAGCCATAATGCTCGAGCATTATGGCTAAGAGACTGGGTTTTTG ( shMsl2 , SIGMA , TRCN0000243429 ) , CCGGAAGGCCGAGAAGAATTCTATCTCGAGATAGAATTCTTCTCGGCCTTTTTTTG ( shMof , GENSCRIPT designed ) , CCGGCTCCAGTCCTCTTCGTCATTGCTCGAGCAATGACGAAGAGGACTGGAGTTTTTG ( shKansl3 , SIGMA , TRCN0000266995 ) , CCGGAAGTGGCGCCTTAGCAACAACCTCGAGGTTGTTGCTAAGGCGCACTTTTTTTG ( shMcrs1 , GENSCRIPT designed ) , CCGGCAACAAGATGAAGAGCACCAACTCGAGACAATTCGGAAGAAATCTGAGCTTTTTG ( Non-targeting control , SIGMA , SHC002 ) . Cells treated with respective shRNAs and scramble control were performed as described earlier in feeder-free W26 mouse ESCs . The cell count was monitored for 6 days post knockdown at 24-hr intervals . In brief , after 4 days of knockdown six sets of 0 . 4 × 104 cells per well were seeded in triplicates in a 12-well gelatinized plate . The cells were grown in ES cell culture medium supplemented with 2000 U/ml LIF and 1 μg/ml puromycin; the medium was changed every 24 hr . For counting , cells were trypsinized and counted using the Neubauer hemocytometer . Detection of alkaline phosphatase , a surface marker and indicator of undifferentiated ESCs , was performed using the following method: feeder-free W26 ESCs were transduced ( 4 days ) with scramble or the shRNAs against the genes of interest . Cells were washed twice with PBS followed by fixation with 4% PFA for 2–3 min . The cells were washed twice with PBS and stained for 20 min with staining solution ( 25 mM Tris-Maleic acid buffer pH 9 . 0 , 0 . 4 mg/ml α-Naphthyl Phosphate ( Sigma ) , 1 mg/ml Fast Red TR Salt ( Sigma ) , 8 mM MgCl2 , 0 . 01% Deoxycholate , 0 . 02% NP40 ) . The reaction was stopped by washing with water followed by two washes with 1 × PBS . Total RNA was extracted from WT26 ESCs and NPCs as biological triplicates using TRIzol Reagent and treated with the TURBO Dnase kit ( Ambion ) . For RNA-seq of knockdowns , feeder-free WT26 ESCs were transduced with shRNAs specific for Msl1 , Msl2 , Mof , Kansl3 and control shRNA as biological triplicates as described above . Briefly , following transduction for 24 hr , cells were washed with PBS thrice to remove the viral supernatant and subjected to puromycin selection ( 1 . 5 µg/ml ) for 24 hr . In the case of Msl1/2 , Mof , control shRNA the cells were maintained in puromycin selection for 4 days and in case of Kansl3 , the cells were maintained in puromycin-selection for 84 hr . An additional set of control shRNA was performed alongside with Kansl3 for 84 hr . Total RNA from all the shRNA-treated cells was extracted using TRIzol Reagent and the samples were treated with DNase using the TURBO DNase kit ( Ambion ) . The quality of the RNA was analyzed using the Bioanalyzer and samples with RIN values between 9 and 10 were used for RNA-seq . For RNA-seq analysis , cDNA libraries were prepared using the Illumina TruSeq Stranded mRNA kit with 3 μg DNase-treated samples . Xist and Huwe1 probes were described previously ( Chow et al . , 2010 ) . Tsix was detected with a DXPas34 plasmid ( Debrand et al . , 1999 ) . Approximately 1 × 105 of F1-21 . 6 ESCs were plated on gelatin-coated coverslips and incubated for 24 to 48 hr . After fixation and permeabilization , coverslips with cells were washed and stored in 70% EtOH at −20°C . Then the coverslips were dehydrated in 80% , 95% , and 100% EtOH ( 5 min each ) and briefly air-dried . FISH probes were labeled by nick translation ( Abbott ) with Spectrum Red-dUTP or Spectrum Green-dUTP following the manufacturer's instructions . Labeled probes were precipitated in the presence of salmon sperm ( 10 μg ) and Cot-1 DNA ( 3 μg ) , denatured and competed with Cot-1 DNA for 45 min at 37°C . Cells were then directly hybridized with labeled probes at 37°C overnight . Next , coverslips were washed three times in 50% formamide/2 × SSC followed by three washes in 2 × SSC at 42°C . Cells were stained with DAPI ( 0 . 2 mg/ml ) . Approximately 1 × 105 of male W26 ESCs and NPCs were plated on gelatin-coated coverslips and incubated for 24 hr . The cells were washed twice with PBS and fixed with pre-warmed 4% formaldehyde for 8 min at 37°C . Next , cells were washed thrice with PBS , 5 min each at room temperature and incubated in Permeabilization buffer ( 1 × PBS , 0 . 2% Triton X-100 ) for 5 min at room temperature . After permeabilization cells were incubated in Blocking buffer ( 1 × PBS , 5% BSA , 0 . 05% Triton-X100 ) for 30 min , stained for 1 hr with primary antibody ( rabbit polyclonal a-NESTIN , 1:500 ) . Next , cells were washed thrice with Wash buffer ( 1 × PBS , 0 . 05% Triton-X100 ) and incubated in 10% goat normal serum solution ( Invitrogen ) for 20 min . Secondary antibody ( goat anti-rabbit Alexa Fluor-488 , 1:1000 ) was added on coverslips and incubated for 45 min . We used a spinning disk confocal microscope ( Observer 1/Zeiss ) with Plan Apochromat 63x1 . 4-oil objective for magnification . 500 ms exposure time was used for all lasers . Sequential z-axis images were collected in 0 . 5 μm steps . ZEN Blue software was used for image analysis . All samples were sequenced by the Deep Sequencing Unit ( MPI-IE , Freiburg ) using Illumina HiSeq2000 . Library preparation was carried out following Illumina standard protocols for paired-end sequencing ( 50 bp reads ) . All raw reads can be found in the GEO database under the accession number GSE51746 . RNA-seq reads were mapped to Ensembl annotation NCBIM37/mm9 using TopHat2 ( Kim et al . , 2013 ) with the options mate-inner-dist , mate-std-dev and library-type ( fr-firststrand ) . The distance between read mates ( mate-inner-dist and mate-std-dev ) were assessed individually for each sequenced library based on the output of the sequencer for average fragment size and CV value . For FPKM value generation , cufflinks ( version 2 . 1 . 1 ) was used for each transcript in each condition ( three replicates for ESC and NPC ) with default parameters; CummeRbund was used for quality checks and data access ( Trapnell et al . , 2013 ) . Based on the distribution of FPKM values , active genes were defined as transcripts with mean FPKM ≥4 ( average over the replicates ) . After mapping of the RNA-seq reads from the shRNA-treated samples ( including scrambled control ) , the reads that mapped to the genome were counted using htseq-count ( doi: 10 . 1101/002824 ) with the stranded option set to reverse . The annotations present in the Mus musculus gtf file from the ENSEMBL release 67 were used as reference for counting . DESeq2 was used for differential expression analysis ( Anders and Huber , 2010 ) . In this analysis , all libraries from knockdown cells were compared in a pairwise manner with its corresponding scrambled shRNA samples . Within the DESeq2 workflow , the cooks–cutoff parameter was set to ‘FALSE’ and the genes with an adjusted p-value ≤0 . 01 were defined as significantly affected . MACS ( version 1 . 4 ) was used for peak calling on every sample individually and on the merged files of two replicates ( Zhang et al . , 2008 ) . Only peaks present in both replicates were considered , using the borders and summits defined by peak calling results for the merged replicates . In addition , peaks with −10log10 ( p-values ) lower than 50 and false-discovery rate values greater than 0 . 1% were excluded from down-stream analyses . We used the RefSeq gene list for genome version mm9/NCBI37 . Unless specified otherwise , alternative transcription start sites were scored as individual TSS in the respective analyses . The list of genes with homologues in different species was downloaded from HomoloGene and subsequently filtered for pairs of mouse and fly genes that belong to the same clusters of homology ID . CpG island information was downloaded from the UCSC Genome Browser ( Wu et al . , 2010 ) , mean observed over expected CpG ratios were extracted for the TSSs ± 0 . 5 kb using UCSC tools . For Figure 2 , a matrix containing the normalized ChIP-seq signals for all peaks was generated as follows: first , the union of peaks was created using mergeBed from the bedtools suite ( Quinlan and Hall , 2010 ) ; then each region was binned to 2 kb and the normalized ChIP values were extracted in 50 bp windows . The ChIP signal values were rank-transformed , converted into euclidean distances using the R function ‘dist’ and subsequently ordered according to their similarity by the ‘hclust’ function ( using Ward's method ) . The resulting dendrogram was pruned to 2 to 10 clusters for which the individual ChIP signals for unscaled regions were extracted ( Figure 2 ) . Visual inspection revealed no striking differences of the binding patterns between the individual clusters for more than 5 clusters . The 3 clusters displayed in the lower part of Figure 4—figure supplement 2B were obtained similarly: first , a matrix was generated that contained the normalized ChIP-seq values of MOF , p300 , H3K4me1 and DNase hypersensitivity sites for all regions of cluster D that did not overlap with ESC enhancers . The regions were then scaled to 1400 bp and mean values were computed for 50 bp bins using the computeMatrix module of deepTools ( Ramirez et al . , 2014 ) . Further processing was done as described above; the resulting dendrogram was pruned to k = 3 and the enrichments of the different factors were computed and visualized for 10 kb regions using the heatmapper module of deepTools . For GO term analyses , we used two approaches: the web interface of DAVID ( Huang da et al . , 2009 ) and GREAT ( McLean et al . , 2010 ) . For DAVID , we determined genes overlapping with the peaks of the individual ChIP-seq samples ( TSS region ± 500 bp ) and supplied the corresponding RefSeq-IDs . The background list contained the union of all TSSs bound by at least one ChIPed protein . We used the Functional Annotation Clustering tool , filtered with the option ‘high stringency’ and manually grouped the returned clusters of gene functions with enrichment scores above 1 . 3 into even broader terms . To assess the GO terms of genes that might be regulated by the TSS-distal binding sites of MOF , MSL1 , MSL2 , KANSL3 and MCRS1 , we used GREAT ( McLean et al . , 2010 ) with the mouse genome as the background data set and default settings . We obtained the top-ranked biological processes of the genes suggested to be cis-regulated by the regions combined in cluster D ( Figure 2 ) . For the analysis of enriched transcription factor binding sites , we used the R package ChIPEnrich ( http://sartorlab . ccmb . med . umich . edu/chip-enrich ) and TRAP ( Thomas-Chollier et al . , 2011 ) . The ChIPEnrich package takes peak regions as input and uses a logistic regression approach to test for gene set enrichments while normalizing for mappability and locus length . We supplied the regions belonging to the individual clusters of binding ( A–E from Figure 2 ) and obtained the corresponding enriched transcription factors . To plot the occurrences of the SMAD3 motif ( V$SMAD3_Q6 , TRANSFAC name M00701; Figure 4—figure supplement 3C ) , TRAP was used with the following command to generate a bedgraph file where the log likelihood of a SMAD3 motif occurrence is stored for the entire genome: ANNOTATEv3 . 04_source/Release/ANNOTATE_v3 . 04 -s mm9 . fa --pssm /transfac . pssm -g 0 . 5 --ttype balanced -name M00701 -d \| awk 'BEGIN{OFS=t}{print $1 , $4+7 , $4+8 , $6}' > SMAD3 . pssm . bedgraph . Heatmaps displaying normalized read densities of ChIP-seq samples , % methylated CpGs and SMAD3 motif score ( Figures 2 , 3C , 4A , Figure 4—figure supplement 2B and 3 ) were generated with the computeMatrix and heatmapper modules of the deepTools package ( Ramirez et al . , 2014 ) with ‘reference-point’ mode . Heatmaps of fractions of overlapped regions as in Figure 3—figure supplement 1 and Figure 4B as well as log2 fold changes ( knockdown/control ) from RNA-seq experiments ( Figure 3E ) were generated with the function ‘heatmap . 2’ from the R gplots package . The values underlying the summary plots such as the meta-gene and meta-enhancer plots in Figures 3D and 4B , Figure 3—figure supplement 2B , Figure 4—figure supplement 1B , 2A , C were generated with the computeMatrix module of the deepTools package using either ‘reference-point’ or ‘scale-regions’ mode and were visualized with the R package ggplots2 . For general assessments of overlaps between bed-files and to extract scores for defined regions the bedTools suite ( Quinlan and Hall , 2010 ) and UCSC tools ( Kuhn et al . , 2013 ) were used . The snapshots of the binding profiles were obtained with IGV browser ( Thorvaldsdottir et al . , 2013 ) . For each knockdown condition for which RNA-seq data had been generated ( see above ) , significantly affected genes were used ( adjusted p-value ≤0 . 01 , see above for differential gene expression analysis ) . Then they were subdivided into TSS- ( ChIP-seq peak overlap with TSS ±1 kb ) , TSS-distal- ( ChIP-seq peaks not overlapping with TSS ±1 kb ) and non-targets ( neither TSS overlap nor part of TSS-distal list ) . A gene was classified as TSS-distally regulated when at least one of the following criteria was true:TSS-distal peaks overlapped its published super or typical enhancer ( Whyte et al . , 2013 ) TSS-distal peaks were predicted by GREAT ( McLean et al . , 2010 ) to regulate the respective geneTSS-distal peaks overlapped with at least one intron Genes were defined as MSL targets when peaks of MOF and MSL1\|MSL2 were overlapping at the TSS ±1 kb or TSS-distal peaks were predicted to regulate the same putative target gene . NSL targets were defined the same way , but with co-occurrences of peaks from MOF and KANSL3\|MCRS1 .	Gene expression is controlled by a complicated network of mechanisms involving a wide range of enzymes and protein complexes . Many of these mechanisms are identical in males and females , but some are not . Female mammals , for example , carry two X chromosomes , whereas males have one X and one Y chromosome . Since the two X chromosomes in females contain essentially the same set of genes , one of them undergoes silencing to prevent the overproduction of certain proteins . This process , which is called X-inactivation , occurs during different stages of development and it must be tightly controlled . An enzyme called MOF was originally found in flies in two distinct complexes—the male-specific lethal ( MSL ) complex , which forms only in males , and the non-specific lethal ( NSL ) complex , which is ubiquitous in both males and females . These complexes are evolutionary conserved and are also found in mammals . While mammalian MOF is reasonably well understood , the MSL and NSL complexes are not , so Chelmicki , Dündar et al . have used various sequencing techniques , in combination with biochemical experiments , to investigate their roles in embryonic stem cells and neuronal progenitor cells in mice . These experiments show that MSL and NSL complexes engage in the regulation of thousands of genes . Although the two complexes often show different gene preferences , they often regulate the same cellular processes . The MSL/NSL-dependent regulation of X chromosome inactivation is a prime example of this phenomenon . The MSL complex reduces the production of an RNA molecule called Xist , which is responsible for the inactivation of one of the two X chromosomes in females . The NSL complex , meanwhile , ensures the production of multiple proteins that are crucial for the development of embryonic stem cells , and are also involved in the repression of X inactivation . This analysis sheds light on how different complexes can cooperate and complement each other in order to reach the same goal in the cell . The knowledge gained from this study will pave the way towards better understanding of complex processes such as embryonic development , organogenesis and the pathogenesis of disorders like cancer .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "genetics", "and", "genomics" ]	2014	MOF-associated complexes ensure stem cell identity and Xist repression
Brown adipose tissue ( BAT ) activation via cold exposure is increasingly scrutinized as a potential approach to ameliorate cardio-metabolic risk . Transition to cold temperatures requires changes in the partitioning of energy substrates , re-routing fatty acids to BAT to fuel non-shivering thermogenesis . However , the mechanisms behind the redistribution of energy substrates to BAT remain largely unknown . Angiopoietin-like 4 ( ANGPTL4 ) , a protein that inhibits lipoprotein lipase ( LPL ) activity , is highly expressed in BAT . Here , we demonstrate that ANGPTL4 is part of a shuttling mechanism that directs fatty acids derived from circulating triglyceride-rich lipoproteins to BAT during cold . Specifically , we show that cold markedly down-regulates ANGPTL4 in BAT , likely via activation of AMPK , enhancing LPL activity and uptake of plasma triglyceride-derived fatty acids . In contrast , cold up-regulates ANGPTL4 in WAT , abolishing a cold-induced increase in LPL activity . Together , our data indicate that ANGPTL4 is an important regulator of plasma lipid partitioning during sustained cold . Adipose tissue can be classified into white adipose tissue ( WAT ) and brown adipose tissue ( BAT ) . Whereas WAT represents the main energy storage organ in the body , BAT is dedicated to the generation of heat via the burning of lipids . BAT is activated during cold exposure , when additional heat production is needed to maintain core body temperature . Heat production by BAT is stimulated via release of norepinephrine by the sympathetic nervous system , causing activation of β-adrenergic signalling and subsequent uncoupling of ATP production from mitochondrial respiration ( Cannon and Nedergaard , 2004 ) . Uncoupling in BAT is mediated by the uncoupling protein UCP1 , which is highly abundant specifically in BAT ( Cannon and Nedergaard , 2004 ) . Studies in the last decade have shown the presence of BAT in humans and have provided preliminary evidence for an inverse relationship between BAT activity and parameters of obesity ( van Marken Lichtenbelt et al . , 2009; Virtanen et al . , 2009; Wang et al . , 2011 ) . As a consequence , interest in BAT function and the possible targeting of BAT for treatment or prevention of metabolic diseases has surged . Upon cold exposure , oxidation of fuels by BAT is dramatically increased . In addition to circulating glucose and free fatty acids , fatty acids derived from circulating triglyceride-rich lipoproteins ( TRLs ) represent a major fuel source for BAT ( Cannon and Nedergaard , 2004 ) . The liberation of fatty acids from TRLs is catalyzed by the enzyme lipoprotein lipase ( LPL ) , which is highly abundant in BAT ( Bartelt et al . , 2011; Kersten , 2014 ) . Cold exposure markedly stimulates LPL activity in BAT , causing a concomitant increase in TRL-derived fatty acid uptake and even uptake of whole lipoprotein particles ( Bartelt et al . , 2011; Khedoe et al . , 2014; Klingenspor et al . , 1996 ) . The increase in fatty acid uptake upon cold exposure , which can be mimicked by pharmacological ß3-adrenergic receptor activation , is highly specific for BAT , suggesting that the body may specifically re-direct lipid fuels to BAT during cold exposure ( Bartelt et al . , 2011; Berbée et al . , 2015; Khedoe et al . , 2014 ) . Both transcriptional and ( post- ) translational regulation has been implicated in the increased LPL activity in BAT upon cold exposure . However , the specific underlying mechanisms have remained elusive ( Carneheim et al . , 1988; Giralt et al . , 1990; Mitchell et al . , 1992 ) . Angiopoietin-like 4 ( ANGPTL4 ) has previously been identified as an inhibitor of LPL activity in muscle and WAT . Alterations of Angptl4 expression in these tissues mediate the changes in LPL activity observed during exercise and fasting , respectively ( Catoire et al . , 2014; Kroupa et al . , 2012 ) . In the initial paper describing the cloning of Angptl4 , we had seen high expression of Angptl4 mRNA in BAT ( Kersten et al . , 2000 ) . Since the exact mechanisms behind regulation of LPL activity in BAT upon cold exposure are currently unclear , we hypothesized that ANGPTL4 may act as an important regulator of LPL-mediated fatty acid uptake into BAT . Accordingly , in the present paper we studied the role of ANGPTL4 in lipid metabolism during cold exposure , taking advantage of Angptl4-deficient ( Angptl4-/- ) mice and Angptl4-overexpressing ( Angptl4-Tg ) mice . ANGPTL4 and LPL protein are co-expressed in BAT ( Figure 1A ) , suggesting a possible role for ANGPTL4 in regulating LPL in BAT . The specificity of the antibody used in the immunoblotting for ANGPTL4 was demonstrated by the lack of observable signal in BAT of Angptl4-/- mice ( Figure 1B - left panel ) . Interestingly , we detected bands for ANGPTL4 at a molecular weight of ∼52 kDa and ∼45 kDa . The expected molecular weight of ANGPTL4 protein is 45 kDa , with a predicted glycosylation site at amino acid Asparagine-177 ( in humans ) or Asparagine 181 ( in mice ) ( Ge et al . , 2004; Kim et al . , 2000; Yang et al . , 2008; Yoon et al . , 2000 ) . Consistent with the notion that these bands represent glycosylated ( ∼52 kDa ) and non-glycosylated forms ( ∼45 kDa ) of ANGPTL4 , the 52 kDa band disappeared following treatment of BAT lysates with the endoglycosidase PGNase , while the 45 kDa band became more intense ( Figure 1B – right panel ) . These data indicate that ANGPTL4 protein is present in BAT in glycosylated and non-glycosylated forms . Immunofluorescent staining on sections from human BAT obtained during surgery ( as qualified based on UCP1 staining ) showed that ANGPTL4 is also expressed in human BAT ( Figure 1C ) . In agreement with high ANGPTL4 expression levels in BAT , Angptl4 mRNA increased markedly upon differentiation of a mouse brown adipocyte cell line ( Figure 1D ) . 10 . 7554/eLife . 08428 . 003Figure 1 . ANGPTL4 expression in BAT is down-regulated upon sustained cold exposure . ( A ) Immunoblot for mouse ANGPTL4 and mouse LPL in lysates of kidney , spleen , heart , muscle , gonodal WAT , inguinal WAT and BAT of a C57BL/6J wild-type mouse . ( B ) Validation of anti-mANGPTL4 antibody in BAT lysates of Angptl4-/- , wild-type and Angptl4-Tg mice ( left panel ) . Detection of glycosylated and non-glycosylated mANGPTL4 following treatment of BAT homogenate of a wild-type mouse with PGNase ( right panel ) . ( C ) Immunofluorescent staining of UCP1 ( upper panel; UCP1 = green , DAPI = blue ) , DAPI only ( middle panel ) and hANGPTL4 ( lower panel; hANGPTL4 = green , DAPI = blue ) in frozen sections ( 5 μm ) of human BAT . ( D ) Angptl4 mRNA in T37i cells after 0 , 3 , 6 or 9 days of differentiation . ( E ) Angptl4 mRNA in BAT lysates of wild-type mice exposed to 4°C or 28°C for 1 or 10 days . ( F ) Immunoblot for ANGPTL4 in BAT homogenates of wild-type mice exposed to 4°C or 28°C for 10 days , following treatment with or without PGNase . ( G ) Lpl mRNA in BAT lysates of wild-type mice exposed to 4°C or 28°C for 1 or 10 days . * Statistically significant compared to control wells or compared to mice exposed to 28°C according to Student’s t-test ( p<0 . 05 ) . Error bars represent ± SEM . n = 8–10 mice per group . DOI: http://dx . doi . org/10 . 7554/eLife . 08428 . 003 We next explored the possible impact of cold on ANGPTL4 expression in BAT . Intriguingly , whereas 1 day of cold exposure did not affect Angptl4 expression , 10 days of cold exposure led to a marked reduction in Angptl4 mRNA ( Figure 1E ) . The reduction in Angptl4 mRNA was paralleled by a marked decrease in ANGPTL4 protein , particularly the glycosylated form of ANGPTL4 ( Figure 1F ) . In contrast , expression of Lpl was mildly elevated after both 1 and 10 days of cold exposure ( Figure 1G ) . To investigate a possible role for ANGPTL4 in BAT function , we exposed Angptl4-/- , wild-type and Angptl4-Tg mice to a cold ( 4°C ) or thermo-neutral ( 28°C ) environment for 10 days in order to activate and recruit BAT ( Figure 2A ) . The Angptl4-Tg mice overexpress Angptl4 under its own promoter and show elevated Angptl4 expression in a variety of tissues , including BAT ( Mandard et al . , 2006 ) . The sustained cold exposure resulted in pronounced changes in BAT morphology , food intake , body weight , weights of BAT and WAT , and body temperature , but no clear differences between the genotypes could be observed ( Figure 2B–F ) . Likewise , expression of the key thermogenic genes Ucp1 and Elovl3 was significantly increased upon cold exposure but not affected by Angptl4 genotype ( Figure 2G ) . 10 . 7554/eLife . 08428 . 004Figure 2 . Down-regulation of ANGPTL4 in BAT upon sustained cold exposure affects plasma TG levels . ( A ) Schematic representation of cold exposure experiment with Angptl4-/- , wild-type and Angptl4-Tg mice . ( B ) Haematoxylin & Eosin staining on BAT sections ( 5 μm ) of Angptl4-/- , wild-type and Angptl4-Tg mice exposed to 4°C or 28°C for 10 days . ( C ) Food intake of Angptl4-/- , wild-type and Angptl4-Tg mice exposed to 4°C for 10 days . ( D ) Weight gain of Angptl4-/- , wild-type and Angptl4-Tg mice exposed to 4°C during 10 days . ( E ) BAT and WAT tissue weights and ( F ) body temperature of Angptl4-/- , wild-type and Angptl4-Tg mice exposed to 4°C or 28°C for 10 days . ( G ) Elovl3 and Ucp1 mRNA expression levels of Angptl4-/- , wild-type and Angptl4-Tg mice exposed to 4°C or 28°C for 10 days . ( H ) Plasma glucose , ( I ) plasma free fatty acids , and ( J ) plasma glycerol levels of Angptl4-/- , wild-type and Angptl4-Tg mice exposed to 4°C or 28°C for 10 days . ( K ) Glycerol levels in medium of differentiated primary white adipocytes from Angptl4-/- , wild-type and Angptl4-Tg mice , serum-starved and treated with 10 μM isoproterenol for 3 hr . ( J ) plasma TG levels of Angptl4-/- , wild-type and Angptl4-Tg mice exposed to 4°C or 28°C for 10 days . ( K ) Fast protein liquid chromatography ( FPLC ) on pooled plasma samples of Angptl4-/- , wild-type and Angptl4-Tg mice exposed to 4°C for 10 days , followed by analysis of TG levels in all fractions . Statistically significant compared to mice of equal genotype at 28°C or between groups as indicated by bars according to two-way ANOVA followed by a post-hoc Tukey HSD test ( p<0 . 05 ) . Error bars represent ± SEM . n = 8–10 mice per group . DOI: http://dx . doi . org/10 . 7554/eLife . 08428 . 004 Following cold exposure , the energy requirements of BAT increase dramatically . Two major fuel sources for BAT are plasma glucose and free fatty acids , both of which were unaltered by cold and Angptl4 genotype ( Figure 2H , I ) . Also , plasma glycerol , which is perhaps a better indicator of adipose tissue lipolysis than free fatty acids , was not different between the three genotypes ( Figure 2J ) . Additionally , ex vivo treatment of differentiated primary white adipocytes with the non-selective β-adrenergic receptor agonist isoproterenol indicated a lack of effect of Angptl4 genotype on adipose tissue lipolysis ( Figure 2K ) . Besides glucose and free fatty acids , circulating TG represent a major fuel for BAT during cold ( Cannon and Nedergaard , 2004 ) . Based on the marked decrease in Angptl4 mRNA levels in BAT upon prolonged cold exposure , we hypothesized that ANGPTL4 might play a role in the metabolism of circulating TG in BAT during cold . In line with this notion , the reduction in plasma TG visible in wild-type and Angptl4-/- mice in response to cold was greatly attenuated in Angptl4-Tg mice ( Figure 2L ) . Lipoprotein profiling by fast protein liquid chromatography ( FPLC ) supported markedly augmented plasma TRL levels in cold-exposed Angptl4-Tg mice ( Figure 2M ) . It is well-established that cold-induced reductions in plasma TG levels are mediated by increased LPL activity in BAT ( Bartelt et al . , 2011; Kersten , 2014; Klingenspor et al . , 1996 ) . Accordingly , we tested whether ANGPTL4 levels influence changes in BAT LPL activity upon prolonged cold exposure . Whereas wild-type mice respond to cold with a reduction in BAT ANGPTL4 mRNA and protein levels , Angptl4-Tg mice maintain higher ANGPTL4 mRNA and protein levels ( Figure 3A , B ) . Mirroring the levels of ANGPTL4 , the cold-induced changes in LPL activity in BAT exhibited a gradient across the three Angptl4 genotypes , with highest LPL activity observed in Angptl4-/- mice and lowest LPL activity in Angptl4-Tg mice ( Figure 3C ) . Furthermore , the marked increase in LPL activity during cold observed in the wild-type mice was significantly blunted in Angptl4-Tg mice ( Figure 3C ) . Interestingly , the two-fold increase in LPL activity by cold in Angptl4-/- mice indicates that part of the induction of LPL activity in BAT is independent of ANGPTL4 ( Figure 3C ) , possibly via an increase in Lpl mRNA , which was observed in all three genotypes ( Figure 3D ) ( Bartelt et al . , 2011; Klingenspor et al . , 1996 ) . Overall , the marked gradient in LPL activity between Angptl4-/- , wild-type , and Angptl4-Tg mice strongly suggests that ANGPTL4 acts as an inhibitor of LPL activity in BAT . 10 . 7554/eLife . 08428 . 005Figure 3 . Down-regulation of ANGPTL4 in BAT upon sustained cold exposure promotes an increase in BAT LPL activity . ( A ) Angptl4 mRNA in BAT of Angptl4-/- , wild-type and Angptl4-Tg mice exposed to 4°C or 28°C for 10 days . ( B ) Immunoblot for ANGPTL4 and LPL in BAT homogenates from Angptl4-/- , wild-type and Angptl4-Tg mice exposed to 4°C or 28°C for 10 days . ( C ) Total LPL activity and ( D ) Lpl mRNA in BAT of Angptl4-/- , wild-type and Angptl4-Tg mice exposed to 4°C or 28°C for 10 days . Statistically significant compared to mice of equal genotype at 28°C or between groups as indicated by bars , according to two-way ANOVA followed by a post-hoc Tukey HSD test ( p<0 . 05 ) . Error bars represent ± SEM . n = 8–10 mice per group . DOI: http://dx . doi . org/10 . 7554/eLife . 08428 . 005 To investigate the role of ANGPTL4 in plasma TG clearance by BAT , we injected cold-exposed Angptl4-/- , wild-type and Angptl4-Tg mice with radiolabelled VLDL-like emulsion particles containing glycerol tri[3H]oleate ( hydrolysable by LPL; TRL-derived fatty acids ) and [14C]cholesteryl-oleate ( not hydrolysable by LPL; TRL-Chol ) ( see Figure 4—figure supplement 1 for experimental set-up ) ( Rensen et al . , 1995 ) . After 15 min , the mice were sacrificed and the tissue distribution of 3H and 14C activity was determined . As expected , cold exposure markedly increased the rate of clearance of the injected VLDL-like particles from the plasma ( Figure 4A , B ) . However , after cold exposure , plasma clearance of glycerol tri[3H]oleate ( TRL FA ) , but not [14C]cholesteryl-oleate ( TRL Chol ) , was significantly slower in Angptl4-Tg mice , indicating that Angptl4-overexpression inhibits LPL-mediated plasma TG clearance in the cold ( Figure 4C , D ) . Cold exposure caused a marked increase in TRL-derived fatty acid uptake into BAT in all three genotypes , indicating that part of the increase in fatty acid uptake is independent of ANGPTL4 ( Figure 4E ) . However , similar to LPL activity , a clear gradient in TRL-derived fatty acid uptake into BAT was observed between the three genotypes , with lowest uptake in Angptl4-Tg mice ( Figure 4E ) . Similar results were obtained for [14C]cholesteryl-oleate , showing a marked decrease in BAT uptake in the Angptl4-Tg mice ( Figure 4F ) . We repeated the plasma TG clearance studies using radiolabelled lipoprotein particles with a larger diameter , resembling postprandial chylomicrons . Again , Angptl4-overexpression markedly reduced uptake into BAT of the TRL-derived fatty acids and the core label cholesteryl-oleyloleate ( Figure 4G , H ) . To visualize the TRL uptake process , we injected hydrophobic fluorescent nanocrystals embedded in lipoprotein particles ( QD-TRLs ) into cold-exposed Angptl4-/- , wild-type and Angptl4-Tg mice ( Bruns et al . , 2009 ) . In agreement with a role for ANGPTL4 in TRL processing in BAT , increased accumulation of QD-TRLs was observed after cold exposure of wild-type mice , but not of Angptl4-Tg mice ( Figure 4I ) . Furthermore , Angptl4-/- mice show an accumulation of QD-TRLs in BAT even when maintained at 28°C ( Figure 4I ) . Together , these data are supportive of a major role for ANGPTL4 as a regulator of LPL activity and concomitant uptake of fatty acids into BAT upon prolonged cold exposure . 10 . 7554/eLife . 08428 . 006Figure 4 . Down-regulation of ANGPTL4 in BAT upon sustained cold exposure promotes an increase in TRL-derived fatty acid uptake by BAT . ( A , B ) Plasma 3H ( A ) and 14C ( B ) activity in wild-type mice exposed to 4°C or 28°C for 10 days intravenously injected with VLDL-like particles labelled with glycerol tri[3H]oleate ( TRL FA ) and [14C]cholesteryl-oleate ( TRL Chol ) . ( C , D ) Plasma 3H ( C ) and 14C ( D ) activity in Angptl4-/- , wild-type and Angptl4-Tg mice intravenously injected with VLDL-like emulsion particles labelled with glycerol tri[3H]oleate ( TRL FA ) and [14C]cholesteryl-oleate ( TRL Chol ) , following exposure to 4°C for 10 days . ( E , F ) 3H activity ( E ) and 14C activity ( F ) in interscapular BAT of Angptl4-/- , wild-type and Angptl4-Tg mice exposed to 4°C or 28°C for 10 days and intravenously injected with VLDL-like particles labelled with glycerol tri[3H]oleate ( TRL FA ) and [14C]cholesteryl-oleate ( TRL Chol ) . ( G , H ) 14C and 3H activity in interscapular BAT of Angptl4-/- , wild-type and Angptl4-Tg mice exposed to 4°C or 28°C for 10 days and intravenously injected with chylomicron-like particles labelled with glycerol tri[14C]oleate ( TRL FA ) ( G ) and [3H]cholesteryl-oleyloleate ( TRL Chol ) . ( I ) Fluorescent image of uptake of intravenously injected QD-TRLs into BAT of Angptl4-/- , wild-type and Angptl4-Tg mice exposed to 4°C or 28°C for 9 days . Image was taken 12 min post-injection . n = 2 mice per group . * Statistically significant compared to mice of equal genotype at 28°C or between groups as indicated by bars , according to two-way ANOVA followed by a post-hoc Tukey HSD test ( p<0 . 05 ) . Error bars represent ± SEM . n = 7 mice per group , unless otherwise indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 08428 . 00610 . 7554/eLife . 08428 . 007Figure 4—figure supplement 1 . ( A , B ) Schematic representation of clearance studies in Angptl4-/- , wild-type and Angptl4-Tg mice , consisting of 10 days of cold exposure or thermo-neutrality , followed by measurement of triglyceride clearance via injection of either radiolabelled VLDL-like particles or chylomicron-like particles . Mice were sacrificed 20 min post-injection to analyze the distribution of the radioactive labels . DOI: http://dx . doi . org/10 . 7554/eLife . 08428 . 007 Whereas BAT utilizes lipids to fuel thermogenesis , WAT provides lipid fuels to be used by BAT . Accordingly , lipid uptake and LPL activity are expected to be regulated differently in BAT as compared to WAT . Indeed , whereas in wild-type mice LPL activity in BAT increased dramatically during cold , LPL activity in WAT remained unchanged , at least in wild-type and Angptl4-Tg mice ( Figure 5A ) . However , LPL activity was increased by cold in Angptl4-/- mice , indicating that ANGPTL4 prevents LPL activity in WAT from going up during cold ( Figure 5A ) . In support of this function for ANGPTL4 , cold exposure caused a marked increase in Angptl4 mRNA and protein levels in WAT ( Figure 5B , C ) . In contrast to Angptl4 , Lpl mRNA levels were not significantly altered in WAT upon sustained cold exposure ( Figure 5D ) . Interestingly , exposure of human subjects to mild cold ( 16°C ) for 48 hr significantly increased plasma ANGPTL4 levels in obese subjects , but not in lean subjects . Considering the higher adipose tissue mass in obese individuals , these data suggest that the increase in ANGPTL4 production in WAT dominates ANGPTL4 levels in human plasma ( Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 08428 . 008Figure 5 . Up-regulation of ANGPTL4 in WAT upon sustained cold exposure suppresses WAT LPL activity and TRL-derived fatty acid uptake . ( A ) Total LPL activity levels and ( B ) Angptl4 mRNA in WAT of Angptl4-/- , wild-type and Angptl4-Tg mice exposed to 4°C or 28°C for 10 days . ( C ) Immunoblot for ANGPTL4 and LPL in WAT homogenates of wild-type mice exposed to 4°C or 28°C for 10 days . ( D ) Lpl mRNA in WAT of Angptl4-/- , wild-type and Angptl4-Tg mice exposed to 4°C or 28°C for 10 days . ( E–H ) Activity of 3H and 14C radiolabels in WAT of Angptl4-/- , wild-type and Angptl4-Tg mice exposed to 4°C or 28°C for 10 days and intravenously injected with radiolabelled VLDL-like ( E , G ) and chylomicron-like particles ( F , H ) . TRL FA uptake ( E , F ) reflects uptake of glycerol tri[3H/14C]oleate , whereas TRL Chol uptake ( G , H ) reflects uptake of the core labels [14C]cholesteryl-oleate or [3H]cholesteryl-oleyloleate . ( I ) Fluorescent image of uptake of intravenously injected QD-TRLs into WAT of Angptl4-/- , wild-type and Angptl4-Tg mice exposed to 4°C or 28°C for 9 days . Image was taken after perfusion of mice with PBS containing 50 IU/mL heparin and upon cryosectioning of tissues . n = 2 mice per group . * Statistically significant compared to mice of equal genotype at 28°C or between groups as indicated by bars , according to two-way ANOVA followed by a post-hoc Tukey HSD test ( p<0 . 05 ) . Error bars represent ± SEM . n = 7–10 mice per group , unless otherwise indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 08428 . 00810 . 7554/eLife . 08428 . 009Figure 5—figure supplement 1 . ( A , B ) Plasma ANGPTL4 levels in 10 obese ( A ) and 10 lean ( B ) individuals before and after exposure to a mild cold ( 16°C ) for 48 hr ( Wijers et al . , 2010 ) . Differences between mild cold and baseline were statistically significant in the obese group ( paired Student’s t-test ( p<0 . 05 ) ) . n = 10 individuals per group . DOI: http://dx . doi . org/10 . 7554/eLife . 08428 . 00910 . 7554/eLife . 08428 . 010Figure 5—figure supplement 2 . ( A ) Total LPL activity levels , ( B ) Angptl4 mRNA , and ( C ) Lpl mRNA in inguinal WAT ( iWAT ) of Angptl4-/- , wild-type and Angptl4-Tg mice exposed to 4°C or 28°C for 10 days . Statistically significant compared to mice of equal genotype at 28°C or between groups as indicated by bars , according to two-way ANOVA followed by a post-hoc Tukey HSD test ( p<0 . 05 ) . Error bars represent ± SEM . n = 7–10 mice per group . DOI: http://dx . doi . org/10 . 7554/eLife . 08428 . 010 To determine if the suppressive effect of ANGPTL4 on LPL activity had any impact on uptake of TRL-derived fatty acids in WAT during cold , we measured fatty acid uptake following injection of radiolabelled VLDL-like and chylomicron-like emulsion particles . Even though the uptake behaviour of the two types of TRL-particles was somewhat different , ANGPTL4 expression caused a clear dose-dependent reduction in uptake of TRL-derived fatty acids and the core labels cholesteryl-oleate/cholesteryl-oleyloleate ( Figure 5E–H ) . In addition , inhibition of TRL processing by ANGPTL4 in WAT was visually confirmed by injection of QD-TRLs , showing increased accumulation of QD-TRLs in WAT of Angptl4-/- mice , as compared to wild-type and Angptl4-Tg mice ( Figure 5I ) . It is noteworthy that in the inguinal fat depot , cold-induced changes in Lpl mRNA , Angptl4 mRNA and LPL activity were similar to the changes observed in the gonadal fat depot described above , whereas changes in uptake of TRL-derived fatty acids upon cold exposure were quite distinct , most likely due to marked activation of browning in inguinal fat ( Figure 6A , B; Figure 5—figure supplement 2 ) . In contrast to BAT and WAT , uptake of TRL-derived fatty acids or cholesteryl-oleate/cholesteryl-oleyloleate from VLDL-like particles and chylomicron-like particles was minimally different between Angptl4-/- , wild-type and Angptl4-Tg mice in liver , skeletal muscle and spleen ( Figure 6A , B ) . Taken together , our data indicate that in BAT down-regulation of ANGPTL4 promotes uptake of plasma TRL-derived fatty acids via enhanced LPL activity , whereas in WAT up-regulation of ANGPTL4 suppresses uptake of plasma TRL-derived fatty acids via inhibition of LPL activity , thereby directing plasma TG to BAT to be used as fuel . 10 . 7554/eLife . 08428 . 011Figure 6 . Uptake of TRL-like particles in liver , spleen and muscle is not affected by Angptl4 genotype . ( A ) 3H and 14C activity in liver , spleen , muscle , inguinal WAT ( iWAT ) and subscapular BAT ( sBAT ) of Angptl4-/- , wild-type and Angptl4-Tg mice exposed to 4°C or 28°C for 10 days and intravenously injected with VLDL-like emulsion particles labelled with glycerol tri[3H]oleate ( TRL FA ) and [14C]cholesteryloleate ( TRL Chol ) . ( B ) 3H and 14C activity in liver , spleen , muscle , iWAT and sBAT of Angptl4-/- , wild-type and Angptl4-Tg mice exposed to 4°C or 28°C for 10 days and intravenously injected with chylomicron-like particles labelled with glycerol tri[14C]oleate ( TRL FA ) and [3H]cholesteryl-oleyloleate ( TRL Chol ) . Statistically significant compared to values of wild-type mice according to Student’s t-test ( p<0 . 05 ) . Error bars represent ± SEM . n = 7 mice per group . DOI: http://dx . doi . org/10 . 7554/eLife . 08428 . 011 We next explored the mechanism accounting for the inverse regulation of ANGPTL4 expression in BAT and WAT . Since many of the cold-induced adaptations are triggered by β-adrenergic signalling , we examined the effect of β-adrenergic activation on ANGPTL4 expression in BAT and WAT using murine BAT ( T37i cells ) and WAT ( 3T3-F442a ) cell lines and primary cells ( Zennaro et al . , 1998 ) . In both BAT and WAT cells , treatment with the non-selective β-receptor agonist isoproterenol consistently resulted in a marked increase in expression of Angptl4 mRNA ( Figure 7A , B ) . Induction of ANGPTL4 by β-adrenergic stimulation was confirmed at the protein level ( Figure 7C ) . From these data , it is evident that activation of β-adrenergic signalling may contribute to the cold-induced up-regulation of ANGPTL4 in WAT , but cannot explain the down-regulation of ANGPTL4 observed in BAT . Levels of Lpl mRNA and protein in white and brown adipocytes were only mildly affected by treatment with β-adrenergic agonists ( Figure 7C , D ) . 10 . 7554/eLife . 08428 . 012Figure 7 . AMPK is activated in BAT , but not WAT , upon sustained cold exposure . ( A ) Angptl4 mRNA in differentiated mouse white adipocytes ( primary adipocytes and 3T3F442a adipocytes ) upon treatment with 10 μM isoproterenol ( ISO ) or control ( CTRL ) for 3 hr . ( B ) Angptl4 mRNA in differentiated brown adipocytes ( primary adipocytes and T37i adipocytes ) treated with 10 μM isoproterenol ( ISO ) orcontrol ( CTRL ) for 3 hr . ( C ) Immunoblot for ANGPTL4 and LPL protein in differentiated 3T3F422a cells treated with 10 μM isoproterenol ( ISO ) or control ( CTRL ) for 3 hr . ( D ) Lpl mRNA in differentiated mouse primary white or brown adipocytes upon treatment with 10 μM isoproterenol ( ISO ) or control ( CTRL ) for 3 hr . ( E ) Immunoblot for AMPKα1 , 2 and phospho-AMPK Thr172 , AMPKα1 , AMPKα2 , AMPKβ1 and AMPKβ2 in tissue lysates of kidney , spleen , heart , muscle , liver , inguinal WAT , gonodal WAT and BAT . Homogenates are identical to the homogenates presented in Figure 1A . ( F ) Immunoblot for AMPKα1 , 2 and phospho-AMPK Thr172 in BAT and WAT lysates of wild-type mice exposed to 4°C or 28°C for 10 days . Statistically significant compared to control samples or between indicated treatments according to Student’s t-test ( p<0 . 05 ) . Error bars represent ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 08428 . 012 We therefore considered mechanisms that are clearly different between BAT and WAT . Previously , it had been shown that 5’AMP-activated protein kinase ( AMPK ) is progressively activated with prolonged cold exposure and that AMPK expression is significantly higher in BAT compared to WAT ( Mulligan et al . , 2007 ) . Consistent with this finding , we found that the catalytic α-subunits of AMPK , as well as phosphorylation of AMPK at Threonine-172 – indicative of AMPK activity – are barely detectable in WAT in the basal state ( Figure 7E ) ( Viollet et al . , 2009 , 2010 ) . Previously , differential tissue expression of isoforms of the α , β , and γ subunits of the AMPK heterotrimer was suggested to determine cellular and systemic responses to different metabolic stressors ( Viollet et al . , 2009 , 2010 ) . Intriguingly , we detected large differences in basal AMPK α- and β-subunit isoform distribution between BAT and WAT and , more specifically , found a high expression of the AMPKα2 catalytic subunit and the AMPKβ1 regulatory subunit in BAT compared to WAT ( Figure 7E ) . The differences in total AMPK expression and subunit distribution may explain why WAT AMPK levels remain weak following sustained cold exposure , whereas a strong increase in AMPK and phospho-AMPK is observed in BAT ( Figure 7F ) . In muscle cells , we found that AMPK activation strongly down-regulates ANGPTL4 mRNA and protein levels ( Catoire et al . , 2014 ) . To examine the consequences of AMPK activation in BAT , we treated differentiated T37i brown adipocytes with multiple AMPK activators , including AICAR , A769662 , metformin , and phenformin hydrochloride . Without exception , AMPK activation markedly down-regulated Angptl4 mRNA expression ( Figure 8A ) . To further confirm the down-regulation of ANGPTL4 by AMPK , we treated different in vitro model systems for BAT with the AMPK activator AICAR . AICAR treatment of differentiated T37i adipocytes , BA adipocytes ( Uldry et al . , 2006 ) or murine primary brown adipocytes resulted in a marked decrease in ANGPTL4 mRNA and protein levels ( Figure 8B , C ) . Part of the inhibitory effect of AMPK activation on Angptl4 expression in differentiated T37i adipocytes could be rescued by siRNA-mediated knock-down of AMPKα1 and AMPKα2 , corroborating the suppressive effect of AMPK on Angptl4 expression ( Figure 8D–F ) . Consistent with the notion that the negative regulation of ANGPTL4 by AMPK is specific for BAT , AICAR treatment markedly reduced ANGPTL4 mRNA and protein levels in BAT explants , but not WAT explants ( Figure 8G , H ) . Based on these data , we propose that the different amount and activation of AMPK between BAT and WAT may be the critical factor in the differential regulation of ANGPTL4 between the two tissues during sustained cold . 10 . 7554/eLife . 08428 . 013Figure 8 . Activation of AMPK down-regulates ANGPTL4 expression specifically in brown adipocytes . ( A ) Angptl4 mRNA in differentiated T37i adipocytes treated for 6 hr with 1 mM AICAR , 100 μM A769662 , 1 mM metformin or 250 μM phenformin hydrochloride . ( B ) Angptl4 mRNA in differentiated primary brown adipocytes , BA adipocytes , or T37i adipocytes treated for indicated times with 1 mM AICAR . ( C ) Immunoblot for ANGPTL4 , LPL , AMPKα1 , 2 and phospho-AMPK Thr172 in differentiated T37i cells treated with control ( CTRL ) or 1 mM AICAR for 3 hr . ( D ) Angptl4 mRNA in differentiated T37i adipocytes treated with CTRL siRNA or siRNA against AMPKα1 and AMPKα2 for 48 hr , followed by incubation with control ( CTRL ) or 1 mM AICAR for 3 hr . ( E ) Angptl4 mRNA in differentiated T37i adipocytes treated with CTRL siRNA or siRNA against AMPKα1 and AMPKα2 for 48 hr , followed by incubation with H2O control medium ( CTRL ) or 100 μM A769662 for 6 hr . ( E ) Ampkα1 and Ampkα2 mRNA in differentiated T37i adipocytes treated with CTRL siRNA or siRNA against AMPKα1 and AMPKα2 for 48 hr . ( F ) Angptl4 mRNA levels in BAT and WAT explants from C57BL/6J wild-type mice ( ∼50 μg ) treated with H2O control medium ( CTRL ) or 1 mM AICAR for 3 hr . ( G ) Immunoblot for ANGPTL4 in BAT and WAT explants from C57BL/6J wild-type mice ( ∼50 mg ) treated with H2O control medium ( CTRL ) or 1 mM AICAR for 3 hr . Statistically significant compared to control samples or between indicated treatments , according to Student’s t-test ( p<0 . 05 ) . Error bars represent ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 08428 . 013 We further pursued the mechanism behind the down-regulation of ANGPTL4 by AMPK . Treatment of brown adipocytes with AICAR and the transcriptional inhibitor actinomycin D showed that both compounds reduced Angptl4 gene expression by nearly the same extent . Co-treatment of brown adipocytes with actinomycin D and AICAR did not result in an additive effect , suggesting that AMPK activation almost completely inhibited Angptl4 gene transcription ( Figure 9A ) . 10 . 7554/eLife . 08428 . 014Figure 9 . Down-regulation of Angptl4 expression by AMPK is likely mediated via inhibition of PPARγ-mediated transcription of Angptl4 . ( A ) Angptl4 mRNA in differentiated T37i adipocytes pre-incubated with 0 . 5 μg/mL actinomycin D ( ActD ) or DMSO control for 1 hr and treated with 1 mM AICAR or control for 3 hr . ( B ) Angptl4 mRNA in differentiated T37i or BA adipocytes treated with DMSO control , 5 μM rosiglitazone , 10 μM Wy14643 , or 5 μM GW742 ( BA adipocytes ) or 5 μM GW501516 ( T37i adipocytes ) for 6 hr . ( C ) Immunoblot for ANGPTL4 in differentiated T37i adipocytes treated with 5 μM rosiglitazone ( Rosi ) , 10 μM Wy14643 ( Wy ) , and 5 μM GW501516 for 24 hr . ( D ) Angptl4 mRNA in differentiated T37i cells treated with DMSO control , 5 μM rosiglitazone ( Rosi ) , 1 mM AICAR or both rosiglitazone and AICAR for 6 hr . ( E ) Relative luciferase activity of HepG2 cells transfected with pGL3-Angptl4 , pSG5-Pparγ , pSG5-Rxr , pcDNA-P300HA vectors , as indicated and treated 16 hr post-transfection with 5 μM rosiglitazone , 1 mM AICAR or both compounds for 9 hr . Data are represented as ± SD . ( F ) Immunoblot of P300 and phospho-P300 ( Ser-89 ) in differentiated T37i brown adipocytes treated with control or 1 mM AICAR for 3 hr . Statistically significant compared to control samples or between indicated samples according to Student’s t-test ( p<0 . 05 ) . Error bars represent ± SEM , unless otherwise specified . DOI: http://dx . doi . org/10 . 7554/eLife . 08428 . 014 Expression of ANGPTL4 is under sensitive control of peroxisome proliferator-activated receptors ( PPARs ) in many tissues , with PPARγ being the dominant regulator of ANGPTL4 in adipose tissue ( Kersten et al . , 2000; Mandard et al . , 2004; Yoon et al . , 2000 ) . We found that PPARγ agonists and to a lesser extent agonists for PPARα and PPARδ highly induce ANGPTL4 mRNA and protein in BA and T37i brown adipocytes ( Figure 9B , C ) . Previous studies have shown that AMPK may inhibit PPARα and PPARγ transcriptional activity ( Leff , 2003; Sozio et al . , 2011 ) . Indeed , we observed that AMPK activation almost completely blocked the induction of Angptl4 mRNA following treatment with the PPARγ agonist rosiglitazone ( Figure 9D ) . Accordingly , we hypothesized that activation of AMPK following prolonged cold exposure may inhibit PPARγ-mediated transcription of the Angptl4 gene . To examine this possibility , a luciferase construct was prepared containing the three conserved PPAR response elements ( PPREs ) of intron 3 of the murine Angptl4 gene . These three PPREs have previously been shown to be responsible for PPAR-mediated up-regulation of Angptl4 ( Kaddatz et al . , 2010; Mandard et al . , 2004 ) . In HepG2 cells transfected with the Angptl4 PPRE construct , rosiglitazone treatment significantly induced luciferase activity , which was further increased upon co-transfection with Pparγ/Retinoid X receptor ( Rxr ) . By contrast , co-treatment with AICAR blunted the increase in luciferase activity ( Figure 9E ) . Interestingly , whereas co-transfection of the PPARγ co-activator P300 ( Gelman et al . , 1999 ) further stimulated luciferase activity in the absence of AICAR , it failed to do so in the presence of AICAR , suggesting that regulation of P300 may be part of the mechanism of inhibition of ANGPTL4 by AMPK ( Figure 9E ) . In support of this possibility , treatment of brown adipocytes with AICAR resulted in phosphorylation of P300 at serine residue 89 , which is known to reduce the capacity of P300 to co-activate nuclear transcription factors , whereas no change in total P300 protein levels was observed ( Figure 9F ) . Collectively , these data suggest that the negative regulation of ANGPTL4 in BAT upon cold exposure may be mediated by the inhibition of PPARγ transcriptional activity by AMPK . The energy requirements of BAT increase manifold during cold exposure . The increased energy demands coincide with a marked increase in LPL activity , stimulating uptake of TRL-derived fatty acids ( Bartelt et al . , 2011; Bertin et al . , 1985; Khedoe et al . , 2014; Klingenspor et al . , 1989 , 1996 ) . Increased LPL activity has been shown to be essential for the lipid-lowering effect of cold exposure , as injection of heparin or tetrahydrolipstatin compromises LPL-dependent uptake of TRLs and TRL-derived fatty acids ( Bartelt et al . , 2011 ) . Since Lpl mRNA in BAT is only moderately increased upon prolonged cold exposure , it has been suggested that the pronounced increase in LPL activity in BAT occurs at the post-translational level ( Giralt et al . , 1990; Klingenspor et al . , 1996 ) . Our data demonstrate that a substantial part of the increase in LPL activity in BAT during prolonged cold exposure is mediated by down-regulation of ANGPTL4 . Overall , our findings reveal a major role for ANGPTL4 in the regulation of lipid partitioning during sustained cold . Specifically , the data implicate ANGPTL4 as an important mediator of preferential shuttling of TRL-derived fatty acids to BAT during cold exposure . Via direct effects on local LPL activity and subsequent fatty acid uptake , the reciprocal regulation of ANGPTL4 in BAT and WAT assures an adequate fuel delivery to BAT during cold exposure . The differential regulation of ANGPTL4 and LPL between BAT and WAT leads to corresponding changes in fatty acid uptake from TRLs , with our data showing a clear dose-dependent and causal relationship between ANGPTL4 expression and TRL-derived fatty acid uptake into both tissues . The importance of ANGPTL4 in the regulation of LPL activity during cold complements the already established role of ANGPTL4 in regulation of LPL during fasting and exercise in WAT and skeletal muscle , respectively ( Catoire et al . , 2014; Kroupa et al . , 2012 ) . ANGPTL4 can thus be viewed as the master regulator of tissue LPL activity and fatty acid uptake during physiological conditions such as fasting , exercise and cold exposure . Our data suggest that the opposite regulation of ANGPTL4 expression during prolonged cold between BAT and WAT may be explained by the differential expression and activation of AMPK between the two tissues . Heterotrimeric AMPK has one catalytic ( α ) , and two regulatory ( β and γ ) subunits , each having distinctive isoforms with a tissue-specific distribution ( Viollet et al . , 2009 , 2010 ) . Tissue-specific combinations of different subunit isoforms may confer tissue-specific properties to AMPK by determining subcellular localization and substrate targeting , thereby controlling cellular and systemic responses to metabolic stressors , including sustained cold ( Viollet et al . , 2009 , 2010 ) . Indeed , with prolonged cold exposure , AMPK becomes progressively activated in BAT and only to a minor extent in WAT ( Bauwens et al . , 2011; Mulligan et al . , 2007 ) . Systemic activation of AMPK has been previously shown to increase the activity of LPL in both heart and muscle and to lower plasma TG levels ( An et al . , 2005; Bergeron et al . , 2001; Buhl et al . , 2002; Catoire et al . , 2014; Geerling et al . , 2014; Ohira et al . , 2009 ) . Furthermore , AMPK activation was found to cause a pronounced reduction in ANGPTL4 expression in muscle cells ( Catoire et al . , 2014 ) . Similar to muscle and heart , we found the AMPKα2 catalytic subunit to be abundantly present in BAT , but not WAT ( Viollet et al . , 2009 ) . Together , the increased AMPK activation and different AMPK subunit expression in BAT as compared to WAT may explain why the repressive effect of cold-induced AMPK activation on ANGPTL4 expression is much more pronounced in BAT than in WAT . We suggest that regulation of ANGPTL4 in WAT during cold may be dominated by activation of β-adrenergic signalling , which may explain the increase in ANGPTL4 expression observed in WAT during sustained cold exposure . We provide evidence that regulation of ANGPTL4 by AMPK occurs at the transcriptional level , affecting PPARγ-mediated transcription of the Angptl4 gene . ANGPTL4 has been repeatedly shown to be a highly sensitive target of all PPAR transcription factors in a variety of tissues and cells and following a variety of physiological stimuli ( Dijk and Kersten , 2014; Georgiadi et al . , 2010; Kersten et al . , 2000; Mandard et al . , 2004 ) . Previously , PPARγ-mediated transcription has been shown to be inhibited by activation of AMPK ( Cheang et al . , 2014; Namgaladze et al . , 2013; Sozio et al . , 2011 ) . A potential link between PPARγ and AMPK may be the modulation of co-activator recruitment to PPARγ by AMPK . A well-established co-activator of PPARγ that has been shown to be regulated by AMPK is P300 ( Gelman et al . , 1999; Leff , 2003; Yang et al . , 2001 ) . P300 is a key regulator of the assembly and mobilization of the transcriptional machinery by connecting transcription factors to the transcriptional machinery and enhancing DNA accessibility ( Vo and Goodman , 2001 ) . AMPK activation enhances P300 degradation and causes phosphorylation of P300 at serine residue 89 , thereby blocking the interaction of P300 with PPARγ and reducing PPARγ transcriptional activity ( Leff , 2003; Lim et al . , 2012; Yang et al . , 2001 ) . Although our in vitro data suggest an involvement of P300 and PPARγ in the suppression of ANGPTL4 by AMPK activation , whether P300 is also involved in in vivo regulation of ANGPTL4 during cold exposure remains to be determined . Despite the unmistakable dependency of BAT LPL activity on ANGPTL4 expression , a modest cold-induced increase in LPL activity and TRL-derived fatty acid uptake is observed in BAT of Angptl4 -/- mice , which may be ( partially ) explained by the moderate increase in Lpl mRNA in BAT in the cold . Alternatively , there may be a role for another , yet to be identified , cold-induced post-translational modulator of LPL . It may be hypothesized that this post-translational modulator is also involved in the rapid increase in LPL activity during acute cold exposure ( Klingenspor et al . , 1996 ) . No overt abnormalities in cold-tolerance were observed in the Angptl4-/- or Angptl4-Tg mice as compared to wild-type mice . This observation , however , does not refute the importance of ANGPTL4 in fuel delivery to BAT during cold exposure . BAT is a well-conserved organ that is postulated to have conferred to mammals the evolutionary advantage to survive cold stressors such as birth or low environmental temperatures ( Cannon and Nedergaard , 2004 ) . It is likely that differential uptake of TG in other organs and altered uptake of other available fuels ( free fatty acids , glucose ) compensate for the reduced uptake of TRL-derived fatty acids by BAT in Angptl4-Tg mice . Strikingly , adipose tissue-specific deletion of LPL in mice does not result in an overt phenotype ( Bartelt et al . , 2012; Garcia-Arcos et al . , 2013 ) . While plasma TG levels are elevated in these mice , no other parameters were altered , indicating that even in mice completely lacking LPL in adipose tissue , alternative mechanisms exist to fuel BAT and WAT ( Bartelt et al . , 2012; Garcia-Arcos et al . , 2013 ) . In conclusion , our data show that regulation of ANGPTL4 is an important factor in directing lipid fuels towards BAT and away from WAT during prolonged cold exposure . Better understanding of the mechanisms underlying fuel re-distribution may pave the way for new strategies to combat metabolic diseases , such as cardiovascular disease and diabetes type 2 , in which a mismatch in regulation of lipid uptake and usage by tissues is an important feature ( Klop et al . , 2013; Young and Zechner , 2013 ) . Three- to four-month old Angptl4-/- , wild-type and Angptl4-Tg mice were either placed at a thermo-neutral temperature ( ∼28°C ) ( n = 7/8 , as indicated in figures ) or at a cold temperature ( ∼4°C ) ( n = 7–10 , as indicated in figures ) for a period of 10 days . All animals are backcrossed on a pure C57Bl/6J background for multiple generations ( >10 ) . Wild-type and Angptl4-Tg mice are littermates . Angptl4-/- mice have been obtained via homologous recombination of embryonic stem cells and lack part of the Angptl4 gene , resulting in a non-functional ANGPTL4 protein ( Köster et al . , 2005; Lichtenstein et al . , 2010 ) . Angptl4-Tg mice over-express the Angptl4 gene in various tissues under its own promoter ( Mandard et al . , 2006 ) . Food intake , body weight and body temperature were monitored daily . Body temperature of cold-exposed mice was monitored via read-out of transponders ( IPTT-300 ) that were injected subcutaneously prior to the experiment ( Bio Medic Data Systems , Seaford , USA ) . The Animal Ethics Committees of Wageningen University and University Medical Center Hamburg-Eppendorf approved all experiments . Radiolabelled VLDL-like emulsion particles were essentially prepared as described previously ( Rensen et al . , 1995 ) . Briefly , 100 mg of lipids ( triolein , egg yolk phosphatidylcholine , lysophosphatidylcholine , cholesteryl-oleate and cholesterol ) were mixed with glycerol tri[3H]oleate and [14C]cholesteryl-oleate ( GE Healthcare , Little Chalfont , UK ) and sonicated ( Soniprep 150 , MSE Scientific Instruments , UK ) . The emulsion was fractionated by consecutive density gradient ultracentrifugation ( Beckman , California , USA ) to yield VLDL-like particles with a diameter of ∼80 nm . Radiolabelled chylomicron-like particles with a diameter of ∼250 nm were prepared from lipids derived from human TRLs of apoCII-deficient subjects ( approved by Ärtzekammer Hamburg , Germany ) as described previously ( Bruns et al . , 2009 ) . Briefly , 10 mg of isolated lipid was mixed with glycerol tri[14C]oleate and [3H]cholesteryl-oleyloleate ( Perkin Elmer , Rodgau , Germany ) in chloroform , after which solvent was removed and TRL particles were formed by sonication in 1 mL of PBS for 10 min at 60°C . Aggregates were removed by filtration through a 450 nm filter ( Millipore ) . TRL particles used for intravital microscopy were prepared similarly , but radiolabels were replaced by hydrophobic fluorescent nanocrystals ( QD-TRLS ) . To study the clearance of radiolabelled TRL-like particles ( 80 and 250 nm ) , Angptl4-/- , wild-type and Angptl4-Tg mice were exposed to cold or thermo-neutral temperature for 10 days ( see Figure 4—figure supplement 1 for an overview ) . Prior to the experiment , animals were fasted for 4 hr . To asses TG clearance , mice were injected intravenously with 200 μL radiolabelled TRL-like particles ( 0 . 2 mg TG for VLDL-like particles , 2 mg TG for chylomicron-like particles ) . Lipid turnover was determined for VLDL-like particles from plasma taken at 2 , 5 , 10 , and 15 min following injection . Total plasma volumes were calculated as 0 . 04706 x body weight ( g ) ( Jong et al . , 2001 ) . 15 min after injection , mice were sacrificed and perfused via the heart with ice-cold PBS containing 50 IU/mL heparin . Multiple organs were collected , weighed and solubilized in Tissue Solubilizer ( Amersham Biosciences , Roosendaal , the Netherlands; for VLDL-like particles ) or Solvable ( Perkin Elmer; for chylomicron-like particles ) overnight . 3H and 14C radioactivity was determined via liquid scintillation counting . Uptake of radioactivity derived from TRL-like particles was calculated as % uptake of the injected radiolabel per gram tissue . For intravital microscopy , interscapular BAT was exposed in anesthetized Angptl4-/- , wild-type and Angptl4-Tg mice and visualized by a confocal microscope with resonant scanner ( Nikon A1R ) . QD-TRLs were injected via a tail vein catheter in anesthetized mice , followed by recording of 30 confocal images per second of the interscapular BAT were recorded for a period of 15 min . The acquired data were edited in Nikon NIS Elements . After recording , mice were perfused with PBS containing 50 IU/mL heparin . Then , BAT , gWAT and iWAT were taken for subsequent cryosectioning , to assess the uptake of QD-TRLs via confocal microscopy . Plasma concentrations of glucose ( Sopachem , Ochten , the Netherlands ) , triglycerides ( TG ) , cholesterol ( Instruchemie , Delfzijl , the Netherlands ) , glycerol ( Sigma-Aldrich , Houten , the Netherlands ) and free fatty acids ( Wako Chemicals , Neuss , Germany; HR ( 2 ) Kit ) were determined following the manufacturers’ instructions . LPL activity in whole tissue homogenates from BAT , WAT and inguinal adipose tissue was measured as described previously ( Ruge et al . , 2004 ) . Briefly , extracts of frozen tissue samples were prepared in 9 ml lysis buffer/g tissue ( 0 . 025 M NH3 , 5 mM Na2EDTA , and per ml: 1 mg bovine serum albumin , 10 mg Triton X-100 , 1 mg SDS , 5 IU heparin , and Complete protease inhibitors [Roche] ) by homogenization with a Polytron homogenizer . Homogenates were spun down for 15 min at 3000 × g to obtain the supernatant used to measure LPL activity . 2 μl of supernatant was assayed , in a total volume of 200 μL , with a 3H-oleic acid-labelled triolein containing substrate emulsion having 100 mg soybean triglycerides and 10 mg egg yolk phospholipids per mL . Incubation was at 25°C for 30–100 min , dependent on the expected level of LPL activity . One milliunit ( mU ) of enzyme activity corresponds to 1 nmol of fatty acid released per min . Total RNA was isolated using TRIzol reagent ( Life Technologies Europe BV , Bleiswijk , the Netherlands ) . RNA from WAT depots was purified using the Qiagen RNeasy Micro kit ( Qiagen , Venlo , the Netherlands ) . RNA was reverse transcribed using a First-Strand cDNA Synthesis Kit ( Thermo Scientific , Landsmeer , the Netherlands ) ( for cells ) or iScript cDNA Synthesis Kit ( Bio-Rad , Veenendaal , the Netherlands ) ( for tissues ) . Real-time PCR was carried out using SensiMiX ( Bioline , GC Biotech , Alphen aan de Rijn , the Netherlands ) on a CFX 384 Bio-Rad thermal cycler ( Bio-Rad ) . TBP and 36B4 were used as housekeeping genes . Primer sequences can be found in Table 1 . 10 . 7554/eLife . 08428 . 015Table 1 . Primer sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 08428 . 015GeneForward primerReverse primerm36b4 ATGGGTACAAGCGCGTCCTGGCCTTGACCTTTTCAGTAAGmAngptl4 GTTTGCAGACTCAGCTCAAGGCCAAGAGGTCTATCTGGCTCTGmLpl GGGAGTTTGGCTCCAGAGTTTGGGAGTTTGGCTCCAGAGTTTmUcp1 CCTGCCTCTCTCGGAAACAATGTAGGCTGCCCAATGAACAmPgc1α AGTCCCATACACAACCGCAGTCGCAACATGCCCTTTCTTGGTGGAGTGGCTGCCTTGGmCidea TGACATTCATGGGATTGCAGACGGCCAGTTGTGATGACTAAGACmElovl3 TTCTCACGCGGGTTAAAAATGGGAGCAACAGATAGACGACCACmPrdm16 CCACCAGCGAGGACTTCACGGAGGACTCTCGTAGCTCGAA Fresh tissues ( WAT and BAT ) were fixed in 4% paraformaldehyde , dehydrated and embedded in paraffin . H&E staining was performed using standard protocols . Frozen human BAT sections obtained during surgery ( 5 µm thick ) were fixated during 15 min in 3 . 7% formaldehyde in PBS , followed by incubation for 45 min at room temperature with a primary antibody ( polyclonal rabbit hANGPTL4 or a polyclonal rabbit UCP1-antibody [kind gift of Dr . B . Cannon , Stockholm University] ) ( Vijgen et al . , 2013; Wu et al . , 2012 ) diluted in 0 . 05% Tween20 in PBS . After three washing steps with PBS , sections were incubated for 45 min at room temperature with the appropriate fluorescently labelled secondary antibodies . The specificity of the antibody for ANGPTL4 was demonstrated previously via immunoblot of human plasma using appropriate peptide controls and was validated by staining of ANGPTL4 in human heart , intestine and muscle ( Alex et al . , 2014; Catoire et al . , 2014; Georgiadi et al . , 2010 ) . Plasma ANGPTL4 levels were measured in plasma samples from a published study in which lean and obese human subjects were exposed to mild cold ( 16°C ) for 48 hr ( Wijers et al . , 2010 ) . Plasma hANGPTL4 levels were measured as described previously ( Kersten et al . , 2009 ) . Briefly , 96-well plates were coated with anti-human ANGPTL4 polyclonal goat IgG antibody ( AF3485; R&D Systems ) and were incubated overnight at 4°C . After blocking , 100 μL of 20-fold diluted human plasma was applied to each well , followed by incubation at room temperature for 2 hr . Next , 100 μL of diluted biotinylated anti-human ANGPTL4 polyclonal goat IgG antibody ( BAF3485; R&D Systems ) was added and incubated for 2 hr . Subsequently , streptavidin-conjugated HRP was added for 20 min , followed by tetramethylbenzidine substrate reagent for 6 min . The reaction was stopped by adding 50 μL of 10% H2SO4 . The absorbance was measured at 450 nm . Tissues were lysed in a mild RIPA-like lysis buffer ( 25 mM Tris-HCl pH 7 . 4 , 150 mM NaCl , 1 mM EDTA , 1% NP-40 and 5% glycerol; Thermo Scientific ) with protease and phosphatase inhibitors ( Roche ) . Cells were lysed directly in 2x Laemmli sample buffer ( LSB ) with DTT . Protein lysates ( 20–30 μg protein per lane ) were loaded on a denaturing gel ( Bio-Rad ) and separated by SDS gel electrophoresis . Proteins were transferred to a PVDF membrane by means of a Transblot Turbo System ( Bio-Rad ) . The primary antibody [rabbit anti–phospho-AMPK antibody ( Thr172 ) , rabbit anti-AMPKα1 , α2 antibody ( Cell Signaling Technology , #2535 and #2532 ) , rabbit anti-mouse/human AMPKα1 ( Cell Signaling Technology , #2795 ) , rabbit anti-mouse/human AMPKβ1 and AMPKβ2 ( Cell Signaling Technology , #4148 and #4178 ) , rabbit anti-mouse AMPKα2 antibody ( Abcam , #ab3760 ) , rat anti-mouse ANGPTL4 antibody ( Adipogen , #Kairos 142–2 ) , goat anti-mouse LPL antibody ( kind gift from André Bensadoun ) , rabbit anti-mouse HSP90 antibody ( Cell Signaling Technology , #4874S ) , or mouse anti-mouse β-tubulin antibody ( Santa-Cruz Biotechnology , #sc23949 ) ] was used at a ratio of 1:1000 ( AMPKα1 , α2 , AMPKα2 , AMPKα1 , AMPKβ1 , AMPKβ2 , ANGPTL4 , HSP90 , β-tubulin ) , 1:2000 ( phospho–AMPK ) or 1:5000 ( LPL ) . Corresponding secondary antibodies ( HRP-conjugated ) ( Sigma-Aldrich ) were used at 1:5000 dilutions . All incubations were performed in Tris-buffered saline , pH 7 . 5 , with 0 . 1% Tween-20 ( TBS-T ) and 5% dry milk , except for anti-AMPKα1 , 2 and anti-phospho-AMPK Thr172 antibody , where 5% bovine serum albumin ( BSA ) was used instead of milk . All washing steps were in TBS-T without dry milk or BSA . Blots were visualized using the ChemiDoc MP system ( Bio-Rad ) and Clarity ECL substrate ( Bio-Rad ) . 3T3-F442a cells ( P8-P14 , Sigma ) were maintained in DMEM ( Lonza ) , supplemented with 10% newborn calf serum and 1% penicillin/streptomycin ( P/S ) under 5% CO2 at 37°C . At confluency , cells were switched to DMEM ( Lonza , Verviers , Belgium ) , supplemented with 10% fetal bovine serum ( FBS ) , 1% P/S and 5 μg/mL insulin ( Sigma-Aldrich ) to stimulate differentiation . During differentiation , medium was changed every 2-3 days . After 10 days of differentiation , cells were switched back to regular medium for 2-3 days , after which experiments were performed . T37i cells ( P31-36; kind gift of Marc Lombès ) were cultured in DMEM/F-12 ( Gibco , Life Technologies , Blijswijk , the Netherlands ) , supplemented with 10% FBS and 1% P/S . Two days post-confluency , cell culture medium was supplemented with 112 ng/mL insulin and 2 nM T3 ( Sigma-Aldrich ) to induce differentiation . After 7 days of differentiation , cells were switched back to regular medium and used for experiments 2-3 days after ( Zennaro et al . , 1998 ) . HepG2 cells ( passage unknown ) were maintained in DMEM ( Lonza ) supplemented with 10% FBS and 1% P/S . At each passage , the cell pellet was filtered through a 40 μm filter to reduce cell clumping . Culture of BA adipocytes was performed as described previously ( Uldry et al . , 2006 ) . Briefly , immortalized brown adipocytes are grown to confluence with differentiation medium ( DMEM , 10% FBS , 20 nM insulin , 1 nM T3 ) . Upon confluence , cells were treated with induction medium ( differentiation medium supplemented with 0 . 5 mM IBMX , 0 . 5 μM dexamethasone , 0 . 125 mM indomethacin ) for two days . After washing , cells were incubated in differentiation medium for another 5 to 7 days . WAT was dissected from C57Bl/6J mice and put in DMEM supplemented with 1% P/S and 1% BSA . Tissues of 3–4 mice were pooled , minced with scissors and digested for 1 hr at 37°C in collagenase-containing medium ( DMEM with 3 . 2 mM CaCl2 , 1 . 5 mg/mL Collagenase type II ( Sigma-Aldrich ) , 10% FBS , 0 . 5% Bovine Serum Albumin ( Sigma-Aldrich ) and 15 mM HEPES , filtered ) . After digestion , cell mixture was passed over a 100 μm cell strainer and centrifuged at 1600 rpm for 10 min . Supernatant was removed and the pellet containing the stromal vascular fraction was re-suspended in erythrocyte lysis buffer ( 155 mM NH4Cl , 12 mM NaHCO3 , 0 . 1 mM EDTA ) and incubated for 2–3 min at room temperature . Following neutralization , cells were centrifuged at 1200 rpm for 5 min . Cells were re-suspended in DMEM containing 10% FBS and 1% P/S and plated . Upon confluence , cells were differentiated following standard protocol of 3T3-L1 cells , as described previously ( Alex et al . , 2013 ) . For isolation of primary brown adipocytes , BAT from 1-month-old pups was used . Tissues of 5–10 pups were pooled , minced with scissors and digested for 30 min in collagenase-containing medium at 37°C ( DMEM w/o serum , 2 mg/mL Collagenase type II , 2% BSA , 25 mM HEPES ) . After digestion , cells were passed through a 70 or 100 μM filter , mature adipocytes were discarded and cells were centrifuged at 800g for 5 min . Cells were re-suspended in differentiation medium ( DMEM , 10% FBS , 20 nM insulin , 1 nM T3 ) and plated . Upon confluence , cells were treated with induction medium ( differentiation medium supplemented with 0 . 5 mM IBMX , 0 . 5 μM dexamethasone , 0 . 125 mM indomethacin ) for two days . After washing , cells were incubated in differentiation medium for another 5 to 7 days . T37i cells were cultured and differentiated as described above . At day 10 of differentiation , mature adipocytes were trypsinized and replated at 70% density . 2 hr post-plating , siRNAs against AMPKα1 and AMPKα2 , or non-target control ( Dharmacon , via Thermo Fisher ) complexed to Lipofectamine RNAimax reagent ( Life Technologies ) were added , according to the manufacturer’s protocol . 48 hr post-transfection , cells were washed and treated as indicated before harvesting for RNA analyses . BAT and WAT tissues were dissected from male C57Bl/6J mice and transferred to DMEM supplemented with 1% P/S and 1% BSA . Tissues of 4 mice were pooled and minced finely with scissors . Approximately 50 mg of tissue was transferred to a well of a 24-wells tissue culture plate and equilibrated for 1 hr in DMEM , supplemented with 1% P/S . After 1 hr , explants were treated as indicated before being harvested for RNA and protein analyses . Isoproterenol , AICAR , metformin , rosiglitazone , Wy14 , 643 , GW510516 , GW742 , dexamethasone , insulin , 3-isobutyl-1-1methylxanthine ( IBMX ) , actinomycin D were purchased from Sigma-Aldrich . A769662 was purchased from Abcam ( Cambridge , United Kingdom ) , phenformin hydrochloride was purchased from Cayman Chemicals ( via SanBio , Uden , the Netherlands ) . A fragment of 1517 bp containing intron 3 of the mouse Angptl4 gene was amplified from DNA of the mouse Angptl4 gene using the primers: Fwd 5’-TTGCTGTCATCTGGCAACTC-3’ and Rev 3’-CACCTAAAGCCTACCCCACA-5’ . The resulting fragment was gel-purified using QIAquick Gel Extraction Kit ( Qiagen ) and subjected to a second PCR to specifically amplify the three functional PPREs of Angptl4 and to introduce Xho1 and Kpn1 restriction sites . Amplification of the 565 bp fragment was done with the following primers: mouse Fwd 5'-atggtaccTTCACACCCTAAGGCTGC-3' , and mouse Rev 3'-atctcgagGGGGGAAGAGGAAGAAAA-5' . Following purification from gel , the fragment was subjected to restriction with Xho1 and Kpn1 enzymes and cloned into the Xho1 and Kpn1 sites of the pGL3 promoter vector ( Promega , Leiden , the Netherlands ) . Presence of the correct insert was validated by sequencing ( EZ-Seq , Macrogen , Amsterdam , the Netherlands ) using a RV3 primer: ‘5-CTAGCAAAATAGGCTGTCCC-3’ . mAngptl4 PPRE pGL3 reporter vector was transfected into the human hepatocellular cell line HepG2 ( ATCC , Manassas , USA ) in the presence or absence of pSG5 vectors expressing Pparγ and Rxr and pcDNA vector expressing HA-P300 ( kind gift of Eric Kalkhoven , University Medical Centre Utrecht , the Netherlands ) . A vector expressing renilla luciferase under a SV40 promoter was co-transfected with all samples to determine transfection efficiency . Transfections were performed using polyethylenimine ( PEI ) ( Polysciences Inc . via Qiagen ) . 16 hr post-transfection cells were incubated with rosiglitazone ( 5 μM ) and/or AICAR ( 1 mM ) for 9 hr . Firefly and renilla luciferase activity were determined using the Dual Glo Luciferase Assay system ( Promega ) , according to manufacturer’s instructions , on a Fluoroskan Ascent apparatus ( Thermo Scientific ) . Data are expressed as mean ± SEM , unless otherwise indicated . Differences were evaluated for statistical significance by student t-test or two-way ANOVA , followed by a post-hoc Tukey HSD test , and considered statistically significant when p<0 . 05 .	The body stores energy in the form of fat molecules . Most of these molecules are stored in white fat cells . Other fat cells , the so-called brown fat cells , consume fats and produce heat to maintain body temperature in cold conditions . The capacity of brown fat cells to consume fats has led researchers to investigate whether brown fat cells might be a key to combat obesity . When an organism is cold , fat is shuttled to the brown fat cells . An enzyme called lipoprotein lipase is involved in a process that allows these fat molecules to be taken up by brown fat cells . However , it was not clear exactly how this process works . A protein called Angiopoietin-like 4 ( ANGPTL4 ) inhibits the activity of lipoprotein lipase in white fat cells and is also found at high levels in brown fat cells . Here , Dijk et al . used genetic and biochemical approaches to study the role of ANGPTL4 in the fat cells of mice . The experiments show that when mice are exposed to cold , the levels of ANGPTL4 decrease in the brown fat cells . This allows the activity of lipoprotein lipase to increase so that these cells are able to take up more fat molecules . However , the opposite happens in white fat cells during cold exposure . The levels of ANGPTL4 increase , which decreases the activity of lipoprotein lipase in white fat cells to allow fat molecules to be shuttled specifically to the brown fat cells . Further experiments suggest that the opposite regulation of ANGPTL4 in brown and white fat cells could be due to a protein called AMPK . This protein is found at higher levels in brown fat cells than in white fat cells and is produced by brown fat cells during cold exposure . Taken together , Dijk et al . show that organs and cells work together to ensure that fat molecules are appropriately distributed to cells in need of energy , such as to brown fat cells during cold . How these findings could be used to stimulate fat consumption by brown fat cells in humans remains open for further investigation .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine" ]	2015	ANGPTL4 mediates shuttling of lipid fuel to brown adipose tissue during sustained cold exposure
Translational readthrough gives rise to low abundance proteins with C-terminal extensions beyond the stop codon . To identify functional translational readthrough , we estimated the readthrough propensity ( RTP ) of all stop codon contexts of the human genome by a new regression model in silico , identified a nucleotide consensus motif for high RTP by using this model , and analyzed all readthrough extensions in silico with a new predictor for peroxisomal targeting signal type 1 ( PTS1 ) . Lactate dehydrogenase B ( LDHB ) showed the highest combined RTP and PTS1 probability . Experimentally we show that at least 1 . 6% of the total cellular LDHB is targeted to the peroxisome by a conserved hidden PTS1 . The readthrough-extended lactate dehydrogenase subunit LDHBx can also co-import LDHA , the other LDH subunit , into peroxisomes . Peroxisomal LDH is conserved in mammals and likely contributes to redox equivalent regeneration in peroxisomes . Translation of genetic information encoded in mRNAs into proteins is carried out by ribosomes . When a stop codon enters the ribosomal A site , release factors bind the stop codon , hydrolyze the peptidyl-tRNA bond , and trigger the release of the polypeptide from the ribosome . If instead of release factor 1 ( eRF1 ) , a near-cognate aminoacyl-tRNA pairs with the stop codon in the ribosomal A site , the stop signal is suppressed . Such decoding of a stop codon as a sense codon is known as translational readthrough . As a consequence , translation continues to the next stop codon resulting in the synthesis of C-terminally extended proteins ( Baranov et al . , 2002; Namy et al . , 2004; Firth and Brierley , 2012 ) . Mutant tRNAs are classic stop codon suppressors , but termination also occurs with less than 100% efficiency in normal physiology . A number of cis-elements on the mRNA , typically 3′ of the stop codon together with trans-acting factors , are known to influence stop codon readthrough ( Firth et al . , 2011 ) . A case of translational readthrough dependent on RNA cis-elements has recently been found and termed programmed translational readthrough ( PTR ) ( Eswarappa et al . , 2014 ) . But it is also known that the stop codon itself and the nucleotides before and after the stop codon affect readthrough . The three stop codons differ in their tendency to be suppressed . In human , UAA is least and UGA is most likely to allow readthrough ( Beier and Grimm , 2001; Baranov et al . , 2002 ) . Studies also show that the nucleotide immediately downstream of the stop codon is biased and can strongly influence readthrough ( McCaughan et al . , 1995 ) . We here define translational readthrough that is entirely dependent on the stop codon and the nucleotides in its immediate vicinity as basal translational readthrough ( BTR ) . Thus BTR is independent of cis-acting elements and also differs from pharmacologically induced readthrough . Induction of readthrough , most prominently by aminoglycoside antibiotics , is an attractive strategy in the treatment of the large number of genetic disorders caused by premature stop codons ( Bidou et al . , 2012; Keeling et al . , 2014 ) . In viruses , readthrough optimizes the coding capacity of compact genomes ( Firth and Brierley , 2012 ) . In the yeast Saccharomyces cerevisiae , the eukaryotic release factor eRF3 can form prion-like polymers , which introduces a level of epigenetic regulation not found in other eukaryotes ( Tuite and Cox , 2003 ) . In fungi , translational readthrough extends cytosolic glycolytic enzymes by a cryptic peroxisomal targeting signal ( Freitag et al . , 2012 ) . In Drosophila , readthrough is known to affect between 200 and 300 proteins ( Jungreis et al . , 2011; Dunn et al . , 2013 ) , and in mammals readthrough has been described for more than 50 individual transcripts ( Geller and Rich , 1980; Chittum et al . , 1998; Yamaguchi et al . , 2012; Dunn et al . , 2013; Eswarappa et al . , 2014; Loughran et al . , 2014 ) . Ribosome profiling and phylogenetic approaches provide powerful methods for the systematic identification of readthrough in mammals ( Jungreis et al . , 2011; Dunn et al . , 2013; Eswarappa et al . , 2014; Loughran et al . , 2014 ) . We wanted to find a physiological role for translational readthrough in humans by identifying C-terminal extensions with targeting signals that would create a functional difference between the normal and the readthrough-extended form . To achieve this aim , we concentrated on proteins deriving from BTR . Based on experimental data , we assigned regression coefficients to all possible nucleotides in the stop codon context ( SCC ) and , using those regression coefficients , estimated the readthrough propensity ( RTP ) of all stop codons in the human genome or transcriptome . We were able to formally derive a new nucleotide consensus for high RTP from the regression coefficients of our model . Then we screened all predicted C-terminal extensions for peroxisomal targeting signals because peroxisomes import most of their matrix proteins through a short targeting signal ( PTS1 ) at the very C-terminus ( Smith and Aitchison , 2013 ) . We here show that lactate dehydrogenase B ( LDHB ) combines a very high translational readthrough with a hidden , yet functional and evolutionarily conserved , PTS1 . This peroxisomal isoform of LDH , containing the readthrough-extended LDHBx subunit , is likely to be involved in the regeneration of redox equivalents for peroxisomal β-oxidation . In order to develop a computational method to assess the RTP of all human SCCs that would allow the identification of genes with high BTR , we focused on SCCs comprising 15 nucleotides including and surrounding the stop codon ( nucleotides −6 to +9 , stop codon at positions 1 to 3 ) . In order to calculate linear regression between the SCCs and their experimental BTR values , we formalized SCCs using a binary vector that represented the stop context in a multi-dimensional vector space ( Figure 1A and Figure 1—figure supplement 1 ) . The three stop codons were condensed into one position , so that the binary vector required 51 dimensions , for the four possible nucleotides in the six positions before and after the stop codon , and for the three stop codons ( 12 × 4 + 3 ) . The vector was combined with experimentally accessible BTR frequencies . For the first approximation model ( LIN ) , we used 66 sequences derived from human nonsense mutations ( Floquet et al . , 2012 ) . The nucleotide sequences of these stop contexts show no bias with respect to RTP , because the contexts and the stop codons evolved independently , and therefore the context nucleotides are random in relation to the stop codon . We calculated a linear regression model for these SCCs and used only the experimental BTR values that had been measured in the absence of aminoglycosides . The model assigns regression coefficients to all possible nucleotides in the stop context ( Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 03640 . 003Figure 1 . Genome-wide in silico analysis of basal translational readthrough ( BTR ) in humans . ( A ) Schematic representation of the readthrough propensity ( RTP ) predictor algorithm . Binary vector representations of stop codon contexts and their experimental readthrough values are used to determine the coefficients of a linear regression model . For prediction of RTP for a given stop codon context , the position-specific regression coefficients are added up . ( B ) RTP for selected human transcripts . LIN denotes first-pass RTP calculations , LINiter iterative improvement of RTP scoring , and LINfs3 and LINfs5 the reduced models . The RTP of all human transcripts can be found in Dataset 1 ( Schueren et al . , 2014 ) . ( C ) Experimental readthrough by dual reporter assay in HeLa cells . Readthrough is expressed as luciferase per Venus signal . The red line marks the background readthrough level obtained from a construct containing two contiguous UAA stop codons separating the Venus and the hRluc . The aminoglycoside geneticin ( 100 µg/ml ) induces translational readthrough . SCC: stop codon context; hRluc: humanized Renilla luciferase . Error bars , SD . DOI: http://dx . doi . org/10 . 7554/eLife . 03640 . 00310 . 7554/eLife . 03640 . 004Figure 1—figure supplement 1 . Schematic representation of the readthrough propensity ( RTP ) prediction procedure . This scheme summarizes how regression coefficients were extracted from experimental basal readthrough ( BTR ) data . ( 1 ) Stop codon contexts ( SCC ) ( positions −6 to +9 , stop codon at position 1 to 3 ) with known experimental BTR values are formalized as binary vectors in 51-dimensional vector space . ( 2 ) The binary vector reserves four entries for the four possible bases in each position ( 4 × 12 ) and three for the stop codon . ( 3 ) In combination with their corresponding experimental readthrough values ( in % ) , they are used to determine the ( 4 ) coefficients of a linear regression model . ( 5 ) For RTP calculation for a given SCC , the position-specific regression coefficients are added up . The values used in the example are from the LINiter model . The algorithm is used to calculate the RTP of SCCs of 42 , 000 unique 3′ transcript termini listed in Dataset 1 ( Schueren et al . , 2014 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03640 . 00410 . 7554/eLife . 03640 . 005Figure 1—figure supplement 2 . Correlation of RTP and BTR in the LINiter model . Scatter plot indicating the correlation between readthrough propensity ( RTP ) and experimental basal translational readthrough ( BTR ) . RTP was obtained by leave-one-out cross-validation . Pearson correlation coefficient 0 . 34 ( p = 0 . 002 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03640 . 005 For a first round of whole-genome RTP prediction , we extracted the SCCs for each transcript from the Ensembl database and calculated RTP by adding up the regression coefficients of all relevant positions . An outline of this algorithm is shown in Figure 1A and in more detail in Figure 1—figure supplement 1 . A sortable list of LIN RTP values for all human transcripts is contained in Dataset 1 ( Schueren et al . , 2014 ) . To expand the data basis of the RTP algorithm and to obtain evidence that the algorithm indeed predicts BTR values , we selected candidate transcripts with high , intermediate , and low RTP and tested them using a dual reporter assay ( Figure 1B and Table 1 ) . For experimental analysis , SCCs spanning 10 nucleotides upstream and downstream of the stop codon were expressed with a 5′/N-terminal yellow fluorescent protein ( Venus ) and a 3′/C-terminal humanized Renilla luciferase ( hRluc ) tag . Stop suppression leads to the expression of hRluc , and Venus served as an internal expression control . Readthrough is expressed as luciferase activity per Venus fluorescence . This approach excludes introns and exon junction complexes and , due to the relatively short stretch of variable nucleotides between the reporters , also does not allow for extensive RNA structures that could modulate readthrough . Consequently , this form of the dual reporter assay focuses on the assessment of BTR not influenced by specific cis-elements . The additional candidates tested showed BTR between 0 . 10% ( ±0 . 006% ) and 2 . 91% ( ±0 . 15% ) relative to the 100% readthrough control expressing the Venus-hRluc fusion protein without an intervening stop codon region ( Figure 1C and Table 1 ) . The aminoglycoside antibiotic geneticin ( G418 ) increased readthrough by between 3 . 25 ( ±0 . 41 ) and 40 . 38 ( ±5 . 33 ) -fold ( Figure 1C ) . Geneticin could only increase the luciferase-per-Venus signal when a stop codon separated Venus and luciferase , indicating that our dual reporter assay faithfully reports readthrough . The finding that experimental readthrough could be increased by treatment with aminoglycosides also excludes alternative mechanisms such as RNA editing or splicing that might explain the relative increase of the luciferase over the Venus signal . The highest levels of induction can only be reached when basal readthrough is low , and , vice versa , a high BTR somewhat limits the maximum induction factor ( Figure 1C ) , suggesting that maximal BTR readthrough is limited to levels below 15% . 10 . 7554/eLife . 03640 . 006Table 1 . Additional experimental dual reporter readthrough data of stop codon context constructs used for the LINiter modelDOI: http://dx . doi . org/10 . 7554/eLife . 03640 . 006Gene symbolStop codon contextReadthrough ( % ) ( SD ) ZNF-574GATCAGTGGC TGA CTCTGCCCGA0 . 31 ( 0 . 020 ) LDHBAAAAGACCTG TGA CTAGTGAGCT1 . 55 ( 0 . 087 ) PPP1R3FATTCTCCCAA TAA AGCTTTACAG0 . 18 ( 0 . 009 ) LDHB [TGAT]AAAAGACCTG TGA TTAGTGAGCT0 . 17 ( 0 . 009 ) LDHB [TAA]AAAAGACCTG TAA CTAGTGAGCT0 . 20 ( 0 . 009 ) LDHB [TAAT]AAAAGACCTG TAA TTAGTGAGCT0 . 17 ( 0 . 009 ) LENG1CCTTACTCAC TGA CTCCTGAGGG0 . 26 ( 0 . 009 ) VASNGCCCTACATC TAA GCCAGAGAGA0 . 12 ( 0 . 004 ) MDH1TTCCTCTGCC TGA CTAGACAATG2 . 91 ( 0 . 147 ) PRDM10CACCAAACCA TGA CTTCCACCCT0 . 13 ( 0 . 005 ) FBXL20CATCATCCTA TGA CAATGGAGGT0 . 10 ( 0 . 006 ) THG1LAGCCAGGCTT TGA CGGAAGAGTC0 . 15 ( 0 . 006 ) EDEM3GGATGAGCTA TGA CTTGCTAAAC0 . 66 ( 0 . 027 ) EDN1AGCACATTGG TGA CAGACCTTCG0 . 25 ( 0 . 008 ) UBQLN1CCAGCCATCA TAG CAGCATTTCT0 . 13 ( 0 . 009 ) IRAK3CAAAAAAGAA TAA ATTCTACCAG0 . 10 ( 0 . 007 ) SLC3A1TACCTCGTGT TAG GCACCTTTAT0 . 18 ( 0 . 008 ) LEPRE1GGATGAGCTA TGA CAGCGTCCAG0 . 27 ( 0 . 010 ) Stop codon constructs expressing plus/minus 10 nucleotides were analyzed in HeLa cells . Next we added our candidate sequences and their experimentally determined readthrough levels to obtain an iterative and extended RTP model ( LINiter ) . Again , we applied this model to all human transcripts ( see Schueren et al . , 2014 for Dataset 1; model parameters are shown in Table 2 ) . We measured the correlation of RTP and experimental BTR by leave-one-out cross-validation during computation of the regression coefficients . For the LINiter model , we obtained a weak but significant Pearson correlation coefficient of 0 . 34 ( p = 0 . 002 ) ( Figure 1—figure supplement 2 ) . To determine the origin of the apparently strong non-linear contribution to RTP , we analyzed the regression coefficients of the LINiter model . Nucleotide positions associated with coefficients of large absolute value contribute most to RTP . The relative contribution of nucleotides within the SCC to the readthrough prediction is shown in Figure 2A . 10 . 7554/eLife . 03640 . 007Table 2 . Regression factors of the LINiter and LINfs3 modelsDOI: http://dx . doi . org/10 . 7554/eLife . 03640 . 007LINiter model ( stop codon context position −6 to +9 ) Base/position−6−5−4−3−2−14A−0 . 000410 . 00130−0 . 00028−0 . 00073−0 . 000710 . 00016−0 . 00037C−0 . 001050 . 001640 . 00075−0 . 000040 . 001330 . 001090 . 00375G0 . 00060−0 . 00077−0 . 000410 . 00193−0 . 000480 . 00043−0 . 00156U/T0 . 00200−0 . 001030 . 00108−0 . 000020 . 00100−0 . 00054−0 . 00067Base/position56789StopA−0 . 000680 . 00276−0 . 000200 . 00105−0 . 00081−0 . 00026TAAC−0 . 00097−0 . 00026−0 . 00062−0 . 000170 . 00148−0 . 00103TAGG−0 . 00008−0 . 000590 . 00245−0 . 000580 . 000140 . 00243TGAU/T0 . 00287−0 . 00076−0 . 000490 . 000840 . 00032LINfs3 model ( Stop and position +4 to +6 ) Base/position456StopA0 . 00006−0 . 000710 . 003060 . 00005TAAC0 . 00351−0 . 000560 . 00021−0 . 00052TAGG−0 . 001110 . 00010−0 . 000930 . 00229TGAU/T−0 . 000640 . 00299−0 . 00053These model weights are ‘raw’ , that is as obtained from the ridge regression procedure . For prediction of RTP , the weights associated with nucleotides within the stop codon context and the corresponding stop codon have to be added up . For calculation of our RTP score , we normalized the model weight vectors ( i . e . , the complete stack of weights ) to Euclidean unit sum which corresponds to a division of weights by 0 . 0088 ( LINiter ) and 0 . 0063 ( LINfs3 ) , respectively . Furthermore , the sequence feature vectors were normalized to Euclidean unit sum which corresponds to a division by the square root of the length ( 3 . 6 and 2 , respectively ) . As a shortcut to this , the sum of raw scores can be divided by 0 . 0317 and 0 . 0126 , respectively . 10 . 7554/eLife . 03640 . 008Figure 2 . Characterization of basal translational readthrough ( BTR ) : consensus and candidates . ( A ) Sequence logo plot of regression coefficients of stop codon contexts ( SCCs ) in the LINiter model . Character size corresponds to regression coefficients . The model treats stop codons as one nucleotide position . Filled/upside-down letters correspond to positive/negative coefficients , respectively . ( B ) Consensus motif for high readthrough propensity ( RTP ) derived from the predictive model . The stop codon together with the nucleotide triplet following the stop codon provides the best predictor for RTP . The consensus was derived by feature selection: starting from LINiter , positions with the least contribution to prediction were successively eliminated as indicated by the gray arrow . Nucleotide positions on the x-axis mark the removed positions upon transition to a reduced model . LINfs3 ( UGA CUA , stop codon underlined ) represents the global minimum of regression error ( filled circle ) . The model LINfs5 , corresponding to a local minimum , additionally encompasses positions +7 and −6 , indicating that these positions could also contribute to high BTR . ( C ) BTR determination of candidates from the genome-wide in silico screen . Dual reporter assays with Venus and humanized Renilla luciferase containing SCCs from AQP4 ( UGA CUA G ) , SYTL2 ( UGA CUA G ) , CACNA2D4 ( UGA CUA T ) , and DHX38 ( UGA CUU G ) . AQP4 , SYTL2 , and CACNA2D4 reveal high BTR in all tissues tested . HT1080 , human fibrosarcoma cell line; U373 , glioblastoma cell line . HEK , human embryonic kidney cells . Error bars , SD . DOI: http://dx . doi . org/10 . 7554/eLife . 03640 . 00810 . 7554/eLife . 03640 . 009Figure 2—figure supplement 1 . Correlation of RTP and BTR in the LINfs3 model . Scatter plot indicating the correlation between readthrough propensity ( RTP ) and experimental basal translational readthrough ( BTR ) . RTP was obtained by leave-one-out cross-validation . Pearson correlation coefficient 0 . 41 ( p = 0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03640 . 00910 . 7554/eLife . 03640 . 010Figure 2—figure supplement 2 . Translational readthrough in humans . Venn diagram indicating experimentally verified human genes and stop codon contexts ( SCCs ) associated with above-average translational readthrough . Genes were identified by ribosome profiling ( Dunn et al . , 2013 ) , by phylogenetic approaches ( Jungreis et al . , 2011; Loughran et al . , 2014 ) , and by in silico profiling ( this study ) . Gene products marked in boldface ( black ) correspond to sequences carrying the consensus motif UGA CUA ( G ) identified in this study and by Loughran et al . , ( 2014 ) . The human genome contains 144 [30] transcripts with the high-RTP motifs UGA CUA [G] . Different experimental strategies lead to the identification of genes with high physiological readthrough rates , but the molecular mechanisms underlying readthrough are likely to vary . DOI: http://dx . doi . org/10 . 7554/eLife . 03640 . 010 The sequence-logo representation of regression factors in Figure 2A indicates that the three or four nucleotides following the stop codon contribute to readthrough . The quantitative manner in which we derived LINiter values allowed us to rationally derive a nucleotide motif permitting high readthrough in humans . We identified the nucleotide positions with the strongest influence on BTR in humans by feature selection , that is by successively eliminating those positions that contribute least to the prediction ( Figure 2B ) . One by one the nucleotide positions with the smallest sum of squared regression coefficients were removed from the model . We find that two reduced models improve the prediction . Models with either five or three relevant context positions in addition to the stop codon correspond to the local and global residual error minimum , respectively . LINfs5 comprises nucleotide position −6 , the stop codon , and positions +4 to +7 , and LINfs3 comprises only the stop codon and positions +4 to +6 , that is the codon following the stop ( Figure 2B ) . The results of this analysis indicate that in humans the stop codon and the three nucleotides immediately downstream of the stop codon have the largest influence on BTR ( LINfs3 ) . The corresponding consensus is UGA CUA ( stop codon underlined ) . Possibly also the nucleotides at positions +7 ( the fourth position after the stop ) and −6 contribute to BTR . The RTP-BTR correlation associated with LINfs3 was 0 . 41 ( p = 0 . 0001 ) ( Figure 2—figure supplement 1 ) . To test if the LINfs3 consensus indeed confers high BTR , we analyzed four additional candidate SCCs . Three high-RTP SCCs were derived from AQP4 , SYTL2 , and CACNA2D4 , and DHX38 was used as a control with a low RTP . AQP4 , SYTL2 , and CACNA2D4 conform with the LINfs3 consensus , whereas DHX38 does not . AQP4 , SYTL2 , and CACNA2D4 showed 2 . 29% ( ±0 . 09% ) , 0 . 99% ( ±0 . 06% ) , and 0 . 61% ( ±0 . 02% ) readthrough in HeLa cells , whereas for DHX38 readthrough was only 0 . 27% ( ±0 . 04% ) ( Figure 2C ) , confirming that LINfs3 SCC indeed allows a very high rate of stop suppression . Next we wanted to test if these conclusions obtained in HeLa cells can be extended to other cell types . We therefore performed dual reporter experiments using the HT1080 fibrosarcoma cell line , the human embryonic kidney cell line ( HEK ) , and the U373 cell line . In all these experiments , the relative distribution of BTR values remained the same , with AQP4 showing the highest and DHX38 the lowest BTR ( Figure 2C ) . The finding that readthrough is lower in CACNA2D4 than in AQP4 and SYTL2 can also be taken as evidence that SCC position +7 ( fourth after the stop ) makes a contribution . Taken together , these experiments show that BTR is indeed a property of the respective SCC , and that readthrough may be differently regulated in different tissues . The linear approximation underlying the LINiter and the LINfs3 models led to the identification of the UGA CUA ( LINfs3 ) consensus conferring high BTR . A partially overlapping set of genes with this consensus was recently tested ( Loughran et al . , 2014 ) . An overview of all experimentally confirmed cases of translational readthrough shown in Figure 2—figure supplement 2 reveals that ribosome profiling , phylogenetic approaches , and RTP screening are complementary approaches . For example , only one of the 42 readthrough genes found by ribosome profiling in foreskin fibroblasts ( Dunn et al . , 2013 ) contains the UGA CUA consensus . The widely varying levels and sequence requirements for efficient stop codon suppression suggest that multiple molecular mechanisms can cause readthrough in mammals . The genome-wide in silico analysis of RTP provides the basis for the identification of the physiological functions of a readthrough protein . We have therefore screened the extensions for possible elements that could confer functional differences between the normal and the extended form of the protein . We screened the extensions for possible transmembrane domains ( Krogh et al . , 2001 ) , for prenylation sites ( Zhang and Casey , 1996 ) , for endoplasmic retention signals ( Zerangue et al . , 2001; Stornaiuolo et al . , 2003 ) , and for glycosylation sites ( Zielinska et al . , 2010; Schwarz and Aebi , 2011 ) . To identify genes with a high BTR and a readthrough extension conferring a biological function , we decided to focus on the detection of proteins carrying a hidden peroxisomal targeting signal type 1 ( PTS1 ) in the extension . This targeting mechanism had been shown to divert a small fraction of cytosolic glycolytic proteins to peroxisomes in fungi ( Freitag et al . , 2012 ) . PTS1 cover more than 90% of the targeting motifs of peroxisomal matrix proteins . The alternative PTS2 is found in only very few matrix proteins , and has even been lost in some organisms ( Lanyon-Hogg et al . , 2010 ) . PTS1 is localized at the very C-terminus of a substrate protein . However , the quintessential PTS1 , Ser-Lys-Leu ( SKL ) , is neither necessary nor sufficient to support matrix protein import into peroxisomes . Variations exist , and amino acids upstream of the terminal tripeptide also contribute to targeting ( Brocard and Hartig , 2006 ) . Moreover , PTS1 does not confer a binary decision ( to import or not to import ) , but is likely to determine an equilibrium between cytosolic and peroxisomal localization . This is best exemplified by the peroxisomal marker protein catalase , a considerable amount of which is not imported into peroxisomes due to an inherently weak PTS1 which is associated with low affinity to the cytosolic PTS1-receptor PEX5 ( Maynard et al . , 2004 ) . We took advantage of these scalable properties of PTS1 and adapted to human PTS1 a prediction algorithm that we had previously developed for plants ( Lingner et al . , 2011 ) . This machine learning-based method has been shown to accurately predict proteins with canonical and non-canonical PTS1 peptides and provides evidence for peroxisome targeting in terms of a posterior probability ( Lingner et al . , 2011 ) . To program the human PTS1 prediction algorithm , we conducted orthology searches on 24 known human PTS1 sequences in metazoa using BLAST against protein and EST databases . The resulting dataset and several thousand metazoan sequences without peroxisomal association were used as positive and negative examples in a discriminative machine learning setup . Here , the sequences were represented by binary vectors encoding the presence or absence of up to 15 C-terminal amino acids . Models were trained and validated using regularized least squares classifiers ( RLSC ) and fivefold cross-validation . A more detailed description of the human PTS1 scoring can be found in the ‘Materials and methods’ section . We calculated the PTS1 posterior probabilities of all predicted C-terminal readthrough extensions derived from the human transcriptome ( see Schueren et al . , 2014 for Dataset 1 ) . Based on the assumption that a protein is more likely to target to peroxisomes by a cryptic PTS1 when the RTP and the extension's PTS1 scores are high , we used the product of RTP LINiter scores and PTS1 posterior probabilities as a predictor of functional peroxisomal targeting by a hidden PTS1 in the extension ( see Schueren et al . , 2014 for Dataset 1 ) . To avoid negative product scores , we scaled RTP between 0 and 1 before multiplication ( now designated RTP+ ) . We identified LDHB , one of the two human lactate dehydrogenase ( LDH ) subunits , at the top ( position 1 of 42 , 069 entries ) of our sorted list of combined RTP+ and PTS1 scores ( see Schueren et al . , 2014 for Dataset 1 ) . The distribution of RTP+ × PTS1 product scores over all human transcripts indicates that other candidates must have considerably lower RTPs and/or targeting efficiencies , because the score drops by 50% over the first 40 of 42 , 069 transcripts ( Figure 3A ) . 10 . 7554/eLife . 03640 . 011Figure 3 . LDHB is extended by translational readthrough . ( A ) Genomic distribution of RTP+ × PTS1 product scores . Product scores are 0 for rank 5015 to 42069 . Green cross: 50% of maximum score . LDHB has the highest product score , exceeding rank 2 by 24% . RTP+ denotes positively scaled LINiter values . ( B and C ) Venus/hRluc dual reporter assay with LDHB wild-type and mutant stop codon contexts . Error bars , SD . ( B ) Wild-type LDHB stop context shows high basal translational readthrough ( BTR ) . Mutational analysis of the LINfs3 consensus of LDHB . Replacement of the stop codon and mutations in positions +4 to +6 reduce readthrough . ( C ) LDHB readthrough induction by the aminoglycoside geneticin . ( D ) Full-length LDHB is extended by readthrough . Western blot of dual tag assay with LDHBx with N-terminal HA- and C-terminal Myc-tag . Molecular mass marker in kDa . DOI: http://dx . doi . org/10 . 7554/eLife . 03640 . 01110 . 7554/eLife . 03640 . 012Figure 3—figure supplement 1 . The LDHB stop context favors readthrough ( Western blot ) . HeLa cells transfected with Venus/hRluc dual reporter constructs were analyzed by Western blot . Wild-type LDHB stop context ( UGA CUA ( stop underlined ) ) allows stop codon readthrough . Mutation of the stop codon ( UAA CUA ) and the downstream bases ( UAA UUA ) reduces readthrough . Geneticin treatment ( 100 µg/ml ) induces translational readthrough in all contexts . Molecular mass marker in kDa . DOI: http://dx . doi . org/10 . 7554/eLife . 03640 . 012 To experimentally confirm high BTR , we expressed the human LDHB SCC in the Venus/hRluc dual reporter assay . Readthrough was 1 . 55% ( ±0 . 09% ) and mutation of the stop codon and/or the consecutive nucleotide strongly suppressed readthrough ( Figure 3B and Figure 3—figure supplement 1 ) . Treatment with geneticin increased readthrough to 4 . 38% ( ±0 . 42% ) ( compare with induction factors in Figure 3C ) . To establish that the full-length protein is extended by stop suppression , LDHB including the extension ( designated LDHBx for ‘extended’ ) and mutants were expressed with N-terminal HA- and C-terminal Myc-tags and analyzed by Western blotting . Full-length LDHB showed aminoglycoside-inducible readthrough , and the loss of readthrough upon exchange of the stop codon or the nucleotide following the stop codon confirms the special function of the LDHB SCC in stimulating translational readthrough ( Figure 3D ) . The identification of LDHB as virtually the only human protein with a high combined readthrough and peroxisomal targeting probability is surprising , because a peroxisomal readthrough-extended LDHBx entails at least one new LDH isoform . On the other hand , LDH activity and isoforms inside peroxisomes were known for more than four decades ( McGroarty et al . , 1974; Osmundsen , 1982; Völkl and Fahimi , 1985; Baumgart et al . , 1996; McClelland et al . , 2003; Gronemeyer et al . , 2013 ) . In the apparent absence of known targeting signals , however , it has not been possible to explain how the protein can enter the peroxisome . Therefore we conducted an investigation to determine whether the extended human LDHBx protein and the predicted PTS1 therein lead to peroxisomal localization . We expressed LDHBx as a fusion protein with an N-terminal enhanced yellow fluorescent protein ( YFP ) and co-labeled cells by immunofluorescence with the peroxisomal marker PEX14 , a peroxisomal membrane protein . YFP-LDHB showed the expected cytosolic localization ( Figures 4A and 5A ) . We hypothesized that a large excess of cytosolic YFP-LDHB masks the peroxisomal localization of LDHBx . To remove cytosolic YFP-LDHB , we permeabilized cells by digitonin before fixation and washed out the cytosol using phosphate-buffered saline ( PBS ) . In agreement with peroxisomal targeting through the cryptic PTS1 , LDHBx is found localized in peroxisomes after removal of the cytosol ( Figures 4B and 5B ) . In control experiments , we show complete removal of cytosolically expressed YFP by cytosol wash-out ( Figure 5—figure supplement 1 ) and peroxisomal localization of a YFP variant fused to PTS1 of the peroxisomal matrix protein ACOX3 ( Figure 5—figure supplement 2 ) . To confirm that LDHB targeting to peroxisomes is dependent on the putative PTS1 in the readthrough extension , we changed the SRL terminus ( PTS1 probability 94 . 3% ) to SSI ( 0 . 002% ) and to SR ( ΔL , 0 . 00001% ) . These mutations blocked YFP-LDHBx targeting to the peroxisome ( Figures 4C and 5C–F ) . Remarkably , exchange of the leaky UGA stop with the tighter UAA reduced peroxisomal localization of YFP-LDHB ( Figure 6A , B ) . Our results show that the high-RTP SCCs as well as the PTS1 in the extension after the stop codon are needed for peroxisome targeting . The extension must be accessible to ribosomal translation and contain a functional PTS1 . It is known that PTS1-dependent targeting guides proteins into peroxisomes and not only to the membrane . The dependence of LDH targeting on the hidden PTS1 and on the nature of the stop codon thus confirms that the protein is indeed inside the peroxisome . As expected , replacing the stop codon by tryptophan-encoding UGG renders LDHBx entirely dependent on the PTS1 ( Figure 6C , D ) . 10 . 7554/eLife . 03640 . 013Figure 4 . LDHBx targets to the peroxisome by translational readthrough and a hidden peroxisomal targeting signal type 1 ( PTS1 ) in the 3′ extension . ( A–C ) Direct fluorescence microscopy of transfected HeLa cells . Immunofluorescence with the peroxisome marker anti-PEX14 ( red ) . ( A ) YFP-LDHB ( green ) mainly localizes to the cytosol . The strong fluorescence signal in the cytosol prevents detection of LDHB in other cellular compartments . ( B ) Upon plasma membrane permeabilization and removal of cytosol ( -CYT ) , a small fraction of LDHB remains co-localized with the peroxisome marker . ( C ) Peroxisomal targeting of LDHB is dependent on the cryptic PTS1 Ser-Arg-Leu ( SRL ) in the extension . Deletion of the L in SRL blocks import into peroxisomes . ( D and E ) Endogenous LDHB is localized to peroxisomes in untransfected wild-type cells . Immunofluorescence with anti-LDHB ( green ) and anti-PEX14 ( red ) antibodies . ( D ) Endogenous LDHB is cytosolic . ( E ) Removal of cytosol ( -CYT ) reveals co-localization with PEX14 . Bar 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03640 . 01310 . 7554/eLife . 03640 . 014Figure 5 . Peroxisome targeting of LDHBx is dependent on a hidden peroxisomal targeting signal in the readthrough extension . Combined direct fluorescence and immunofluorescence in HeLa cells . ( A ) YFP-LDHBx expression: LDHBx is mainly cytosolic . ( B ) LDHBx targets to the peroxisome . Cells were permeabilized with digitonin , and cytosol was removed by washing with phosphate-buffered saline . ( C–F ) Mutation of the cryptic PTS1 in the extension blocks peroxisomal targeting of LDHBx . ( C and D ) Deletion of the amino acid L of the SRL in the PTS1 readthrough extension gives a wild-type cytosolic localization of LDHB and blocks import into the peroxisome completely . ( E and F ) Similarly , the SRL-to-SSI substitution does not interfere with cytosolic expression of the LDHB but completely blocks peroxisomal localization of LDHBx[SSI] . Bar 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03640 . 01410 . 7554/eLife . 03640 . 015Figure 5—figure supplement 1 . Permeabilization by digitonin allows complete removal of cytosol . Combined direct fluorescence with anti-PEX14 immunofluorescence . HeLa cells were transfected with the empty vector expressing YFP in the cytosol . ( A ) Cytosolic expression of YFP . ( B ) Complete removal of cytosolic after cell permeabilization and washing with phosphate-buffered saline . Bar 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03640 . 01510 . 7554/eLife . 03640 . 016Figure 5—figure supplement 2 . Cell permeabilization and removal of cytosol maintains peroxisomal integrity and co-localization of peroxisome marker ( positive control ) . Combined direct fluorescence with anti-PEX14 immunofluorescence in HeLa cells . ( A and B ) Cells were transfected with a construct expressing the PTS1 of ACOX3 fused to the C-terminus of the YFP variant Venus . ( B ) Co-localization of PTS1 and PEX14 after removal of cytosol following cell permeabilization . Bar 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03640 . 01610 . 7554/eLife . 03640 . 017Figure 6 . Peroxisome targeting of LDHBx is dependent on the stop codon . Combined direct fluorescence and immunofluorescence in HeLa cells . ( A and B ) Exchange of UGA stop codon with the tighter stop UAA ( YFP-LDHBx[TAA] ) reduces peroxisomal localization of LDHB . ( C and D ) When UGA is replaced by tryptophan-coding UGG ( LDHBx[TGG] ) , a larger proportion of LDHB is targeted to the peroxisome , and peroxisome localization becomes obvious without removal of the cytosol . ( B , D ) Cytosol was removed after cell permeabilization with digitonin . Bar 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03640 . 017 To obtain more direct evidence for the readthrough-dependent low abundance targeting of human LDHB to peroxisomes , we analyzed untransfected wild-type cells by immunofluorescence with anti-LDHB and anti-PEX14 antibodies . LDHB appears distributed in the cytosol ( Figure 4D ) . After cytosol depletion , however , the remaining LDHB signal is mainly peroxisomal ( Figure 4E ) . A small portion of LDHB may localize to other cellular locations protected against cytosol removal . We confirmed these results in human skin fibroblasts , COS-7 cells ( monkey kidney fibroblast line ) , the human glioblastoma cell line U118 , and freshly prepared rat cardiomyocytes ( Figure 7 ) . Our data are in agreement with readthrough-dependent targeting of about 1 . 6% of the LDHB to peroxisomes mediated by the cryptic PTS1 in the extension . Remarkably , treatment of untransfected wild-type HeLa cells with geneticin increased LDHBx levels in the peroxisome ( induction factor 1 . 89 , n = 28 , t test p < 0 . 0001 ) suggesting elevated peroxisomal LDHBx levels as a general pharmacological consequence of aminoglycoside treatment . 10 . 7554/eLife . 03640 . 018Figure 7 . Endogenous LDHB is localized to peroxisomes in wild-type cells . Immunofluorescence in wild-type cultured cells ( A–F ) or freshly prepared ( G and H ) cells with antibodies recognizing LDHB ( secondary antibody Alexa488-coupled ) and the peroxisome marker PEX14 ( secondary antibody Cy3-coupled ) . ( A and B ) COS-7 cells , ( C and D ) human skin fibroblasts , ( E and F ) U118 glioblastome , and ( G and H ) primary rat cardiomyocytes . ( B , D , F , H ) Cytosol was removed after permeabilization with digitonin ( -CYT ) . Bar 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03640 . 018 Next we wanted to test if there is evidence for differential regulation of translational readthrough of LDHB in different cell types . We expressed LDHB and mutant dual reporter constructs in COS-7 cells , U118 cells , and HEK cells . Readthrough of LDHB ranged between 1 . 55% ( ±0 . 09% ) in HEK and HeLa and 1 . 88% ( ±0 . 14% ) in COS-7 . Surprisingly , in U118 cells LDHB readthrough is increased to 5 . 09% ( ±1 . 03% ) ( Figure 8 ) . Geneticin induced readthrough by factors ranging between 1 . 32 ( ±0 . 09 ) and 2 . 82 ( ±0 . 27 ) ( Figure 8 ) . LDHB stop suppression is thus not restricted to special tissues , and may be differently regulated in different cell types . 10 . 7554/eLife . 03640 . 019Figure 8 . Evidence for regulation of readthrough . LDHB stop codon readthrough in various mammalian cell types . COS-7 , HEK , and U118 cells were transfected with LDHB and mutant dual reporter constructs and analyzed by Venus fluorescence and luciferase assays . Readthrough is expressed as hRLuc/Venus signal . Readthrough is induced by 100 µg/ml geneticin . DOI: http://dx . doi . org/10 . 7554/eLife . 03640 . 019 Analysis of animal LDHB orthologs in vertebrates shows that PTS1 in the extension is exclusively and strictly conserved in mammals , supporting the notion of a functional extension in these proteins and an evolutionarily conserved targeting of LDHBx to peroxisomes in mammals ( Figure 9 ) . 10 . 7554/eLife . 03640 . 020Figure 9 . LDHBx extensions including hidden PTS1 are strictly conserved in mammals . Alignments of LDHBx termini from mammals and non-mammalian vertebrates . PTS1 extension is boxed . The conserved readthrough PTS1 extension is found exclusively in mammals and marks the mammalian–non-mammalian border in vertebrates . Alligator mi . : Alligator mississippiensis; Canis lupus famil . : Canis lupus familiaris; Meleagris gall . : Meleagris gallopavo; Oryctolagus cun . : Oryctolagus cuniculus; Taeniopygia gut . : Taeniopygia guttata . DOI: http://dx . doi . org/10 . 7554/eLife . 03640 . 020 LDHB together with lactate dehydrogenase A ( LDHA ) can form five tetrameric LDH isoforms , of which two are homotetramers and three are heterotetramers ( Boyer et al . , 1963; Markert , 1963 ) , and peroxisomes have the unusual ability to import folded and even oligomeric proteins ( McNew and Goodman , 1996; Lanyon-Hogg et al . , 2010 ) . We therefore wanted to test if peroxisomal LDHBx piggy-backs LDHA into peroxisomes . For this purpose we adapted a two-hybrid assay previously used to analyze co-import of subunits of the dimeric peroxisomal hydrolase Lpx1 in a heterologous system ( Thoms et al . , 2011 ) . When LDHA was expressed as a fusion protein with N-terminal YFP without co-expression of any form of LDHB , the protein localized to the cytosol as expected ( Figure 10A ) . However , when we co-expressed YFP-LDHA with CFP-LDHBx[TGG] , that is cyan fluorescent protein ( CFP ) fused to the readthrough form of LDHB , we found YFP-LDHA in peroxisomes ( Figure 10B ) . This experiment shows that the readthrough form of LDHB , LDHBx , can interact with LDHA , and that LDHBx is capable of carrying LDHA into the peroxisome . To show that co-import of LDHA is dependent on the hidden targeting signal in LDHBx , we mutated the targeting signal to SSI , or we deleted the terminal leucine . Either LDHBx PTS1 mutation blocked co-import of LDHA ( Figure 10—figure supplement 1 ) . The peroxisome is thus accessible to all four new LDH isoforms containing LDHBx . To support our data on LDHBx-LDHA co-import , we drew a structural model of the LDH-1 tetramer , the fundamental all-B isoform of LDH ( Figure 10—figure supplement 2 ) . The C-terminal amino acid leucine is extended by three amino acids not resolved in the structure , and , in LDHBx , by an additional seven amino acids . The model shows that this extension protrudes from the tetramer and is located distal to the protomer-interaction site , confirming that oligomerization is not hampered by the extension . The protruding LDHBx extension carrying the PTS1 is also accessible on the tetramer surface for PEX5 binding and import into the peroxisome . 10 . 7554/eLife . 03640 . 021Figure 10 . Piggy-back co-import of LDHA by LDHBx into peroxisomes . Direct fluorescence of YFP-labeled LDHA ( green ) in the absence or presence of CFP-labeled LDHBx[TGG] ( red ) combined with immunofluorescence with a peroxisome marker ( blue ) . ( A ) YFP-LDHA localization is mainly in the cytosol when expressed in the absence of LDHBx . ( B ) LDHA is imported into peroxisomes when co-expressed with LDHBx[TGG] . Cytosol was removed after permeabilization with digitonin . Bar 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03640 . 02110 . 7554/eLife . 03640 . 022Figure 10—figure supplement 1 . Mutation of the cryptic targeting signal SRL in LDHBx blocks co-import of LDHA into peroxisomes . PTS1 mutated by deletion of the leucine ( ΔL ) or substitution of RL by SI ( SSI ) blocks co-import of LDHA into peroxisomes . ( A and B ) Co-expression of YFP-LDHA with CFP-LDHBx[TGG , ΔL] . ( C and D ) Co-expression of YFP-LDHA with CFP-LDHBx[TGG , SSI] . ( B and D ) Cytosol was removed after permeabilization with digitonin . Bar 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03640 . 02210 . 7554/eLife . 03640 . 023Figure 10—figure supplement 2 . Tetrameric lactate dehydrogenase ( space fill model ) from human heart ( LDH-1 , all-B isoform ) . The individual subunits are shown in different colors , and the last resolved amino acid at the C-terminus ( Leu in position 331 ) of all subunits is shown in yellow . This structural model shows that the termini protrude from the compact tetramer . The readthrough LDH contains at least one subunit LDHBx that is extended by seven amino acids containing the PTS1 . For import into the peroxisome , at least one C-terminus has to bind to the soluble PTS1 receptor PEX5 . The structure also shows that the added PTS1 is unlikely to block oligomerization of the protein because the C-termini are far away from the interaction surface of the protomers . The PTS1 has to be protruding from the compact oligomer in an unstructured manner to be buried in the TPR domain pocket of PEX5 . Therefore , it is also unlikely that the PTS1 extensions generated by readthrough fold back onto the protein to induce a conformational change that would interfere with the subunit interaction . The structural model is derived from structure 1IOZ ( Read et al . , 2001 ) in the International Protein Database ( www . pdb . org ) and was rendered using Jmol , an open-source Java viewer for chemical structures in 3D ( http://www . jmol . org/ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03640 . 023 LDH is an enzyme with several isoforms , which has also been instrumental in devising the enzyme isoform concept per se . The identification of the classic muscle and heart subunits LDH-M ( LDHA ) and LDH-H ( LDHB ) in the late 1950s was followed by the identification of a testes-specific LDHA variant , LDHC ( Boyer et al . , 1963; Goldberg et al . , 2009 ) . Now we find that readthrough-extended LDHBx is encoded by the well-known LDHB gene by translational stop suppression and can give rise to new isoforms . Peroxisomal LDH is a novel isoform of LDH containing at least one readthrough-extended LDHBx subunit . LDHB readthrough and readthrough-dependent peroxisomal localization are evident in various human cell types , suggesting that the LDHBx subunit is expressed and localized to peroxisomes in all tissues that express LDHB . LDHBx exemplifies a new mechanism of post-transcriptional diversification of the genome's coding potential in mammals . The 1 . 6% LDHBx stop codon readthrough that we find in our experiments corresponds to the 1 . 5–2% LDH activity found in association with peroxisomes ( McGroarty et al . , 1974; Osmundsen , 1982; Baumgart et al . , 1996 ) , suggesting that cellular suppression of the stop codon is the only pathway for LDHB into peroxisomes . Assuming that peroxisomes fill approximately 1–2% of the cell volume , translational readthrough ensures almost equal concentration of LDH in cytosol and in peroxisomes . Fatty acid β-oxidation reactions are the hallmark of peroxisomes in most cell types and organisms . In mammalian peroxisomes , β-oxidation is involved in the degradation of very long chain fatty acids ( VLCFA ) and biogenetic reactions such as the synthesis of bile acids ( Lodhi and Semenkovich , 2014 ) . Therefore patients with peroxisomal disorders accumulate VLCFA and bile acid intermediates ( Braverman et al . , 2013 ) . During fatty acid oxidation and other peroxisomal processes , nicotinamide adenine dinucleotide ( NAD+ ) is reduced to NADH . However , the pathway of NAD+ regeneration inside peroxisomes is not clear ( Kunze and Hartig , 2013 ) . For efficient β-oxidation to occur , it is necessary that a redox shuttle system exists for NAD+ regeneration , because peroxisomes are impermeable to NAD+/NADH ( Visser et al . , 2007 ) . The identification of LDH inside the peroxisome suggested the existence of a lactate/pyruvate shuttle involved in the regeneration of redox equivalents ( Baumgart et al . , 1996; McClelland et al . , 2003; Gladden , 2004 ) . In the absence of a peroxisomal targeting signal , however , peroxisomal LDH was not universally accepted by researchers . Lactate/pyruvate shuttling could either occur directly through the peroxisomal membrane ( Visser et al . , 2007 ) or make use of monocarboxylate shuttles in the peroxisomal membrane ( McClelland et al . , 2003 ) . Generally , functional LDHBx targeting to peroxisomes highlights the role of intracellular lactate shuttle mechanisms ( Brooks , 2009 ) . In liver peroxisomes , pyruvate production is catalyzed by alanine-glyoxylate aminotransferase , an important enzyme in glyoxylate detoxification . Glyoxylate , however , is itself a substrate of LDH ( Salido et al . , 2012 ) . Therefore , peroxisomal LDH may also be involved in peroxisomal glyoxylate metabolism . Peroxisomal LDH is not the first glycolytic enzyme found in peroxisomes . Trypanosomes have sequestered the full set of glycolytic enzymes in specialized peroxisomes called glycosomes ( Gualdrón-López et al . , 2012 ) . And recently , in fungi , part of the glycolytic pathway upstream of pyruvate including glyceraldehyde-3-phosphate dehydrogenase and 3-phosphoglycerate kinase , was shown to be localized to peroxisomes by alternative splicing and/or translational readthrough ( Freitag et al . , 2012 ) . It is compelling that fungi as well as mammals use stop codon suppression to localize a small fraction of glycolytic enzymes to peroxisomes . We hypothesize that both translational readthrough as well as PTS1 evolve easily , and so can divert a low and steady amount of these enzymes to peroxisomes . A small fraction of cytosolic LDHB is imported into peroxisomes . This fraction is likely to be constant with respect to the overall LDHB expression levels in given tissue . We speculate that the peroxisomal LDHB shunt helps to coordinate redox processes between the cytosol and the peroxisome . Importantly , our study reveals a new pharmacological effect of readthrough-inducing drugs such as the commonly prescribed aminoglycosides , as they will increase LDHB readthrough and peroxisome import of LDHBx . It is not known at the moment whether translational readthrough is regulated in humans . The very high readthrough of approximately 5% in a glioblastoma cell line suggests that readthrough is differentially regulated in different tissues . Future experiments will show if the increased LDHB readthrough we find in this cell line are a cancer-associated dysregulation linked to the Warburg effect ( Hsu and Sabatini , 2008 ) , or if it just matches a higher abundance of peroxisomes in these cells to ensure an equal concentration of LDH in cytosol and peroxisomes in these cells as suggested above . It is also possible that glial cells generally have a higher demand for peroxisomal LDH that could be involved in neuronal/glial lactate metabolism . The first mammalian readthrough proteins were identified by chance ( Geller and Rich , 1980; Chittum et al . , 1998; Yamaguchi et al . , 2012 ) . Recently , two powerful and complementary methods have been employed in the genome-wide identification of readthrough-extended proteins . Ribosome profiling can recognize translating ribosomes in 3′UTRs and thereby identify readthrough and other recoding events outside known coding regions ( Ingolia et al . , 2011; Dunn et al . , 2013 ) . Phylogenetic approaches such as those implemented in PhyloCSF ( Lin et al . , 2011 ) evaluate the coding potential of sequences before and after the stop codon to help predict readthrough and are particularly powerful when genome sequences from closely related species are available ( Jungreis et al . , 2011; Loughran et al . , 2014 ) . Ribosome profiling , however , depends on gene expression , and can identify readthrough events only when the cell type in question is actually analyzed . Ribosome profiling may also fail to identify short readthrough extensions . Phylogenetic approaches , on the other hand , may miss readthrough when it is not conserved in the given dataset or when sufficiently dense datasets are not available , or when the extensions are too short to provide a basis for phylogenetic comparison . Our approach to systems-level identification of translational readthrough is based on the formalization of SCCs and a linear regression model with experimental readthrough values . The majority of the input sequences have been derived from patient nonsense mutations . In consequence , these sequences are biased neither by preselection by any pre-determined RTP or experimental readthrough levels , or by the SCCs , because the contexts did not evolve together with the respective stop codons . The algorithm we develop in this paper is limited to six nucleotide positions before and after the stop codon . This approach excludes the identification of extended RNA secondary structures involved in PTR and other recoding events ( Baranov et al . , 2002; Firth and Brierley , 2012; Eswarappa et al . , 2014; Loughran et al . , 2014 ) . The identification of the LINfs3 consensus and the human genes associated with this consensus justifies this approach . The LINfs3 motif , derived by feature selection , encompasses the stop codon and the first codon after the stop: UGA CUA . Our analysis suggests that positions +7 and −6 might also further contribute to readthrough . We have tested five of the 144 candidates in the genome with the UGA CUA motif and confirmed their high BTR . Highest BTR appears to correlate with a G in position +7 ( UGA CUA G ) within the LINfs5 consensus . This motif is found 30 times in the human genome and has recently been shown to support high translational readthrough ( Loughran et al . , 2014 ) . The motifs for high BTR are distinct from the consensus UGA CAR YYA ( R = A/G , Y = C/U ) found in some viruses and yeast ( Namy et al . , 2001; Harrell et al . , 2002 ) but resembles the alphavirus-like high readthrough stop codon context ( Li and Rice , 1993 ) . Interestingly , the same stop suppression context in the LAMA3 gene has been shown to alleviate the disease severity of an otherwise fatal nonsense mutation in a patient with junctional epidermolysis bullosa , the major and most devastating form of epidermolysis bullosa ( Pacho et al . , 2011 ) . The existence of the consensus motif UGA CUA is the origin of the non-linear contribution to RTP in our models . This is supported by the finding that correlation of BTR and RTP for LINfs3 is higher than for the LINiter model , so that the reduced number of parameters in LINfs3 provides a better model fit . This finding implies that with the currently small dataset , compact linear models should be preferred over non-linear models with many parameters . The identification of the few relevant nucleotide positions will help to create datasets with fully specified BTR for a wide range of SCCs and cell types . A larger training set of sequences with verified readthrough rates will allow the development of non-linear approximation models . LDHBx shows an unusually high readthrough of 1 . 6% , and its stop context UGA CUA G ( stop codon underlined ) matches the LINfs3 consensus . The 18-nucleotide extension in LDHBx is unlikely to contain a extensive secondary structure that would suggest a combined effect of BTR and PTR . The identification of LDHBx and the recently discovered readthrough form of vascular endothelial growth factor A , VEGF-Ax ( Eswarappa et al . , 2014 ) , thus mark two extreme and separable cases of physiological stop suppression: LDHBx appears independent of cis-factors beyond the SCC and marks a prototypical example of BTR . In contrast , the readthrough of VEGF-Ax is relatively independent of its SCC but instead requires a more distantly located cis-element ( Eswarappa et al . , 2014 ) . The distinction between PTR and BTR , however , is not exclusive . A thorough analysis of readthrough in OPRK1 and OPRL1 indicates that readthrough levels of more than 30% can be obtained by a combination of cis-elements and UGA CUA-based BTR ( Loughran et al . , 2014 ) . The era of systematic analysis of translational readthrough in humans is only beginning . We expect that a combination of in silico modeling and screening , ribosome profiling , phylogenetic methods , and mass spectrometry will help to identify the ‘extensome’ , the complete set of readthrough-extended proteins in mammals . To predict the RTP of gene transcripts , we developed a linear regression model based on the SCCs and their experimentally determined basal readthrough values . The SCC comprises the stop codon itself ( positions +1 to +3 ) and the nucleotide sequences surrounding the stop codon ( −6 to +9 ) . For the first-pass model ( LIN ) , we re-analyzed 66 SCCs with known experimental basal readthrough values ( Floquet et al . , 2012 ) . The stop codons evolved independently of their contexts ( Table 3 ) . Nucleotide sequences were represented by indicator vector coding . Here , 12 × 4 binary vector entries are used to indicate the presence [1] or absence [0] of a nucleotide ( A , C , G , or U ) at a particular position ( −6 to −1 , +4 to +9 ) surrounding the stop codon . Three further entries are reserved to indicate the type of stop codon ( UAA , UAG , or UGA , positions +1 , +2 , or +3 ) . The resulting feature vectors of all sequences were normalized to Euclidean unit length . 10 . 7554/eLife . 03640 . 024Table 3 . Nucleotide frequencies in each position of the stop codon contextDOI: http://dx . doi . org/10 . 7554/eLife . 03640 . 024NucleotideACGUPosition−60 . 28920 . 25300 . 26510 . 1928−50 . 32530 . 26510 . 14460 . 2651−40 . 15660 . 22890 . 34940 . 2651−30 . 24100 . 33730 . 24100 . 1807−20 . 26510 . 18070 . 20480 . 3494−10 . 24100 . 25300 . 26510 . 241040 . 22890 . 31330 . 33730 . 120550 . 26510 . 25300 . 14460 . 337360 . 27710 . 21690 . 25300 . 253070 . 25300 . 31330 . 28920 . 144680 . 32530 . 16870 . 24100 . 265190 . 18070 . 27710 . 27710 . 2651Stop codonsUAAUAGUGA1 to 30 . 19280 . 33730 . 4699The nucleotide and stop codon frequencies for positions −6 to −1 and 4 to 9 were calculated for the 81 sequences used in the RTP predictor ( LINiter model ) . For the estimation of the regression model coefficients , we performed a regularized least-squares ( ‘rigde’ ) regression ( Hoerl and Kennard , 1970 ) . Let X be the n × d matrix of n sequence feature vectors with dimensionality d and y be the ( n-dimensional ) vector of readthrough values associated with the sequences . Then the weight vector w = ( XTX + k × I ) −1 × XTy represents the solution of the linear least-squares problem and y = wTx corresponds to the RTP value y for a sequence feature vector x . To evaluate the influence of the regularization parameter k , we performed a leave-one-out cross-validation ( loo-cv ) with k = {10i\|i = −3 , −2 . 7 , . . . 0 , . . . , 3} for all model types . The minimum loo-cv error in terms of the sum of squared deviations of predictions from known readthrough values was 4 . 75 × 10−7 for k = 100 . 3 ( approximately 1 . 995 ) . For genome-wide prediction of readthrough propensities for human transcripts , we downloaded all 215 , 621 coding sequences from the Ensembl BioMart ( Flicek et al . , 2012 ) using the Homo sapiens Genes v74 section ( GRCh37 . p13 ) plus 300 nucleotides downstream of the CDS end ( ensembl . org , November 2013 ) . Transcripts corresponding to identical protein products , short sequences ( <15aa protein-coding ) and incomplete ( e . g . , missing or mislocated stop codon ) or insufficiently sequenced ( i . e . , undetermined nucleotides ) DNA were removed . Sequences with identical 3′/C-termini ( nucleotide positions −45 to +303 ) were aggregated to one representative sequence , resulting in 42 , 069 unique transcripts . ORF extensions were identified by detection of an in-frame stop codon within 300 nucleotides downstream of the annotated stop codon . To obtain a more comprehensive model for RTP prediction , we included 15 sequences and their corresponding experimentally determined readthrough values from this study in the prediction model ( see Schueren et al . , 2014 for Dataset 1 ) . The regression coefficients for the iterative model considering all 12 stop context positions ( LINiter ) were computed as described in the previous section . The minimum regression error was 6 . 24 × 10−6 at k = 100 . 3 . A sequence logo representation of the regression coefficients for this model is displayed in Figure 2A . The sequence logo was created using the enoLOGOS web server ( Workman et al . , 2005 ) . Furthermore , we evaluated reduced model sizes by stepwise elimination of context positions carrying no or little information for RTP prediction ( feature selection ) . Starting from the complete mode ( LIN ) , we removed the position corresponding to the minimum sum of squared regression coefficients . Regression error and coefficients were then calculated for the remaining positions ( including the stop codon ) as described above . This procedure was repeated until only the stop codon position was left . Figure 2B shows the development of the regression error for reduced model sizes by stepwise elimination of positions . Here , a first local minimum can be identified for model LINfs5 with five positions remaining ( −6 , stop , +4 to +7 ) and the global minimum corresponds to model LINfs3 with three positions besides the stop codon ( stop , +4 to +6 ) . To identify cryptic peroxisomal localization signals in readthrough extensions , we adapted a peroxisomal targeting signal type 1 ( PTS1 ) detection algorithm that was previously developed for plant proteins ( Lingner et al . , 2011 ) . For this purpose , we used 24 known human PTS1 proteins ( ACOT4 , ACOX1 , ACOX2 , ACOX3 , AGXT , AMACR , BAAT , CRAT , DAO , EHHADH , GNPAT , HAO1 , HAO2 , HSD17B4 , IDE , MLYCD , PRDX5 , ACOT8 , CROT , PECI , ECH1 , LONP2 , PECR , and PIPOX ) and performed orthology searches on metazoan protein and EST sequences using a bidirectional best BLAST hit strategy . Starting from each human protein sequence , we identified significant BLAST hits ( e-value < 10−10 ) to metazoan sequences within the ‘nr’ and ‘dbEST’ database . Then , the best hit of each organism was searched against the human proteome and sequences not re-identifying the starting sequence were removed . Afterwards , the starting sequences and putative orthologs were pooled and sequences with uncommon PTS1 tripeptides , that is tripeptides which occurred less than three times , were removed from the set . The resulting set of sequences was used as positive examples for training machine learning models as previously published ( Lingner et al . , 2011 ) . Briefly , a regularized least-square classification algorithm was trained using indicator vector representations of up to 15 C-terminal amino acids of positive and negative example sequences . A set of negative example sequences was created by extracting all metazoan sequences without peroxisomal association from the Swiss-Prot section of UniProt ( http://www . uniprot . org ) in November 2011 . The best model ( 15 C-terminal amino acids ) was determined by fivefold cross-validation and yielded a prediction accuracy of 0 . 996 and 0 . 863 in terms of the area under the receiver operating characteristic ( ROC ) curve ( auROC ) and the area under the precision/recall curve ( auPRC ) , respectively . When a stop codon was considered in the PTS1 prediction , the stop codon was scored as an undefined amino acid ( ‘X’ ) without a contribution to the PTS1 posterior probability . The multiple alignment of genomic sequences for the LDHB SCC ( position −36 to +48 ) was downloaded from the Ensembl database ( www . ensembl . org ) in November 2013 . The ‘21 amniota vertebrates’ alignment was used and split into mammalian and non-mammalian species . Sequences without residues in the extension region were deleted and the non-mammalian alignment was augmented by LDHB sequences from the NCBI nucleotide database ( http://www . ncbi . nlm . nih . gov/nuccore ) in November 2013 . In total , the alignments comprise 13 mammals and nine non-mammalian vertebrates: Homo sapiens ( human ) , Mus musculus ( mouse ) , Rattus norvegicus ( rat ) , Oryctolagus cuniculus ( rabbit ) , Pan troglodytes ( chimpanzee ) , Gorilla gorilla ( gorilla ) , Pongo abelii ( orangutan ) , Macaca mulatta ( rhesus macaque ) , Felis catus ( cat ) , Canis familiaris ( dog ) , Equus caballus ( horse ) , Bos taurus ( cow ) , Ovis aries ( sheep ) , Xenopus tropicalis ( western clawed frog ) , Anolis carolinensis ( anole lizard ) , Ficedula albicollis ( flycatcher ) , Taeniopygia guttata ( zebra finch ) , Gallus gallus ( chicken ) , Meleagris gallopavo ( turkey ) , Alligator mississippiensis ( American alligator ) , Salmo salar ( salmon ) , and Danio rerio ( zebrafish ) . The genomic sequences were translated into amino acid sequences using the ‘EMBOSS Transeq’ web server ( http://www . ebi . ac . uk/Tools/st/emboss_transeq/ ) . Species trees were obtained from the Interactive Tree Of Life ( iTOL ) website ( http://itol . embl . de/ ) and visualized with the Phylip package ( Felsenstein , 1989 ) . JalView software ( Waterhouse et al . , 2009 ) was used to visualize the alignments and to compute alignment quality and consensus . Here , the quality score of an alignment column is inversely proportional to the average cost of all pairs of mutations in terms of BLOSUM 62 substitution scores and the consensus reflects the fraction of the most frequent residue for each column of the alignment . Plasmids used in this study are listed in the table in Supplementary file 1 . Oligonucleotides used in this study are listed in the table in Supplementary file 2 . The dual reporter vector pDRVL ( PST1360 ) encoding an N-terminal Venus tag and a C-terminal hRluc tag was derived from pEXP-Venus-hRluc ( a gift from Ania Muntau and Sören Gersting ) by introducing a short multicloning site ( MCS ) containing BstEII , ClaI , BspEI , and BsiWI restriction sites . pDRVL was created by ligating pre-annealed oligonucleotides OST963 and OST964 into the XhoI site of pEXP Venus-hRluc . Dual reporter constructs PST1384–1385 , 1387 , 1393–1396 , 1418–1426 , 1430 , 1435 , 1437 , 1493 , 1494 , 1497 , 1504 , and PST1444 were derived from pDRVL by insertion of pre-annealed oligonucleotides OST1081–1084 , 1086–1087 , 1117–1124 , 1144–1145 , 1148–1157 , 1160–1165 , 1158–1159 , 1190–1191 , 1198–1199 , 1229–1230 , JH59–60 , JH61–62 , JH67–68 , and JH81–82 into BspEI and BstEII sites , as listed in Supplementary file 2 . For cloning of pEYFP-LDHBx ( PST1388 ) , the LDHB open reading frame including the stop codon and the 18-nucleotide 3′ extension , was PCR-amplified from pOTB7-LDHB using primers OST1053 and 1054 and inserted into EcoRI and XbaI sites of pEYFP-C1 . The stop codon variants pEYFP-LDHBx[TGG] ( PST1389 ) , pECFP-LDHBx[TGG] ( PST1440 ) , pEYFP-LDHBx[TAA] ( PST1410 ) , pEYFP-LDHBx[TAAT] ( PST1411 ) , and pEYFP-LDHBx[TGAT] ( PST1409 ) were created by amplifying LDHBx using primer OST1053 with reverse primers OST1055 , 1127 , 1128 , and 1129 , respectively . Similarly , the PTS1 mutation variants pEYFP-LDHBx[ΔL] ( PST1407 ) , pECFP-LDHBx[TGG , ΔL] ( PST1512 ) ( deletion of the last amino acid in the cryptic PTS1 SRL ) , and pEYFP-LDHBx[SSI] ( PST1408 ) , pECFP-LDHBx[TGG , SSI] ( PST1513 ) ( substitution of the PTS1 SRL by SSI ) were created using forward primer OST1053 and reverse primers OST1125 , 1263 , 1126 , and 1264 , respectively . LDHA was amplified from human cDNA using primers OST1130 and 1131 and cloned into EcoRI and XbaI sites of pEYFP-C1 to yield pEYFP-LDHA ( PST1434 ) . For cloning of pEXP Venus-PTS1 ( PST1209 ) , primers OST801 and 802 ( encoding the PTS1 of ACOX3 ) were annealed and inserted into pENTR-TOPO-D . Then the PTS1 tag was transferred to pEXP-N-Venus using LR clonase II ( Invitrogen , Carlsbad , California ) . Full-length dual reporter constructs pcDNA3 . 1-HA-LDHBx-Myc and variants were cloned by amplifying LDHB and stop codon variants from PST1388 ( LDHB wt ) , PST1389 ( LDHB [TGG] ) , PST1409 ( LDHB [TGAT] ) , PST1410 ( LDHB [TAA] ) , and PST1411 ( LDHB [TAAT] ) , using primers OST1202 and 1203 and cloning into NheI and BamHI restriction sites of pcDNA3 . 1/Myc-His ( − ) A . All plasmids were confirmed by DNA sequencing . HeLa cells and human skin fibroblasts were maintained in low glucose Dulbecco's minimal essential medium ( DMEM ) , HEK cells , HT1080 , U118 , U373 and COS-7 cells in high glucose DMEM . Culture media were supplemented with 1% ( wt/vol ) glutamine , 5–10% ( vol/vol ) heat inactivated fetal calf serum ( FCS ) , 100 units/ml penicillin , and 100 µg/ml streptomycin . For U118 cells , 1% non-essential amino acids and 1% pyruvate were added to the media . Cells were transfected using Effectene transfection reagent ( Qiagen , Germany ) as described by the manufacturer . Plasmids were diluted in Buffer EC and Enhancer and incubated for 5 min at room temperature . Effectene was added and incubated for 10 min at room temperature . Prewarmed medium was added to the HeLa cells and to the transfection mixture which was then added to cells and incubated at 37°C in a humidified 5% CO2 incubator for 24 hr . Then , 6 hr after transfection , transfection reagent was removed , and , where indicated , geneticin ( G418 ) was added at a concentration of 100 µg/ml . Cells were washed with PBS and lysed by Renilla Luciferase Assay Lysis Buffer ( Promega , Madison , Wisconsin ) according to the manufacturer's manual . Cells were spun down ( 14 krpm , 2 min , 4°C ) and supernatants were stored at −80°C . For Venus fluorescence measurement , cell lysates were diluted 1:25 in PBS and analyzed at 485 nm excitation , 530 nm emission ( sensitivity: 130 ) using a Synergy Mx plate reader ( Biotek , Winooski , Vermont ) . PBS was used as a blank control for fluorescence measurements . Undiluted lysates ( 20 µl ) were used to measure hRluc luminescence by the Renilla Luciferase Assay System ( Promega ) and the Synergy Mx plate reader ( Biotek ) . An automated injector was used to add 100 µl Renilla Luciferase Assay Reagent . Luminescence was read 2 s after injection and integrated over 10 s ( sensitivity: 150 ) . Renilla Luciferase Assay Reagent was used as a blank control for hRluc luminescence measurements . Each construct was analyzed in three to seven biological replicates and each biological sample was measured in triplets . To obtain readthrough rates , the ratio of hRluc/Venus fluorescence was calculated , and the readthrough of pDRVL was set to 100% . The ratio ( y ) and standard deviation of fluorescence ( x1 ) and luminescence ( x2 ) signal for each replicate were calculated using uncertainty propagation ( σy = [σ2x1 × ( dy/dx1 ) 2 + σ2x2 × ( dy/dx2 ) 2]0 . 5 ) . Let wi = 1/σi2 be the weight of a readthrough value from replicate i with σi being the error of the ratios . Then the weighted mean xm of the replicates and its error σxm were calculated according to xm = ( Σi ( xiwi ) /Σiwi ) and σxm = ( Σiwi ) −0 . 5 . Transfected LDHB and LDHA fusion constructs were detected in HeLa cells by combined direct fluorescence and immunofluorescence experiments . Endogenous LDHB was analyzed in HeLa , U118 , and COS-7 cells , and in primary rat cardiomyocytes by immunofluorescence . Approximately 1 × 105 cells were seeded on cover slips or on laminin-coated ( Sigma , St . Louis , Missouri ) glass slides for HEK cells and cardiomyocytes and transfected as indicated . For removal of cytosol , cells were treated with 0 . 02% ( wt/vol ) digitonin ( Invitrogen ) for 5 min at room temperature . Cells were fixed with 10% ( wt/vol ) formaldehyde for 20 min , and permeabilized with 0 . 5% Triton X-100 for 5 min . After blocking for 20 min at 37°C with 10% BSA , antigens were labeled with primary antibodies at 37°C for 1 hr . Antibody dilutions were 1:200 for anti-PEX14 rabbit polyclonal antibodies ( ProteinTech , Chicago , Illinois ) and 1:500 for anti-LDHB mouse monoclonal antibodies ( Abnova , Taiwan ) . Secondary antibody labeling ( 1:200 ) was done for 1 hr with antibodies labeled with Cy3 and/or Alexa647 ( Jackson Immuno Research , West Grove , Pennsylvania ) and/or Alexa488 ( MoBiTech , Germany ) . Cover slips were mounted with Mowiol containing 0 . 01 mg/ml 4′ , 6-diamidino-2-phenylindole ( DAPI ) . DAPI was omitted in cases where cells had been transfected with CFP-expressing plasmids . Fluorescence microscopy was done using a 100× oil objective ( 1 . 3 NA ) with a Zeiss Imager M1 fluorescence wide field scope equipped with the Zeiss Axiocam HRm Camera and Zeiss Axiovision 4 . 8 acquisition software . z-Stacks with 30 images and 0 . 25 µm spacing were recorded and subjected to deconvolution . Where necessary , linear contrast enhancements were applied ( Axiovision ) . To quantify induction of endogenous LDHB by geneticin , fluorescence images from samples prepared with anti-LDHB and anti-PEX14 antibodies were recorded under identical conditions and subjected to deconvolution . The LDHB/PEX14 intensities were measured , and the same threshold ratios were applied to all channel pairs ( ImageJ ) . Induction is expressed as the ratio of LDHB/PEX14 ratios with and without geneticin treatment , respectively . Cells were lysed in RIPA lysis buffer ( 20 mM Tris–HCl , pH 7 . 4 , 150 mM sodium chloride , 2 mM EDTA , 1% NP40 , 1 mM DTT , 0 . 1 mM PMSF , Complete protease inhibitors [Roche , Switzerland] ) 24 hr after transfection . Proteins were separated by SDS-PAGE on a 12% gel , transferred to a nitrocellulose membrane , and probed with primary and secondary antibodies . The following antibodies were used: anti-HA rabbit polyclonal ( Abcam , UK ) , anti-Myc mouse monoclonal ( Cell Signaling , UK ) , anti-luciferase mouse monoclonal ( Millipore ) , anti-GFP mouse monoclonal ( Living Colors , Mountain View , California ) , and anti-actin mouse monoclonal ( Sigma ) . HRP-conjugated goat anti-rabbit IgG and donkey anti-mouse IgG ( Jackson Immuno Research ) were used as secondary antibodies . We also used 1:1000 dilutions of primary antibody and 1:5000 dilutions of secondary antibody . Reactive bands were revealed with Lumi-light and Lumi-light plus Western blotting substrate ( Roche ) . Images were scanned using Luminescent image analyzer LAS 4000 . Dataset 1 . Spreadsheet containing predicted readthrough extensions , RTP scores ( LIN , LINiter , LINfs5 , LINfs3 ) , PTS1 scores , predictions of ER retentions signals , glycosylation motifs , transmembrane domains , and transmembrane topology , and the LINiter+ × PTS1 product scores for all human transcript termini . Publicly available at the Dryad Digital Repository with the doi 10 . 5061/dryad . j2n18 ( Schueren et al . , 2014 ) .	Amino acids are the building blocks of proteins , and the order of the amino acids in a protein is determined by the order in which ‘codons’ appear in a messenger RNA molecule . Most codons represent a specific amino acid , but there are also three stop codons that are used to mark the end of a protein . When the cellular machinery that ‘translates’ the messenger RNA molecule into a protein encounters a stop codon , it stops and releases the completed protein . Sometimes , however , the stop codon is not interpreted as a stop signal , and the translation of the messenger RNA molecule continues until another stop codon is encountered . This process is known as readthrough . Some organisms , in particular viruses and fungi , use readthrough to produce a wider range of proteins than their genomes would otherwise allow . While readthrough also occurs in higher organisms such as mammals , it is not known if the resulting proteins perform extra functions that the original protein does not perform . A number of factors affect whether readthrough occurs when an mRNA template is being translated . For example , each of the three stop codons has a different likelihood of having its stop signal misinterpreted , and the mRNA sequence that surrounds the stop codon can also affect the likelihood of readthrough . Schueren et al . have developed a computational model that estimates how common this form of translational readthrough is in the human genome . The model was based on the identity of the stop codons themselves and the surrounding mRNA sequence . This model was then combined with another model that identifies proteins that are targeted to a structure inside a cell called the peroxisome , which is where a number of essential energy-releasing reactions take place . The combined model enabled Schueren et al . to identify proteins that both perform functions in the peroxisome and are likely to be formed by readthrough . The combined model suggested a protein that is a part of lactate dehydrogenase: an enzyme that speeds up chemical reactions that are important for the cell to produce energy . Low levels of lactate dehydrogenase had previously been found in the peroxisome , despite it apparently lacking a specific sequence of amino acids that proteins need to have to enter the peroxisome . However , Schueren et al . confirmed experimentally that readthrough does occur for the lactate dehydrogenase component identified by the model , revealing that it contains a ‘hidden’ peroxisome-targeting region . Furthermore , when more translational readthrough occurred , more lactate dehydrogenase was found in the peroxisomes . This unusual way that lactate dehydrogenase enters the peroxisome is an example of how the cell optimizes the used of the genetic information encoded in the genome and in messenger RNA . Translational readthrough always ensures that a certain proportion of lactate dehydrogenase will be brought to the peroxisome . The computational model developed here will be a valuable tool to identify other such proteins produced from genomes , including the human genome and those of other species .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2014	Peroxisomal lactate dehydrogenase is generated by translational readthrough in mammals
Human positive transcription elongation factor b ( P-TEFb ) phosphorylates RNA polymerase II and regulatory proteins to trigger elongation of many gene transcripts . The HIV-1 Tat protein selectively recruits P-TEFb as part of a super elongation complex ( SEC ) organized on a flexible AFF1 or AFF4 scaffold . To understand this specificity and determine if scaffold binding alters P-TEFb conformation , we determined the structure of a tripartite complex containing the recognition regions of P-TEFb and AFF4 . AFF4 meanders over the surface of the P-TEFb cyclin T1 ( CycT1 ) subunit but makes no stable contacts with the CDK9 kinase subunit . Interface mutations reduced CycT1 binding and AFF4-dependent transcription . AFF4 is positioned to make unexpected direct contacts with HIV Tat , and Tat enhances P-TEFb affinity for AFF4 . These studies define the mechanism of scaffold recognition by P-TEFb and reveal an unanticipated intersubunit pocket on the AFF4 SEC that potentially represents a target for therapeutic intervention against HIV/AIDS . At many genes in humans—including the integrated HIV genome as well as loci that regulate development and mediate responses to stress—RNA polymerase II initiates transcription but forms a stable paused complex after the synthesis of 30–50 nucleotides ( Lin et al . , 2011; Levine , 2012; Luo et al . , 2012a; Zhou et al . , 2012 ) . These paused polymerases are poised for rapid , synchronous efficient transcription . For these genes , escape of the paused polymerase from the promoter-proximal region and elongation of the mRNA are rate-limiting regulated processes . Promoter escape requires positive transcription elongation factor b ( P-TEFb ) , a heterodimeric protein kinase composed of CDK9 and cyclin T1 ( CycT1 ) subunits . P-TEFb triggers promoter escape by directly or indirectly stimulating phosphorylation of the RNA polymerase II C-terminal domain and the associated factors , NELF ( negative elongation factor ) and DSIF ( DRB sensitivity inducing factor ) . Consequently , recruitment of active P-TEFb to the paused polymerase complex serves as an important checkpoint for gene expression ( Levine , 2012; Luo et al . , 2012b; Zhou et al . , 2012 ) . P-TEFb cycles between inactive and active complexes ( Zhou and Yik , 2006 ) . Recent studies of gene fusions in myeloid leukemias ( Lin et al . , 2010; Yokoyama et al . , 2010 ) , as well as complexes recruited to the HIV promoter by the HIV Tat protein ( He et al . , 2010; Sobhian et al . , 2010; Jager et al . , 2012 ) , uncovered a family of Super Elongation Complexes ( SECs ) that bring together active P-TEFb and other transcription elongation factors . The SECs act at many normal human genes to stimulate mRNA elongation not only by triggering promoter escape but also by limiting proteolytic degradation of transcription elongation factors and increasing the processivity of RNAP II ( He et al . , 2010; Lin et al . , 2010; Biswas et al . , 2011; Liu et al . , 2012 ) . The SECs also couple to the PAF complex , which stimulates efficient transcript elongation ( He et al . , 2011 ) . In addition to P-TEFb , the SECs contain subunits in the AF4 , ELL , and ENL/AF9 protein families . Despite the major roles of SECs in metazoan gene expression and human disease , little is known about the architecture of these complexes . To define the interactions that mediate SEC assembly and to understand how HIV Tat recruits P-TEFb ( Wei et al . , 1998 ) in the context of these large protein complexes , we mapped contacts among SEC subunits ( Figure 1A ) ( Chou et al . , 2012 ) . Here , we report the structural and functional analysis of P-TEFb in complex with the cognate binding site on the SEC scaffold protein , AFF4 . These studies reveal that AFF4 recognizes P-TEFb by binding the CycT1 subunit on the side opposite from CDK9 . AFF4 is positioned to make direct contacts with HIV-1 Tat . Tat increases the affinity of P-TEFb for AFF4 by over an order of magnitude in vitro and rescues P-TEFb binding of AFF4 interface mutants in vivo . These results suggest that the SEC scaffold is an unanticipated direct partner of HIV-1 Tat , and an intersubunit Tat-binding pocket in the AFF4-P-TEFb complex may afford an unexpected site to target with selective inhibitors of HIV transcription . 10 . 7554/eLife . 00327 . 003Figure 1 . AFF4 binds CycT1 distal to CDK9 . ( A ) Schematic model of the SEC . AFF4 is an intrinsically disordered scaffold that binds partners via 20–50 residue segments . ( B ) Ribbon diagram showing the strand–helix–helix arrangement of AFF4 ( blue ) bound to CycT1 ( yellow ) remote from CDK9 ( teal ) . AFF4 adopts an extended conformation with no intramolecular tertiary contacts . AMPPNP ( spheres ) is bound to CDK9 . DOI: http://dx . doi . org/10 . 7554/eLife . 00327 . 00310 . 7554/eLife . 00327 . 004Figure 1—figure supplement 1 . Electron density for AFF42–73 . Fo-Fc omit map ( 3 . 5 σ , gray ) for residues 34–66 of AFF4 . Anomalous difference map ( 5 σ , red ) shows the positions of the methionine residues in AFF4 . The anomalous difference Fourier was calculated using X-ray data recorded from a crystal grown with Seleno-methionine labeled AFF4 . DOI: http://dx . doi . org/10 . 7554/eLife . 00327 . 00410 . 7554/eLife . 00327 . 005Figure 1—figure supplement 2 . A crystal contact formed by AFF42–21 . ( A ) An isolated helix from the aff4 N-terminus , packs loosely against αE and αI of one CDK9 subunit ( chain C ) and makes contacts with the β2-β3 loop of a symmetry-related CDK9 molecule . ( B ) Interactions between CDK9 residues ( teal sticks ) and the isolated aff4 helix H0 ( blue ) . Anomalous difference map ( 3 . 5 σ , red ) shows the position of SeMet11 . The occurrence of this helix in only one of three independent complexes in the crystals and the lack of electrostatic complementarity with CDK9 suggest that the helix may depend on the crystal environment and the kinase subunit is not the functional partner of this sequence . DOI: http://dx . doi . org/10 . 7554/eLife . 00327 . 005 The AF4 proteins , AFF1–4 , form intrinsically disordered scaffolds that bind other transcription elongation factors not through folded domains but rather through dispersed short binding sites in the first 750 amino acids ( Chou et al . , 2012; Leach et al . , 2013 ) ( Figure 1A ) . Residues 2–73 of AFF4 , for example , are sufficient to bind P-TEFb through the CycT1 subunit , and a peptide encompassing AFF42–73 folds upon binding CycT1 . Using this binding site , we determined the 2 . 9-Å-resolution crystal structure of AFF42–73-P-TEFb-AMPPNP ( R/Rfree = 0 . 207/0 . 245; Figure 1 , Figure 1—figure supplements 1 and 2 , and Table 1 ) . 10 . 7554/eLife . 00327 . 006Table 1 . X-ray data collection and refinement statistics for AFF4-P-TEFb-AMPPNPDOI: http://dx . doi . org/10 . 7554/eLife . 00327 . 006Data collectionAFF4-P-TEFb-AMPPNPSpace groupP212121Cell dimensions: a , b , c100 . 691 , 126 . 298 , 195 . 626Resolution ( Å ) 50-2 . 94 ( 2 . 99–2 . 94 ) Unique reflections54 , 189 ( 2664 ) Rsym ( % ) 9 . 3 ( >100 ) I/σ ( I ) 23 . 2 ( 1 . 3 ) Completeness ( % ) 100 . 0 ( 100 . 0 ) Redundancy8 . 1 ( 7 . 5 ) Temperature ( K ) 100Mosaicity ( ° ) 0 . 45–0 . 6RefinementResolution ( Å ) 48 . 7-2 . 94No . reflections53 , 775Rwork/Rfree0 . 207/0 . 245No . atoms/B-factors ( Å2 ) CDK9 , molecule 1 , 2 , 32558 ( 111 . 9 ) , 2533 ( 116 . 3 ) , 2558 ( 121 . 6 ) Cyclin T1 , molecule 1 , 2 , 32003 ( 121 . 3 ) , 2024 ( 123 . 1 ) , 2001 ( 118 . 5 ) AFF4 , molecule 1 , 2 , 3248 ( 156 . 3 ) , 421 ( 161 . 1 ) , 243 ( 160 . 3 ) Water19 ( 90 . 1 ) Root mean square deviationsBond lengths ( Å ) 0 . 004Bond angles ( ° ) 0 . 666Ramachandran plot†Favored ( % ) 94 . 7Allowed ( % ) 4 . 48Disallowed ( % ) 0 . 78Protein Data Bank ID4IMY*Values in parentheses are for the highest resolution shell . †Values from MOLPROBITY . In all three independent copies of the complex in the crystals , AFF4 residues Leu34-Ile66 are ordered , binding to the second cyclin domain of CycT1 opposite CDK9 ( Figures 1B , 2 ) . In one complex , an isolated helix containing AFF4 residues 3–21 bridges symmetry-related CDK9 molecules in the crystals ( Figure 1—figure supplement 2 ) . We focus on the shared features of the three independent complexes . Consistent with the coupling of folding to binding , AFF4 lacks intramolecular tertiary contacts as it snakes across the CycT1 surface . An extended segment in AFF4 tracks antiparallel to the H3′-H4′ loop in CycT1 , and two short AFF4 helices fit into shallow grooves in CycT1 between H3′ and H5′ and the surface formed by H2′ and H3′ ( Figure 2A ) . Compared to the separated models , 1457 Å2 of AFF4 and 1251 Å2 of CycT1 accessible surface are buried in the complex . Twenty-six of the 33 ordered residues in AFF4 contact CycT1 , emphasizing the extensive recognition determinants in the scaffold . 10 . 7554/eLife . 00327 . 007Figure 2 . Basis for AFF4 scaffold recognition by P-TEFb . ( A ) AFF4 residues 34–66 ( blue spheres ) fill grooves on CycT1 ( yellow surface ) . ( B ) Chemical complementarity mediates AFF4 binding . Exposed hydrophobic residues of CycT1 ( yellow surface ) are buried in the AFF4 complex . Hydrogen bonds ( black dotted lines ) also mediate binding . ( C ) AFF4 Phe35 is buried in a hydrophobic pocket formed by aromatic and nonpolar residues on the surface of CycT1 . ( D ) The C-terminus of the CycT1 cyclin domain ( gray in the ellipse ) adjusts to make contacts with AFF4 ( blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00327 . 00710 . 7554/eLife . 00327 . 008Figure 2—figure supplement 1 . Example interactions between AFF4 and P-TEFb . ( A ) Main-chain hydrogen bonds between the extended aff4 peptide 34–41 and CycT1 . ( B ) Hydrophobic interactions with CycT1 Trp207 and hydrogen bonds of CycT1 Trp207 and Asp169 anchor the aff4 α1–α2 loop . DOI: http://dx . doi . org/10 . 7554/eLife . 00327 . 00810 . 7554/eLife . 00327 . 009Figure 2—figure supplement 2 . Conserved AFF4 sequences mediate P-TEFb recognition . Multiple sequence alignment of amino acids 1–78 ( AFF4 numbering ) for human and mouse AF4 family members . DOI: http://dx . doi . org/10 . 7554/eLife . 00327 . 009 Many hydrophobic and aromatic residues in CycT1 mediate contacts with AFF4 ( Figure 2B–C and Figure 2—figure supplement 1 ) . A pocket between CycT1 Leu163-Val164-Arg165 , Trp221 , and Tyr224 , for example , anchors AFF4 Phe35 , which forms a classic edge-to-face interaction ( Burley and Petsko , 1985 ) with the Tyr ( Figure 2C ) . CycT1 Trp210 adjusts to create a site that buries AFF4 Pro38 and allows a short antiparallel β-sheet to form between AFF4 Tyr39-Val41 and CycT1 Asn 209-Glu211 ( Figure 2—figure supplement 1A ) . CycT1 Trp207 shifts to form a hydrogen bond with the Gly57 carbonyl in the loop between the two AFF4 helices , additional van der Waals contacts with the loop backbone and nonpolar contacts with AFF4 residues Leu56 and Tyr59 ( Figure 2—figure supplement 1B ) . Eight intermolecular hydrogen bonds help establish the chemical complementarity between AFF4 and CycT1 . Twenty-one of the 26 interacting residues in AFF4 are conserved in AFF1–3 ( Figure 2—figure supplement 2 ) . In comparison to the structure of P-TEFb alone ( CDK91–330/CycT11–259; PDB ID 3BLQ; Baumli et al . , 2008 ) , the last 20 residues present in the CycT1 subunit undergo a major rearrangement upon AFF4 binding ( Figure 2D ) . AFF4 overlaps with the position of residues 243–259 in the superimposed P-TEFb cyclin subunit . To accommodate AFF4 , helix H5′ straightens and residues 251–256 form a helical segment that packs against CycT1 helices H1 and H2' and forms part of the AFF4 binding surface . In addition to burying hydrophobic residues in the first AFF4 helix , the adjustments in CycT1 helix H5′ mediate formation of a hydrogen-bonded ion pair between AFF4 Arg51 and CycT1 Glu246 . CycT1 Trp256 , the last residue ordered in the AFF4 complex , moves over 13 Å upon AFF4 binding . To probe the basis for CycT1 binding , we measured the effects of interface mutations on the stability of the AFF4 complex in vitro . The AFF4 2–363 and 33–67 fragments bound to CycT1 with similar affinities ( Table 2 ) , suggesting that the ordered AFF4 segment in the crystal structure captures the major CycT1 binding determinants . Mutations throughout the CycT1 interaction surface reduced AFF4 binding ( Figure 3A and Table 3 ) . The Trp210Ala and Trp207Ala substitutions in CycT1 had the largest effects , respectively , reducing AFF4 affinity by 21- and 58-fold . These results point to these CycT1 Trp residues as interaction hotspots and suggest that contacts all along the interface observed in the crystal structure mediate AFF4 recognition . 10 . 7554/eLife . 00327 . 010Table 2 . Binding affinities of AFF4 segmentsDOI: http://dx . doi . org/10 . 7554/eLife . 00327 . 010Direct bindingCompetition assayAFF432–67AFF432–67AFF42–73AFF42–363CycT1104 ± 17 nM102 ± 10 nM130 ± 18 nM115 ± 15 nMP-TEFb36 ± 6 nM36 ± 4 nM10 ± 1 nM7 ± 1 nMTat-P-TEFb8 . 8 ± 0 . 8 nM4 . 5 ± 0 . 6 nM0 . 85 ± 0 . 15 nM0 . 6 ± 0 . 1 nMDissociation constants measured by direct binding of fluorescein-labeled AFF432–67 and by competition with unlabeled AFF4 segments . The increased affinity of AFF4 for P-TEFb compared to CycT1 may be due to structural changes in the cyclin subunit or additional interactions with the CDK9 kinase subunit . The similar affinities of AFF42–73 and AFF42–363 for all the cyclin-containing species suggest that AFF42–73 encompasses the binding sites for P-TEFb and Tat-P-TEFb . 10 . 7554/eLife . 00327 . 011Figure 3 . AFF4 interface mediates P-TEFb recognition . ( A ) Mutations of CycT1 contact residues reduce AFF4 affinity . Fluorescence polarization of fluorescein-labeled AFF432–67 ( 5 nM ) is plotted as a function of the concentration of the indicated CycT1 variant . ( B ) Transcriptional effects of AFF4 tandem Ala mutants . Stimulation of Tat-independent transcription from the HIV LTR was measured in extracts of cells cotransfected with a luciferase reporter construct and an expression vector for the indicated di-Ala AFF4 variant . Activity was normalized to the level of AFF4 expression . Values represent the mean of three independent assays . Tandem alanine substitutions cover the first 72 residues of AFF4 . Horizontal lines correspond to the mean stimulation of di-Ala substitutions in residues 3–32 ( left; 113 . 9 ± 5 . 1 ) and 33–66 ( right; 73 . 6 ± 4 . 9 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00327 . 01110 . 7554/eLife . 00327 . 012Figure 3—figure supplement 1 . Expression levels of AFF4 variants . ( A ) Representative Western blot probed with anti-FLAG antibodies to measure the level of FLAG-tagged aff4 variants in HeLa cell lysates . ( B ) Expression levels of aff4 variants . Values are the average of three biological replicates . The tandem alanine mutants showed similar levels of expression . DOI: http://dx . doi . org/10 . 7554/eLife . 00327 . 01210 . 7554/eLife . 00327 . 013Table 3 . Dissociation constants of AFF432–67 for Cyclin T1 mutantsDOI: http://dx . doi . org/10 . 7554/eLife . 00327 . 013CycT1 variantKd ( nM ) Wild-type104 ± 17Y175A228 ± 18E211A356 ± 29D169A438 ± 41F176A645 ± 58R165A1592 ± 171W210A2190 ± 246W207A6050 ± 871 To examine the functional roles of the N-terminal 72 residues of AFF4 , we measured the effects of tandem alanine mutations on the stimulation of expression of a luciferase reporter gene driven by the HIV-1 promoter in HeLa cells ( He et al . , 2011 ) ( Figure 3B and Figure 3—figure supplement 1 ) . Consistent with the structure , mutations of AFF4 residues that contact P-TEFb reduced Tat-independent transcriptional stimulation . Three of the most deleterious variations—alanine substitutions at Pro33/Leu34 , Phe35 ( Ala36 was maintained ) , and Met55/Leu56—shorten hydrophobic side chains that are buried in the interface . Tandem alanine substitutions ( Glu45/Asp46 and Gly57/Asn58 ) that remove side chains that cap and stabilize the AFF4 helices also reduced luciferase expression . In contrast , residues such as Glu61/Met62 and Ile71/Pro72 , which are more tolerant of di-alanine substitutions , are more solvent exposed or flexible . Tandem alanine replacements in residues 3–32 , which flank the ordered CycT1 contacts in the crystal structure , showed significant but generally smaller effects on AFF4 transcriptional stimulation activity ( Figure 3B ) . These results implicate the P-TEFb binding site , as well as the flanking flexible sequences , in the function of the AFF4 N-terminal segment . The HIV-1 Tat protein was shown nearly 15 years ago to bridge P-TEFb and the TAR RNA element near the 5′ end of nascent HIV transcripts to recruit the active CDK9 kinase to the HIV promoter ( Wei et al . , 1998 ) . The discovery that Tat recruits P-TEFb as part of a larger SEC ( He et al . , 2010; Sobhian et al . , 2010 ) raised the question of how Tat distinguishes the SEC from free P-TEFb , particularly since Tat binds P-TEFb in isolation . Moreover , Tat shows specificity for SECs containing AFF4 and AFF1 ( He et al . , 2010; Sobhian et al . , 2010 ) . What accounts for this specificity ? The AFF4-P-TEFb crystal structure provides a simple and unsuspected explanation—Tat binds in a position to make direct contacts with the scaffold ( Figure 4A ) . Superposition of the CycT1 subunits of the structures of Tat-P-TEFb ( Tahirov et al . , 2010 ) and AFF4-P-TEFb shows that Tat is positioned to pack against helix 2 of AFF4 . Tat Met1 , Lys28 ( which is reversibly acetylated in vivo; Kiernan et al . , 1999; Ott et al . , 2011 ) and Phe32 , in particular , are predicted to interact with AFF4 Glu61 , Met62 , Phe65 , and Ile66 . The disordered C-terminus of the AFF4 peptide also neighbors Glu2 and His13-Gly15 of Tat . These seven residues of Tat are crucial for transcriptional activation , even though they are exposed to solvent in the Tat-P-TEFb complex ( D'Orso et al . , 2012 ) . 10 . 7554/eLife . 00327 . 014Figure 4 . AFF4 binds in position to make direct contacts with HIV-1 Tat . ( A ) Superposition of the AFF4-P-TEFb complex and the Tat-P-TEFb complex using the cyclin subunit ( yellow ) shows the close proximity of AFF4 ( blue ) and Tat ( red ) . Tat Lys28 ( pink ) , where acetylation stimulates function , as well as other residues essential for Tat transcriptional activation ( D'Orso et al . , 2012 ) that are exposed to solvent in the Tat-P-TEFb complex ( bright red ) are positioned adjacent to AFF4 . ( B ) Tat enhances AFF4 binding in vitro . Fluorescence polarization of fluorescein-labeled AFF432–67 ( 5 nM ) is plotted as a function of the concentration of CycT1 ( blue circles ) , P-TEFb ( red squares ) , and Tat-P-TEFb ( green triangles ) . ( C ) Alanine substitutions in the P-TEFb binding site of AFF4 reduce CycT1 binding but not associations of other SEC subunits in HeLa cells . Western blots show associations of each indicated factor with different FLAG-tagged AFF4 variants ( top ) . Lysates were immunoprecipitated with an anti-FLAG antibody . Expression of Tat ( right ) rescues defects in CycT1 binding , except for the E61/M62 double alanine mutant . This mutant in the predicted AFF4-Tat interface shows equal small defects in P-TEFb binding in the absence ( left ) and presence ( right ) of Tat . ( D ) AFF4 ( blue ) and CycT1 ( yellow ) create an intersubunit pocket where Tat ( red ) can bind with minor structural adjustments . The program DoGSiteScorer ( Volkamer et al . , 2012 ) assigns this cleft a high druggability score ( 0 . 83 out of 0–1 . 0 ) and shows that it contains the most nonpolar surface of any pocket in the AFF4-P-TEFb structure . DOI: http://dx . doi . org/10 . 7554/eLife . 00327 . 01410 . 7554/eLife . 00327 . 015Figure 4—figure supplement 1 . Thermodynamic cycle for AFF4 and Tat binding to P-TEFb . The enhancement of Tat affinity for P-TEFb by aff4 is equal to the ratio of aff4 affinity for P-TEFb in the presence and absence of Tat . DOI: http://dx . doi . org/10 . 7554/eLife . 00327 . 015 The juxtaposition of AFF4 and Tat on the P-TEFb surface predicts that the scaffold enhances Tat binding . Direct measurements of Tat affinity are problematic , however , because Tat is unstable in vitro in the absence of partners and difficult to maintain in an active form . To overcome this problem , we took advantage of a thermodynamic cycle that illustrates that AFF4 and Tat mutually influence the P-TEFb affinity of each other by the same amount ( Figure 4—figure supplement 1 ) . In quantitative in vitro assays ( Figure 4B and Table 2 ) , the purified AFF42–73 peptide bound Tat-P-TEFb ( Kd = 0 . 85 ± 0 . 15 nM ) ∼11 times more tightly than P-TEFb ( Kd = 10 ± 1 nM ) . The AFF42–363 segment also bound Tat-P-TEFb ( Kd = 0 . 6 ± 0 . 1 nM ) ∼11 times more tightly than P-TEFb ( Kd = 7 ± 1 nM ) . These results support the model of direct Tat interactions with AFF4 and suggest that the AFF42–73 peptide includes the principal residues that contact Tat ( Figure 4 and Figure 4—figure supplement 1 ) . To test the importance of the AFF4 interactions for P-TEFb and Tat recognition in vivo , we measured the effects of mutations in AFF4 on the binding of SEC subunits and Tat in HeLa cells . Cells were transfected with wild-type or mutant AFF4 containing a C-terminal 3× FLAG tag , and immunoprecipitations using an anti-FLAG antibody were probed for the presence of associated factors . Alanine substitutions in P-TEFb contacts such as AFF4 Phe35 , Tyr59/Asp60 , and Lys63/Asp64 reduced binding of P-TEFb but not the ELL2 , AF9 , or ENL subunits of the SEC ( Figure 4C ) . These defects were rescued by overexpressing a stably integrated gene for Tat , which strengthened the AFF4-P-TEFb association . In contrast , tandem alanine substitutions for AFF4 Glu61/Met62 , which are more exposed in the P-TEFb interface and predicted to contact Tat , caused a small reduction ( ∼30% ) in the binding of P-TEFb that was not rescued by overexpressing Tat ( Figure 4C ) . The specificity of these mutational effects supports the model that Tat interacts directly with AFF4 . In keeping with the tenuous contacts of the AFF43–21 helix with CDK9 , AFF42–73 ( Kd = 10 ± 1 nM ) binds only 3 . 6 times more tightly than AFF4 32–67 ( Kd = 36 ± 4 nM ) to P-TEFb ( Table 2 ) . In addition , the CDK9 kinase subunit structure is little changed upon AFF4 binding to P-TEFb ( CDK915–360 RMSD = 0 . 54 Å vs 1BLQ ) . In vivo , the Glu13Ala/Arg14Ala AFF4 mutant in the heart of the CDK9 interface in the crystals is associated with a <25% decrease in transcription stimulation ( Figure 3B ) and little change in P-TEFb binding ( Figure 4C ) . To measure if recruitment of P-TEFb to the SEC scaffold regulates the kinase , we assayed the effects of AFF4 on the in vitro phosphorylation of an RNA polymerase II CTD substrate by purified P-TEFb and Tat-P-TEFb ( Bitoun et al . , 2007 ) . AFF42–73 inhibited CTD phosphorylation by approximately twofold in a purified system ( Figure 5 and Figure 5—figure supplements 1 and 2 ) . By comparison , Tat stimulated P-TEFb by sevenfold , and addition of AFF2-73 did not further stimulate the kinase activity of the Tat-P-TEFb complex . Taken together , these results point away from CDK9 as the primary physiological partner of AFF43–21 and suggest that AFF4 functions as a SEC scaffold but not an allosteric activator of CDK9 . 10 . 7554/eLife . 00327 . 016Figure 5 . Kinase activity of P-TEFb and P-TEFb-Tat complexes with AFF4 . ( A ) Autoradiogram showing phosphorylation of GST-CTD ( 500 ng ) by P-TEFb and P-TEFb-Tat with and without excess ( 0 . 28 μM ) AFF42-73 in the presence of low ( 50 μM ) ATP . ( B ) Phosphorylation of 500 ng GST-CTD by P-TEFb and P-TEFb-Tat with and without excess ( 0 . 28 μM ) AFF42–73 in the presence of saturating ( 500 μM ) ATP . AFF4 reduces the activity of P-TEFb twofold and has little influence on the kinase activity of Tat-P-TEFb . Tat-P-TEFb is , however , sevenfold to tenfold more active than P-TEFb . Lane 3 in panels ( A and B ) is a control without GST-CTD . ( C ) Quantitation of the radioactive GST-CTD in ( A and B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00327 . 01610 . 7554/eLife . 00327 . 017Figure 5—figure supplement 1 . SDS polyacrylamide gel of P-TEFb and P-TEFb-Tat at the same ratio as they were used in the kinase assay . DOI: http://dx . doi . org/10 . 7554/eLife . 00327 . 01710 . 7554/eLife . 00327 . 018Figure 5—figure supplement 2 . Western blots of kinase reaction products from panel B . Phosphorylated CTD was detected with anti-phoshoSer2 and anti-phoshoSer5 antibodies . The GST-CTD was phosphorylated on Ser2 and Ser5 . However , the Ser2 phosphorylation was detected only on the full-length CTD , while Ser5 phosphorylation was detected disproportionately on proteolytic fragments of the CTD compared to the full-length CTD domain . DOI: http://dx . doi . org/10 . 7554/eLife . 00327 . 018 The AFF4-P-TEFb crystal structure reveals that a high density of contacts in residues 34–66 of the AFF4 scaffold mediates binding to the CycT1 subunit of P-TEFb . These contacts are largely conserved in AF4 family members and can be perturbed physically and functionally by mutations ( Figure 3 and Figure 2—figure supplement 2 ) . Consistent with these results , tandem alanine substitutions of Pro33/Leu34 , Val41/Thr42 , Arg51/Ile52 , and Met55/Leu56 in the CycT1 binding site ( but not Arg3/Glu4 or Glu25/Asp26 in the preceding segment ) also reduce P-TEFb binding in vivo ( Chou et al . , 2012 ) . The impacts of alanine substitutions on transcriptional stimulation by AFF4 ( Figure 3B ) show that residues in the P-TEFb interface , as well as helix stabilizing residues , play crucial roles . The additional sensitivity of transcriptional stimulation to alanine substitutions of disordered residues flanking the CycT1 binding site suggests that the flexibility and potential interactions with other ligands are important for function . HIV-1 Tat binds adjacent to AFF4 , increases the affinity of AFF4 for P-TEFb by over an order of magnitude , and rescues P-TEFb binding to AFF4 interface mutants in vivo . The AFF4 Glu61/Met62 double alanine substitution in the proposed Tat interface blocks this rescue . These results suggest that direct contacts between Tat and the AFF4 scaffold in complex with P-TEFb mediate the selective recruitment of the SEC to the HIV promoter . The functional AFF4-Tat interface , including the acetylated Lys28 in Tat as a potential regulator of affinity ( Kiernan et al . , 1999 ) , supports the idea that Tat evolved to function within the SEC . In the crystal structure of the ( low-affinity ) Tat-P-TEFb subcomplex ( Tahirov et al . , 2010 ) , Tat binds to a relatively open groove . In the context of the AFF4-P-TEFb structure , however , AFF4 creates an unanticipated pocket for Tat ( Figure 4D ) . This pocket may ultimately provide a suitable therapeutic target for the development of small-molecule inhibitors of Tat binding that selectively block HIV transcription . CDK9 ( 1–330 ) and cyclin T1 ( 1–264 ) were cloned into a modified pFastBac Dual donor plasmid , and HIV-1 Tat ( 1–86 ) was cloned into the pFastBac1 donor plasmid using the Bac-To-Bac system from Life Technologies ( Carlsbad , CA ) . CDK9 was cloned with a Tobacco Etch Virus ( TEV ) protease cleavable N-terminal His-tag while CycT1 and Tat remained un-tagged . Each virus genome was transfected into Sf9 cells to generate the baculovirus according to manufacturer's protocol . Each baculovirus was amplified ( 2× ) , plaque purified , and amplified ( 3× ) to obtain a stock with 108 plaque forming units ( PFU ) /ml . Test infections were screened for expression levels of the target proteins by Western blot analysis , and the highest expressing virus stock was used . For large-scale production of P-TEFb , 4 l of High5 cells at 1 × 106 cells/ml in ESF921 medium ( Expression Systems , Reno , NV ) were infected with 20 ml of virus stock per liter of culture . The flasks were incubated for 52 hr at 27°C on a rotary shaker . Cells were harvested by centrifugation for 30 min at 350xg in a Beckmann JLA-8 . 1 rotor , washed quickly in 50 mM Tris pH 7 . 5 , 150 mM NaCl , and centrifuged at 350×g for 10 min . The supernatant was removed , and the cell pellets were frozen in liquid nitrogen . For large-scale production of P-TEFb-Tat , insect cells were coinfected with 20 ml each of CDK9 , CycT1 , and Tat virus stocks . The remaining steps were the same as for production of P-TEFb . AFF42–73 was cloned into a modified pET28 plasmid . The recombinant protein includes a N-terminal TEV-protease-cleavable His-tag . The plasmid was transformed into Rosetta ( DE3 ) pLysS . 3 l of transformed Escherichia coli were grown at 37°C to OD600 = 0 . 6 , and expression was induced with 0 . 3 mM IPTG at 18°C for 16 hr . Cells were harvested , and pellets were frozen in liquid nitrogen . Pellets from 4 l infected High5 cells were resuspended in 75 ml lysis buffer ( 20 mM Na-HEPES pH 7 . 4 , 10 mM NaCl , 1 mM DTT ) with 1 × Roche Complete Protease Inhibitor and 0 . 5 mM AEBSF . Cells were lysed with a Dounce homogenizer . The lysate was brought up to 0 . 2 M NaCl by adding 3 . 0 ml 5 M NaCl , incubated on ice for 10 min , and centrifuged at 5800xg in a SS34 rotor . The supernatant was saved , and the pellet extracted again with 30 ml lysis buffer in the homogenizer . The supernatants of the two centrifugations were combined , cleared by centrifugation at 210 , 000×g in an ultracentrifuge , and filtered through a 0 . 8-μm syringe filter . The cleared lysate was loaded onto a 5 ml His-Trap HP column ( GE Healthcare , Piscataway , NJ ) equilibrated in buffer A ( 20 mM Na-HEPES pH 7 . 4 , 0 . 3 M NaCl , 10% glycerol , 1 mM DTT , 20 mM imidazole ) . After washing for 10 column volumes with buffer A + 1 M NaCl , followed by 10 column volumes of buffer A , P-TEFb was eluted with a gradient from 0% A to 100% B ( 20 mM Na-HEPES pH 7 . 4 , 0 . 3 M NaCl , 10% glycerol , 0 . 5 M imidazole , 1 mM DTT ) . The eluted P-TEFb was dialyzed for 3 hr against 2 L buffer A . TEV-protease was added to the protein at a 1:25 ( wt/wt ) ratio , and the digest was incubated for 1 hr at room temperature and 4°C overnight . The digest was loaded on a His-Trap HP column , and P-TEFb lacking the His tag eluted in the flow-through of the column , while undigested protein and TEV-protease eluted later in the imidazole gradient . The yield was ∼0 . 8 mg P-TEFb/L High5 cell culture . The Tat-P-TEFb complex was purified in a similar way . The cell lysate was purified over a 5 ml His-Trap HP column , dialyzed , and digested with TEV protease as described above . The digested complex was diluted with 1 . 1 volumes of 20 mM Na-HEPES pH 7 . 3 , 1 mM DTT , 1% β-octyl-glucoside to a final concentration of 0 . 14 M NaCl , and 0 . 5% β-octyl-glucoside and applied to a Source S anion exchange column equilibrated in 20 mM Na-HEPES pH 7 . 3 , 10% glycerol , and 1 mM DTT . The column was developed with a linear gradient to 20 mM Na-HEPES pH 7 . 3 , 0 . 75 M NaCl , 10% glycerol , and 1 mM DTT . P-TEFb-Tat eluted as single peak at about 0 . 25 M NaCl . To purify AFF42–73 , 30 g of E . coli cell pellet was resuspended in 125 ml lysis buffer ( 25 mM Tris/HCl pH 7 . 5 , 0 . 2 M NaCl , and 1 mM DTT ) . Lysozyme was added to 0 . 1 mg/ml final concentration and incubated for 30 min . After adding Roche Complete Protease Inhibitor without EDTA , 0 . 5 mM AEBSF , and DNaseI ( 5 units/ml ) , the cells were lysed by sonication . The lysate was centrifuged for 1 hr at 17 , 000 rpm in a SS34 rotor . The supernatant was filtered through a 0 . 8-μm syringe filter and applied to a 5 ml His-Trap column . The protein was purified as described for P-TEFb . The separately purified P-TEFb and AFF42–73 were combined at a 1:1 . 4 ( mol/mol ) ratio , concentrated to 0 . 6 ml , and injected onto an analytical Superdex S200 gel exclusion column equilibrated with 25 mM Na-HEPES pH 7 . 4 , 0 . 2 M NaCl , and 1 mM DTT . The center fractions of the eluted three-protein peak were used for crystallization . The purified AFF4-P-TEFb complex was concentrated to 10 mg/ml using Millipore Ultrafree centrifugal devices . Crystals were grown from 1 . 0 µl protein combined with 0 . 5 µl silver bullet condition 30 ( Hampton Research , Aliso Viejo , CA; 0 . 33% wt/vol Gly-Phe , 0 . 33% wt/vol Gly-Tyr , and 0 . 33% wt/vol Leu-Gly-Gly in 20 mM Na-HEPES pH 6 . 8 ) and 0 . 5 µl reservoir solution equilibrated against 16–18% PEG 3350 , 0 . 1 M Na-HEPES pH 7 . 0 . After equilibrating for 24 hr , diluted microseeds from previous crystallization experiments were added with a hair . Seeding produced large single crystals ( 0 . 3 × 0 . 2 × 0 . 2 mm ) with and without 1 mM AMPPNP , 5 mM MgCl2 . The presence of all three proteins in the crystals was confirmed by gel electrophoresis and mass spectrometry of dissolved crystals . Crystals were soaked in cryoprotectant ( 25% PEG 3350 , 0 . 1 M Na-HEPES pH 7 . 0 , 10% glycerol , 1:4 silver bullet 30 ) and flash frozen in liquid nitrogen . X-ray data were collected at Beamline 8 . 3 . 1 at the Advanced Light Source at the Lawrence Berkeley National Laboratory ( MacDowell et al . , 2004 ) . The best data were collected from a crystal that was grown in the presence of 1 mM AMPPNP and 5 mM MgCl2 . The reflections were processed using HKL2000 ( Otwinowski and Minor , 1997 ) ( Table 1 ) . The structure was determined by molecular replacement with PHENIX ( Adams et al . , 2010 ) using P-TEFb from the P-TEFb-Tat complex ( PDB ID 3MI9 ) as the search model . The asymmetric unit contains three complexes . Initial refinement using PHENIX was performed with model restraints , as well as noncrystallographic symmetry restraints . Model restraints were removed in later stages of refinement . AFF4 was built manually and the model was adjusted using Coot ( Emsley and Cowtan , 2004 ) . The model was refined using gradient minimization with weight optimization and maximum-likelihood targets , TLS refinement , and individual atomic B-factor refinement . The model was checked against composite omit maps . Density was missing for residues 1–7 and 88–95 in CDK9 mol1 and mol3 , and residues 1–7 and 89–97 in CDK9 mol2 . Density also was absent for residues 1–7 and 253–264 in CycT1 mol1 , residues 1–7 and 257–264 in CycT1 mol2 and mol3 , and residues 2–33 and 67–73 in AFF4 , mol1 and mol3 . For AFF4 in molecule 2 residues 2 , 22–33 , and 67–73 are missing . The register of the AFF4 sequence in the electron density was confirmed by locating the Se atoms in crystals grown with SeMet-labeled AFF4 . AFF42−73 was labeled with SeMet ( Van Duyne et al . , 1993 ) and purified as described above . Optimized crystallization conditions were seeded with unlabeled microcrystals . Anomalous differences were calculated from data collected at 12 , 657 eV , the peak of Se fluorescence measured from the crystal . Buried surface areas were calculated with PISA ( Krissinel and Henrick , 2007 ) . Surface pockets were identified with the program DoGSiteScorer ( Volkamer et al . , 2012 ) . Figures of molecular structures were prepared with PyMOL Version 1 . 5 ( Schroedinger LLC , New York , NY ) . Protein binding was measured using fluorescence anisotropy of a 36-residue segment of AFF4 ( residues 32–67 ) encompassing the protein–protein contacts in the crystal structure . The AFF4 peptide was synthesized at the University of Utah DNA/Peptide Facility using the following sequence: C-FAM-GABA-SPLFAEPYKVTSKEDKLSSRIQSMLGNYDEMKDFIG-amide where FAM indicates 5-carboxyfluoroscein and GABA indicates a γ-amino-butyric acid spacer . Varying amounts of purified CycT1 , P-TEFb , and Tat-P-TEFb were incubated for 30 min with 5 nM labeled peptide at room temperature in the dark in 25 mM HEPES pH 7 . 5 , 100 mM NaCl , 10% glycerol , 0 . 1% NP40 , and 0 . 5 mM Tris ( 2-carboxyethyl ) phosphine ( TCEP ) . Competition titration experiments of unlabeled peptides were performed using 75 nM CycT1 , 45 nM P-TEFb , and 6 nM Tat-P-TEFb in 25 mM HEPES pH 7 . 5 , 100 mM NaCl , 10% glycerol , 0 . 1% NP40 , 0 . 5 mM TCEP , and 5 nM fluorescent peptide . Fluorescence anisotropy was measured using a Victor 3V ( Perkin Elmer ) multi-label plate reader . Data points represent the average of six independent measurements . Binding curves were fit to the single-site binding equation using Prism version 5 . 0c ( Graphpad Software ) . HeLa cells were cultured in Dulbecco's modified Eagle's medium supplemented with 10% FBS at 37°C in a humidified atmosphere with 5% CO2 . Cells were seeded at 5 × 105 cells/ml in TC-treated 96-well plates one day prior to plasmid transfection using the 25-kDa linear polyethyleneimine reagent ( Sigma-Aldrich , St . Louis , MO ) . Cells were cotransfected with 100 ng of an HIV-LTR firefly luciferase reporter construct ( He et al . , 2010 ) and 350 ng of pCDNA3 . 1 containing AFF4 variants with a C-terminal 3× FLAG tag . Following stimulation for 48 hr with the indicated ligands , the cells were lysed in passive lysis buffer ( Promega , Fitchburg , WI ) for 5 min at 25°C . The cell lysates were incubated with firefly luciferase substrate , and luminescence was measured on a SpectraMax L microplate reader ( Molecular Devices , Sunnyvale , CA ) . The relative luminescence was normalized to the concentration of AFF4 in the cell determined by Western blotting using an anti-FLAG primary antibody . HeLa cells were seeded in 10-cm TC-treated plates at 2 . 2 × 105 cells/ml , Incubated for 24 hr and transfected with pCDNA3 . 1 ( 1 μg ) encoding an AFF4 variant with a C-terminal 3× FLAG tag . After incubating for 2 days , cells were collected in PBS , washed , and resuspended in hypotonic buffer ( 20 mM Tris-HCl pH 7 . 4 , 10 mM NaCl , and 3 mM MgCl2 ) . After 15 min on ice , 0 . 4% Triton X-100 was added , and the cell suspension was mixed and centrifuged at 3000×g for 10 min . The pellet was resuspended in nuclear extraction buffer ( 100 mM Tris-HCl pH 7 . 4 , 100 mM NaCl , 1% Triton X-100 , 1 mM EDTA , 10% glycerol , 0 . 7% Tween 20 , and protease inhibitors [AEBSF , leupeptin and E64] ) for 30 min on ice followed by centrifugation at 14 , 000×g for 30 min to yield nuclear extract . Anti-FLAG agarose beads ( Sigma-Aldrich ) were incubated in the nuclear extract at 4°C for 2 hr and washed with nuclear extraction buffer . The beads were eluted with 0 . 1 M glycine-HCl pH 3 . 5 , and the neutralized eluate was analyzed by Western blotting with the indicated antibodies . Kinase assays were performed in LoBind tubes ( Eppendorf ) in 20 μl reactions containing 50 mM HEPES pH 7 . 3 , 50 mM NaCl , 1 mM DTT , 10 mM MgCl2 , and 0 . 05 mM or 0 . 5 mM ATP . For assays with 0 . 05 mM ATP , 0 . 15 μCi [32P]-γ-ATP was used , and for reactions with 0 . 5 mM ATP , 1 . 5 μCi [32P]-γ-ATP was used . Reactions in conventional tubes gave distinct less reproducible results . Purified recombinant P-TEFb or P-TEFb-Tat was pre-incubated with 500 ng purified recombinant GST-CTD ( 52 C-terminal domain repeats from human RNA polymerase ) in the absence or presence of 0 . 3 µM purified AFF42–73 for 15 min at 20°C . After addition of ATP , the kinase reactions were stopped at different times by addition of 5 μl of 5× SDS sample buffer . The samples were analyzed by SDS-polyacrylamide gel electrophoresis , followed by measurement of the radioactive protein bands on a Typhoon phosphorimager ( GE Healthcare ) . For Western blots of kinase reactions performed with non-radioactive ATP , 0 . 2 µg of GST-CTD from each reaction was loaded per lane of a 4–20% SDS-PAGE gel , transferred onto a PVDF membrane ( Immobilon-FL; Millipore , Billerica , MA ) and processed by standard Western Blot procedures . The primary antibodies were ab5095 ( αpS2; Abcam ) and ab5131 ( αpS5; Abcam ) at 1:1000 dilution for both . The secondary antibody was a fluorescently labeled goat anti-rabbit antibody ( Odyssey; LI-COR Biosciences , Lincoln , NE ) at 1:20 , 000 dilution .	The rates at which many genes are expressed as proteins are limited by the efficiency of a process called transcriptional elongation . This process takes place as the stretch of DNA that defines the gene is transcribed into an RNA molecule and it is catalyzed by an enzyme called RNA polymerase II . However , this enzyme can become trapped , and another enzyme called P-TEFb ( positive transcription elongation factor b ) is needed to release it . P-TEFb and other elongation factors therefore have an important role in gene expression . The human immunodeficiency virus ( HIV ) is a retrovirus that hijacks the gene expression processes in human immune cells to replicate the RNA genome of the virus . To do this , the virus produces a protein called Tat that recruits P-TEFb as part of a multi-protein machine called the super elongation complex . This ensures that the process of transcriptional elongation , and hence the overall replication process , is highly efficient . There are gaps , however , in our knowledge of the architecture of the super elongation complex , which is known to be organized on a flexible scaffold . In turn , the molecular basis for the interaction between HIV-1 Tat and P-TEFb within the super elongation complex is not well understood . Now Schulze-Gahmen et al . show that only one of the two subunits in P-TEFb—a cyclin known as CycT1—binds to the AFF4 scaffold protein in the super elongation complex . In addition to assisting with the expression of hundreds of human genes , super elongation complexes containing P-TEFb-AFF4 are hijacked in various forms of cancer and viral infections , including HIV/AIDS . Schulze-Gahmen et al . show that AFF4 can directly contact HIV-1 Tat , which binds to the P-TEFb-AFF4 complex much more strongly than it binds to P-TEFb alone . This suggests that HIV-1 Tat evolved to work within the super elongation complex . Moreover , Schulze-Gahmen et al . reveal that HIV-1 Tat binds to a cleft between the P-TEFb enzyme and the AFF4 protein , which raises the possibility that this cleft could be used as a target for anti-HIV/AIDS drugs .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "structural", "biology", "and", "molecular", "biophysics" ]	2013	The AFF4 scaffold binds human P-TEFb adjacent to HIV Tat
Cyanobacteria were responsible for the oxygenation of the ancient atmosphere; however , the evolution of this phylum is enigmatic , as relatives have not been characterized . Here we use whole genome reconstruction of human fecal and subsurface aquifer metagenomic samples to obtain complete genomes for members of a new candidate phylum sibling to Cyanobacteria , for which we propose the designation ‘Melainabacteria’ . Metabolic analysis suggests that the ancestors to both lineages were non-photosynthetic , anaerobic , motile , and obligately fermentative . Cyanobacterial light sensing may have been facilitated by regulators present in the ancestor of these lineages . The subsurface organism has the capacity for nitrogen fixation using a nitrogenase distinct from that in Cyanobacteria , suggesting nitrogen fixation evolved separately in the two lineages . We hypothesize that Cyanobacteria split from Melainabacteria prior or due to the acquisition of oxygenic photosynthesis . Melainabacteria remained in anoxic zones and differentiated by niche adaptation , including for symbiosis in the mammalian gut . Among the geochemical changes that have occurred over the past few billion years , perhaps the most dramatic was the transformation of the Earth’s atmosphere and upper oceans into oxygen-rich environments ( Bekker et al . , 2004 ) . Cyanobacteria are presumed responsible for this geochemical revolution , as they comprise the sole lineage known to have innovated the production of oxygen as a byproduct of photosynthesis ( Mulkidjanian et al . , 2006 ) . Further , nitrogen fixation by Cyanobacteria is central to the Earth’s nitrogen cycle ( Vitousek et al . , 2002 ) . Cyanobacteria are inferred to be one of the earliest branching bacterial lineages ( Altermann and Kazmierczak , 2003; Bekker et al . , 2004 ) and have diversified across environments—land , fresh , and salt water , and all levels of the photic zone ( Dworkin , 2006 ) . Via endosymbiosis , Cyanobacteria became the chloroplasts of plants ( Sagan , 1967 ) , a role that underlines their broad evolutionary importance . The evolutionary history of the Cyanobacteria phylum is only partially resolved: no related taxa , from which a common ancestor could be inferred , have been described . However , recent culture-independent 16S rRNA gene surveys of microbial communities have revealed a novel clade sibling and basal to Cyanobacteria ( Ley et al . , 2005 ) . The presence of organisms related to Cyanobacteria in the gut is notable because they are widely shared across individuals ( Consortium HMP , 2012 ) , where members of this group can comprise up to 20% of the total sequences recovered from stool ( Dethlefsen and Relman , 2011 ) , as well as shared across various mammalian species ( Ley et al . , 2008 ) . As Cyanobacteria are photosynthetic organisms , it has been assumed that these sequences represent genomic material derived from ingestion of chloroplasts or Cyanobacterial cells ( Turnbaugh et al . , 2009; Koenig et al . , 2011; Consortium HMP , 2012 ) . However , given their large evolutionary separation from Cyanobacteria and the lack of cultured representatives , no conclusion as to the roles of these predominant organisms of the human gut has been possible . Related bacteria deep-branching from the Cyanobacteria have also been detected in water and other anoxic environments including sediments ( Ley et al . , 2005 ) . An earlier phylogenetic reconstruction based on full-length 16S rRNA gene sequences indicated that this water-soil-sediment-derived clade is distinct from that made up entirely of gut-derived sequences ( Ley et al . , 2005 ) . This mapping of habitats onto the two clades suggested that niche adaptation had shaped the groups’ evolution and hence their phylogeny , but beyond this observation the lineage remained enigmatic . New sequencing methods and bioinformatics advances provide a route for genomic analysis of uncultured organisms from complex microbial communities ( Dick et al . , 2009; Iverson et al . , 2012; Wrighton et al . , 2012 ) . We show that these methods can yield complete and near-complete genomes from relatively low abundance organisms , without the need for single cell genomic approaches . Here , we analyze eight curated genomes from bacteria from intestinal samples and an aquifer sediment to evaluate their metabolisms and roles in their respective habitats . The analysis provides clues as to the ancestral state of the lineage that gave rise to these organisms and to the Cyanobacteria . We generated metagenomes from three fecal samples obtained from three healthy adult humans ( termed A , B , and C; Table 1 ) . In addition , we identified genome fragments that derived from an organism deeply branching with respect to Cyanobacteria in a microbial community metagenomic dataset from a subsurface aquifer ( Wrighton et al . , 2012 ) . The microbial communities in the human fecal samples and the subsurface differ substantially . The bacterial communities of human fecal samples A , B , and C are typical of human fecal microbiota , as they are predominantly composed of members of the Firmicutes and Bacteroidetes , with a few other phyla represented ( Figure 1A–C ) ( Consortium HMP , 2012 ) . The subsurface sample , on the other hand , has a greater phylum-level phylogenetic diversity with the most abundant members belonging to Proteobacteria and the candidate phyla OD1 and OP11 ( Figure 1D ) . For both sample types , the abundances of genomes from the Cyanobacteria sibling clade were less than 5% of the total community . 10 . 7554/eLife . 01102 . 003Table 1 . Samples from which Melainabacteria genomes were recoveredDOI: http://dx . doi . org/10 . 7554/eLife . 01102 . 003SampleEnvironment# of readsAbundance%# genomes recoveredAGut109 , 557 , 61641 Complete , 1 PartialBGut124 , 163 , 24832 CompleteCGut112 , 578 , 26421 Complete , 1 Near Complete , 1 PartialACDAquifer232 , 878 , 9790 . 71 Near CompleteSample , number of reads sequenced , and estimates of the abundance of Melainabacteria in the communities based on 16S rRNA gene survey and coverage information . ACD20 was assembled from three samples ( see Wrighton et al . , 2012 ) . 10 . 7554/eLife . 01102 . 004Figure 1 . Community composition of samples containing Melainabacteria . ( A–C ) The relative composition of the human fecal samples A , B , and C , and ( D ) the aquifer community members . In A and D estimated percent relative abundance of the community is plotted , and in B and C , coverage is plotted , but estimated percent relative abundance is noted on the figure for select members . Organisms are classified at the phylum level . The human fecal sample A community is dominated by Prevotella copri DSM 18205 , which accounts for more than 40% of the sequencing reads and is represented by several strains . Sequencing depth was not sufficient for human fecal sample C to accurately estimate roughly 25% of the community abundance , which includes MEL . C3 . Aspects of the community composition of the aquifer sample are discussed in Wrighton et al . ( 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01102 . 004 Despite the relatively low abundance of these genomes in the samples ( Table 1 ) , recently developed algorithms that improve the assembly and manual curation of metagenomic data ( Sharon et al . , 2013 ) allowed us to recover two genomes from sample A ( MEL . A1 , MEL . A2 ) , two from sample B ( MEL . B1 , MEL . B2 ) , and three genomes from sample C ( MEL . C1 , MEL . C2 , MEL . C3 ) for a total of seven distinct genomes reconstructed from human fecal samples ( Tables 1 and 2 ) . Through genome curation , we were able to establish linkage among all scaffolds for four of these genomes ( complete genomes; Table 2 ) . Completeness was confirmed by validating assembly graph connectivity , and also by considering expected genome features such as single copy genes . Correctness was confirmed by re-assembly of potentially mis-assembled regions such as scaffold ends , and by considering the ‘phylogenetic profile’ of genes in each scaffold . Our curation method verified unique paired read placement throughout the reconstructed genomes , a requirement consistent with standard methods of isolate genomics . All scaffolds identified as deriving from an organism with some similarity to Cyanobacteria , based on the phylogenetic profile of the encoded genes , were incorporated into the closed , complete genomes . Additional small scaffolds were identified and incorporated using paired read placement . The phylogenetic signal for novelty was robust , because essentially all other genomic fragments ( excluding phage and plasmids ) shared high similarity with genomes of previously sequenced organisms . 10 . 7554/eLife . 01102 . 005Table 2 . Melainabacteria genomes recovered in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 01102 . 005Sample IDCoverageGenome statusSize ( bp ) %GCScaffoldsN50Coding features16S rRNA genesACD2030xNear Complete2 , 979 , 54833 . 519133 , 3612 , 819NDMEL . A173xComplete1 , 867 , 33632 . 911 , 867 , 3361 , 8322MEL . A25 . 5xPartial1 , 192 , 45530 . 68816 , 6131 , 386NDMEL . B162xComplete2 , 302 , 30735 . 321542 , 1172 , 2192MEL . B244xComplete2 , 308 , 20536 . 326375 , 3762 , 2222MEL . C126 . 5xComplete2 , 053 , 64234 . 141 , 742 , 0552 , 1202MEL . C227 . 5xNear Complete2 , 159 , 32735 . 334146 , 2322 , 1042MEL . C36xPartial1 , 323 , 47829 . 99315 , 8781 , 472NDND = not determined . See the section Genome assembly in ‘Materials and methods’ for an explanation of Genome Status . The assembled genomes range from 1 . 9 to 2 . 3 Mbp and encode 1 , 800 to 2 , 230 genes . Additionally , we analyzed the binned genome , hereafter , ACD20 , ( Tables 1 and 2 ) from the aquifer dataset ( Wrighton et al . , 2012 ) . The ACD20 genome is larger than the genomes recovered from fecal samples—3 . 0 Mbp encoding 2 , 819 genes . Additional genome details are provided in Tables 1 and 2 . We used all eight genomes for phylogenetic analyses and four representative genomes ( three from the gut plus the sediment genome ) for the metabolic analyses that follow . Corroborating earlier findings ( Ley et al . , 2005 ) , a 16S rRNA gene sequence-based phylogeny built with publically available sequences places the unknown lineages , represented in part by the gut and aquifer bacteria , basal to photosynthetic Cyanobacteria ( Figure 2A ) . Three subgroups are revealed , one of which comprises sequences obtained from animal guts ( Figure 2B ) . The 16S rRNA gene sequences of the gut and aquifer bacteria share no more than 84% identity to Cyanobacterial sequences , consistent with placement of these organisms in a new candidate phylum ( <85% identity , as suggested previously [Hugenholtz et al . , 1998] ) . The bacterial tree has been described as a polytomy due to the inability of 16S rRNA gene phylogenies to capture any specific branching order for the phyla ( Pace , 1997 ) , so this phylum is unusual in its robustly supported relationship to Cyanobacteria . To further substantiate this evolutionary relationship , we constructed a phylogeny of concatenated ribosomal protein sequences ( Figure 3A , B ) . The result shows that the eight new genomes form a monophyletic lineage that branches deeply from the Cyanobacterial lineage , with ACD20 basal to the group ( Figure 3B ) . Importantly , a common ancestor for these organisms and photosynthetic Cyanobacteria is well supported ( 100/100 bootstrap bipartitions ) in both trees . With the sum of this evidence , we designate these bacteria as the new candidate phylum Melainabacteria , where ‘melaina’ refers to the Greek nymph of dark waters . 10 . 7554/eLife . 01102 . 006Figure 2 . 16S rRNA gene phylogeny of Melainabacteria and Cyanobacteria . Trees were built using 16S rRNA gene sequences from MEL . A1 , MEL . B1 , and MEL . B2 ( the 16S rRNA sequence of ACD20 was not recovered ) . ( A ) 16S rRNA gene phylogeny tree with five representative sequences from each phylum obtained from the Greengenes May 2011 database ( DeSantis et al . , 2006 ) . Bootstrap values greater than 50% are indicated . ( B ) 16S rRNA gene phylogeny built using one representative sequence from each order within Cyanobacteria from the Greengenes database ( May 2011 ) ( DeSantis et al . , 2006 ) besides orders YS2 , SM1D11 , and mle1-12 from which all sequences were used . For Melainabacteria , the habitats from which the sequences were predominantly derived are indicated and colored according to isolation source ( blue = environmental ( non-gut ) ; brown = gut ) . Cyanobacteria are displayed in green . Bootstrap values greater than 70% are indicated by a black square . DOI: http://dx . doi . org/10 . 7554/eLife . 01102 . 00610 . 7554/eLife . 01102 . 007Figure 3 . Concatenated ribosomal protein phylogeny of the Melainabacteria and Cyanobacteria . Maximum likelihood phylogeny and trait-based comparison of the eight novel organisms and 80 Cyanobacteria based on a concatenated protein alignment of 16 core ribosomal proteins from 733 taxa . In ( A ) the complete tree is shown at the phylum level and in ( B ) only the cyanobacterial-melainabacterial portion of the tree is shown . Bootstrap values >50% are indicated . Cyanobacteria branches are colored blue and Melainabacteria branches , red . The complete tree with all taxa shown is provided in Figure 3—figure supplement 1 . The protein alignment on which this tree is based is provided in Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 01102 . 00710 . 7554/eLife . 01102 . 008Figure 3—source data 1 . Concatenated protein alignment of 16 core ribosomal proteins from 733 taxa and the eight Melainabacteria described here . DOI: http://dx . doi . org/10 . 7554/eLife . 01102 . 00810 . 7554/eLife . 01102 . 009Figure 3—figure supplement 1 . Complete phylogeny of 733 taxa and the eight Melainabacteria based on a concatenated protein alignment of 16 core ribosomal proteins . Melainabacteria branches are shown in red and Cyanobacteria branches in blue . Bootstrap values >50% are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 01102 . 009 The newly described melainabacterial genomes contain other genes that reinforce an ancestry shared with Cyanobacteria . For instance , the genomes encode the 30S ribosomal protein S1 rpsA gene , rather than the homolog ypfD , which is exclusive to the Firmicutes ( Danchin , 2009 ) . Moreover , two of the gut genomes ( MEL . B1 and MEL . B2 ) encode the A type of RnpB ( E-values 5 . 676e-59 , 2 . 235e-60 respectively ) , which is found in all Bacteria except the Firmicutes and Tenericutes ( Haas et al . , 1996; Zwieb et al . , 2011 ) . Three of the four complete genomes also have homologs to the S-layer like COG , CyOG00138 ( e . g . , Anabaena variabilis: Q3MBT3 ) , a protein found only in Cyanobacteria ( Mulkidjanian et al . , 2006 ) . Although these genes are not exclusive to Cyanobacteria , they are important phylogenetic markers , support a shared ancestry with Cyanobacteria , and refute an ancestry with the Firmicutes , a possibility that arises when considering the metabolic genes of the Melainabacteria genomes ( see below ) . Oxygenic photosynthesis is perhaps the most exceptional characteristic of Cyanobacteria , as all other bacterial photosyntheses are anoxygenic . The Melainabacteria genomes appear to entirely lack genes for photosynthesis ( photosystem I and photosystem II , thylakoid membranes , succinate dehydrogenase , and the cytochrome b6f complex ) , indicating that none have the capacity for oxygenic or anoxygenic photosynthesis . Also absent are genes for soluble ( e . g . , plastocyanin or photosynthetic ferredoxin ) and membrane-affiliated electron carrier proteins ( e . g . , cytochromes , quinones , Fe-S , or flavin ) . Additionally , we found no genes for aerobic respiration including terminal quinol and cytochrome c oxidases , terminal reductases involved in anaerobic respiration ( e . g . , fumarate , nitrate ) , or carbon fixation pathways . Together these findings suggest that these members of the Melainabacteria are not capable of phototrophy or respiratory metabolism . Despite being non-photosynthetic , the melainabacterial genomes encode homologs of the circadian rhythm regulators RpaA and RpaB and the high intensity light sensor NblS . The histidine kinase NblS in Cyanobacteria preserves the photosynthetic machinery by regulating its expression and degradation under high and blue/UV-A conditions ( van Waasbergen et al . , 2002 ) . RpaA and RpaB are found in all Cyanobacteria ( Mulkidjanian et al . , 2006 ) , where they regulate the circadian clock KaiABC ( Hanaoka et al . , 2012 ) and link energy transfer between the antennae and the photosystem ( Ashby and Mullineaux , 1999 ) . The four complete genomes of Melainabacteria lack SasA , which typically functions as the sensor to the response regulators RpaA and RpaB ( Hanaoka et al . , 2012 ) , as well as any photosynthetic machinery , which suggests that these proteins may have another function . Similarly , the NblS homolog in the gut and aquifer Melainabacteria most likely has a different function . The current KEGG database ( Kanehisa et al . , 2012 ) indicates that RpaA and NblS are exclusive to Cyanobacteria . RpaB is likely also exclusive to Cyanobacteria; however , given this gene’s similarity to other response regulators it could be considered present in two other bacterial genomes , Variovorax paradoxus and Desulfomicrobium baculatum . In all cases , the three Melainabacteria genes have highest homology to those present in Cyanobacteria ( Supplementary file 1 ) . Like Cyanobacteria , Melainabacteria are inferred to have a Gram-negative cell envelope . This conclusion is based on the large number of genes for lipopolysaccharide ( LPS ) , Lipid A biosynthesis , and O-antigen polymerases and transporters found in the curated genomes described here ( Supplementary file 1 ) . LPS , Lipid A , and O-antigen are components of the Gram-negative outer membrane; in contrast , Gram-positive bacteria lack an outer membrane and therefore lack these structures ( Purves et al . , 2003 ) . This finding is significant , because many genes in the newly reconstructed genomes share closest sequence similarity to Gram-positive members of the Firmicutes ( see below ) . As previously mentioned , the genomes encode a homolog to an S-layer-like protein , suggesting that the cell envelope has an S-layer . S-layers and O-antigens have both been previously observed on photosynthetic Cyanobacteria ( Hoiczyk and Hansel , 2000 ) . Taken together , the cell envelope is likely similar to that of Cyanobacteria , consistent with a shared vertical ancestry . Based on the lack of a linked electron transport chain , aerobic or anaerobic respiratory complexes ( see above ) , and a complete TCA cycle ( see below ) , and the presence of fermentative and degradative enzymes , we infer that Melainabacteria are obligate anaerobic fermenters . We predict that Melainabacteria can use a wide variety of carbon compounds , including hemicellulosic compounds ( only ACD20 ) , polysaccharides , oligosaccharides , and simple sugars , as well as organic acids , amino acids , and fatty acids to yield hydrogen , lactate , acetate ( ACD20 ) , formate ( Gut ) , hydrogen , possibly butyrate ( ACD20 ) and ethanol ( Gut ) ( Figure 4 ) . Specific sugars predicted to be fermented are glucose , fructose , sorbitol , mannose , trehalose , starch , glycogen , hemicellulose , and amylose , and the relevant enzymes are α-galactosidase , β-galactosidase , α-glucosidase , β-glucosidase , β-glucuronidase , β-fructofranosidase ( sucrase ) , α-mannosidase , pullulanase , α-amylase , and endo-1 , 4-beta-xylanase . These enzymes facilitate the utilization of a variety of sugar compounds by degrading the compounds into simpler sugars that can enter the main Embden-Meyerhof-Parnas ( EMP ) glycolytic pathway . This pathway contains not the classical ATP-dependent enzyme , but a pyrophosphate-dependent phosphofructokinase , a gene found in diverse organisms capable of anaerobic glycolysis ( Mertens , 1991 ) . This difference in phosphoryl donor specificity may confer an energetic advantage to the cell when glycolysis is the primary source of ATP ( Figure 4 , gene 3 ) . The genomes have the genes necessary for hexose interconversion in the EMP pathway via the pentose phosphate pathway , such that ribose , arabinose , xylulose , and other five-carbon sugar or sugar-alcohols may be utilized . Unlike most Cyanobacteria , which use internal carbon pools for fermentation ( Stal and Moezelaar , 1997 ) , Melainabacteria likely acquire sugars and sugar-alcohols from the environment via a variety of cytoplasmic membrane permeases ( Figure 4 ) . Overall , the cells are inferred to have a lifestyle analogous to anaerobic obligate fermentative bacteria known to play an important role in carbon transformation in gut ( Mackie , 2002 ) and subsurface systems . 10 . 7554/eLife . 01102 . 010Figure 4 . The physiological and metabolic landscape of Melainabacteria . Metabolic predictions for Melainabacteria based on genes identified in Figure 4—source data 1 . Genes in pathways detected in the genomes of the subsurface and at least one gut genome ( white box ) , only in the subsurface genome ( grey box ) , only in at least one gut genome ( orange box ) , and genes missing from pathways in all genomes ( red box ) . Glycolysis proceeds via the canonical Embden-Meyerhof-Parnas ( EMP ) pathway with the exception of fructose-6-phosphate 1-phosphotransferase ( EC:2 . 7 . 1 . 90 , gene 3 ) . Names of pathways and fermentation end-products are bolded and ATP generated by substrate-level phosphorylation are noted . All Melainabacteria genomes sampled lack electron transport chain components ( including cytochromes ( Cyto ) , succinate dehydrogenase ( sdh ) , flavins , quinones ) , terminal respiratory oxidases or reductases , and photosystem I or II ( PS1 , PS2 ) . The genomes also lack a complete TCA cycle ( absent enzymes noted by red boxes ) , with the TCA enzymes instead linked to the fermentation of amino acids and organic acids denoted ( pathways , blue arrows ) . Ferredoxin ( Fd , green text ) is important for hydrogen ( H2 ) production via hydrogenases ( yellow background box ) . Proton translocation mechanisms ( green background box ) may be achieved by the activity of trimeric oxaloacetate ( OAA ) decarboxylase and sodium-hydrogen antiporter , pyrophosphate ( PPi ) hydrolysis with pyrophosphatases , 11 subunit NADH dehydrogenase , and an annotated NiFe hydrogenase ( green enzyme ) . Annotations for the gene numbers are in Figure 4—source data 2 . The complete metabolic comparison of the Melainabacteria can be accessed at http://ggkbase . berkeley . edu/genome_summaries/81-MEL-Metabolic-Overview-June2013 . DOI: http://dx . doi . org/10 . 7554/eLife . 01102 . 01010 . 7554/eLife . 01102 . 011Figure 4—source data 1 . Examination of enzymes ( steps ) in near-complete KEGG based modules shared among or unique to subsurface ACD20 and gut Melainabacteria genomes MEL . A1 , MEL . B1 , and MEL . B2 . Analysis is based on the KEGG Module database ( Kanehisa and Goto , 2000 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01102 . 01110 . 7554/eLife . 01102 . 012Figure 4—source data 2 . Gene annotations corresponding to the numbers in Figure 4 . If the gene occurs in both the ACD20 and gut genomes , the reported annotation is based on ACD20 . DOI: http://dx . doi . org/10 . 7554/eLife . 01102 . 012 Nearly 30% of the 920 core conserved orthologous genes ( COGs ) match most closely to genes belonging to members of the phylum Firmicutes that have a fermentative-based metabolism , compared to 15% with closest matches to cyanobacterial genes ( Figure 5A ) . These non-cyanobacterial genes are spread throughout the genomes ( Figure 5—figure supplement 1 ) , arguing against acquisition via a recent lateral transfer event or a chimeric assembly artifact . The COGs whose best match are to COGs of Firmicutes are enriched in functions related to metabolism ( Figure 5B ) , including metabolism of carbohydrates , amino sugars , nucleotides , amino acids , and vitamins . It should be noted that while the best match to these genes are within the Firmicutes phylum , the abundance of Firmicutes genomes in databases may have inherently biased this result . These results corroborate extensive divergence in the metabolic lifestyles of Cyanobacteria and Melainabacteria . 10 . 7554/eLife . 01102 . 013Figure 5 . MEL-COG phylum and functional assignments . ( A ) Assignment of the 920 MEL-COGs ( Figure 5—source data 1 ) to their best matching phyla . ( B ) Functional assignment of COGs by phylum assignment . Only COGs with functional assignments were considered . Number of MEL-COGs with no/multiple functional assignments are 532/42 , 136/12 , and 87/10 for All , Firmicutes , and Cyanobacteria , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 01102 . 01310 . 7554/eLife . 01102 . 014Figure 5—source data 1 . List of 920 MEL-COGs , including their assigned phylum and KEGG Orthology ( KO ) identifier . DOI: http://dx . doi . org/10 . 7554/eLife . 01102 . 01410 . 7554/eLife . 01102 . 015Figure 5—figure supplement 1 . Distribution of MEL-COGs with best hits from different phyla across the MEL-A1 genome . Gene distribution across the genome does not support large-scale recombination , lateral transfer events , or a chimeric genome assembly accounting for the presence of genes with greater similarity to genes from phyla other than Cyanobacteria . DOI: http://dx . doi . org/10 . 7554/eLife . 01102 . 015 Both gut and subsurface Melainabacteria studied here have trimeric confurcating FeFe hydrogenases with motifs required for catalytic activity ( Figure 4 , Supplementary file 1 ) . One of the three confurcating hydrogenases in ACD20 , ACD20_18461 , has all the necessary residues for functionality ( L1 , L2 , and L3 motifs ) , while the other two , ACD20_9246_G0007 and ACD20_9246_G0010 , contain all three motifs but have a replacement of a serine by cysteine in motif 1 ( TSCSPGW rather than TSCCPAW ) and have replaced the cysteine in motif 3 with isoleucine . The gut Melainabacteria FeFe hydrogenase has a complete L1 motif but lacks L2 and L3 motifs . These hydrogenases may indicate an ecological role in H2 production in both gut and subsurface systems: syntenic homologs for the three subunits have been identified in genomes of obligate syntrophs and fermenters known to produce high molar ratios of H2 ( Sieber et al . , 2012 ) . The production of H2 typically requires a syntrophic association with an H2-consuming partner to maintain low partial hydrogen pressures . Therefore , in addition to being anaerobic fermenters , the Melainabacteria may be H2 producers living in syntrophy with archaeal methanogens or bacterial acetogens in the human gut and with respiring organisms in the subsurface . With Prochlorococcus , marine Synechococcus , and UCYN-A as exceptions , the vast majority of Cyanobacteria have a complete TCA cycle ( Zhang and Bryant , 2011 ) . In contrast , the Melainabacteria genomes reported here encode no more than four unlinked genes from the TCA and reverse TCA cycles ( Figure 4 ) . We also confirmed that the genomes lack alternative enzymes first identified in Synechococcus species PCC 7002 that functionally complete the TCA cycle ( Zhang and Bryant , 2011 ) . Moreover , the genomes have an NAD+ not the NADP+ dependent isocitrate dehydrogenase found in most Cyanobacteria . The absence of a complete TCA cycle necessitates an external requirement for dicarboxylic acids , which could be imported by the two dicarboxylic acid transporters found in each genome . This observation , along with the need for an H2-consuming partner , could guide development of growth media for future isolation efforts . We infer that the TCA cycle enzymes in the genomes reported here function to link to nitrogen metabolism and energy generating pathways . Both the ACD20 and MEL . B2 genomes encode isocitrate dehydrogenase and fumarase , whose end-products are important intermediates in nitrogen assimilation and amino acid pathways . Unlike the genomes of the gut bacteria , ACD20 can augment glycolytic ATP substrate-level phosphorylation with additional ATP generation using TCA cycle intermediates . For example , unidirectional aspartate ammonia-lyase , fumarase , and malic enzyme can convert amino acids ( alanine and aspartate ) and organic acids to pyruvate and ultimately acetate with ATP generation . Also , like citrate fermentation in Klebsiella pneuemoniae ( Dimroth , 1980; Dimroth and Schink , 1998 ) , ACD20 may use citrate lyase in conjunction with the combined action of membrane-bound oxaloacetate decarboxylase ( EC 4 . 1 . 1 . 3 ) to pump sodium ions . Combined with a sodium-hydrogen antiporter , a hydrogen gradient can be generated and used to drive cellular processes ( Figure 4 , blue arrows ) . Hence , the TCA cycle does not appear to function as it does in most Cyanobacteria , where it generates reducing equivalents for oxidative phosphorylation , but rather links to nitrogen metabolism and organic acid fermentation for energy generation . In the absence of a complete electron transport chain ( above ) , the gut and aquifer bacteria studied here appear to augment the ATP produced by substrate-level phosphorylation by membrane energization . This membrane potential can be capitalized on for ATP synthesis by the F-type ATP synthase , used for flagellar motion ( see below ) , or used for solute transport . Given that the human gut associated Melainabacteria only produce ATP by substrate-level phosphorylation in glycolysis , these alternative modes of generating an energized membrane may be important to their overall energy balance . The Melainabacteria can generate a membrane potential by four mechanisms: ( i ) a sodium gradient generated by decarboxylation of oxaloacetate ( using an oxaloacetate decarboxylase and sodium-hydrogen antiporter , see TCA cycle ) ; ( ii ) membrane bound H+-translocating pyrophosphatases , which use some of the energy liberated during inorganic pyrophosphate ( PPi ) hydrolysis to drive proton translocation across the cytoplasmic membrane ( Schocke and Schink , 1998 ) ; ( iii ) an 11-subunit complex I dehydrogenase ( Battchikova et al . , 2011 ) ; and ( iv ) a putatively annotated six subunit NiFe membrane-bound hydrogenase that lacks required hydrogen binding motifs ( Vignais and Billoud , 2007; Marreiros et al . , 2013 ) . Notably , these genomes all lack membrane-associated Rhodobacter nitrogen fixation ( Rnf ) and formate dehydrogenase complexes found in the genomes of obligate fermentative organisms ( McInerney et al . , 2007; Biegel et al . , 2011 ) . Given the high demand for reduced ferredoxin in the cell , we have considered that both the complex I and the annotated NiFe hydrogenase may use the proton-motive force to produce reduced ferredoxin , which is a required electron donor for the FeFe hydrogenase and nitrogenase systems . The Melainabacteria genomes encode complete pathways for biosynthesis of vitamins B2 ( riboflavin ) , B3 ( nicotinamide ) , B7 ( biotin ) , and B9 ( dihydrofolate ) . The gut types additionally make vitamin B5 ( pantoate ) . We are unsure if the subsurface bacterium ACD20 can make vitamin B5 as it appears to lack the final enzyme required in the synthesis of vitamin B5 , 2-dehydropantoate 2-reductase . ACD20 and one of the human gut types ( MEL . A1 ) may also be able to synthesize vitamins K1 and K2 . Cyanobacteria are capable of synthesizing the B and K vitamins as well ( Kanehisa et al . , 2012 ) . Germ-free animals raised aseptically , and which lack gut microbiota , have an increased nutrient requirement for B and K vitamins , suggesting that under normal conditions the mammalian gut microbiota are a source of these vitamins for the host ( Backhed et al . , 2005 ) . Hence , Melainabacteria may represent one of the microbial sources of the B and K vitamins for their hosts . Nitrogen fixation is a capacity common among Cyanobacteria and is accomplished via a nitrogenase complex ( Zehr et al . , 2003 ) . While the human gut-derived melainabacterial genomes lack the genes required for a functional nitrogenase complex ( nifD , nifK , and nifH ) , the ACD20 genome encodes these genes and the nifE , nifV , nifS , nifU , nifB , and nifB/X genes involved in nitrogen fixation . We confirmed that the ACD20 NifH protein sequence contains the required [4Fe/4S] cluster , all motifs for functionality , and a conserved lysine in position 15 responsible for ATP interaction . Therefore , it seems likely that ACD20 has the capacity to fix nitrogen . This ability has been proposed to account for the increased dominance of Geobacter species under ammonium limiting conditions created during acetate stimulated U ( VI ) bioremediation of the same subsurface aquifer ( Mouser et al . , 2009 ) . The intestinal relatives do not have nitrogenase capabilities . ( Figure 4 , Supplementary file 1 ) . When placed in phylogenetic context with 865 nifH gene sequences ( Zehr et al . , 2003 ) , the nifH gene from ACD20 ( Figure 6 , red ) does not cluster with the primary cyanobacterial nifH genes ( Figure 6 , green ) in nifH group I , but is affiliated with nifH group III . Group III is composed of sequences from phylogenetically distant organisms , many of which are obligate anaerobes ( e . g . , Clostridium species , sulfate-reducers , and methanogens ) ( Zehr et al . , 2003 ) ( Figure 6 ) as well as nifH from some Cyanobacteria ( e . g . , Microcoleus chthonoplastes PCC7420 and Anabaena variabillis ATCC 29413 , which have secondary copies of nifH; Figure 6 , shown in green in group III ) . The phylogenetic placement within the nifH cluster III was robust to alignment curation method ( manual or automatic with GBLOCKS ) , and the ACD20 nifH sequence was never monophyletic with cyanobacterial sequences . The best hit to the ACD20 nifH sequence is a group III nitrogenase annotated from Methanoregula boonei 6A8 ( ABS56522 ) . These results indicate that the melainabacterial nitrogenase is not related to the primary nitrogenase in Cyanobacteria . 10 . 7554/eLife . 01102 . 016Figure 6 . The phylogeny of the Melainabacteria nitrogenase . A maximum likelihood phylogenetic tree constructed with 865 nitrogenase nifH genes from sequenced genomes ( Zehr et al . , 2003 ) is shown . nifH groups I and III are shown . The ACD20 nifH ( in group III ) is denoted in red , while photosynthetic cyanobacterial nifH sequences ( in groups I and III ) are denoted in green . Relative to group I , group III is characterized by deep bifurcations and long-branch lengths ( Zehr et al . , 2003 ) , which are represented in the constructed tree by low-bootstrap values ( <50 ) for internal branch positions in group III . ACD20 sequences are monophyletic ( but with low bootstrap support ) with nifH sequences from anaerobic Clostridium and Fusobacterium species . DOI: http://dx . doi . org/10 . 7554/eLife . 01102 . 016 Further distinctions between the ACD20 and cyanobacterial nitrogen metabolism include how nitrogen is assimilated . The large and small subunit of the glutamine oxoglutarate aminotransferase ( GOGAT ) system , responsible for nitrogen assimilation in Melainabacteria , are NADP+-based . This finding distinguishes Melainabacteria from known Cyanobacteria , which use either a 3Fe-4S ferredoxin-dependent monomeric enzyme or a two subunit NADH-dependent GOGAT ( Muro-Pastor et al . , 2005 ) . While a common nitrogenase could have existed in the ancestor of Melainabacteria and Cyanobacteria , the extant capacity to fix and assimilate nitrogen appears to have been acquired independently in Cyanobacteria and this sibling lineage . Unlike Cyanobacteria , the organisms studied here are flagellated . All four of the analyzed genomes contain genes for flagella production , and all but one of the analyzed gut genomes ( MEL . A1 ) contains the requisite genes to produce a fully functional flagellum ( Supplementary file 1 ) . The flagella are composed of the M , S , P , and L rings , as expected given a Gram-negative cell envelope ( DePamphilis and Adler , 1971 ) . At least one copy of the flagellin protein in each Melainabacteria genome contains the eight amino acid sequence recognized by Toll-like receptor 5 ( Andersen-Nissen et al . , 2005 ) , indicating an ability to interact with the host immune system ( Figure 7 ) . Trees built using flagellum-related gene sequences show that the genes branch deeply with the Firmicutes and Spirochaetes ( Figure 8 ) , arguably the two most basal bacterial lineages ( Daubin et al . , 2002; Ciccarelli et al . , 2006 ) . This result suggests that the common ancestor of Cyanobacteria and Melainabacteria may have been flagellated . Given that no flagellated Cyanobacteria are known and cyanobacterial motility is accomplished by gliding or twitching ( Schaechter , 2010 ) , flagella may have been non-essential to the cyanobacterial lifestyle and hence lost . However , it is not possible to rule out the alternative of flagella being acquired by Melainabacteria after the divergence from Cyanobacteria . 10 . 7554/eLife . 01102 . 017Figure 7 . Putative TLR5 activation region in Melainabacteria flagellin genes . Protein sequence alignment of residues 88–103 ( Escherichia coli coordinates ) for the flagellin genes . The range of residues required for TLR5 activation ( Andersen-Nissen et al . , 2005 ) are indicated by the top bracket . Sequences are organized by similarity within these residues . Species whose flagellin are reported ( Andersen-Nissen et al . , 2005 ) to be recognized ( R ) or unrecognized ( UR ) by TLR5 are noted . Based on the visualization of the alignment , flagellin genes predicted to be recognized or unrecognized by TLR5 are indicated; genes of ambiguous TLR5 recognition status are unmarked . DOI: http://dx . doi . org/10 . 7554/eLife . 01102 . 01710 . 7554/eLife . 01102 . 018Figure 8 . Phylogeny of flagella-related genes . Supertree ( cladogram ) of 13 bootstrap ML trees of the flagellar genes shared among the four analyzed genomes . The phylum ( or more specific taxonomic identifier ) of each species is listed: ( F ) Firmicutes , ( S ) Spirochaetes , ( E-P ) Epsilonproteobacteria , ( MEL ) Melainabacteria , ( A ) Aquificae , ( T ) Thermotogae , ( D-P ) Deltaproteobacteria , ( Pl ) Planctomycetes , ( B ) Bacteroidetes , ( A-P ) Alphaproteobacteria , ( B-P ) Betaproteobacteria , ( G-P ) Gammaproteobacteria . In all 13 individual trees , Melainabacteria branched with Firmicutes and Spirochaetes . DOI: http://dx . doi . org/10 . 7554/eLife . 01102 . 018 The gut–associated clade of the Melainabacteria , which diverged from the sediment-associated lineage , appears to have undergone genome reduction relative to ACD20 ( Figure 3 , Table 2 ) . The gut types lack genes for chemotaxis , production of some amino acids ( aspartic acid , asparagine , phenylalanine , arginine , histidine , tyrosine , and , in MEL . A1 and MEL . B1 , alanine ) , a type I secretion system , nitrogen fixation , and genes for additional energy generation by substrate-level phosphorylation and for the production of acetate and perhaps butyrate as fermentation end-products . Note that in ACD20 , genes for four of the five steps in butyrate synthesis were identified . The fifth step , which could be carried out by butyrate kinase and phosphotransbutyrylase or butyryl-coenzyme A ( CoA ) : acetate CoA-transferase ( as occurs in anaerobic bacteria ) was not detected . An alternative enzyme , analogous to the process used by some Archaea ( e . g . , Pyrococcus species ) , acetate CoA ligase ( Mai and Adams , 1996 ) , found in ACD20 , may substitute to convert butyryl CoA to butyrate . Genome reduction has been observed in symbiotic bacteria and may have occurred in Melainabacteria during adaptation to life in the animal gut . The lack of the filament , filament cap , hook-filament junction , and L and P rings required for flagellum biosynthesis in the genome assembled into a single scaffold ( MEL . A1 ) indicates that these genes were likely not missed in the genome sequencing and assembly but rather that further genome reduction may be ongoing in the gut-associated clade . To update our view of the ecological niches for Melainabacteria , we searched for 16S rRNA gene sequences in recent datasets ( Figure 9—source data 1 ) . We detected the water-soil-sediment ( non-animal associated ) clade in a wide variety of environments ( Figure 9 ) , with the highest abundances in municipal water . An analysis of the Human Microbiome Project ( Consortium HMP , 2012 ) and other datasets ( Figure 9—source data 1 ) showed that across the human body , they occur as rare members of skin , airway , and mouth communities , but are most abundant in fecal samples ( Figure 9B ) . The gut types were also detected in fecal samples from a range of other mammalian species , with the highest levels in herbivores , consistent with a role in fermentation of dietary substrates ( Figure 9C ) . Within the herbivores , Melainabacteria are more abundant in feces obtained from foregut than hindgut fermenters ( two tailed t-test , p=0 . 026 ) . Within the Human Microbiome Project dataset ( Consortium HMP , 2012 ) , approximately 10% of the samples contained melainabacterial 16S rRNA gene sequences , providing a rough estimate as to what fraction of the American population carries Melainabacteria . When comparing three global human populations ( Yatsunenko et al . , 2012 ) , we observed the highest abundances in fecal samples obtained from predominantly vegetarian Malawian and Venezuelan individuals ( Figure 9C ) . These observations suggest that Melainabacteria also play a role in fermentation of dietary plant polysaccharides in humans . 10 . 7554/eLife . 01102 . 019Figure 9 . The prevalence of members of Melainabacteria in different environments , including distinct human body habitats . In all three panels , the relative abundances of Melainabacteria in different samples types are plotted as box plots ( log10 transformed; i . e . , 10−1 = 0 . 1% ) . Data were obtained from the QIIME database , derive from a variety of studies , and are publically available ( Figure 9—source data 1 ) : ( A ) soil , sediment , and water sites , ( B ) different human body sites ( GI = gastrointestinal ) , ( C , left ) mammal stool classified by host diet , ( C , right ) country of origin for human stool . UD = undetermined . DOI: http://dx . doi . org/10 . 7554/eLife . 01102 . 01910 . 7554/eLife . 01102 . 020Figure 9—source data 1 . 16S rRNA gene sequence datasets used to analyze the sources of Melainabacteria . DOI: http://dx . doi . org/10 . 7554/eLife . 01102 . 020 The reconstruction of genomes from uncultivated bacteria from human gut and subsurface environmental metagenomes has enabled us to describe Melainabacteria , a novel candidate phylum sibling to Cyanobacteria , and to further elucidate the evolutionary history of one of the Earth’s most important bacterial phyla . Our results suggest that while photosynthesis probably developed in Cyanobacteria after the separation from Melainabacteria , the evolution of light related capabilities may have been enabled by regulators present in their common ancestor . The ability to fix nitrogen appears to have developed separately in Cyanobacteria and the water-soil-sediment clade of Melainabacteria . The role of the Melainabacteria in the human gut is one of an obligate fermenter , and its enrichment in human subjects and animals with a plant-rich diet likely relates to a prominent role of the Melainabacteria in the processing of plant fibers . As plant fibers have been minimized in Western diets , the Melainabacteria may be regarded as part of the microbiota that is disappearing from modernized populations ( Blaser and Falkow , 2009 ) . The metabolic reconstructions made from these genomes should guide efforts into obtaining Melainabacteria in culture , which would allow a better understanding of this important human symbiont . The three fecal samples from healthy adults ( A , B , and C ) were collected under Cornell University IRB ( Protocol ID 1108002388 ) from the United Kingdom Adult Twin Registry ( TwinsUK ) . Samples were collected in 15 ml conical tubes , refrigerated for 1–2 days , and then stored at −80°C at King’s College London until being shipped on dry ice to Cornell University , where they were subsequently stored at −80°C . Approximately 100 mg of sample was processed with PowerSoil DNA isolation kit ( MoBio Laboratories Ltd , Carlsbad , CA ) to isolate genomic DNA . Construction of three shotgun genomic libraries and sequencing were carried out at the WM Keck Center for Comparative and Functional Genomics , Roy J Carver Biotechnology Center , University of Illinois at Urbana-Champaign . The barcoded DNAseq libraries were prepared with Illumina’s ‘TruSeq DNAseq Sample Prep kit’ ( Illumina , San Diego , CA ) . The final libraries were quantitated with Qubit ( Life Technologies , Grand Island , NY ) , and the average size was determined on an Agilent bioanalyzer High-Sensitivity DNA chip ( Agilent Technologies , Wilmington , DE ) , diluted to 10 nM and pooled . The 10 nM dilution was further quantitated by qPCR on an ABI 7900 ( Life Technologies ) . The pooled libraries were sequenced on one lane of a flowcell for 101 cycles from each end of the fragments on a HiSeq2000 using TruSeq SBS sequencing kit version 3 , and the fastq files were generated with Casava1 . 8 . 2 . Overall , sequencing yielded a total of 346 million reads ( 34 . 6 Gbp , almost 12 Gbp per sample on average ) . Details relating to the collection , sequencing , and data analysis for the ACD20 genome are provided in ( Wrighton et al . , 2012 ) . This publication focused on members of the OP11 , OD1 , PER , ACD80 , and BD1-5 candidate bacterial phyla and included no discussion of the ACD20 genome . We reconstructed four complete ( MEL . A1 , MEL . B1 , MEL . B2 , and MEL . C1 ) , one near-complete ( MEL . C2 ) , and two partial ( MEL . A2 and MEL . C3 ) genomes from the three human fecal samples . An eighth near-complete genome ( ACD20 ) was recovered from a different dataset ( see Wrighton et al . ( 2012 ) for genome assembly details ) . Potential metagenomic sequences belonging to Melainabacteria were determined by similarity to Cyanobacteria or by being completely novel . These scaffold fragments were linked to other fragments by coverage , %GC , and paired-end read information using the assembly curation steps and scripts previously described ( Sharon et al . , 2013 ) . Genome MEL . A1 was assembled from human fecal sample A using Velvet ( Zerbino and Birney , 2008 ) with parameters optimized based on the expected genome coverage . Genome MEL . A2 was recovered from an IDBA-UD ( Peng et al . , 2012 ) assembly for human fecal sample A based on a phylogenetic profile of hits for the scaffolds and genes . Genomes MEL . B1 and MEL . B2 were reconstructed from human fecal sample B and genomes MEL . C1 , MEL . C2 , and MEL . C3 from human fecal sample C . These genomes were assembled using the IDBA-UD assembler . Identification of scaffolds belonging to the target Melainabacteria genome was aided by utilizing similarity to genes in MEL . A1 . Scaffolds not belonging to MEL . C1 or MEL . C2 were identified as belonging to genome MEL . C3 when more than 50% of the best hits for its genes were to genes in the other Melainabacteria genomes and not from other published genomes . Genome completeness was assessed as follows: Complete genomes have a complete set of single copy genes ( Raes et al . , 2007 ) ( Figure 10 ) , and the linkage between all scaffolds is established; Near complete genomes have a complete set of single copy genes ( Figure 10 ) , and the linkage between almost all scaffolds is established; Partial genomes lack a complete set of single copy genes , and/or scaffold linkage is lacking . The newly sequenced gut genomes are available at http://ggkbase . berkeley . edu/mel/organisms . 10 . 7554/eLife . 01102 . 021Figure 10 . Single copy gene inventory from reconstructed genomes . Data are based on single copy genes ( numbers in circles indicate the number of copies found ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01102 . 021 A core group of 16 syntenic ribosomal proteins was selected based on published metrics of lateral gene transfer frequencies ( rpL2 , 3 , 4 , 5 , 6 , 14 , 15 , 16 , 18 , 22 , 24 , and rpS3 , 8 , 10 , 17 , 19 ) ( Sorek et al . , 2007; Wu and Eisen , 2008 ) . Reference datasets were obtained from the PhyloSift database ( Darling et al . , 2012 ) . The NCBI and JGI IMG databases were mined for the 16 ribosomal proteins from recently sequenced genomes from the Cyanobacteria , Chloroflexi , Nitrospira , and TM7 phyla . The 16 syntenic ribosomal protein genes were identified in the eight new genomes , translated , and included for phylogenetic placement . The complete dataset contained 733 taxa . Each individual protein dataset was aligned using Muscle version 3 . 8 . 31 ( Edgar , 2004 ) and then manually curated to remove end gaps and single-taxon insertions . Model selection for evolutionary analysis was determined using ProtTest3 ( Darriba et al . , 2011 ) for each single protein alignment . The curated alignments were concatenated to form a 16-protein , 733 taxa alignment with 3 , 082 unambiguously aligned positions . A maximum likelihood phylogeny for the concatenated alignment was conducted using Phyml ( Guindon and Gascuel , 2003 ) under the LG+αG model of evolution and with 100 bootstrap replicates . 16S rRNA gene sequences from MEL . A1 , MEL . B1 , and MEL . B2 were aligned using NAST ( McDonald et al . , 2012 ) . An aligned full-length sequence set was created for five representative sequences from each phylum in the Greengenes ( May 2011 ) database ( DeSantis et al . , 2006 ) plus the three Melainabacteria sequences . For the Cyanobacteria-Melainabacteria specific tree , the aligned sequence set was composed of the three Melainabacteria sequences and one representative sequence from each order within Cyanobacteria except for orders YS2 , SM1D11 , and mle1-12 , from which we included all sequences as these are most closely related to the Melainabacteria sequences . Phylogenetic trees were constructed using maximum likelihood implemented in RAxML ( Stamatakis , 2006 ) and using the GTR+γ+I model of evolution and 100 bootstrap resamplings . The trees were rooted using five archaeal sequences . Trees were visualized using the Interactive Tree of Life ( iTOL ) ( Letunic and Bork , 2007 , 2011 ) . RnpB RNAs were identified by searching for matches to 5′ GAGGAAAGUCC 3′ , which is highly conserved in RNase P RNAs ( Haas et al . , 1991 ) , as well as the surrounding intergenic region . Matches were analyzed in Bcheck ( Yusuf et al . , 2010 ) to determine the structure and type of the RnpB RNAs . The following four genomes were used for the metabolic analyses: ACD20 , MEL . A1 , MEL . B1 , and MEL . B2 . Two partial genomes ( MEL . A2 and MEL . C3 ) were excluded from this analysis because of reliance on clusters of orthologs from all participating genomes . MEL . C1 and MEL . C2 were not included in these analyses because of their similarity to MEL . A1 and MEL . B1 , respectively . Clusters of orthologs ( 920 MEL-COGs in total , Figure 5—source data 1 ) for four of the Melainabacteria genomes ( ACD20 , MEL . A1 , MEL . B1 , and MEL . B2 ) were constructed by ( 1 ) BLASTing each protein from each genome against the proteome of the other genomes individually , ( 2 ) connecting proteins between pairs of genomes by reciprocal best BLAST hits , and ( 3 ) forming clusters of these reciprocal best hits when groups of four proteins ( one from each genome ) had at least two proteins connected to all three other members of the group , and the rest were connected to at least two other members of the group . The COGs were annotated with ( I ) phylogenetic origin according to best matching phylum in a BLAST search of NCBI’s nr database and ( II ) KEGG category according to the BLAST match of the ACD20 homolog in the KEGG database . Phylogenetic assignments were decided based on majority voting for all four genes of the MEL-COG ( one gene from each of the four Melainabacteria genomes analyzed [ACD20 , MEL . A1 , MEL . B1 , and MEL . B2] ) . First , a phylum was assigned for each single gene . Then , the phylum assigned to majority of the four MEL-COG genes was assigned to the MEL-COG . If none of the MEL-COG genes received a significant hit to an existing gene in another phylum , the MEL-COG was assigned ‘No hit’ . If more than one phylum shared the highest number of assignments , or several MEL-COG genes had a ‘Multiple’ assignment from #3 below , the MEL-COG was assigned ‘Multiple’ . Phylum assignment to each individual gene in a MEL-COG was accomplished using homology to genes belonging to other phyla using the following workflow: ( 1 ) MEL-COG genes were BLASTed against NCBI’s nr database; ( 2 ) The top two hits with e-value ≤ 1e-5 were collected; and ( 3 ) The alignments of these genes with the MEL-COG gene were compared , considering both alignment-length and % identity: each alignment was represented by the total number of identical positions , calculated as ( alignment-length ) × ( % identity ) . The top hit was considered to be a significantly better match if its number of identical positions was at least 5% more than the number of identical positions for the second best matching gene . In that case , the phylum of the best hit was assigned to the MEL-COG gene; otherwise the MEL-COG gene received the ‘Multiple’ assignment . The ACD20 representative in each MEL-COG was BLASTed against the Kyoto Encyclopedia of Genes and Genomes ( KEGG ) database ( 5 ) . The KEGG term for the best hit was assigned to the MEL-COG . If no hits were found , the MEL-COG assignment was ‘unknown’ , else all pathways to which the KEGG term belongs were identified , as well as their KEGG category . Assignment was: ( a ) ‘Undecided’—if the KEGG term did not belong to any pathway; ( b ) ‘Multiple’—if the pathways to which the KEGG term belongs were from multiple categories; or , ( c ) the category of the term’s pathway , if all pathways belonged to the same category . KEGG annotations for the ACD20 , MEL . A1 , MEL . B1 , and MEL . B2 genomes were uploaded to the KEGG database . Pathways were visually inspected for completeness , reactants , and products . All annotations were confirmed by manual inspection , including confirmation of active residues and phylogenetic tree analyses . Phylogenetic trees were constructed using protein alignments as described in detail ( Wrighton et al . , 2012 ) . Confirmation and cross-genome comparisons were constructed by using the LIST and GENOME SUMMARY features in the ggKbase website ( http://ggkbase . berkeley . edu/genome_summaries/81-MEL-Metabolic-Overview-June2013 ) . Genomic information from the Melainabacteria dataset is stored in the publically accessible ggKbase database ( http://ggkbase . berkeley . edu/mel/organisms ) . Phylogenetic analyses of the nifH gene constructed with a database of 865 nifH sequences from genome-sequenced Bacteria and Archaea ( Zehr et al . , 2003 ) . Two separate phylogenetic trees were constructed with and without the use of GBLOCKS in the pipeline reported previously ( Wrighton et al . , 2012 ) . Figure 6 is based upon the GBLOCKS alignment . Maximum likelihood phylogenetic trees were produced using RAxML version 7 . 3 . 0 using the rapid bootstrap analysis and the general time reversible model of nucleotide substitution with optimization and categorization of per-site substitution rates on 500 distinct trees ( raxmlHPC -f a -m GTRCAT -x 1234 -N 500 ) ( Stamatakis , 2006 ) and visualized in iTOL ( Letunic and Bork , 2007 , 2011 ) . Phylogenetic trees were built from the 13 flagellum-related genes in the MEL-COG list ( the ACD20 homolog is listed ) : ACD20_20398 . 28785 . 13G0015 ( flgG ) , ACD20_20398 . 28785 . 13G0016 ( flgG ) , ACD20_20398 . 28785 . 13G0018 ( flgC ) , ACD20_20398 . 28785 . 13G0021 ( fliF ) , ACD20_20398 . 28785 . 13G0023 ( fliH ) , ACD20_20398 . 28785 . 13G0024 ( flhA ) , ACD20_20398 . 28785 . 13G0025 ( fliI ) , ACD20_20398 . 28785 . 13G0028 ( flgE ) , ACD20_26563 . 5896 . 13G0002 ( flhA ) , ACD20_26723 . 8006 . 14G0006 ( fliP ) , ACD20_26723 . 8006 . 14G0007 ( fliQ ) , ACD20_26723 . 8006 . 14G0009 ( fliR ) , ACD20_29089 . 28969 . 14G0027 ( flhB ) . For each gene , the ACD20 homolog was used in a pBLAST search ( e-value <10−5 ) against the 41 flagellated bacterial species used in ( Liu and Ochman , 2007 ) . Matching protein sequences across the queried species were aligned using MUSCLE version 3 . 6 ( Edgar , 2004 ) . Protein alignments were converted to DNA alignments . Phylogenetic trees were produced by maximum likelihood using RAxML version 7 . 3 . 0 , executing the rapid bootstrap analysis and the general time reversible model of nucleotide substitution with optimization and categorization of per-site substitution rates ( 50 rate categories ) on 1000 distinct trees ( raxmlHPC -e 0 . 1 -f a -c 50 -m GTRCAT -x 92957 -N 1000 -p 3212 ) ( Stamatakis , 2006 ) . A supertree was built using heuristic searches in Clann ( Creevey and McInerney , 2005 ) . Flagellin genes were recovered from the ACD20 , MEL . A1 , MEL . B1 , and MEL . B2 genomes . The flagellin gene MEL . B1 . 001_31 was excluded as this sequence is incomplete and does not align well to other flagellin sequences . Using MUSCLE ( 3 . 8 ) multiple sequence alignment ( Edgar , 2004 ) , these sequences were aligned with representative flagellin genes from species whose flagellins are known to be either recognized or unrecognized by the mammalian Toll-like receptor 5 ( Andersen-Nissen et al . , 2005 ) ( gene names and Uniprot IDs are given ) : E . coli fliC ( Q0GJI9 ) , Bacillus subtilis hag ( P02968 ) , Salmonella Typhimurium fliC ( P06179 ) , Vibrio anguillarum flaC ( Q56574 ) , Listeria monocytogenes flaA ( Q02551 ) , Bartonella bacilliformis fla1 ( P35633 ) , Campylobacter jejuni flaA ( Q46113 ) , Helicobacter pylori flaA ( P0A0S1 ) , Helicobacter felis flaA ( Q9XB38 ) , Wolinella succinogenes flaG ( Q79HP6 ) . For visualization purposes , the sequences were ordered by similarity within the TLR5 activation domain ( Andersen-Nissen et al . , 2005 ) . BOXSHADE ( http://www . ch . embnet . org/software/BOX_form . html ) was used to display the protein alignment . We built a reference dataset of environmental ( non-gut ) and gut associated Melainabacteria 16S rRNA gene sequences ( Figure 2 ) obtained from the Greengenes database ( DeSantis et al . , 2006 ) and queried the dataset at 97% ID against publicly available 16S rRNA gene datasets obtained from http://www . microbio . me/qiime ( Figure 9—source data 1 ) . Similar samples types were combined ( e . g . , forest soil , grassland soil , shrubland combined as ‘soil’ ) , and Melainabacteria sequence reads were tallied . Samples with zero Melainabacteria sequence reads were removed . Samples types ( i . e . , air ) with fewer than five reads were not plotted .	Microbes are ubiquitous in the world and exist in complex communities called microbiomes that have colonized many environments , including the human gut . Until modern techniques for sequencing nucleic acids became available , many of the organisms found in these microbiomes could not be studied because they could not be cultured in the laboratory . However , advances in sequencing technology have made it possible to study the evolution and properties of these microbes , including their impact on human health . Bacteria belonging to the phylum Cyanobacteria had a significant effect on the prehistoric Earth because they were the first organisms to produce gaseous oxygen as a byproduct of photosynthesis , and thus shaped the Earth’s oxygen-rich atmosphere . Early plants took up these bacteria in a symbiotic relationship , and plastids—the organelles in plant cells that perform photosynthesis and produce oxygen–are the descendants of Cyanobacteria . Organisms evolutionarily related to Cyanobacteria have been found in the human gut and in various aquatic sources , but these bacteria have not been studied because it has not been possible to isolate or culture them . Now , Di Rienzi , Sharon et al . have used modern sequencing techniques to obtain complete genomes for some of these bacteria , which they assign to a new phylum called Melainabacteria . By analyzing these genomes , Di Rienzi , Sharon et al . were able to make predictions about the cell structure and metabolic abilities of Melainabacteria . Like Cyanobacteria , they have two membranes surrounding the bacterial cell; unlike Cyanobacteria , however , they have flagella that propel them through liquid or across surfaces . Most interestingly , Melainabacteria are not able to perform photosynthesis , but instead produce energy through fermentation and release hydrogen gas that can be consumed by other microorganisms . The genome of the bacteria isolated from water reveals that it has the capacity to fix nitrogen . Cyanobacteria can also fix atmospheric nitrogen , but the protein complexes used by the two phyla are not related , which suggests that nitrogen fixation evolved after the evolutionary divergence of Cyanobacteria and Melainabacteria . By exploring previously published datasets of bacterial communities , Di Rienzi , Sharon et al . found that Melainabacteria are common in aquatic habitats . They are also prevalent in the guts of herbivorous mammals and humans with a predominantly vegetarian diet . Melainabacteria from the human gut also synthesize several B and K vitamins , which suggests that these bacteria are beneficial to their host because in addition to aiding with the digestion of plant fibers , they are also a source of vitamins .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2013	The human gut and groundwater harbor non-photosynthetic bacteria belonging to a new candidate phylum sibling to Cyanobacteria
Synaptic vesicle ( SV ) release probability ( Pr ) is a key presynaptic determinant of synaptic strength established by cell-intrinsic properties and further refined by plasticity . To characterize mechanisms that generate Pr heterogeneity between distinct neuronal populations , we examined glutamatergic tonic ( Ib ) and phasic ( Is ) motoneurons in Drosophila with stereotyped differences in Pr and synaptic plasticity . We found the decoy soluble N-ethylmaleimide sensitive factor attachment protein receptor ( SNARE ) Tomosyn is differentially expressed between these motoneuron subclasses and contributes to intrinsic differences in their synaptic output . Tomosyn expression enables tonic release in Ib motoneurons by reducing SNARE complex formation and suppressing Pr to generate decreased levels of SV fusion and enhanced resistance to synaptic fatigue . In contrast , phasic release dominates when Tomosyn expression is low , enabling high intrinsic Pr at Is terminals at the expense of sustained release and robust presynaptic potentiation . In addition , loss of Tomosyn disrupts the ability of tonic synapses to undergo presynaptic homeostatic potentiation . Ca2+-dependent fusion of synaptic vesicles ( SVs ) is the primary mechanism for neurotransmission and is mediated by the soluble N-ethylmaleimide sensitive factor attachment protein receptor ( SNARE ) family ( Jahn and Scheller , 2006; Söllner et al . , 1993; Sudhof , 2004; Weber et al . , 1998 ) . Following an action potential , SNARE proteins located on the SV and plasma membrane zipper into an energetically favorable coiled-coil bundle to induce SV fusion ( Jahn and Scheller , 2006; Söllner et al . , 1993; Südhof and Rothman , 2009 ) . Neurotransmitter release results in a postsynaptic response that varies in size depending on the strength of the synapse , which can be regulated from both pre-and post-synaptic compartments . The postsynaptic cell controls sensitivity to neurotransmitters by governing receptor field composition , while the presynaptic neuron establishes the probability ( Pr ) of SV fusion ( Citri and Malenka , 2008; Körber and Kuner , 2016; Bliss et al . , 2003; Yang and Calakos , 2013 ) . Highly stereotyped differences in Pr exist across neurons , with many neuronal populations broadly classified as tonic or phasic depending on their spiking patterns , Pr and short-term plasticity characteristics ( Atwood and Karunanithi , 2002; Dittman et al . , 2000; Lnenicka and Keshishian , 2000 ) . How cell-intrinsic properties establish differences in presynaptic Pr between neuronal classes , and how release strength is further refined via plasticity , remain incompletely understood . The Drosophila melanogaster larval neuromuscular junction ( NMJ ) provides a robust genetic system for characterizing mechanisms mediating synaptic communication and tonic versus phasic release properties ( Aponte-Santiago et al . , 2020; Aponte-Santiago and Littleton , 2020; Genç and Davis , 2019; Lu et al . , 2016; Newman et al . , 2017; Wang et al . , 2021 ) . Larval body wall muscles are co-innervated by two glutamatergic motoneuron populations that drive locomotion , including the tonic-like Ib and phasic-like Is subtypes ( Aponte-Santiago et al . , 2020; Harris and Littleton , 2015; Jan and Jan , 1976; Lnenicka and Keshishian , 2000 ) . Tonic Ib terminals display lower initial Pr and sustained release during stimulation , whereas phasic Is terminals show higher intrinsic Pr and rapid depression ( Lu et al . , 2016; Newman et al . , 2017 ) . The Drosophila NMJ also undergoes robust presynaptic homeostatic potentiation ( PHP ) that rapidly increases Pr to compensate for disruptions to postsynaptic glutamate receptor ( GluR ) function ( Böhme et al . , 2019; Gratz et al . , 2019; Li et al . , 2018; Müller et al . , 2012; Ortega et al . , 2018; Weyhersmüller et al . , 2011 ) . In addition to intrinsic release differences , the Ib and Is subtypes display distinct capacity for PHP ( Newman et al . , 2017; Genç and Davis , 2019 ) . How tonic and phasic neurons differentially regulate Pr during normal synaptic communication and plasticity is largely unknown . The highly conserved SNARE regulatory protein Tomosyn negatively controls SV release and has been proposed to participate in synaptic plasticity ( Ben-Simon et al . , 2015; Chen et al . , 2011; Gracheva et al . , 2006; McEwen et al . , 2006 ) . Tomosyn has an N-terminal WD40 repeat domain and a C-terminal SNARE motif with homology to the SV v-SNARE Synaptobrevin 2 ( Syb2 ) ( Fasshauer et al . , 1998; Hatsuzawa et al . , 2003; Hattendorf et al . , 2007; Pobbati et al . , 2004; Williams et al . , 2011 ) . Tomosyn inhibits presynaptic release by binding the t-SNAREs Syntaxin1 ( Syx1 ) and SNAP-25 to prevent Syb2 incorporation into the SNARE complex fusion machinery ( Hatsuzawa et al . , 2003; Lehman et al . , 1999; Sakisaka et al . , 2008; Williams et al . , 2011 ) . To further examine the role of Tomosyn in synaptic transmission and plasticity , we used CRISPR to generate mutations in the sole Drosophila tomosyn gene . Structure-function analysis revealed the SNARE domain is critical for release inhibition , while the scaffold region promotes enrichment of Tomosyn to SV-rich sites . Despite enhanced evoked release , tomosyn mutants fail to maintain high levels of SV output during sustained stimulation due to rapid depletion of the immediately releasable SV pool . Tomosyn is highly enriched at Ib synapses and generates tonic neurotransmission properties characterized by low Pr and sustained release in this population of motoneurons . Indeed , optogenetic stimulation and optical quantal analysis demonstrate an exclusive role for Tomosyn in regulating intrinsic release strength in tonic motoneurons . PHP expression primarily occurs at tonic synapses and is abolished in tomosyn mutants , suggesting Tomosyn is also essential for acute PHP expression . Together , these data indicate Tomosyn mediates the tonic properties of Ib motoneurons by suppressing Pr to slow the rate of SV usage , while decreasing Tomosyn suppression enables Pr enhancement during PHP . Conversely , the absence of Tomosyn in Is motoneurons facilitates phasic release properties by enabling an intrinsically high Pr that quickly depletes the releasable SV pool , resulting in rapid synaptic depression and reduced capacity for PHP . The Drosophila Tomosyn homolog is highly conserved with other Tomosyn proteins , displaying high sequence conservation in critical domains including the C-terminal SNARE motif . This region enables the formation of a SNARE complex composed of the Tomosyn C-terminus and the t-SNAREs Syx1A and SNAP-25 ( Fasshauer et al . , 1998; Hatsuzawa et al . , 2003; Pobbati et al . , 2004; Williams et al . , 2011 ) . BLOSUM62 alignment using the C-terminal tail of the yeast homolog Sro7 as an outgroup indicates the Tomosyn SNARE motif forms a phylogenetically distinct group from other v-SNAREs despite their shared affinity for t-SNAREs ( Figure 1A ) . Homology modeling suggests Drosophila Tomosyn forms a SNARE complex that is similar in structure to mammalian Tomosyn ( Figure 1B ) . A conserved ERG sequence within the C-terminal SNARE motif enables zippering of Tomosyn with t-SNAREs in a complex that prevents association with the SV fusion clamp Complexin , in contrast to SNARE complexes containing Syb2 ( Figure 1C and D ) . Similar to other species , Drosophila tomosyn is alternatively spliced at exon 13 to generate two splice variants , Tomosyn13A and Tomosyn13B , that encode distinct regions of the WD40 repeat scaffold . The sequence of exon 13A is highly conserved across insect genomes , while the 13B exon sequence is poorly conserved ( Figure 1E ) . As such , Tomosyn13A is likely the more functionally conserved isoform . Iterative homology modeling of Tomosyn13A suggests it forms a double-barrel structure with three disordered loops projecting from the core WD40 scaffold domain ( Figure 1F ) , as predicted for mammalian Tomosyn-1 and Tomosyn-2 proteins ( Williams et al . , 2011 ) . Exon 13 encodes one of the loops protruding from the WD40 core , indicating alternative splicing regulates secondary features of Tomosyn beyond its SNARE-binding properties . Together , these data suggest Drosophila Tomosyn shares conserved features with its mammalian counterparts . To assay the function of Tomosyn at Drosophila synapses , CRISPR/Cas9 was used to generate two mutant alleles of the tomosyn gene on the X-chromosome ( Figure 1E ) . A deletion mutant of tomosyn was generated using homology-directed repair to replace the entire coding sequence with a DsRed cassette ( tomosynNA1 ) . A frame shift mutant with an early stop codon ( tomosynFS1 ) was also isolated using a gRNA that targets the 5’ end of the tomosyn coding region . TomosynNA1 mutants were primarily used in this study , though both alleles displayed similar phenotypes ( Figure 2 and Figure 2—figure supplement 1 ) . TomosynNA1 males are homozygous viable and eclose as adults at similar rates to a genetic background control ( n≥95 eclosed flies; Chi-square test , p=0 . 9163 ) . Homozygous adult females eclosed less frequently , suggesting the existence of sex-divergent roles . Tomosyn has been suggested to inhibit SV SNARE complex formation by competing for t-SNARE binding with the v-SNARE Syb2 . To determine if Drosophila Tomosyn plays a similar role in negatively regulating SNARE complex formation , SDS-resistant SNARE complex ( 7S complex ) abundance was assayed . Western blots of control and tomosynNA1 brain lysates with Syntaxin1A antisera demonstrated a 2 . 7-fold increase in SNARE complex levels in the absence of Tomosyn ( Figure 1G ) , consistent with Tomosyn inhibition of SNARE assembly . To characterize synaptic transmission at Tomosyn-deficient synapses , two-electrode voltage clamp ( TEVC ) recordings were performed at 3rd instar muscle 6 NMJs in larval segment A4 . TomosynNA1 null hemizygous males displayed a 62 % increase in the amplitude of evoked excitatory junctional currents ( eEJCs ) in 0 . 3 mM extracellular Ca2+ , indicating Tomosyn suppresses neurotransmitter release ( Figure 2A–C ) . A similar 51 % increase in evoked release was found in the tomosynFS1 frameshift mutant ( Figure 2—figure supplement 1A-C ) . Enhanced release in tomosyn mutants could result from a larger postsynaptic response to single SVs ( quantal size ) or fusion of a larger number of SVs per stimulus ( quantal content ) . Quantal size as measured by miniature excitatory junctional current ( mEJC ) amplitude was unchanged in tomosynNA1 ( Figure 2D–F ) . Instead , tomosyn mutant terminals released 70 % more SVs across the active zone ( AZ ) population in response to a single stimulus ( Figure 2G and H ) , with an average increase in quantal content from 84 SVs released in controls to 143 in tomosynNA1 . In higher extracellular Ca2+ ( 2 . 0 mM ) saline , evoked responses in tomosynNA1 remained larger and displayed a slower evoked charge transfer that resulted in a 43 % increase in EJC area ( Figure 2I–M ) . The enhancement in delayed SV release is consistent with Drosophila RNAi knockdowns and tomosyn mutants in other species ( Chen et al . , 2011; Gracheva et al . , 2006; McEwen et al . , 2006; Sakisaka et al . , 2008 ) . TomosynNA1 mutants also showed a 3 . 5-fold increase in the rate of stimulation-independent spontaneous miniature release events ( Figure 2N and O ) , a phenotype not reported in C . elegans or mammalian Tomosyn-1 mutants though present in mammalian Tomosyn-2 mutants ( Geerts et al . , 2015 ) . To confirm the elevated mini frequency was not due to a second-site mutation , tomosynNA1 mutants were crossed with a deficiency line ( Df ( 1 ) ED7161 ) spanning the tomosyn locus . This allelic combination showed similar increases in spontaneous release ( Figure 2N and O ) , as did the tomosynFS1 allele ( Figure 2—figure supplement 1D , E ) . TomosynNA1/Df ( 1 ) 7 , 161 trans-heterozygous null females showed even larger evoked responses compared to tomosynNA1 hemizygous males or controls ( Figure 2A–C ) . Together with the reduction in homozygous female viability , sex-specific differences in Tomosyn function are likely to account for these enhanced phenotypes . Tomosyn null males were used for the remainder of the study to avoid phenotypic sex differences . To test conservation of Tomosyn function , Drosophila Tomosyn or human Tomosyn-1 transgenes were pan-neuronally expressed in the tomosynNA1 background using the Gal4/UAS system . Both homologs rescued the increased evoked and spontaneous release phenotypes in tomosynNA1 ( Figure 2P and Q ) . Unexpectedly , rescue with human Tomosyn-1 suppressed SV release to below control levels and below rescue with Drosophila Tomosyn . Immunohistochemistry against a Myc epitope attached to the transgenic Tomosyn proteins revealed human Tomosyn-1 was more abundant in presynaptic terminals than Drosophila Tomosyn ( Figure 2R and S ) , suggesting dosage-sensitive inhibition of SV fusion is likely to account for the enhanced suppression . Together , these data indicate Tomosyn suppresses both evoked and spontaneous SV release at Drosophila NMJs , with these properties retained in human Tomosyn-1 . To identify critical domains within Tomosyn that mediate suppression of SV release , full-length or truncation mutants were expressed in tomosynNA1 using the Gal4/UAS system ( Figure 3 ) . Both Tomosyn13A and 13B splice variants restored neurotransmitter release in tomosynNA1 ( Figure 3A and B ) . Eliminating the SNARE motif from either splice isoform abolished rescue , while expressing the Tomosyn SNARE domain alone only partially rescued enhanced release ( Figure 3A ) . Although the SNARE motif is necessary for Tomosyn function , the failure of the SNARE-only construct to fully rescue release defects suggests the scaffold domain also contributes . The scaffold domain could be independently required or act together with the SNARE motif to inhibit fusion . Co-expression of the scaffold and SNARE domains as independent transgenes failed to reconstitute full-length Tomosyn function , indicating these domains must be linked and act cooperatively to decrease SV release ( Figure 3A ) . To determine whether the N-terminal scaffold domain regulates Tomosyn localization , anti-Myc immunocytochemistry was performed on the panel of Myc-tagged rescue constructs . Full-length Tomosyn was present throughout the periphery of presynaptic boutons as observed for other SV proteins ( Figure 3C and D ) , consistent with the presence of Tomosyn on SVs in C . elegans and mammals ( Geerts et al . , 2017; McEwen et al . , 2006 ) . Both Tomosyn13A and 13B co-localized with the SV protein Synapsin to a greater extent than with the neuronal plasma membrane marker anti-HRP ( Figure 3C and D ) . Tomosyn13A and the 13 A scaffold domain alone ( Tom13A-∆SNARE ) showed similar SV enrichment , suggesting the SNARE motif is dispensable for SV localization . The Tomosyn13B scaffold domain ( Tom13B-∆SNARE ) was slightly less efficient at localizing this isoform to SV rich sites . In contrast to the scaffold domain , the Tomosyn SNARE motif alone showed reduced co-localization with SVs . Together , these data indicate the scaffold domain functions to enhance Tomosyn SV localization . To determine whether Tomosyn bidirectionally modulates SV release , the protein was overexpressed in a wildtype background . Full-length Tomosyn13A suppressed evoked and spontaneous release by 33% and 40% , respectively ( Figure 3—figure supplement 1A , B ) . Tomosyn13B overexpression was less effective at reducing release , although the 13B scaffold alone modestly decreased mini frequency ( Figure 3—figure supplement 1B ) . Overexpression of the mammalian Tomosyn scaffold alone reduces SV fusion ( Yamamoto et al . , 2010; Yizhar et al . , 2007; Yizhar et al . , 2004 ) , suggesting the Tomosyn13B scaffold may have similar properties . Overexpression of the remaining Tomosyn truncation mutants , including the SNARE motif alone , failed to inhibit evoked or spontaneous release . These data indicate Tomosyn13A acts as a bidirectional modulator of presynaptic output and requires both the scaffold and SNARE domains to fully regulate SV release . Beyond its role as a decoy SNARE , Tomosyn has been suggested to decrease release by binding to the Ca2+ sensor Synaptotagmin 1 ( Syt1 ) and reducing its ability to activate fusion ( Yamamoto et al . , 2010 ) . If a Syt1/Tomosyn interaction mediates release inhibition in Drosophila , loss of Syt1 should eliminate Tomosyn’s ability to decrease SV fusion . To test this model , neurotransmitter release in tomosyn; syt1 double null mutants ( tomosynNA1; syt1AD4/N13 ) was characterized . Most of the evoked response in tomosynNA1 was Syt1-dependent , as double mutants had a large reduction in evoked release compared to controls ( Figure 4A ) . However , tomosyn;syt1 evoked responses were 86 % larger than those of syt1 mutants alone ( Figure 4A and B ) , indicating Tomosyn inhibits release independent of Syt1 . Syt1 mutants also show increases in the slower asynchronous phase of evoked fusion ( Jorquera et al . , 2012; Yoshihara and Littleton , 2002 ) . Asynchronous release was increased 1 . 7-fold in tomosyn;syt1 double mutants compared to syt1 alone ( Figure 4C–F ) , indicating Tomosyn reduces both synchronous and asynchronous SV fusion . Similar to Tomosyn suppression of spontaneous release at wild-type synapses , the elevated mini frequency normally observed in syt1 was enhanced 2 . 8-fold in tomosyn;syt1 double mutants ( Figure 4G and H ) . Together , these data indicate Syt1 and Tomosyn regulate evoked and spontaneous SV fusion through independent mechanisms . Another member of the Synaptotagmin family , Syt7 , regulates evoked release by controlling the size and usage of the fusogenic SV pool in Drosophila ( Guan et al . , 2020 ) . Like tomosyn , Syt7 null mutants ( syt7M1 ) show increased quantal content , suggesting Syt7 and Tomosyn may restrict SV availability and fusion via a shared pathway . To test this hypothesis , we generated and characterized tomosynNA1;;;syt7M1 double mutants . Both evoked release and mini frequency at syt7M1 mutant NMJs was enhanced by loss of Tomosyn ( Figure 4I–L ) , indicating the proteins act through independent mechanisms to negatively regulate SV fusion . In addition , increased evoked release in tomosyn;;;syt7 double mutants indicate presynaptic output can still be enhanced beyond that observed in the absence of Tomosyn alone . We next assayed if Tomosyn inhibition of SV release is Ca2+-sensitive by recording evoked responses across a range of extracellular [Ca2+] . Loss of Tomosyn enhanced release across the entire Ca2+ range but did not alter the Ca2+ cooperativity of release ( Figure 4—figure supplement 1A , B ) . Paired-pulse stimulation in Ca2+ conditions that matched first pulse EJC amplitudes between tomosynNA1 and controls revealed facilitation was also preserved in the absence of Tomosyn ( Figure 4—figure supplement 1C , D ) . At interstimulus intervals ( ISI ) of 25 and 50 msec , tomosyn mutants displayed enhanced paired-pulse facilitation ( PPF ) . At 75 ms ISI , PPF is preserved but not significantly enhanced ( Figure 4—figure supplement 1E , F ) . The preservation of PPF suggests Tomosyn is unlikely to reduce fusogenicity of individual SVs . Given tomosyn mutants do not decrease the Ca2+ dependence of fusion or PPF , Tomosyn inhibits SV release through a Ca2+-independent mechanism . Loss of Tomosyn also leads to enhanced synchronous and asynchronous release , together with elevated rates of spontaneous SV fusion . These data indicate Tomosyn controls SV supply independent of the specific route for SV release , likely by sequestering free t-SNAREs to reduce fusogenic SNARE complex formation . The enhanced evoked response in tomosyn mutants could reflect an increased number of AZs per NMJ , a higher number of docked SVs per AZ , or an increase in individual SV fusogenicity . Given tomosyn mutants display enhanced PPF , increased SV fusogenicity is unlikely . To determine if AZ number or SV docking is increased , immunocytochemistry and transmission electron microscopy ( TEM ) were used to characterize the morphology and ultrastructure of tomosynNA1 NMJs . Immunostaining for the AZ scaffold protein Bruchpilot ( BRP ) demonstrated AZ number was unchanged at tomosyn synapses ( Figure 5A and B ) . Additionally , BRP abundance at individual AZs was not altered ( Figure 5C ) , indicating release site number and AZ scaffold accumulation are not affected . To further probe NMJ morphology , bouton size and number were analyzed in tomosyn mutants . Despite a slightly smaller bouton area at Ib NMJs , total NMJ area was unchanged due to a mild increase in the number of boutons per NMJ ( Figure 5D–F ) . Is terminals showed no morphological differences from controls ( Figure 5D and E ) . Immunostaining for Syt1 revealed total SV abundance was not altered in tomosyn mutants ( Figure 5G–H ) . Together , these data indicate morphological defects or AZ number are unlikely to account for enhanced SV release . TEM was used to determine whether enhanced SV docking contributes to increased SV release in tomosyn mutants . Despite no gross changes to bouton ultrastructure ( Figure 6A ) , Ib terminals had a 52 % increase in the number of docked SVs per AZ in tomosynNA1 ( Figure 6B and C ) . Increased docking was observed over the entire length of the AZ , with no change in the absolute fraction of docked SVs along the 400 nm trajectory from the electron dense T-bar center ( Figure 6D and E ) . The average distance between neighboring SVs was also unchanged ( Figure 6F ) , suggesting SV clustering is not altered . Tomosyn mutants also showed a larger number of SVs within 100 and 150 nm concentric circles positioned over the AZ center ( Figure 6G and H ) . Average SV diameter ( Figure 6I ) and SV density were unchanged ( Figure 6J ) , indicating Tomosyn does not affect SV formation or total SV number . Together , these data suggest Tomosyn suppresses release by decreasing SV availability at AZs , with enhanced SV docking occurring in the absence of the protein . Endogenous activity at larval NMJs is controlled by central pattern generators ( CPGs ) that trigger intermittent high frequency motoneuron bursting ( 5–40 Hz ) to drive locomotion ( Jan and Jan , 1976; Lu et al . , 2016; Pulver et al . , 2015 ) . To examine how elevated release in tomosyn mutants change during different rates of neuronal firing , synaptic output was compared between low ( 0 . 33 Hz ) and high ( 10 Hz ) frequency stimulation in 2 mM extracellular Ca2+ . Consistent with the enhanced single eEJC phenotype , low frequency stimulation ( 0 . 33 Hz ) resulted in a 4 . 3-fold increase in the EJC area of the first evoked response , followed by subsequent depression that stabilized at a 1 . 4-fold increased quantal content per action potential in tomosynNA1 ( Figure 7A and B ) . At 10 Hz stimulation , tomosyn mutants displayed more robust synaptic depression , with quantal content quickly dropping below control levels ( Figure 7A and B ) . Control NMJs had lower initial quantal content at 10 Hz and showed a gradual depression in release that was eventually equivalent to synaptic output of tomosynNA1 terminals by the 30th stimulus ( Figure 7C ) . The size of the immediately releasable SV pool ( IRP ) , approximated by the cumulative number of quanta released within 30 stimuli , showed no difference between tomosynNA1 and controls ( Figure 7D ) . However , the depression rate was dramatically enhanced in tomosyn ( Figure 7E ) , indicating Tomosyn normally restricts release from the IRP . To approximate the size of the larger readily releasable SV pool ( RRP ) and the SV recycling rate , 10 Hz stimulation was continued for 1 , 500 stimuli to reach steady state where the number of SVs released equals the number of recycled SVs ( Thanawala and Regehr , 2016 ) . The recycling rate in tomosynNA1 mutants was not significantly different from controls , though the RRP size was increased by 42 % ( Figure 7—figure supplement 1A-D ) . Together these data indicate Tomosyn is required to support sustained release by limiting the number of fusogenic SVs . Tonic Ib and phasic Is motoneurons differ in their ability to sustain release during stimulus trains , with Ib synapses showing continued release and Is terminals displaying high initial Pr and rapid depression ( Aponte-Santiago and Littleton , 2020; Lu et al . , 2016 ) . Given the phasic release character of tomosyn mutant synapses ( Figure 7A–E ) , we examined if Tomosyn differentially regulates release from Ib and Is motoneuron populations . To probe endogenous expression of Tomosyn , the GFP variant mClover3 was inserted into the tomosyn 13 A genomic locus ( tomosyn13A-Clover ) using CRISPR ( Figure 1E ) . Immunostaining for Tomosyn13A-Clover revealed a 2 . 1-fold enrichment of endogenous Tomosyn in Ib terminals relative to Is NMJs ( Figure 7F and G ) . To determine whether this difference in expression resulted in functional changes in neurotransmitter release between the two classes of motoneurons , optogenetics was used to isolate Ib and Is evoked responses using motoneuron-specific Gal4 drivers to express UAS-channelrhodopsin2 ( ChR2 ) ( Aponte-Santiago et al . , 2020; Dawydow et al . , 2014; Pérez-Moreno and O’Kane , 2019 ) . Optogenetic stimulation of Ib synapses in tomosynNA1 mutants showed a 3 . 8-fold increase in evoked EJC area ( Figure 7H and I ) . In contrast , optogenetic stimulation of tomosynNA1 Is terminals revealed no differences in evoked output , indicating enhanced release in tomosyn mutants is solely contributed from increased SV fusion at Ib terminals . These data indicate higher expression of Tomosyn in Ib motoneurons results in greater intrinsic release suppression . Is and Ib motoneuron populations also show stereotyped difference in single AZ Pr , with Is having intrinsically higher Pr than Ib . To determine whether Tomosyn differentially regulates Pr , optical quantal analysis was performed in the tomosynFS1 null mutant . This mutant lacks the DsRed reporter cassette found in tomosynNA1 and has less background fluorescence during live imaging . To detect single SV release events at individual AZs , membrane-tethered GCaMP7s was expressed in postsynaptic muscles along with a tagged GluR subunit ( GluRIIA-RFP ) to identify individual PSDs as previously described ( Akbergenova et al . , 2018 ) . Nerve stimulation in control animals indicated Is motoneurons showed a 2 . 4-fold higher average AZ Pr ( 0 . 17 ± 0 . 007 ) than Ib motoneurons ( 0 . 07 ± 0 . 004 ) . In contrast , tomosynFS1 mutants displayed higher Pr at Ib AZs than Is due to increased Ib Pr and no effect on Is Pr ( Figure 7J–L ) . Together , these data indicate Tomosyn suppresses release from tonic synapses and contributes to the intrinsic release differences between these motoneuron subclasses . Loss of Tomosyn in Ib neurons changes both the initial Pr and short-term depression properties such that tonic Ib terminals display phasic release similar to Is motoneurons . Reductions in postsynaptic GluR function at Drosophila NMJs trigger a rapid and robust increase in presynaptic quantal content that homeostatically compensates for decreased quantal size ( Davis et al . , 1998; Frank et al . , 2006; Li et al . , 2018; Petersen et al . , 1997 ) . Given Tomosyn is a key regulator of quantal content , and prior data suggest PHP is more robust at tonic Ib synapses ( Newman et al . , 2017 ) , we assayed if Tomosyn is required for PHP in tonic motoneurons . An allosteric inhibitor of Drosophila GluRs ( Gyki ) was used to acutely reduce quantal size and induce PHP as previously described ( Nair et al . , 2020 ) . Following addition of Gyki into the extracellular saline , quantal size as measured by mini amplitude was reduced in both control and tomosynNA1 mutants ( Figure 8A–C ) . Mini frequency was not significantly changed following Gyki application , indicating spontaneous fusion events remained detectable ( Figure 8—figure supplement 1A , B ) . Control animals compensated for the reduction in quantal size with a 62 % increase in quantal content that preserved the original evoked response amplitude ( Figure 8D–F ) . In contrast , tomosyn NMJs showed no significant enhancement in quantal content after Gyki application , indicating PHP expression is impaired . Loss of PHP could result from an inability to support higher levels of release , or Tomosyn could be a key effector for PHP with post-translational modification decreasing its inhibitory function . To test if impaired PHP in tomosyn mutants is due to release saturation , the quantal content of potentiated NMJs in 0 . 35 mM extracellular [Ca2+] was compared to non-potentiated NMJs at 1 . 5 mM [Ca2+] . In elevated Ca2+ , quantal content was greater than after Gyki-induced potentiation in low Ca2+ for controls ( 46 % increase ) and tomosynNA1 ( 61 % increase ) , indicating lack of potentiation in tomosyn is not due to release saturation ( Figure 8—figure supplement 1C ) . Together with the observation that tomosyn , syt7 double mutants displayed even higher levels of evoked release than tomosyn mutants alone in low Ca2+ ( Figure 4J ) , these data indicate Tomosyn is required for normal expression of Gyki-induced PHP and represents a key effector for enhancing presynaptic output during this form of plasticity . Ib and Is motoneurons also differ in their ability to express PHP , with tonic Ib neurons showing more robust PHP in GluRIIA mutants ( Newman et al . , 2017 ) . To monitor how differential Tomosyn expression in Ib and Is motoneurons affects expression of PHP in real time , optical quantal mapping was used to monitor AZ Pr at individual release sites before and after acute Gyki application . Because Gyki reduces the fluorescent change ( ∆F ) from quantal release by decreasing postsynaptic Ca2+ influx from GluRs ( Figure 8—figure supplement 1D ) , transgenic animals expressing the more sensitive GCaMP variant GCaMP8s ( Zhang et al . , 2020 ) fused to a myristoylation domain for membrane tethering were generated to ensure SV release events could still be detected after Gyki application . Control Ib terminals showed a rapid and robust 1 . 8-fold increase in average AZ Pr 15 min after Gyki incubation ( Figure 8G–J ) . Enhanced SV release occurred across the majority of the AZ population . In addition , previously silent AZs were recruited during evoked stimulation following Gyki application ( Figure 8G ) . In contrast to the robust effect at Ib synapses , Is terminals showed no significant change in AZ Pr or recruitment of silent AZs following Gyki application ( Figure 8G and K–M ) , indicating this form of PHP is predominantly expressed from Ib motoneurons . Tomosyn mutants displayed no significant increase in AZ Pr from either Ib or Is terminals following Gyki application ( Figure 8G–M ) . Together , these data indicate Gyki-induced PHP is Tomosyn-dependent and occurs exclusively at tonic Ib terminals . Loss of Tomosyn generates synaptic responses and a lack of PHP at tonic Ib NMJs that is similar to that observed in phasic Is neurons , indicating Tomosyn levels represent a key presynaptic mechanism for generating tonic versus phasic presynaptic output . The findings reported here indicate the conserved presynaptic release suppressor Tomosyn functions in setting presynaptic output and plasticity differences for a tonic/phasic pair of motoneurons that co-innervate Drosophila larval muscles . CRISPR-generated mutations in Drosophila tomosyn revealed synchronous , asynchronous and spontaneous SV release are all elevated in the absence of the protein . While single evoked responses were enhanced , rapid depression of release was observed during train stimulation , suggesting loss of Tomosyn biases synapses toward a more phasic pattern of SV release . To directly test whether Tomosyn plays a unique role in tonic synapses , Ib and Is motoneurons were separately stimulated using optogenetics to measure their isolated contributions . These experiments revealed a 4-fold increase in output from Ib neurons with no change to Is release . Optical quantal analysis confirmed the Ib specific effect of Tomosyn and demonstrated enhanced evoked responses in tomosyn is due to higher intrinsic Pr across the entire AZ population . Endogenously-tagged Tomosyn was more abundant at Ib synapses than Is , consistent with Tomosyn’s role in regulating Ib release . Together , these data indicate the intrinsically high Pr and rapid depression normally found in Is motoneurons is due in part to a lack of Tomosyn inhibition of SV usage at phasic synapses . High-frequency stimulation experiments demonstrate Tomosyn does not regulate the size of the immediately releasable SV pool ( IRP ) but rather regulates IRP usage to ensure sustained availability of SVs during prolonged stimulation , as the IRP is strongly biased towards early release in tomosyn mutants . We propose a model where Drosophila synapses are more phasic in release character by default , with tonic release requiring higher levels of Tomosyn to generate a fusion bottleneck that enables extended periods of stable release by slowing the rate of SV usage . How Tomosyn normally suppresses SV release has been unclear ( Sakisaka et al . , 2008; Yamamoto et al . , 2010; Yizhar et al . , 2007; Yizhar et al . , 2004 ) . The most widely hypothesized mechanism is that Tomosyn competes with Syb2 for binding t-SNAREs . By forming fusion-incompetent SNARE complexes that must be disassembled by NSF , a pool of t-SNAREs is kept in reserve and can be mobilized by alleviating Tomosyn inhibition . Indeed , enhanced SNARE complex formation was found in Drosophila tomosyn mutants , consistent with the model that Tomosyn’s SNARE domain acts as a decoy SNARE to inhibit productive SNARE complex assembly . Expression of the Tomosyn scaffold alone failed to rescue the null phenotype , while overexpression of the scaffold had no effect on evoked release . As such , these data indicate that while the scaffold is required for full Tomosyn function , it does not directly inhibit fusion . Our observations are consistent with the mechanism proposed in C . elegans , but differ from studies in cultured mammalian cells suggesting the scaffold acts as an independent release suppressor by inhibiting Syt1 ( Burdina et al . , 2011; Yamamoto et al . , 2010; Yizhar et al . , 2007 ) . Characterization of Drosophila tomosyn/syt1 double mutants demonstrated Tomosyn suppresses release independent of Syt1 , arguing the scaffold must serve a function that enhances the inhibitory activity of the SNARE domain independent of Syt1 . Indeed , we found the Tomosyn SNARE motif was mislocalized without the WD40 scaffold , arguing this region indirectly supports Tomosyn’s inhibitory activity by ensuring proper localization so the SNARE domain can compete for t-SNARE binding . Similar to studies in C . elegans and mammals , we find Drosophila Tomosyn co-localized with other SV proteins ( Geerts et al . , 2017; McEwen et al . , 2006 ) . Human Tomosyn transgenes also rescued elevated evoked and spontaneous release in tomosyn mutants , indicating functional conservation of its inhibitory properties . Overexpression of either Drosophila or human Tomosyn in a wildtype background also decreased release , demonstrating presynaptic output can be bi-directionally controlled by varying Tomosyn expression levels . In addition to intrinsic release differences between tonic and phasic motoneurons , we found Tomosyn also controls presynaptic homeostatic potentiation ( PHP ) . This form of synaptic plasticity occurs when presynaptic motoneurons upregulate Pr and quantal content to compensate for decreased GluR function and smaller quantal size ( Böhme et al . , 2019; Frank , 2014; Genç and Davis , 2019; Goel et al . , 2019; Gratz et al . , 2019 ) . Inducing PHP with the allosteric GluR inhibitor Gyki revealed Tomosyn is required for expression of this form of acute PHP at Ib terminals . Removing Tomosyn inhibition at Ib synapses generates a ~ 4 fold enhancement in evoked release , more than sufficient to compensate for a twofold reduction in evoked response size from two equally contributing motoneurons . Indeed , AZ Pr mapping revealed Ib synapses potentiate in the presence of Gyki while Is terminals showed no change , indicating enhanced release from Ib is sufficient to homeostatically compensate for Gyki-induced decreases in quantal size . Although future studies will be required to determine the molecular cascade through which Tomosyn mediates PHP expression , prior work indicates PKA phosphorylation of Tomosyn reduces its SNARE binding properties and decreases its inhibition of SV release ( Baba et al . , 2005; Ben-Simon et al . , 2015; Chen et al . , 2011 ) . Given Gyki-induced PHP expression requires presynaptic PKD ( Nair et al . , 2020 ) , an attractive hypothesis is that PKD phosphorylates Tomosyn and reduces its ability to inhibit SNARE complex formation . Similar to tomosyn mutants , this could promote SV availability by generating a larger pool of free t-SNAREs to support enhanced docking of SVs at AZs . Increased docking would elevate single AZ Pr by increasing the number of fusion-ready SVs upon Ca2+ influx , similar to the effect we observed with quantal imaging . Despite the importance of Tomosyn in regulating release character between tonic and phasic motoneurons , tomosyn null mutants are viable into adulthood . As such , the entire range of Tomosyn expression can be used by distinct neuronal populations in vivo to set presynaptic output . Tonic Ib terminals shift towards phasic release with no effect on Is output in tomosyn null mutants , resulting in a collapse of presynaptic release diversity between these two neuronal subgroups . Like tomosyn , null mutants in syt7 are viable and show dramatically enhanced evoked release ( Fujii et al . , 2021; Guan et al . , 2020 ) . Tomosyn/syt7 double mutants show even greater increases in release output , arguing multiple non-essential presynaptic proteins can independently fine tune synaptic strength within the presynaptic terminal . Together , these experiments demonstrate Tomosyn is a highly conserved release inhibitor that varies in expression between distinct neuronal subtypes to regulate intrinsic Pr and plasticity , providing a robust mechanism to generate presynaptic diversity across the nervous system . Drosophila melanogaster were cultured on standard medium between 22°C and 25°C . Third instar larvae were used for all in vivo and immunostaining experiments . Adult brain extracts were used for western blot analysis . Males were preferentially used in this study to facilitate genetic crossing schemes and avoid sex-specific phenotypic differences . Tomosyn null mutants used in the study include tomosynNA1 ( this study ) , tomosynFS1 ( this study ) , and Df ( 1 ) ED7161 ( Bloomington Drosophila Stock Center ( BDSC ) #9217 ) . Strains used for rescue experiments include elavC155-GAL4 ( BDSC#8765 ) , UAS-Tom13A-6xMyc ( this study ) , UAS-Tom13A-∆SNARE-6xMyc ( this study ) , UAS-Tom13B-6xMyc ( this study ) , UAS-Tom13B-∆SNARE-6xMyc ( this study ) , UAS-4xMyc-TomSNARE ( this study ) , and UAS-HumanTom1-6xMyc ( this study ) . Double mutant experiments were performed with syt1AD4 ( DiAntonio and Schwarz , 1994 ) , syt1N13 ( Littleton et al . , 1993 ) , syt7 control ( Guan et al . , 2020 ) , and syt7M1 ( Guan et al . , 2020 ) . For single neuron optical stimulation experiments , the Ib-specific Gal4 driver GMR94G06 ( BDSC #40701 ) and the Is-specific Gal4 driver GMR27F01 ( BDSC #49227 ) was used to drive expression of UAS-ChR2-T159C ( Dawydow et al . , 2014 ) in Ib or Is motoneurons innervating larval muscle 1 . For AZ Pr mapping experiments , Mef2-Gal4 ( BDSC #27390 ) , 44H10-LexA ( provided by Gerry Rubin ) , LexAOp-myr-jGCaMP7S ( this study ) , UAS-myr-jGCaMP8s ( this study ) , GluRIIA-RFP ( provided by Stephan Sigrist ) , and GluRIIB-GFP ( provided by Stephan Sigrist ) transgenic lines were used . To generate tomosynNA1 , two guide RNAs ( gRNAs ) flanking the tomosyn locus were selected using the CRISPR Optimal Target Finder ( Gratz et al . , 2014 ) . These gRNAs were fused with the pCFD4 expression vector ( Addgene #49411 ) ( Port et al . , 2014 ) according to the Gibson assembly protocol using NEBuilder HighFidelity DNA Assembly Cloning Kit ( E5520 ) . Gibson assembly was used to generate a donor construct encoding a floxed P3> DsRed reporter cassette ( Addgene #51434 ) flanked with homology arms directly outside of the tomosyn gene isolated using PCR . These constructs were co-injected into vasa-Cas9 embryos ( BDSC #56552 ) and DsRed positive transformants were selected by BestGene Inc ( Chino Hills , CA , USA ) . To generate tomosynFS1 , the pCFD4 gRNA construct was injected without a donor , and frame shift mutants were identified by PCR and sequencing . The Cas9 chromosome was removed from both lines by backcrossing to w-/- ( BDSC #3605 ) . For both tomosynNA1 and tomosynFS1 , unmodified progeny of the CRISPR-injected embryos were used as genetic background controls . To generate tomosyn13A-Clover , gRNAs targeting exon13A of tomosyn were cloned into pCFD5 ( Addgene #73914 ) ( Port and Bullock , 2016 ) and co-injected with a donor plasmid by BestGene Inc The donor was made by amplifying homology arms from the genome by PCR and fusing them by Gibson assembly with a cDNA coding for 6xHis-mClover3 ( Addgene #74252 ) ( Bajar et al . , 2016 ) in frame with exon 13 A . To generate rescue constructs , the relevant cDNAs were synthesized by GENEWIZ , Inc ( South Plainfield , NJ , USA ) and cloned into pBid-UASc ( Addgene #35200 ) ( Wang et al . , 2012 ) using EcoRI and XbaI . These constructs were injected into embryos containing the VK27 attP acceptor site by BestGene , Inc ( BDSC #9744 ) . Positive transformants were selected and balanced . The fluorescent Ca2+ sensor GCaMP7s was tethered to the plasma membrane with an N-terminal myristoylation ( myr ) sequence . A cDNA encoding the first 90 amino acids of Src64b , containing a myristoylation target sequence , was PCR amplified from the pBid-UAS-myr plasmid ( Akbergenova et al . , 2018 ) and fused with the GCaMP7s cDNA ( Addgene # 104463 ) ( Dana et al . , 2019 ) and EcoRI/XbaI digested pBid-LexA ( a gift from Brian McCabe ) using Gibson assembly . pBid-UAS-myr-jGCaMP8s was made by fusing a GCaMP8s cDNA ( Addgene # 162374 ) ( Zhang et al . , 2020 ) with BglII/XbaI digested pBid-UAS-myr using Gibson assembly . These constructs were injected by BestGene , Inc into embryos containing the attP2 acceptor site and positive transformants were isolated ( BDSC #8622 ) . NCBI BLAST was used to identify homologs of Drosophila nSyb and Tomosyn in C . elegans , N . vectensis , M . lignano , O . sinensis , C . teleta , A . planci , D . rerio , M . musculus , and H . sapiens . The C-terminal tail of S . cerevisiase Sro7 was used as the outgroup . UCSC Genome Browser’s Cons 124 feature was used to assess sequence conservation with Drosophila tomosyn as the reference sequence ( htpps://genome . ucsc . edu/ ) . The Póle Rhône-Alpes de Bioinformatique ( PRABI; https://npsa-prabi . ibcp . fr ) coiled-coil prediction tool was used to identify the C-terminal SNARE domain of each protein and the BLOSUM62 algorithm of the Matlab 2020a seqpdist function was used to create sequence alignment . Phylogenetic trees were generated with the seqlinkage Matlab function . Protein sequences used for alignment and phylogenetic tree construction: Western blotting of adult head lysates ( ten heads per/lane ) was performed using standard laboratory procedures with mouse anti-Syx1a ( 8C3 , 1:1000; Developmental Studies Hybridoma Bank ( DSHB , Iowa City , IA ) ) anti-Myc ( GeneTex: GTX29106 , 1:1000 ) and mouse anti-Tub ( Sigma: T5168 , 1:1 , 000 , 000 ) . The boiling step was omitted to preserve the 7 S complex . IR Dye 680LT-conjugated goat anti-mouse ( 1:10 , 000 , LICOR; 926–68020 ) was used as the secondary antibody . Visualization was performed with a LI-COR Odyssey Imaging System ( LI-COR Biosciences , Lincoln , MA , USA ) and analysis was performed using the Plot Lanes and Measure Areas function of FIJI image analysis software ( Schindelin et al . , 2012 ) . Lanes with poor protein loading were excluded from analysis as described in the source data and statistics supplementary file . Immunostaining for AZ and bouton counting was performed on wandering 3rd instar larvae dissected in Ca2+-free HL3 . 1 and fixed for 7 min in Ca2+-free HL3 . 1 containing 4 % PFA . Larvae were blocked and permeabilized for 1 hr in PBS containing 0 . 1 % Triton X-100 , 2 . 5 % NGS , 2 . 5 % BSA and 0 . 1 % sodium azide . Larvae were incubated overnight with primary antibody at 4 °C and 2 hr in secondary antibody at room temperature . Samples were mounted on slides with Vectashield ( Vector Laboratories , Burlingame , CA ) . Antibodies used for immunolabeling were: rabbit anti-GFP at 1:1000 ( ab6556; Abcam , Cambridge , UK ) , mouse anti-BRP at 1:500 ( Nc82; DSHB ) , mouse anti-Synapsin at 1:500 ( 3C11; DSHB ) , rabbit anti-Myc at 1:500 ( GTX29106; GeneTex , Irvine , CA , USA ) , rabbit anti-Syt1 ( gift of Noreen Reist ) at 1:500 , and DyLight 649 conjugated anti-HRP at 1:1000 ( #123-605-021; Jackson Immuno Research , West Grove , PA , USA ) . Secondary antibodies for morphology and co-localization experiments were used at 1:500: goat anti-rabbit Alexa Fluor 488-conjugated antibody ( A-11008; Thermofisher ) and goat anti-mouse Alexa Fluor 546-conjugated antibody ( A-11030; ThermoFisher ) . The secondary antibody used for anti-GFP staining was goat anti-rabbit Alexa Flour 488-conjugated antibody ( A-11008; Thermofisher ) used at 1:500 . Immunoreactive proteins were imaged at segments A3 and A4 of muscle fiber four for all experiments , except for anti-GFP staining , which was imaged at muscles 6/7 . Images were acquired on a PerkinElmer Ultraview Vox spinning disk confocal microscope system using a 63 x oil immersion objective . Ib and Is terminals were identified based on bouton and NMJ size , with Is having characteristically smaller boutons and total NMJ size . NMJ morphology , staining intensity , and co-localization between channels were analyzed using Volocity 6 . 3 . 1 software . Postsynaptic currents from the indicated genotypes were recorded from 3rd instar larvae at muscle fiber 6 ( unless otherwise noted ) of segment A4 using two-electrode voltage clamp with a −80 mV holding potential in HL3 . 1 saline solution ( in mM , 70 NaCl , 5 KCl , 10 NaHCO3 , 4 MgCl2 , 5 trehalose , 115 sucrose , 5 HEPES , pH 7 . 18 ) as previously described ( Jorquera et al . , 2012 ) . Final [Ca2+] was adjusted to the level indicated . All electrophysiology experiments were performed at room temperature . Inward currents recorded during TEVC are labeled on a reverse axis in the figures for simplicity . Asynchronous release contribution was approximated by fitting the weighted average of two logarithmic regressions with separate time constants to the normalized cumulative charge transfer of evoked responses as previously described ( Jorquera et al . , 2012 ) . The Ca2+ cooperativity of release was determined from the Hill coefficient of a 4-parameter logistic regression of evoked responses fit to the linear range ( 0 . 1–0 . 75 mM Ca2+ ) . Data acquisition and analysis was performed using Axoscope 9 . 0 and Clampfit 9 . 0 software ( Molecular Devices , Sunnyvale , CA , USA ) . mEJCs were analyzed with Mini Analysis software 6 . 0 . 3 ( Synaptosoft , Decatur , GA , USA ) . Motor nerves innervating the musculature were severed and placed into a suction electrode . Action potential stimulation was applied at 0 . 33 Hz ( unless indicated ) using a programmable stimulator ( Master8 , AMPI; Jerusalem , Israel ) . Optogenetic experiments were performed in the same way with the following modifications . Postsynaptic currents were recorded from 3rd instar larvae at segment A4 of muscle fiber 1 . Evoked postsynaptic currents were generated using the Master8 stimulator and an LED driver ( LED-D1B , THORLABS , Newton , NJ , USA ) to generate 470 nm light pulses from an attached LED ( M470F3 , THORLABS , Newton , NJ , USA ) . Ib and Is currents were separately evoked by driving expression of ChR2 ( UAS-ChR2-T159C , provided by Robert Kittel ) under the control of GMR94G06-Gal4 ( BDSC #40701 ) or GMR27F01-Gal4 ( BDSC# 49227 ) , respectively . Gyki was diluted fresh each day in HL3 . 1 to a final concentration of 10 µM . The final Ca2+ concentration was adjusted to the level indicated . The Gyki solution was bath applied to fully dissected larvae for 15 minutes as previously described ( Nair et al . , 2020 ) . Subsequent recordings were performed in the continued presence of bath applied Gyki . Gyki was used instead of the GluR blocker Philanthotoxin-433 ( PhTX ) , as PhTX requires a partially dissected preparation capable of muscle contraction for PHP induction . This more intact preparation is not compatible with imaging AZ release before and after PHP . In contrast , PHP expression can occur in a fully stretched preparation following Gyki application . AZ Pr mapping experiments were performed on a Zeiss Axio Imager equipped with a spinning-disk confocal head ( CSU-X1; Yokagawa , Japan ) and ImagEM X2 EM-CCD camera ( Hamamatsu , Hamamatsu City , Japan ) as previously described ( Akbergenova et al . , 2018 ) . For Pr mapping of tomosynFS1 , myristoylated-GCaMP7s was expressed in larval muscles with GMR44H10-LexA ( provided by Gerald Rubin ) . Individual PSDs were visualized at segments A2-A4 of muscle fiber four by expression of GluRIIA-RFP and GluRIIB-GFP ( hereafter referred to as GluR ) under control of their endogenous promoters ( provided by Stephan Sigrist ) . An Olympus LUMFL N 60 X objective with a 1 . 10 NA was used to acquire GCaMP7s imaging data at 8 Hz . Third instar larvae were dissected in Ca2+-free HL3 containing 20 mM MgCl2 . After dissection , preparations were maintained in HL3 with 20 mM MgCl2 and 1 . 0 mM Ca2+ for 5 min . A dual channel multiplane stack was imaged at the beginning of each experiment to identify GluR-positive PSDs . Single focal plane videos were then recorded while motoneurons innervating the muscles were stimulated with a suction electrode at 0 . 3 Hz for 3 min . GluR PSD position was re-imaged every 25 s during experimentation . The dual channel stack was merged with single plane images using the max intensity projection algorithm from Volocity 6 . 3 . 1 software . The position of all GluR PSDs was then added to the myr-GCaMP7s stimulation video . GluR positive PSDs were detected automatically using the spot finding function of Volocity and equal size ROIs were assigned to the PSD population . In cases where the software failed to label visible GluR PSDs , ROIs were added manually . GCaMP7s peak flashes were detected and assigned to ROIs based on centroid proximity . Evoked events were identified as frames with three or more simultaneous GCaMP events across the arbor . The time and location of Ca2+ events were imported into Excel or Matlab for further analysis . Evoked GCaMP events per ROI were divided by the number of stimulations to calculate AZ Pr . AZ Pr experiments with Gyki were performed in the same way with the following modifications . Mef2-Gal4 ( BDSC #27390 ) was used to drive expression of UAS-myr-GCaMP8s in larval muscles . Dissected preparations were maintained in HL3 containing 10 mM MgCl2 and 0 . 5 mM Ca2+ and imaged at muscle fibers 6/7 . The HL3 solution was exchanged for an identical solution containing 10 µM Gyki and incubated for 15 min . A second imaging session was recorded at each NMJ after Gyki incubation . AZ locations were identified by labeling peaks for all events and regions of highest peak densities were assigned as ROIs . Release events were assigned to ROIs using the centroid proximity algorithm in Volocity 6 . 3 . 1 . TomosynNA1 and control 3rd instar larvae were dissected in Ca2+-free HL3 saline and fixed in 1 % glutaraldehyde , 4 % formaldehyde and 0 . 1 M sodium cacodylate buffered saline ( CBS ) with 1 mM magnesium chloride for 10 min at room temperature as previously described ( Akbergenova and Bykhovskaia , 2009 ) . After fixative exchange , samples were microwaved in a BioWave Pro Pelco ( Ted Pella , Inc , Redding , CA , USA ) using the following fixation protocol: ( 1 ) 100 W 1 min , ( 2 ) 1 min off , ( 3 ) 100 W 1 min , ( 4 ) 300 W 20 s , ( 5 ) 20 s off , ( 6 ) 300 W 20 s . Steps 4–6 were repeated twice more . Samples were then washed in CBS and stained en bloc for 30 min in 1 % osmium tetroxide . Following another CBS wash , samples were stained en bloc for 30 min in 2 % uranyl acetate and briefly incubated in sequentially anhydrous solutions of ethanol and then pure anhydrous acetone . Epoxy resin infiltration was performed by incubating the dehydrated samples in a series of acetone/epoxy mixtures , with the acetone percentage decreasing in each successive step ( Embed 812; Electron Microscopy Sciences ) . Thin sections ( 40–50 nm ) were collected on Formvar/carbon-coated copper slot grids and stained on grid for ~5 min with lead citrate . Sections were imaged at ×49 , 000 magnification at 120 kV using a Tecnai G2 electron microscope ( FEI , Hillsboro , OR , USA ) equipped with a charge-coupled device camera ( Gatan , Pleasanton , CA , USA ) . Micrographs of type Ib boutons from segment 4 of muscle fibers 6/7 were analyzed using Volocity 6 . 3 . 1 . SV centers were annotated as points , T-bar bases as single pixel ROIs , and electron densities as contoured lines . Distances between these features were calculated using the Measure Distance function to determine SV spacing , SV number , and docked SV number ( SVs with centers<50 nm to the electron dense AZ ) . Statistical analysis and graphing were performed with GraphPad Prism ( San Diego , CA , USA ) . In two cases , outliers were identified and removed using the default settings of the Identify Outlier function in Prism9 ( mini frequency of elav-Gal4 , tomosynNA1 in Figure 2Q , excluded mini frequency was 23 . 3 Hz; mini frequency of syt7M1 in Figure 4L , excluded mini frequency was 6 . 20 Hz ) . Electrophysiological traces were generated using the plot function in Matlab R2020A ( MathWorks , Natick , MA , USA ) . Statistical significance was determined using Student’s t test for comparisons between two groups , or a One-way ANOVA followed by Tukey’s multiple comparisons test for comparisons between three or more groups unless noted . In the figures , the center of each distribution is plotted as the median value and reported in the figure legends as the median , mean ± SEM , n . In the main text , the centers and n are reported as mean ± SEM , n . In all cases , n represents the number of individual NMJs analyzed unless otherwise noted . The number of larvae used per group in each experiment is indicated in the figure legends . Asterisks in the figures denote p-values of: * , p ≤ 0 . 05; , p ≤ 0 . 01; * , p ≤ 0 . 001; and **** , p ≤ 0 . 0001 . The Source Data and Statistical Analysis excel file contains individual spreadsheets labeled with figure number and includes all primary source data and statistical comparisons .	Nerve cells transmit messages in the form of electrical and chemical signals . Electrical impulses travel along a neuron to the junction between two neighbouring cells , the synapse . There , chemical messengers called neurotransmitters are released from one cell and detected by the next , which can either excite or inhibit the recipient cell . Synapses differ in their ability to propagate signals and their signalling activity also fluctuates at times . Moreover , synaptic connections can be strengthened or weakened in a process called plasticity , which is a key part of learning new skills and recovering from a brain injury . It is thought that synaptic signalling might be amped up or dialled down to change the output of the connection between two cells , but exactly how this happens remains unclear . To investigate why synapses differ and how their signalling capabilities change , Sauvola et al . examined the connections between neurons and muscle cells in developing fruit flies . In fruit fly larvae , two types of neurons – called tonic Ib and phasic Is neurons – form synapses with muscle cells . But their synapses have different signalling properties: Ib synapses are weaker than Is synapses . Sauvola et al . hypothesised that a protein called Tomosyn – which is thought to restrict chemical signalling at the synapse – might be more active at weaker Ib synapses . Sauvola et al . found that Tomosyn was indeed more abundant at Ib synapses than at Is synapses , appearing to reflect their differences in signalling properties . In flies engineered to lack the Tomosyn protein , Ib synapses became four times stronger than usual , while Is synapses hardly changed . This supports the idea that Tomosyn restricts the release of neurotransmitters at typically weak Ib synapses . Further experiments showed Ib synapses in flies lacking Tomosyn also lost their malleability and ability to become strengthened during synaptic plasticity . Though the precise molecular interactions need further investigation , the findings suggest that Tomosyn is required for some forms of synaptic plasticity by controlling how much chemical signal neurons release . In summary , this work advances our understanding of synaptic signalling and brain plasticity , showing once again how the brain can change itself .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2021	The decoy SNARE Tomosyn sets tonic versus phasic release properties and is required for homeostatic synaptic plasticity
Follicle rupture , the final step in ovulation , utilizes conserved molecular mechanisms including matrix metalloproteinases ( Mmps ) , steroid signaling , and adrenergic signaling . It is still unknown how follicles become competent for follicle rupture/ovulation . Here , we identify a zinc-finger transcription factor Hindsight ( Hnt ) as the first transcription factor regulating follicle’s competency for ovulation in Drosophila . Hnt is not expressed in immature stage-13 follicle cells but is upregulated in mature stage-14 follicle cells , which is essential for follicle rupture/ovulation . Hnt upregulates Mmp2 expression in posterior follicle cells ( essential for the breakdown of the follicle wall ) and Oamb expression in all follicle cells ( the receptor for receiving adrenergic signaling and inducing Mmp2 activation ) . Hnt’s role in regulating Mmp2 and Oamb can be replaced by its human homolog Ras-responsive element-binding protein 1 ( RREB-1 ) . Our data suggest that Hnt/RREB-1 plays conserved role in regulating follicle maturation and competency for ovulation . Ovulation is a complex process of releasing fertilizable oocytes from mature follicles and is essential for animal reproduction ( Espey and Richards , 2006 ) . To ensure successful ovulation , a follicle must be developed to full maturity to be competent to receive an ovulatory stimulus and to activate proteolytic systems for follicle rupture . Several proteolytic systems have been found to regulate follicle rupture in vertebrates , including matrix metalloproteinase ( Mmp ) , plasminogen activator/plasmin , and ADAMS-TS ( Curry and Smith , 2006; Takahashi et al . , 2013 ) . In addition , a surge of luteinizing hormone ( LH ) serves as a master regulator to initiate the ovulation event and activates the EGF/EGFR-Ras-MAPK signaling pathway to propagate the ovulatory signal from outer granulosa cells to inner cumulus cells in the preovulatory follicles ( Conti et al . , 2012; Fan et al . , 2009 , 2011 , 2012; Hsieh et al . , 2007 ) . However , molecular mechanisms coupling the Ras-MAPK pathway to the activation of proteolytic systems for follicle rupture are largely unknown . Ovulation in Drosophila utilizes conserved molecular mechanisms and involves a follicle rupture process to release mature oocytes from the ovary . Drosophila have two ovaries , connected at their posterior ends by bilateral oviducts ( Figure 1 ) . Each ovary contains ~16 ovarioles , where egg chambers are assembled in the germarium at the anterior and develop through 14 characteristic stages toward the posterior end ( Spradling , 1993 ) . Each egg chamber contains one oocyte and 15 nurse cells surrounded by a layer of somatic follicle cells . In stage-14 egg chambers ( also named mature follicles ) , all nurse cells are degraded , leaving an oocyte surrounded by follicle cells; Matrix metalloproteinase 2 ( Mmp2 ) is upregulated in posterior follicle cells ( Figure 1; Deady et al . , 2015 ) . In addition , Oamb ( octopamine receptor in mushroom body ) , encoding an α-adrenergic receptor-like G-protein-coupled receptor for octopamine ( OA ) , is also upregulated in all follicle cells of stage-14 egg chambers ( Lee et al . , 2003; Deady and Sun , 2015 ) . OA , released from terminal nerves that innervate ovaries , activates Oamb receptor in stage-14 follicle cells , which induces calcium rise and activates Mmp2 ( Deady and Sun , 2015; Heifetz et al . , 2014; Middleton et al . , 2006; Monastirioti , 2003 ) . Mmp2 enzymatic activity leads to degradation of posterior follicle cells and release of the encapsulated oocyte ( called follicle rupture; Figure 1; Deady et al . , 2015 ) . The rest of the follicle cells remain at the end of the ovariole to form a corpus luteum ( Deady et al . , 2015 ) . Local adrenergic signaling has also been suggested to regulate mammalian ovulation but no molecular mechanisms have been illustrated ( Kannisto et al . , 1985; Schmidt et al . , 1985 ) . In parallel to progesterone signaling in mammalian ovulation , ecdysteroid signaling is also activated in stage-14 follicle cells and is essential for Drosophila ovulation; ecdysteroid signaling modulates OA/Oamb-induced Mmp2 activation , but does not affect Oamb expression nor Mmp2 expression ( Knapp and Sun , 2017 ) . Thus , it is currently unknown what induces Mmp2 and Oamb expression in stage-14 follicle cells and how these follicles become fully competent for ovulation . The zinc-finger transcription factor Hindsight ( Hnt; encoded by gene pebbled ) contains 14 C2H2 zinc-finger domains and is homologous to mammalian Ras-responsive element-binding protein 1 ( RREB-1 ) . Both Hnt and RREB-1 bind to similar DNA sequences , and human RREB-1 can functionally replace Hnt in attenuating expression of nervy and hnt itself in Drosophila salivary gland ( Ming et al . , 2013 ) . RREB-1 functions downstream of the Ras-MAPK pathway to either suppress or promote Ras target genes in multiple tissues including colon , thyroid , and pancreatic cancers ( Kent et al . , 2010 , 2013; Mukhopadhyay et al . , 2007; Thiagalingam et al . , 1996; Zhang et al . , 2003 ) . Hnt is expressed in a variety of tissues in development and plays multiple developmental roles including control of embryonic germ band retraction ( Yip et al . , 1997; Reed et al . , 2001 ) , regulation of retinal cell fate and morphogenesis ( Pickup et al . , 2002; Wilk et al . , 2004; Pickup et al . , 2009; Oliva and Sierralta , 2010; Oliva et al . , 2015 ) , maintenance of tracheal epithelial integrity ( Wilk et al . , 2000 , 2004 ) , and differentiation of spermathecae and intestinal stem cells ( Sun and Spradling , 2013; Baechler et al . , 2015 ) . Hnt is also expressed in follicle cells of stage 7-10A egg chambers , where it functions as a downstream target of Notch signaling to suppress Hedgehog signaling and to induce the mitotic/endocycle transition ( Sun and Deng , 2007 ) . Hnt continues its expression in anterior follicle cells throughout late oogenesis . In contrast , Hnt expression in mainbody follicle cells is downregulated from stage 10B to stage 13 and re-upregulated in stage-14 ( Deady et al . , 2015 ) , where its role is unknown . Moreover , few downstream targets of Hnt have been identified and its relationship to Ras signaling is also unknown . Here , we characterized the dynamic expression of Hnt in stage-14 follicle cells . By using molecular and genetic tools , we demonstrated that Hnt expression in stage-14 follicle cells is essential for follicle rupture partly by upregulation of Oamb and Mmp2 expression in these follicle cells . Thus , Hnt functions as an essential transcription factor to prime follicles to be competent for follicle rupture/ovulation . In addition , Hnt’s role in follicle rupture can be replaced by human RREB-1 . Our data , along with the involvement of Ras-MAPK signaling in mammalian ovulation , lead us to propose that Hnt/RREB-1 has a conserved role in regulating follicle rupture/ovulation downstream of Ras-MAPK signaling pathway . Hnt is not expressed in stage-13 follicle cells except those at the anterior region; however , it is upregulated in all stage-14 follicle cells and the corpus luteum ( Deady et al . , 2015 ) . Upon closer examination , we found three distinct patterns of Hnt expression throughout stage-14 egg chambers: ( I ) high Hnt expression in anterior and posterior but low/no Hnt in the middle follicle cells ( ‘A/P-Hnt’ egg chambers; Figure 2A and F , Figure 2—figure supplement 1A–B ) ; ( II ) high Hnt expression in all follicle cells ( ‘high-Hnt’ egg chambers; Figure 2B and G ) ; ( III ) low Hnt expression in all follicle cells ( ‘low-Hnt’ egg chambers; Figure 2C and H; also see Figure 2—figure supplement 1C–D ) . To determine the developmental sequence of the aforementioned three types of stage-14 egg chambers , we analyzed the expression patterns of Hnt against the expression of a stage-14 follicle-cell Gal4 driver ( 44E10-Gal4; renamed as FC1 for simplicity ) ( Deady and Sun , 2015 ) and the number of nurse cell nuclei in these egg chambers . 86% of A/P-Hnt egg chambers had medium-level GFP expression driven by FC1 , while more than 73% of high-Hnt and 80% of low-Hnt egg chambers had high-level GFP expression ( Figure 2A–D ) . This indicates that A/P-Hnt egg chambers are the youngest , which is consistent with the observation that A/P-Hnt egg chambers typically have more residual nurse-cell nuclei than the other two types of egg chambers ( Figure 2E ) . In addition , high-Hnt egg chambers still contained one or two nurse-cell nuclei , while low-Hnt egg chambers typically did not contain nurse-cell nuclei and were skinner and dehydrated ( Figure 2E and Figure 2—figure supplement 1D ) . Because nurse-cell nuclei are progressively degraded starting around stage-12 by a non-cell-autonomous mechanism to generate fully matured egg chambers , which have no nurse cell nuclei and are dehydrated ( Drummond-Barbosa and Spradling , 2004; Timmons et al . , 2016 ) , the above analysis demonstrates that high-Hnt egg chambers are at the intermediate stage , while low-Hnt egg chambers are the most mature egg chambers . This conclusion was further supported by additional analysis using a late stage-14 follicle-cell Gal4 driver ( 47A04-Gal4; renamed as FC2 for simplicity ) ( Deady and Sun , 2015 ) . Consistent with the previous result , both A/P- and high-Hnt egg chambers had no or minimal GFP expression driven by FC2 , while low-Hnt egg chambers had highest GFP expression and fewest nurse-cell nuclei ( Figure 2F–J ) . Altogether , these analyses demonstrate that Hnt is first upregulated in posterior follicle cells , filled in across the entire egg chamber , and then overall downregulated in follicle cells of fully matured egg chambers ( Figure 2K ) . Therefore , we propose to categorize stage-14 egg chambers into three distinct stages and rename A/P- , high- , and low-Hnt egg chambers as stage-14A , stage-14B , and stage-14C egg chambers , respectively ( Figure 2K ) . The dynamic Hnt expression in stage-14 follicle cells prompted us to investigate its function in follicle maturation and ovulation . To bypass the early requirement of Hnt in follicle cell differentiation ( Sun and Deng , 2007 ) , we used RNA interference ( RNAi ) to deplete Hnt expression specifically in stage-14 follicle cells with FC1 or FC2 Gal4 driving UAS-hntRNAi expression . While FC1 started to be expressed in stage-14A follicle cells , it was weak to deplete Hnt expression in stage-14A follicle cells ( Figure 3—figure supplement 1A–D ) , but became progressively more efficient in older follicle cells with two independent hntRNAi lines ( Figure 3A–D and Figure 3—figure supplement 1E–H ) ; more than 80% of stage-14C egg chambers had no detectable Hnt expression in their follicle cells . In contrast , FC2 started to be expressed in stage-14B follicle cells and most effectively depleted Hnt expression in stage-14C follicle cells except with hntRNAi2 , which only showed strong reduction in ~43% of egg chambers ( Figure 3E–H and Figure 3—figure supplement 1I–P ) . Females with RNAi-mediated hnt depletion in stage-14 follicle cells ( named hntRNAi females for simplicity ) were then assayed for egg-laying ability . hntRNAi females laid significantly fewer eggs than control females after mating ( Figure 3I ) . This phenotype was manifested by using both stage-14 follicle-cell Gal4 drivers and with two independent hntRNAi lines , which demonstrates that Hnt expression in stage-14 follicle cells was essential for normal egg laying . The decrease in egg-laying number was not caused by an oogenesis defect , as plenty of stage-14 egg chambers were observed before and after egg-laying experiments ( Figure 3J and Figure 3—figure supplement 2 ) . The egg-laying process consists of ovulation ( the release of egg from the ovary into the oviduct ) , egg transportation through the oviduct , and oviposition ( the release of egg in the uterus to the outside environment ) . To determine which step in the egg-laying process was affected in hntRNAi females , we utilized our previously developed method to estimate the average time required for each step in the egg-laying process ( Sun and Spradling , 2013; Deady and Sun , 2015; Knapp and Sun , 2017 ) . Consistent with our previous data , control females spent 12–14 min to ovulate an egg , less than a minute to transport egg through the oviduct , and eight - 10 min to hold an egg in the uterus and oviposit ( Figure 3K–L and Supplementary file 1 ) . In contrast , hntRNAi females spent more than 25 min to ovulate an egg , which was significantly longer than the control females ( Figure 3K–L and Supplementary file 1 ) . These data demonstrate that Hnt in stage-14 follicle cells is required for normal ovulation . Ovulation consists of a breakdown of posterior follicle cells and a subsequent rupture of oocyte into the lateral oviduct ( Figure 1 ) , which is induced by octopaminergic signaling and can be recapitulated in an ex vivo culture system ( Deady and Sun , 2015 ) . The requirement of Hnt for normal ovulation led us to hypothesize that Hnt is required for OA-induced follicle rupture . Consistent with this idea , about 45% stage-14 egg chambers isolated according to FC1 expression from control females ruptured in response to OA stimulation , whereas fewer than 10% of egg chambers from hntRNAi females ruptured in response to OA stimulation ( Figure 4A–D ) . In addition , more than 85% stage-14 egg chambers isolated according to FC2 expression from control females ruptured in response to OA stimulation ( Figure 4E and H ) , consistent with the fact that FC2 is expressed in more mature egg chambers than FC1 ( Figure 2 ) . In contrast , egg chambers isolated from hntRNAi1 or hntRNAi2 females with FC2 ruptured at the rate of 10% and 33% , respectively ( Figure 4F–H ) . Consistent with this result , follicles isolated from hnt transheterozygous mutant females also showed significant reduction in OA-induced follicle rupture in comparison to control follicles ( Figure 4—figure supplement 1 ) . These results demonstrate that Hnt is required in stage-14 follicle cells for follicle rupture . Mmp2 is essential for follicle rupture , and its enzymatic activity is activated by OA stimulation ( Deady and Sun , 2015 ) . To determine whether Hnt regulates follicle rupture by influencing Mmp2 activity , we assayed Mmp2 enzymatic activity in egg chambers from control and hntRNAi females after OA stimulation ex vivo . After a three-hour incubation with OA , ~60% of control egg chambers isolated according to FC1 expression had posterior gelatinase activity ( Figure 5A and D ) , whereas only ~25% of hntRNAi egg chambers had posterior gelatinase activity ( Figure 5B–D ) . In addition , about 90% of control egg chambers isolated according to FC2 had posterior gelatinase activity , in contrast to 25% and 47% of hntRNAi1 and hntRNAi2 egg chambers , respectively ( Figure 5E–H ) . The proportion of follicles with posterior gelatinase activity was correlated to the proportion of follicles that ruptured , and both were significantly decreased in hntRNAi egg chambers , which strongly supports that Hnt controls follicle rupture by regulating Mmp2 activity . To further support this notion and to avoid the possibility that the above observed phenomenon is an artifact of ex vivo culture , we determined whether Hnt indeed regulates Mmp2 activity in vivo . One of the known substrates of Mmp2 is the basement-membrane ( BM ) protein collagen IV , encoded by viking ( vkg ) , during imaginal disc morphogenesis and fatbody dissociation at pupal development ( Srivastava et al . , 2007; Jia et al . , 2014 ) . Vkg is detected in the basement membrane of follicle cells throughout oogenesis and we reasoned it could be a substrate of follicular Mmp2 as well . We found that 70% of FC2-expressing egg chambers had lost follicular Vkg protein in a large posterior area ( open-BM configuration ) , 26% had lost Vkg protein in a small posterior area ( broken-BM configuration ) , and 4% had intact Vkg protein ( intact-BM configuration ) at their posterior end ( Figure 5I–J and O ) . When tissue inhibitor of matrix metalloproteinase ( Timp , encoding an inhibitor of Mmp enzymatic activity ) was overexpressed in stage-14 follicle cells using FC2 , the BM configuration was dramatically shifted toward intact-BM configuration ( Figure 5K and O ) , indicating that Mmp activity is responsible for the degradation of Vkg at the posterior end of stage-14 egg chambers . In addition , RNAi-mediated Mmp2 depletion in stage-14 follicle cells showed a similar trend as overexpression of Timp , although less effectively ( Figure 5L and O ) , demonstrating that Mmp2 is , at least partially , responsible for the Vkg degradation . Furthermore , hnt depletion in stage-14 follicle cells also shifted BM configuration toward broken- and intact-BM configuration as Mmp2 depletion ( Figure 5M–O ) . Altogether , these data demonstrate that Hnt regulates Mmp2 activity , which is responsible for Vkg degradation at the posterior end of stage-14 egg chambers during ovulation . OA binds to Oamb receptor in stage-14 follicle cells , which leads to a rise of intracellular Ca2+ concentration and subsequent activation of Mmp2 ( Deady and Sun , 2015 ) . To elucidate the mechanism of Hnt in regulating Mmp2 activity , we sought to determine whether Hnt interferes with OA/Oamb-Ca2+-Mmp2 pathway upstream and/or downstream of Ca2+ rise . In comparison to OA , Ca2+ ionophore ionomycin was sufficient to induce more than 95% control egg chambers to rupture at the end of a three-hour culture , regardless whether egg chambers were isolated according to FC1 or FC2 expression ( Figure 6A , D , E and H ) . In contrast , ionomycin was still not sufficient to induce follicle rupture in hntRNAi egg chambers ( except those with FC2 driving hntRNAi2 expression; Figure 6B–D and F–H ) , despite that it was able to induce Ca2+ rise in follicle cells ( Figure 6—figure supplement 1A–D , Videos 1–3 ) . The incompetency of hntRNAi egg chambers to ionomycin stimulation indicates that Hnt regulates components downstream of Ca2+ rise in the OA/Oamb-Ca2+-Mmp2 pathway; the almost normal response to ionomycin but defective response to OA in hntRNAi2 egg chambers with FC2 indicates that Hnt also regulates components upstream of Ca2+ rise in the OA/Oamb-Ca2+-Mmp2 pathway . Consistent with this idea , OA was not sufficient to induce Ca2+ rise in hntRNAi egg chambers with FC2 ( Figure 6—figure supplement 1E–H , Videos 4–6 ) . Since Hnt is first upregulated in posterior follicle cells ( Figure 2 ) , where Mmp2 is expressed ( Deady et al . , 2015 ) , and then swept across the entire follicle cells ( Figure 2 ) , where Oamb is expressed ( Lee et al . , 2003; Deady and Sun , 2015 ) , we hypothesize that Hnt regulates both Mmp2 and Oamb expression in stage-14 follicle cells . To investigate the role of Hnt in Mmp2 expression , we examined Mmp2 expression using a Mmp2::GFP fusion gene in the endogenous locus . Consistent with our previous report , Mmp2::GFP was detected in posterior follicle cells of stage-14 egg chambers , most prominently in stage14B and 14C ( Figure 6I and Figure 6—figure supplement 2A ) . Mmp2::GFP formed a gradient that peaked at the posterior tip and gradually decreased toward the anterior . In contrast , there was marked reduction of Mmp2::GFP intensity in hntRNAi egg chambers ( Figure 6J–K and Figure 6—figure supplement 2B–C ) . More than 80% of FC1- or FC2-expressing control egg chambers had moderate or high-level of Mmp2::GFP expression in their posterior follicle cells , while fewer than 30% of hntRNAi egg chambers ( 32% in the case of FC2 driving hntRNAi2 ) had moderate or high-level of Mmp2::GFP expression ( Figure 6L–M ) . Due to technical challenges , we were unable to quantify Mmp2 protein level directly using western blotting . However , we speculated that Hnt might regulate Mmp2 transcription . Therefore , we used real-time RT-PCR to quantify Mmp2 mRNA level in control and hntRNAi egg chambers . Consistent with this hypothesis , Mmp2 mRNA levels were significantly decreased in hntRNAi egg chambers in comparison to the control ( Figure 6N–O ) . Altogether , these data demonstrate that Hnt regulates Mmp2 expression at the transcriptional level . We noticed that hntRNAi2 egg chambers with FC2 Gal4 had slightly weaker reduction of Mmp2 mRNA and protein expression ( Figure 6M and O ) and responded normally to ionomycin stimulation ( Figure 6H ) , but were defective in OA-induced Ca2+ rise , Mmp2 activation , and follicle rupture ( Figures 4H and 5H , and Figure 6—figure supplement 1E–H ) . This suggests that components upstream of Ca2+ rise , for example Oamb , are defective in these egg chambers . Consistent with this hypothesis , Oamb mRNA was reduced two or more fold in hntRNAi egg chambers regardless the Gal4 drivers or hntRNAi lines ( Figure 7A–B ) . Therefore , Hnt is also required for Oamb expression in stage-14 follicle cells . Next , we aimed to rescue the rupture defect of hntRNAi egg chambers by overexpression of Oamb . Oamb overexpression was not able to restore the competency to OA-induced rupture in hntRNAi egg chambers with FC1 or hntRNAi1 egg chambers with FC2 , but it was able to do so in hntRNAi2 egg chambers with FC2 ( Figure 7C–D ) , consistent with the ionomycin experiment ( Figure 6D and H ) . These data suggest that the major defect in hntRNAi2 egg chambers with FC2 is the disruption of Oamb expression , while hntRNAi1 egg chambers with FC2 or hntRNAi egg chambers with FC1 have more defects than Oamb alone , such as Mmp2 expression . The combination of later FC2 and weaker hntRNAi2 may explain why only Oamb is majorly affected in this genetic manipulation . In addition , we noticed that egg chambers with Oamb overexpression alone initiated rupture before being able to perform the ex vivo culture ( i . e . few intact egg chambers could be isolated ) . This is likely due to its high Oamb expression , which leads to high sensitivity to very low amount of endogenous OA released during egg chamber isolation . Nevertheless , all these data support the notion that Hnt transcriptionally upregulates Mmp2 expression in posterior follicle cells and then Oamb expression in all follicle cells to make stage-14 egg chambers to be competent to respond to OA-induced follicle rupture . To address whether the role of Hnt in stage-14 follicle cells can be replaced by its human homolog RREB-1 , we first aimed to rescue the defects of hntRNAi egg chambers with hnt overexpression using hntEP55 ( see Materials and methods ) . To our surprise , overexpression of hnt in hntRNAi egg chambers did not rescue their defect in OA-induced follicle rupture ( Figure 8—figure supplement 1A and C ) . In addition , these females laid similar numbers of eggs as hntRNAi females ( Figure 8—figure supplement 1B and D ) . Surprisingly , Hnt protein was still depleted despite using FC1 or FC2 Gal4 driving hntEP55 expression ( Figure 8—figure supplement 1E–L ) , indicating that hntRNAi is sufficient to disrupt overexpressed hnt mRNA . This was further validated in a flip-out Gal4 system , in which Hnt protein was greatly reduced in cells with both hntRNAi and hntEP55 ( Figure 8—figure supplement 1M–P ) . Despite the failure to rescue ovulation in hnt-depleted females , it is interesting to note that overexpression of hnt alone with FC1 or FC2 Gal4 driver enhanced and suppressed OA-induced follicle rupture , respectively ( Figure 8—figure supplement 1A and C ) , suggesting that dynamic upregulation and downregulation of Hnt in stage-14 follicle cells may be required for normal function of these cells . Next , a functional RREB-1::GFP fusion gene was overexpressed in hntRNAi females with FC1 to see whether RREB-1 could rescue the ovulation defect of hntRNAi females . RREB-1 is successfully overexpressed in hntRNAi egg chambers , and overexpression of RREB-1 did not affect Hnt expression in control nor hntRNAi egg chambers ( Figure 8—figure supplement 2A–F ) . hntRNAi2/RREB-1::GFP females showed significant increase of egg-laying number in comparison to hntRNAi2 females , indicating a rescue of ovulation defect ( Figure 8A ) . This is supported by the result that hntRNAi2/RREB-1::GFP females spent 13 min , in comparison to 27 min in hntRNAi2 females , to ovulate an egg , close to that in control females ( Supplementary file 1 ) . In contrast , females with hntRNAi1/RREB-1::GFP laid significantly fewer eggs than females with hntRNAi1 alone ( Figure 8A ) and spent even longer time to ovulate an egg ( Supplementary file 1 ) . In addition , we noticed that these females frequently have eggs in the oviduct ( Supplementary file 1 ) , which may be caused by more frequent and uncoordinated follicle rupture leading to egg jamming in the oviduct . The persistence of egg in the oviduct may feedback to the ovary to inhibit further ovulation in vivo . To more directly investigate the role of RREB-1 in ovulation , we isolated stage-14 egg chambers and performed OA-induced follicle rupture ex vivo . Excitingly , RREB-1::GFP overexpression was sufficient to rescue the rupture defect of hntRNAi egg chambers ( Figure 8B–F ) , whereas overexpression of UAS-GFP was insufficient ( Figure 8—figure supplement 3 ) . In addition , overexpression of RREB-1 alone led to increased OA-induced follicle rupture , similar to overexpression of hnt with FC1 ( Figure 8B and Figure 8—figure supplement 1A ) . Consistent with the rescue of follicle rupture , both Mmp2 and Oamb mRNA expression was rescued to normal or even higher level by overexpression of RREB-1 ( Figure 8G–H ) . Therefore , RREB-1 can replace Hnt’s role in upregulating Mmp2 and Oamb expression in follicle cells . Altogether , our data demonstrate that zinc-finger transcription factor Hnt/RREB-1 may play conserved roles in promoting follicle maturation and ovulation competency . Work in this study demonstrated for the first time that Hnt has a dynamic expression pattern in stage-14 follicle cells and is a key factor for the final maturation of stage-14 egg chambers ( Figure 9 ) . Oocyte maturation has been well studied in Drosophila and other species ( Eichhorn et al . , 2016; Kronja et al . , 2014; Von Stetina and Orr-Weaver , 2011 ) ; however , the maturation of follicle cells surrounding the oocyte in the stage-14 egg chamber is poorly defined at the molecular level ( Duhart et al . , 2017; Klusza and Deng , 2011; Spradling , 1993 ) . According to Hnt expression in stage-14 egg chambers , we define the stage-14 egg chambers into three sub stages . Hnt is first upregulated in posterior follicle cells of stage-14A egg chambers , which is likely corresponding to Hnt’s role in upregulating Mmp2 expression in these follicle cells ( Figure 9 ) . Then Hnt is upregulated in all main-body follicle cells of stage-14B egg chambers , which is likely corresponding to Hnt’s role in upregulating Oamb expression ( Figure 9 ) . The sequential upregulation of Mmp2 and Oamb is fully consistent with the fact that FC1-expressing egg chambers , in comparison to FC2-expressing egg chambers , are less efficient for OA-induced follicle rupture , but fully competent to respond to ionomycin-induced follicle rupture ( Figures 4D , H , 6D and H ) . The orchestrated upregulation of Mmp2 and Oamb , and possibly other components in the OA/Oamb-Ca2+-Mmp2 pathway , by Hnt makes the final stage-14C egg chambers fully competent for ovulation . Components in the ecdysteroid signaling pathway , including the enzyme Shd for steroid production and Ecdysone receptor ( EcR ) , also changes its expression pattern from stage-13 to stage-14 ( Knapp and Sun , 2017 ) . It is unknown whether Hnt is also responsible for such changes; however , it is unlikely that ecdysteroid signaling upregulates Hnt in stage-14 follicle cells , because ecdysteroid signaling does not affect Mmp2 and Oamb expression ( Knapp and Sun , 2017 ) . Hnt is upregulated in follicle cells from stage 7 to stage 10A , which depends on Notch signaling ( Sun and Deng , 2007 ) ; however , Notch signaling is not active in stage-14 follicle cells and is unlikely to upregulate Hnt at this stage . Thus , the developmental signal for Hnt upregulation in stage14 and the transition from stage 13 to stage 14 is still unknown . Mmp2 , along with Mmp1 , are the only genes in the fly genome encoding matrix metalloproteinase and are crucial for extracellular matrix homeostasis during normal development , wound repair , and cancer metastasis ( Page-McCaw , 2008; Stevens and Page-McCaw , 2012 ) . Unlike Mmp1 , whose expression is tightly regulated by Jun-related kinase ( JNK ) signaling ( Uhlirova and Bohmann , 2006 ) , regulation of Mmp2 expression is largely unknown . Our work clearly defines the role of Hnt in regulating Mmp2 expression and basement membrane remodeling during ovulation . Hnt directly binds to two adjacent Hnt-binding sequences in the regulatory region of hnt and nervy genes and attenuates their expression ( Ming et al . , 2013 ) . Such Hnt-binding motifs are not found in the gene region of Mmp2 and Oamb . Thus , Hnt may indirectly regulate Mmp2 expression in posterior follicle cells . In addition , other transcriptional regulators must exist to coordinate with Hnt to restrict Mmp2 expression to posterior follicle cells . Hnt’s role in regulating Mmp2 expression and extracellular matrix homeostasis may not be restricted to posterior follicle cells . It has been shown that Hnt has a general role in regulating epithelial integrity in multiple organ systems and developmental stages . During retinal morphogenesis , hnt mutant photoreceptor cells frequently delaminate from retinal epithelium and are unable to maintain their integrity ( Pickup et al . , 2002 ) . In the tracheal system , hnt mutant tracheal epithelium disintegrate to form sacs and vesicles from collapsed dorsal trunk and branches ( Wilk et al . , 2000 ) . During oogenesis , Hnt is essential for proper cell adhesion and collective cell migration in stage-9 egg chambers . Ectopic expression of Hnt in the cluster of border cells leads to dissociation of the border-cell cluster ( Melani et al . , 2008 ) . In addition , genetic modifier screens identify basement-membrane components Vkg and Laminin as Hnt’s genetic interactors ( Wilk et al . , 2004 ) . All of these studies suggest that Hnt plays general roles in regulating epithelial integrity and extracellular matrix homeostasis in multiple organ systems . It will be interesting to see whether the regulation of Mmp2 by Hnt also occurs in other Hnt-expressing or Mmp2-expressing tissues/organs . Drosophila Hnt and mammalian RREB-1 are functionally conserved in many aspects . Both Drosophila Hnt and mammalian RREB-1 are required for proper cell migration ( Melani et al . , 2008 ) . Human RREB-1 binds to similar DNA sequences in Drosophila salivary gland polytene chromosomes as Hnt and rescues the germ band retraction phenotype in hnt mutant embryos ( Ming et al . , 2013 ) . In addition , our current work shows that human RREB-1 is able to rescue Oamb and Mmp2 expression in stage-14 follicle cells and OA-induced follicle rupture/ovulation phenotype in hntRNAi females ( Figure 8 ) . The role of RREB-1 in mammalian ovulation has not been studied so far , however , RREB-1 is detected in granulosa cells in mouse ovaries by microarray analysis ( Fan et al . , 2009 ) . In addition , mammalian RREB-1 functions downstream of the Ras-MAPK signaling pathway in multiple occasions ( Kent et al . , 2010; 2013; Mukhopadhyay et al . , 2007; Thiagalingam et al . , 1996; Zhang et al . , 2003 ) , and the Ras-MAPK signaling pathway is involved in mammalian ovulation ( Fan et al . , 2009 ) . It is possible that RREB-1 may function in granulosa cells to regulate Mmp expression and ovulation downstream of Ras-MAPK pathway in mammals . Flies were reared on standard cornmeal and molasses food at 25°C , and all RNAi-mediated depletion experiments were performed at 29°C with UAS-dcr2 . Two stage-14 follicle-cell specific Gal4 drivers from the Janelia Gal4 collection ( Pfeiffer et al . , 2008 ) were used in this study: R44E10-Gal4 ( FC1 ) and R47A04-Gal4 ( FC2 ) . The following RNAi lines were used: UAS-hntRNAi1 ( V3788 ) and UAS-hntRNAi2 ( V101325 ) from the Vienna Drosophila Resource Center; and UAS-Mmp2RNAi ( Uhlirova and Bohmann , 2006 ) . UAS-Oamb . K3 ( Lee et al . , 2009 ) , UAS-Timp ( Page-McCaw et al . , 2003 ) , hntEP55 ( a P-element insertion line containing UAS sequence in the promoter region of hnt; Bloomington Drosophila Stock Center , BDSC# 5358 ) , UAS-RREB1::GFP ( Ming et al . , 2013 ) were used to overexpress Oamb , Timp , Hnt , and RREB1 , respectively . Oamb . K3 is the Oamb isoform expressed in wild-type stage-14 follicle cells ( Deady and Sun , 2015 ) . hntXE81 and hntEH704a are loss-of-function hnt alleles , while hntpeb ( BDSC# 80 ) is a temperature-sensitive hnt allele ( Wilk et al . , 2004 ) . Animals bearing hntpeb were raised at room temperature , and newly emerged adult flies were shifted to the 29°C restrictive temperature . For generating flip-out actin-Gal4 clones ( Pignoni and Zipursky , 1997 ) , hsFLP;;act <CD2<Gal4 , UAS-RFP/TM3 , Sb ( derived from BDSC# 30558 ) was used to cross to hntEP55 or hntEP55; UAS-hntRNAi and adult flies were heat shocked in a 37°C water bath for 45 min . UASp-GFP::act79B; UAS-mCD8::GFP was crossed to Gal4 lines and used to visualize Gal4 expression pattern . UAS-RFP was recombined to Gal4 drivers and used for isolating stage-14 egg chambers for ex vivo culture . UAS-GCaMP5G ( Akerboom et al . , 2012 ) was used to visualize calcium responses in follicle cells ( BDSC# 42037 ) . Protein trap lines vkg::GFPCC00791 ( Buszczak et al . , 2007 ) and Mmp2::GFP ( Deady et al . , 2015 ) were used for Vkg and Mmp2 expression , respectively . Control flies for all experiments were prepared from crossing Gal4 drivers to Oregon-R . Egg laying and egg-laying time analyses were performed as previously described ( Deady and Sun , 2015; Knapp and Sun , 2017 ) . Five virgin females ( five-to-six days old with one day of wet yeast feeding ) were placed with ten Oregon-R males in one bottle to lay eggs on grape juice-agar plates with a drop of wet yeast paste for two days in 29°C . After each day ( 22 hr in 29°C ) of egg laying , grape juice-agar plates were removed and replaced with a new one . Typically , five bottles for each genotype are performed in each experiment . After egg laying , ovaries were dissected and mature follicles in female ovaries were counted . Virgin females were dissected before mating for a ‘pre-egg laying’ mature follicle count to ensure normal oogenesis occurred . The average number of eggs laid per female per day was used to calculate the average time to lay one egg , as described previously . The egg-laying time was further proportioned into the amount of time an egg spent in the ovary ( ovulation time ) , in the oviduct ( oviduct time ) , and in the uterus ( uterus time ) according to the distribution of females with eggs in their reproductive tract six hours after mating . For this assay , ten virgin females and fifteen males are mated in a vial with dry yeast at 29°C . Typically , two to three vials for each genotype were performed in each experiment . After a six-hour mating , the flies were frozen at −80°C for approximately four minutes , and then dissected to examine the location of an egg within the reproductive tract . The ex vivo follicle rupture assay was performed as described previously ( Deady and Sun , 2015 ) . In brief , 5–6 day-old virgin females fed with wet yeast for 2–3 days were used to isolate stage-14 egg chambers in Grace’s insect medium ( Caisson Lab , Smithfield , UT ) . Within one hour , isolated mature follicles from ~10 females were separated into groups of ~30 egg chambers , then cultured in culture media ( Grace’s medium , 10% fetal bovine serum , and 1X penicillin/streptomycin ) supplemented with 20 μM OA ( Sigma , St . Louis , MO ) , or 5 μM ionomycin ( Cayman Chemical Co . , Ann Arbor , MI ) . All cultures were performed at 29°C , the same condition as flies were maintained , to enhance Gal4/UAS efficiency . One data point represents the percent of ruptured follicles per experimental group ( ~30 egg chambers ) . Data were represented as mean percentage ± standard deviation ( SD ) . In situ zymography for detecting gelatinase activity was performed as previously reported with minor modifications ( Deady and Sun , 2015 ) . 20–25 μg/mL of DQ-gelatin conjugated with fluorescein ( Invitrogen , Eugene , OR ) was added into the culture media with 20 μM OA for three hours . Mature follicles with posterior fluorescein signal were directly counted , and data represented as percent of follicles with posterior fluorescein signal . Follicles with lateral fluorescein signal , which is likely induced by damage during dissection , are not counted as Mmp2 activity , because Mmp2 is only expressed in posterior follicle cells ( Deady et al . , 2015 ) . For quantitative RT-PCR , total RNA was extracted from 60 stage-14 egg chambers isolated from 10 flies using Direct-zol RNA MicroPrep Kit ( Zymo Research , Orange , CA ) . cDNA synthesis , real-time PCR amplification and primers of Oamb . K3 and Mmp2 were described previously ( Knapp and Sun , 2017 ) . The data are presented as mean ± SEM from three biological replicates , except for RREB-1 rescue experiment , in which one single biological experiment was presented . Immunostaining was performed following a standard procedure , including ovary dissection , fixation in 4% EM-grade paraformaldehyde for 15 min , blocking in PBTG ( PBS + 0 . 2% Triton + 0 . 5% BSA + 2% normal goat serum ) , and primary and secondary antibody staining diluted in PBTG . For vkg::GFP analysis , stage-14 egg chambers were first isolated from ovaries in cold Grace’s medium before fixation . Mouse anti-Hnt ( 1:75; Developmental Study Hybridoma Bank ) , mouse anti-GFP ( 1:2000; Invitrogen ) , rabbit anti-GFP ( 1:4000; Invitrogen ) , and rabbit anti-RFP ( 1:1000; MBL international ) were used as primary antibodies , and Alexa 488 , 546 , and 633 goat anti-mouse and goat anti-rabbit ( 1:1000 , Invitrogen ) were used as secondary antibodies . Images were acquired using a Leica TCS SP8 confocal microscope or Leica MZ10F fluorescent stereoscope with a sCOMS camera ( PCO . Edge ) , and assembled using Photoshop software ( Adobe Inc . , Mountain View , CA ) and ImageJ . To visualize calcium response to ionomycin and octopamine , egg chambers expressing GCaMP5G and hntRNAi were isolated into an imaging chamber . Images were acquired on a Zeiss Axio Zoom microscope at 0 . 2 FPS , and 10 μL of ionomycin or octopamine were added to the solution after frame five to a final concentration of 5 μM or 20 μM , respectively . A ROI in the center of the main-body follicle cells was selected and the integrated intensity was measured . F0 was defined as the average baseline intensity ( first five frames ) , and ΔF/F0 is reported . Statistical tests were performed using Prism 7 ( GraphPad , San Diego , CA ) . For comparison of more than two means , one-way ANOVA with post hoc Fisher’s Least Significant Difference test was used . For comparison of distribution , Chi square test was used except in Figure 3D and H , where Fisher’s exact test was used . In addition , Z-score test was used for egg-laying time analysis in Figure 3K–L and Supplementary file 1 .	The release of an egg from the ovary of a female animal is a process known as ovulation . Animals as different as humans and fruit flies ovulate in largely similar ways . Yet the systems involved in controlling ovulation are still not well understood . An egg cell develops within a collection of cells that help the egg to form properly . Together , this unit is called a follicle . During ovulation , connections between the egg and the rest of the follicle break down and the egg is eventually ejected . Ovulation happens in response to a hormone signal from the brain . In humans , this hormone is called luteinizing hormone , whereas in flies it is called octopamine . Specialized protein molecules on the surface of the follicle cells receive these hormone signals , but can only cause ovulation in mature follicles . It was not clear what allows only mature follicles to ovulate . Deady et al . have now used the fruit fly Drosophila melanogaster to examine ovulation to identify how the process is controlled . The results showed that a protein called Hindsight primes follicle cells for ovulation . When a follicle reaches its final stage ( called stage 14 in flies ) , the gene for Hindsight becomes active and produces the protein . This protein then activates other genes . One of the activated genes makes a protein that receives the hormone signal , while another makes a protein that breaks down follicle cells and allows the egg to be released . The findings of Deady et al . reveal that Hindsight is needed for ovulation in flies . Further experiments then showed that the gene for equivalent human protein can be transplanted into flies and can still prime follicles for ovulation . This indicates that the genes in humans and flies may perform the same tasks . Studying ovulation is an important part of understanding female fertility and could help scientists to understand more about human reproduction . These results may also lead to new contraceptives and improved approaches for treating infertility .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2017	The zinc-finger transcription factor Hindsight regulates ovulation competency of Drosophila follicles
Damaged mitochondria can be selectively eliminated by mitophagy . Although two gene products mutated in Parkinson’s disease , PINK1 , and Parkin have been found to play a central role in triggering mitophagy in mammals , how the pre-autophagosomal isolation membrane selectively and accurately engulfs damaged mitochondria remains unclear . In this study , we demonstrate that TBC1D15 , a mitochondrial Rab GTPase-activating protein ( Rab-GAP ) , governs autophagosome biogenesis and morphology downstream of Parkin activation . To constrain autophagosome morphogenesis to that of the cargo , TBC1D15 inhibits Rab7 activity and associates with both the mitochondria through binding Fis1 and the isolation membrane through the interactions with LC3/GABARAP family members . Another TBC family member TBC1D17 , also participates in mitophagy and forms homodimers and heterodimers with TBC1D15 . These results demonstrate that TBC1D15 and TBC1D17 mediate proper autophagic encapsulation of mitochondria by regulating Rab7 activity at the interface between mitochondria and isolation membranes . Autophagosomes enclose seemingly random portions of the cytoplasm to supply nutrients during starvation or they can specifically engulf cellular debris to maintain quality control ( Mizushima et al . , 2011 ) . Although these two pathways share a number of core biochemical steps , the morphology of the isolation membranes can differ greatly . Cup-shaped isolation membranes surround cytosol and organelles upon starvation ( Baba et al . , 1994 ) . However , when large aggregates of debris accumulate or during xenophagy ( Kageyama et al . , 2011 ) , autophagosomes form locally on the surface of the substrate and grow in contour with the often irregularly shaped structures that can be much larger than the membrane cups formed during starvation . Furthermore , for large aggregates of mitochondria , several autophagosomes may form simultaneously on different sides of the cargo and fuse together to engulf them ( Yoshii et al . , 2011 ) . Ubiquitinated protein aggregates may be tagged by LIR ( LC3-interacting region ) domain-containing proteins that bind LC3 family proteins on isolation membranes to help recruit autophagosomes to the debris ( Birgisdottir et al . , 2013 ) . However , what links isolation membrane expansion to cargo size is unknown . We examined autophagosome morphogenesis during Parkin-mediated mitophagy ( Youle and Narendra , 2011 ) where multiple mitochondria can be engulfed in large aggregates ( Narendra et al . , 2010a; Yoshii et al . , 2011 ) . A broad range of mitochondrial outer membrane proteins become ubiquitinated by Parkin ( Chan et al . , 2011; Sarraf et al . , 2013 ) , which appears to trigger the recruitment of autophagy-related ( Atg ) proteins to mitochondria and activate autophagosome assembly ( Itakura et al . , 2012 ) . Although mitophagy is accomplished in part by canonical autophagy gene products , how autophagosomes selectively recognize and assemble around damaged mitochondria remains unclear . So far , no mitochondrial proteins have been identified to mediate mitophagy downstream of Parkin . TBC1D15 has been shown to bind mitochondria through Fis1 ( Onoue et al . , 2013 ) , to bind the Atg8 family member GABARAP ( Behrends et al . , 2010 ) , and has been found in a screen of Parkin substrates ( Sarraf et al . , 2013 ) . TBC1D15 is one of the TBC ( Tre-2/Bub2/Cdc16 ) family proteins that function as Rab GTPase-activating proteins ( Rab-GAPs ) ( Fukuda , 2011; Frasa et al . , 2012 ) . The membrane trafficking activity of Rab proteins is increased by guanine nucleotide exchange factors ( GEFs ) , which accelerate the exchange of GDP for GTP and is decreased by GAPs , which facilitate the Rab GTPase activity . Fis1 , conserved from yeast to mammals , is C-terminally anchored in the mitochondrial outer membrane ( Mozdy et al . , 2000 ) with an N-terminal tetratricopeptide-repeat ( TPR ) domain exposed to the cytosol ( Suzuki et al . , 2003 ) . Although mammalian Fis1 had been thought to be a mitochondrial fission factor in the same way as yeast Fis1 , Fis1 knock out mammalian cells do not display a defect in mitochondrial fission ( Otera et al . , 2010 ) . Also in contrast to yeast , mammalian Drp1 recruitment to mitochondria requires Mff , MIEF1/MiD51 , and/or MiD49 ( Otera et al . , 2010; Palmer et al . , 2011; Zhao et al . , 2011 ) . Interestingly , the Fis1 null worm and Fis1 knock out mammalian cells have excessive LC3 accumulation following stress induced by mitochondrial inhibitors such as antimycin A in a PINK1-dependent manner ( Shen et al . , 2014 ) suggesting Fis1 may have a role in mitophagy . In this study , we show that TBC1D15 and Fis1 act in concert to control autophagosome morphology during Parkin-mediated mitophagy but not during starvation-induced autophagy . This step is dependent on the small GTPase Rab7 . We also demonstrate that TBC1D15 must bind LC3 homologue proteins as well as Fis1 to coordinate Rab7 activity to shape the nascent autophagosome isolation membrane providing mechanistic insight into autophagosome morphogenesis during mitophagy . We recently found that Fis1 null C . elegans and mammalian cells display aberrant LC3 accumulation ( Shen et al . , 2014 ) . However , the molecular mechanism underlying the association between LC3 accumulation and the loss of Fis1 remains unclear . As TBC1D15 binds to Fis1 ( Onoue et al . , 2013 ) and is a Rab-GAP potentially involved in autophagy ( Behrends et al . , 2010 ) , we made a TBC1D15 gene knock out ( KO ) cell line using TALENs ( Transcription activator-like effector nucleases [Gaj et al . , 2013] ) and compared the LC3 accumulation phenotype with that of FIS1−/− cells . We designed TALEN binding pairs that target exon 9 of the TBC1D15 gene because it is shared by all conceivable TBC1D15 splicing isoforms . One TBC1D15−/− clone harbors a 14-bp deletion in one allele and a large deletion in the other allele , both of which are frame-shifting and would cause mRNA decay ( Figure 1—figure supplement 1 ) . We confirmed the knock out of TBC1D15 expression by immunoblotting ( Figure 1A ) . As TBC1D15 was previously reported to mediate mitochondrial fission and bind to Fis1 ( Onoue et al . , 2013 ) , we checked the expression level of several proteins that have been previously linked to mitochondrial fission pathways in the TBC1D15−/− cells as well as in FIS1−/− , MFF−/− , DRP1−/− and the corresponding WT HCT116 cells ( Figure 1A ) . Although each KO cell line has complete deletion of the target protein , expression and/or stability of the other fission-related proteins was not affected; FIS1−/− and TBC1D15−/− cells possess normal levels of TBC1D15 and Fis1 , respectively . 10 . 7554/eLife . 01612 . 003Figure 1 . TBC1D15 is dispensable for mitochondria and peroxisome morphologies . ( A ) Total cell lysates prepared from the indicated HCT116 cell lines were analyzed by immunoblotting . For comparison , different amounts of proteins ( 1:3 ratio ) were applied . An asterisk indicated non-specific crossreactive bands . ( B ) The indicated cell lines were analyzed by immunofluorescence microscopy using anti-Cytochrome c antibody for mitochondria and anti-PMP70 antibody for peroxisome staining . Images are displayed as z-stacks of 6 confocal slices . Magnified images are also shown for mitochondrial morphologies . Scale bars , 10 μm . ( C ) Quantification of mitochondrial morphologies in ( B ) . Percentages of cells harboring fragmented , tubular , or elongated mitochondria are shown . Tubular and elongated denote normal tubular mitochondria seen in WT cells and highly connected mitochondrial network , respectively . The error bars represent ±SD from three independent replicates . Over 50 cells were counted in each of three replicate wells . DOI: http://dx . doi . org/10 . 7554/eLife . 01612 . 00310 . 7554/eLife . 01612 . 004Figure 1—figure supplement 1 . TBC1D15 gene knock out by TALENs . Highlighted TALEN-TBC1D15-L and TALEN-TBC1D15-R indicate the TALEN binding pair that targets exon 9 of the TBC1D15 gene . DNA sequencing confirmed the nucleotide deletions in both alleles of TBC1D15−/− #74 clone . DOI: http://dx . doi . org/10 . 7554/eLife . 01612 . 004 It has been reported that several mitochondrial fission components such as Mff , Drp1 , and Fis1 are localized not only on mitochondrial membranes but also on peroxisomal membranes ( Li and Gould , 2003; Kobayashi et al . , 2007; Koch and Brocard , 2012 ) . We compared the mitochondrial and peroxisomal morphology of TBC1D15−/− cells with that of FIS1−/− , MFF−/− , and DRP−/− cells . In agreement with previous work ( Otera et al . , 2010 ) , mitochondrial morphology , as well as peroxisomal morphology in FIS1−/− cells , was similar to that of WT cells ( Figure 1B ) . In contrast to moderately elongated mitochondria in TBC1D15 siRNA-treated cells reported previously ( Onoue et al . , 2013 ) , complete depletion of TBC1D15 by knock out resulted in no obvious mitochondrial morphology changes ( Figure 1B , C ) . Furthermore , peroxisome shape in TBC1D15−/− cells was also indistinguishable from that of WT cells ( Figure 1B ) . In sharp contrast , MFF−/− and DRP−/− cells display elongated mitochondria and peroxisomes ( Figure 1B ) , indicating that Mff and Drp1 play important roles in regulating both mitochondria and peroxisome morphology , consistent with the previous findings ( Smirnova et al . , 1998; Koch et al . , 2003; Gandre-Babbe and van der Bliek , 2008 ) . Quantification of mitochondrial morphology demonstrated that the fission defect in MFF−/− cells is similar to , but not as strong as , that caused by DRP1 deletion ( Figure 1C ) . Therefore , it appears that both TBC1D15 and Fis1 are dispensable for mitochondrial and peroxisomal fission in human cells . FIS1−/− cells accumulate LC3 during Parkin-mediated mitophagy ( Shen et al . , 2014 ) . To assess autophagosomes in the absence of TBC1D15 , we made stable cell lines expressing YFP-LC3 and mCherry-Parkin and treated the cells with valinomycin , a potassium ionophore that dissipates mitochondrial inner membrane potential , to trigger Parkin-mediated mitophagy . 3 hr of valinomycin treatment to depolarize mitochondria-stimulated Parkin translocation onto mitochondria with similar efficiencies in all cell lines tested ( Figure 2A , B ) . WT cells have several small dots , crescent-shaped , or spherical YFP-LC3 puncta representing isolation membranes and autophagosomes , respectively , that were observed on or near fragmented mitochondria , consistent with previous observations ( Figure 2A and Narendra et al . , 2008 ) . However , upon mitophagy induction in FIS1−/− cells , YFP-LC3 accumulates excessively in foci , which often appears interconnected with one another , localized on or near mitochondria ( Figure 2A , C and Figure 2—figure supplement 2; Shen et al . , 2014 ) . Interestingly , TBC1D15−/− cells display expanded LC3-labeled structures very similar to those in FIS1−/− cells suggesting that Fis1 and TBC1D15 are involved at the same step in autophagy ( Figure 2A , C and Figure 2—figure supplement 2 ) . This phenotype requires both Parkin ( Figure 2—figure supplement 3A ) and PINK1 ( Figure 2—figure supplement 3B ) expression and the loss of the inner mitochondrial membrane potential ( Figure 2—figure supplement 3C ) , indicating that it occurs upon mitophagy induction downstream of PINK1-induced Parkin translocation to mitochondria . Expression of N-terminally 3×FLAG-tagged Fis1 rescued the LC3 accumulation in FIS1−/− cells ( Figure 2—figure supplement 4 ) , confirming the effect was caused by the loss of Fis1 . Similarly , N-terminally HA-tagged WT TBC1D15 expressed in TBC1D15−/− cells reverted the excessive LC3 accumulation to that of WT after mitophagy induction ( Figure 2D , E ) . 10 . 7554/eLife . 01612 . 005Figure 2 . Loss of Fis1 or TBC1D15 causes LC3 accumulation during mitophagy . ( A ) The indicated cell lines stably expressing YFP-LC3 and mCherry-Parkin were treated with valinomycin for 3 hr and subjected to confocal immunofluorescence microscopy with anti-TOMM20 antibody . Scale bars , 10 μm . ( B ) Quantification of mCherry-Parkin translocation to mitochondria after 3 hr of valinomycin treatment . Partial or complete translocation to mitochondria in each cell was scored as separate phenotypes . Partial and complete denote that Parkin translocates to some of or all mitochondria , respectively . The error bars represent ±SD from three independent experiments . Over 100 cells were counted in each of three separate wells . ( C ) YFP-LC3 morphologies in ( A ) were quantified . Percentages of cells harboring diffuse , punctate or accumulated YFP-LC3 are shown . The error bars represent ±SD from three independent replicates . Over 100 cells were counted in each replicate . For the criteria of LC3 morphology , see Figure 2—figure supplement 1 . ( D ) YFP-LC3 and mCherry-Parkin stably expressing TBC1D15−/− cells in the absence or presence of HA-tagged TBC1D15 WT or Δ221-250 mutant were treated with valinomycin for 3 hr . Cells were subjected to immunofluorescence microscopy with anti-HA antibody . Scale bars , 10 μm . ( E ) The YFP-LC3 morphology of cells in ( D ) was quantified . The error bars represent ±SD from three independent replicates . Over 50 cells were counted in each well . ( F and G ) YFP-LC3 and mCherry-Parkin stably expressing WT ( F ) and TBC1D15−/− ( G ) cells were treated with valinomycin for 3 hr and then subjected to immunoelectron microscopy with anti-GFP antibody . The square in panel a shows enlarged areas in panel b Scale bars , 500 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 01612 . 00510 . 7554/eLife . 01612 . 006Figure 2—figure supplement 1 . Image examples of different LC3 morphologies . HCT116 cells stably expressing YFP-LC3 ( green ) and mCherry-Parkin were treated with valinomycin for 3 hr and then subjected to immunostaining with anti-TOMM20 ( red ) antibody . Diffuse , YFP-LC3 distributes throughout the cytosol with few dots that are not on mitochondria; Punctate , many discrete YFP-LC3 dots , crescent-shaped , or circular signals on or near mitochondria; Accumulated , highly interconnected YFP-LC3 accumulation or large YFP-LC3 clumps on or near mitochondria . DOI: http://dx . doi . org/10 . 7554/eLife . 01612 . 00610 . 7554/eLife . 01612 . 007Figure 2—figure supplement 2 . YFP-LC3 accumulation in FIS1−/− and TBC1D15−/− cells during mitophagy . Low magnification images of cells stably expressing YFP-LC3 and mCherry-Parkin treated with valinomycin for 3 hr . The cells were subjected to immunofluorescence microscopy with anti-TOMM20 antibody . Scale bars , 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01612 . 00710 . 7554/eLife . 01612 . 008Figure 2—figure supplement 3 . Excessive LC3 accumulation depends on PINK1/Parkin-mediated mitophagy . ( A ) The indicated HCT116 cells stably expressing YFP-LC3 ( without mCherry-Parkin ) were treated with valinomycin for 3 hr followed by immunofluorescence microscopy with anti-TOMM20 antibody . Scale bars , 10 μm . Of note , HCT116 cells do not express enough endogenous Parkin to induce mitophagy . ( B ) WT , FIS1−/− and TBC1D15−/− cells stably expressing YFP-LC3 and mCherry-Parkin were treated with control ( NTC ) or PINK1 siRNA . The percentages of cells harboring diffuse , punctate or accumulated YFP-LC3 after 3 hr of valinomycin treatment were quantified . The error bars represent ±SD from three independent replicates . Over 100 cells were counted in each of three replicates . ( C ) The indicated cells stably expressing YFP-LC3 and mCherry-Parkin without valinomycin treatment were subjected to immunofluorescence microscopy with anti-TOMM20 antibody . Scale bars , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01612 . 00810 . 7554/eLife . 01612 . 009Figure 2—figure supplement 4 . LC3 accumulation in FIS1−/− cells was rescued by Fis1 re-expression . ( A ) YFP-LC3 and mCherry-Parkin stably expressing FIS1−/− cells with or without 3xFLAG-Fis1 were treated with valinomycin for 3 hr . Cells were subjected to immunofluorescence microscopy with anti-FLAG antibody . ( B ) YFP-LC3 morphologies in cells in ( A ) were quantified . Percentages of cells harboring diffuse , punctate or accumulated YFP-LC3 are shown . The error bars represent ±SD from three independent replicates . Over 50 cells were counted in each replicate . DOI: http://dx . doi . org/10 . 7554/eLife . 01612 . 00910 . 7554/eLife . 01612 . 010Figure 2—figure supplement 5 . ULK1 and Atg14 recruitment in WT , FIS1−/− and TBC1D15−/− cells during mitophagy . The indicated cell lines stably expressing GFP-mouseULK1 ( A ) or GFP-Atg14 ( B ) and mCherry-Parkin were treated with or without valinomycin for 1 hr and subjected to immunofluorescence microscopy with anti-Cytochrome c antibody . Images are displayed as z-stacks of 6 confocal slices . Magnified images are also shown . Scale bars , 10 μm . The right graph shows the number of GFP-ULK1 or GFP-Atg14 dots and/or cup-shaped signals near mitochondria . The error bars represent ±SD from three replicates . Over 50 cells were counted in each well . DOI: http://dx . doi . org/10 . 7554/eLife . 01612 . 01010 . 7554/eLife . 01612 . 011Figure 2—figure supplement 6 . Atg16L1 and DFCP1 recruitment in WT , FIS1−/− and TBC1D15−/− cells during mitophagy . ( A ) The indicated cell lines stably expressing mCherry-Parkin were treated with or without valinomycin for 3 hr and subjected to immunofluorescence microscopy with anti-Atg16L1 and anti-Cytochrome c antibodies . ( B ) The indicated cell lines stably expressing GFP-mouseDFCP1 and mCherry-Parkin were treated with or without valinomycin for 3 hr and subjected to immunofluorescence microscopy with anti-Cytochrome c antibody . Images are displayed as z-stacks of 6 confocal slices . Magnified images are also shown . Scale bars , 10 μm . The right graph shows the number of Atg16L1 or GFP-DFCP1 dots near mitochondria . The error bars represent ±SD from three independent replicates . Over 50 cells were counted in each replicate . DOI: http://dx . doi . org/10 . 7554/eLife . 01612 . 011 To clarify the nature of LC3 accumulation in TBC1D15−/− cells in more detail , we conducted immunoelectron microscopy . Valinomycin treatment for 3 hr caused the formation of preautophagomes with closely apposed membranes that were labeled with gold particles attached to YFP-LC3 near mitochondria in WT cells ( Figure 2F ) . Although TBC1D15−/− cells also have similar cup-shaped membrane structures , many of them displayed much thicker lumens than those in WT cells ( Figure 2G , panel a and b ) . In addition , larger membrane capsules that contain mitochondria and possibly other membrane organelles were observed only in TBC1D15 −/− cells ( Figure 2G , panel c and d ) . Furthermore , some LC3 was associated with small diameter membrane structures that localize very close to mitochondria ( Figure 2G , panel e ) and may have out of plane contiguity to preautophagosomes ( Figure 2G , panel e and f ) . These results strongly suggest that the LC3-labeled structures seen in TBC1D15−/− cells are not simply LC3 protein aggregates , but express LC3 associated along membrane bilayers . We compared the localization of other preautophagosomal marker proteins and proteins involved in autophagy induction among WT , FIS1−/− , and TBC1D15−/− cells during mitophagy . GFP-ULK1 that functions in the most upstream ATG1/ULK kinase complex and GFP-Atg14 that is an autophagy specific subunit of the class III PI3K complex formed dots and/or cup-shaped structures around mitochondria following valinomycin treatment , although the number of dots did not further increase upon the loss of Fis1 or TBC1D15 ( Figure 2—figure supplement 5 ) . Moreover , we analyzed the recruitment of endogenous Atg16L1 that is only present on the preautophagosome membrane and GFP-DFCP1 that can serve as an omegasome marker . Although valinomycin induced the formation of both Atg16L1 and DFCP1 dots on mitochondria , neither of them increased further in FIS1−/− or TBC1D15−/− cells ( Figure 2—figure supplement 6 ) . Thus , Fis1 and TBC1D15 seem to regulate LC3 accumulation downstream of autophagy initiation . To observe autophagosomal structures in thicker optical slices , we reconstructed z-stacked confocal microscopic images comprising six sequential optical x-y sections taken at 0 . 8 μm z-intervals of cells stably expressing YFP-LC3 and mCherry-Parkin after 3 hr treatment with valinomycin . WT cells treated with valinomycin display perinuclear-clustered fragmented mitochondria and many YFP-LC3 puncta associated with them ( Figure 3A ) . In sharp contrast , upon Parkin-induced mitophagy , TBC1D15−/− cells display more expanded LC3 accumulation than WT cells and also generate thin LC3-labeled tubular structures interconnected to one another , which are very similar to that seen in FIS1 −/− cells ( Figure 3A ) . These tubular structures were frequently observed near the cell cortex . Live cell imaging showed that the LC3-containing tubules emanating from the LC3 foci are highly mobile ( Video 1 ) . Immunoelectron microscopy confirmed that long YFP-LC3-labeled tubules observed in TBC1D15−/− cells are membrane associated ( Figure 3B ) 10 . 7554/eLife . 01612 . 012Figure 3 . Tubular LC3 expands along microtubules . ( A ) The indicated cells stably expressing YFP-LC3 and mCherry-Parkin were treated with valinomycin for 3 hr followed by immunofluorescence microscopy with anti-Cytochrome c antibody . Confocal images were acquired as z-stacks comprising 6 sequential sections with 0 . 8 μm z-intervals . Scale bars , 10 μm . ( B ) TBC1D15−/− cells prepared as in ( A ) were subjected to immunoelectron microscopy with anti-GFP antibody . White arrowheads indicate YFP-LC3-labeled tubules . High magnification image is shown in the lower panel . Scale bars , 500 nm . ( C ) The indicated cells prepared as in ( A ) were subjected to immunostaining with anti-Tubulin antibody . YFP-LC3 and Tubulin staining are merged in the right panels . Magnified images are shown for TBC1D15−/− cells . Scale bars , 10 μm . ( D ) The indicated cells stably expressing YFP-LC3 and mCherry-Parkin were treated with valinomycin in the presence or absence of nocodazole for 3 hr . YFP-LC3 and mCherry-Parkin images are merged in the right panels . Scale bars , 10 μm . ( E ) YFP-LC3 morphologies of cells in ( D ) were quantified . The error bars represent ±SD from three independent replicates . Over 50 cells were counted in each well . DOI: http://dx . doi . org/10 . 7554/eLife . 01612 . 01210 . 7554/eLife . 01612 . 013Figure 3—figure supplement 1 . Tubular YFP-LC3 morphologies in TBC1D15−/− cells . WT and TBC1D15−/− cells stably expressing YFP-LC3 and mCherry-Parkin were treated with valinomycin for 3 hr . Cells were stained with ( A ) Alexa Fluor 635 Phalloidin for actin , ( B ) anti-KDEL antibody for the ER , ( C ) anti-LAMP2 antibody for lysosomes , and ( D ) anti-GM130 antibody for Golgi staining . Z-stack confocal images are shown . YFP-LC3 and organelle staining are merged in the right panels . Scale bars , 10 μm . Magnified images are also shown for Actin and ER . DOI: http://dx . doi . org/10 . 7554/eLife . 01612 . 01310 . 7554/eLife . 01612 . 014Video 1 . Tubular LC3 movement in TBC1D15−/− cells . TBC1D15−/− cells stably expressing YFP-LC3 ( green ) and mCherry-Parkin ( red ) were treated with valinomycin for 3 hr and then took z-stack pictures . DOI: http://dx . doi . org/10 . 7554/eLife . 01612 . 014 Although these LC3-labeled tubules were not associated with ER or actin filaments ( Figure 3—figure supplement 1A , B ) , a portion of them tightly colocalized with microtubules stained for Tubulin ( Figure 3C ) . To examine whether microtubules are important for generating LC3 tubular structures in TBC1D15−/− cells , we added nocodazole , which interferes with the polymerization of microtubules , together with valinomycin to cells for 3 hr . Although WT cells have similar LC3 puncta with or without nocodazole treatment , LC3-labeled tubular structures and LC3 accumulation in TBC1D15−/− cells completely disappeared upon nocodazole treatment ( Figure 3D , E ) . Thus , intact microtubules are essential for LC3-labeled membrane expansion in TBC1D15−/− cells , a step upstream of autophagosome fusion with lysosomes where previous work indicated dynein-facilitated autophagy ( Kimura et al . , 2008 ) . Although some late endosomes/lysosomes colocalize with LC3 ( Figure 3—figure supplement 1C ) consistent with the results of immunoelectron microscopy ( Figure 2G , panel f ) , the morphologies of lysosomes and the Golgi apparatus were similar between WT and TBC1D15−/− cells ( Figure 3—figure supplement 1C , D ) . As the excessive LC3 accumulation in the absence of TBC1D15 indicates either an augmentation in autophagy induction or a defect in autophagosomal flux , we asked whether the loss of TBC1D15 alters the rate of Parkin-mediated mitophagy . For this purpose , YFP-Parkin stably expressing cells were treated with valinomycin for various times and total cell lysates were analyzed by immunoblotting . More endogenous lipidated LC3 ( LC3-II ) accumulated in TBC1D15−/− cells as compared to that of WT cells as also seen in FIS1−/− cells ( Figure 4A ) . This is consistent with the greater appearance of YFP-LC3 accumulation by microscopic observation ( Figure 2A ) . Although Mfn1 , a mitochondrial outer membrane protein , was completely degraded within 2 hr in WT , FIS1−/− and TBC1D15−/− cells , ubiquitination of TOMM20 , another outer membrane protein was retarded in FIS1−/− and TBC1D15−/− cells ( Figure 4A; TOMM20 asterisk ) . In WT cells , PINK1 is barely detectable under normal growing conditions because of the degradation through N-end rule pathway ( Yamano and Youle , 2013 ) , whereas it accumulates on the outer membrane upon addition of valinomycin ( Figure 4A ) as seen also with CCCP ( Matsuda et al . , 2010; Vives-Bauza et al . , 2010; Narendra et al . , 2010b ) . The accumulated PINK1 starts to decrease after 8 hr of valinomycin likely as mitochondria are eliminated by Parkin-mediated mitophagy . On the other hand , accumulated PINK1 in FIS1−/− or TBC1D15−/− was retained longer than in WT cells ( Figure 4A ) , suggesting , in conjunction with the delayed TOMM20 degradation , that autophagic clearance and/or proteasomal degradation is delayed in FIS1−/− or TBC1D15−/− cells . The rate of degradation of TOMM20 , TIMM23 , and mitochondrial HSP60 in FIS1−/− or TBC1D15−/− cells , however , was not different as compared to that of WT cells at the 8-hr time points ( Figure 4A ) . On the other hand , when cells were treated with valinomycin for longer times up to 40 hr , most of the TOMM20 was degraded and protein levels of the matrix HSP60 and the mitochondrial DNA-encoded COXII ( MT-COXII ) decreased in WT cells , whereas they were retained in FIS1−/− and TBC1D15−/− cells ( Figure 4B , C ) . These results indicate that the Fis1-TBC1D15 pair is not essential for mitophagy but the deletion of either impedes clearance of damaged mitochondria , despite the greater magnitude of LC3 accumulation following Parkin translocation . We also found that , although Fis1 is degraded during mitophagy , TBC1D15 escapes the degradation ( Figure 4D ) . 10 . 7554/eLife . 01612 . 015Figure 4 . Loss of Fis1 or TBC1D15 impedes clearance of damaged mitochondria . ( A and B ) YFP-Parkin stably expressing cells were treated with valinomycin for indicated times . Total cell lysates were subjected to immunoblotting . I and II denote cytosolic and lipidated LC3B , respectively . An asterisk indicates ubiquitinated TOMM20 . ( C ) Indicated protein amounts as in ( B ) were quantified . The amount of protein without valinomycin treatment was set to 100% . The error bars represent ±SD from three independent experiments . ( D ) YFP-Parkin stably expressing WT HCT116 cells were treated with or without valinomycin for 40 hr . Total cell lysates were analyzed by immunoblotting . DOI: http://dx . doi . org/10 . 7554/eLife . 01612 . 015 Mitochondrial fission has been reported to be involved in mitophagy ( Twig et al . , 2008; Tanaka et al . , 2010; Mao et al . , 2013 ) . Therefore , we assessed if fission defects generally cause LC3 accumulation . The excessively long mitochondria found in MFF−/− or DRP−/− cells are converted from an elongated tubular network to larger , more rounded , or bubble-like structures relative to WT cells upon 3 hr valinomycin treatment ( Figure 5A ) . However , Parkin translocation and mitochondrial colocalization with YFP-LC3 in MFF−/− and DRP1−/− cells were similar to those in WT cells ( Figure 5A–C ) without the LC3 accumulation and/or tubulation seen in FIS1−/− and TBC1D15−/− cells ( Figure 2A and Figure 3A ) . These results suggest that LC3 accumulation is a unique feature following the loss of Fis1 or TBC1D15 and is not caused by defects in mitochondrial fission , consistent with the finding that Fis1 and TBC1D15 do not appear to regulate mitochondrial morphology ( Figure 1B ) . 10 . 7554/eLife . 01612 . 016Figure 5 . Effect of mitochondrial fission and starvation induction on LC3 accumulation in WT , FIS1−/− and TBC1D15−/− cells . ( A ) WT , MFF−/− or DRP1−/− cells stably expressing YFP-LC3 and mCherry-Parkin were treated with valinomycin for 3 hr and subjected to immunofluorescence microscopy with anti-TOMM20 antibody . Scale bars , 10 μm . ( B ) Quantification of mCherry-Parkin translocation to mitochondria after 3 hr of valinomycin treatment . Partial and complete denote that Parkin translocates to some of and all mitochondria , respectively . The error bars represent ±SD from three independent replicates . Over 50 cells were counted in each replicate . ( C ) YFP-LC3 morphologies of cells in ( A ) were quantified . Percentages of cells harboring diffuse , punctate or accumulated YFP-LC3 are shown . The error bars represent ±SD from three independent replicates . Over 100 cells were counted in each replicate . ( D ) WT , FIS1−/− , and TBC1D15−/− cells stably expressing YFP-LC3 were grown in starvation media . Z-stacks of confocal images are shown . Scale bars , 20 μm . ( E ) The number of YFP-LC3 dots in cells under growth or starvation conditions was quantified . The error bars represent ±SD from three independent replicates . Over 50 cells were counted in each well . ( F ) Total cell lysates from cells grown in normal or starvation media for 6 hr were subjected to immunoblotting . LC3-I and LC3-II denote cytosolic and lipidated LC3B , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 01612 . 016 We also investigated the effect of loss of Fis1 or TBC1D15 on starvation-induced autophagy by incubating cells in media lacking amino acids . Upon starvation , the number of YFP-LC3 puncta greatly increased , but indistinguishably between WT , FIS1−/− , and TBC1D15−/− cells ( Figure 5D , E ) . Additionally , immunoblotting of endogenous LC3 demonstrated that although more of the lipidated LC3 was generated upon starvation , the amount was not further increased by the depletion of Fis1 or TBC1D15 ( Figure 5F ) . Therefore , TBC1D15 and Fis1 , which form a complex on mitochondria , seem to govern autophagosome biogenesis specifically during mitophagy . To investigate whether the Fis1 binding of TBC1D15 is important for LC3 accumulation , we introduced into TBC1D15−/− cells HA-tagged TBC1D15 lacking amino acids 221-250 ( HA-TBC1D15 Δ221-250 ) , which was reported to be deficient in Fis1 binding ( Onoue et al . , 2013 ) . Unexpectedly , mitochondrial localization of TBC1D15 Δ221-250 was observed under normal growing conditions ( Figure 6A ) and HA-TBC1D15 Δ221-250 could rescue LC3 accumulation in TBC1D15−/− cells during mitophagy ( Figure 2D , E ) . However , TBC1D15 appears to bind Fis1 as reported ( Onoue et al . , 2013 ) because in a FIS1−/− background , TBC1D15 localizes to the cytosol ( Figure 6A ) . Furthermore , Fis1 overexpression can recruit overexpressed cytosolic WT TBC1D15 , but not TBC1D15 Δ221-250 , to the mitochondria ( Figure 6B ) . These results raise the possibility that there is another protein that can facilitate TBC1D15 Δ221-250 binding to Fis1 . 10 . 7554/eLife . 01612 . 017Figure 6 . Identification of TBC1D17 as a Fis1 and TBC1D15 binding protein . ( A ) TBC1D15−/− cells and those stably expressing HA-TBC1D15 WT and HA-TBC1D15 ( Δ221-250 ) , and FIS1−/− cells stably expressing HA-TBC1D15 WT were subjected to immunostaining with anti-TOMM20 and anti-HA antibodies . Scale bars , 10 μm . ( B ) HA-TBC1D15 WT or HA-TBC1D15 ( Δ221-250 ) with or without Fis1 was transiently overexpressed ( OE ) in HeLa cells . Cells were subjected to immunostaining with anti-HA and anti-Fis1 antibodies . Scale bars , 20 μm . ( C ) YFP-TBC1D17 together with pcDNA vector or Fis1 was transiently overexpressed ( OE ) in HeLa cells . Cells were subjected to immunostaining with anti-Fis1 and anti-Cytochrome c antibodies . Scale bars , 20 μm . ( D ) YFP or YFP-Fis1 was co-overexpressed with HA-TBC1D15 or HA-TBC1D17 in HEK293 cells . The cell extracts were subjected to pull down assays with GFP-Trap . 5% input and bound fractions were analyzed by immunoblotting with anti-HA ( upper panel ) and anti-GFP ( lower panel ) antibodies . ( E ) YFP , YFP-TBC1D15 , or YFP-TBC1D17 was co-overexpressed with HA-TBC1D15 ( upper panel ) or HA-TBC1D17 ( lower panel ) in HEK293 cells . The cell extracts were subjected to pull down assays with GFP-Trap . 5% input and bound fractions were analyzed by immunoblotting with anti-GFP and anti-HA antibodies . ( F ) HA-TBC1D15 ( Δ221-250 ) together with Fis1 and YFP-TBC1D15 WT or YFP-TBC1D17 WT were transiently overexpressed ( OE ) in HeLa cells . Cells were subjected to immunostaining with anti-HA and anti-Fis1 antibodies . Images of HA and Fis1 staining were merged in the right panels . Scale bars , 20 μm . ( G ) Schematic model of Fis1 , TBC1D15 , and TBC1D17 binding . Homo- or hetero-dimer of TBC1D15 can interact with Fis1 dimer on the mitochondrial outer membrane ( OMM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01612 . 01710 . 7554/eLife . 01612 . 018Figure 6—figure supplement 1 . TBC1D17 gene knock out by TALENs . Highlighted TALEN-TBC1D17-L and TALEN-TBC1D17-R indicate TALEN binding pair that targets the intron–exon 5 junction of TBC1D17 gene . DNA sequencing confirmed nucleotide deletions in both alleles of TBC1D17−/− #74 clone and of TBC1D15/17 DKO #58 clone . DOI: http://dx . doi . org/10 . 7554/eLife . 01612 . 01810 . 7554/eLife . 01612 . 019Figure 6—figure supplement 2 . TBC1D17 is dispensable for normal mitochondria and peroxisome morphologies . ( A ) The indicated cell lines were analyzed by immunofluorescence microscopy using anti-Cytochrome c antibody for mitochondria and anti-PMP70 antibody for peroxisome staining . Images are displayed as z-stacks of 6 confocal slices . Magnified images are also shown for mitochondrial morphologies . Scale bars , 10 μm . ( B ) Quantification of mitochondrial morphologies in ( A ) . Percentages of cells harboring fragmented , tubular or elongated mitochondria were shown . Tubular and elongated denote normal tubular mitochondria seen in WT cells and highly connected mitochondrial network , respectively . The error bars represent ±SD from three independent replicates . Over 50 cells were counted in each replicate . DOI: http://dx . doi . org/10 . 7554/eLife . 01612 . 01910 . 7554/eLife . 01612 . 020Figure 6—figure supplement 3 . YFP-LC3 accumulation in TBC1D17−/− and TBC1D15/17 DKO cells during mitophagy . ( A ) TBC1D17−/− and TBC1D15/17 DKO cells stably expressing YFP-LC3 and mCherry-Parkin were treated with valinomycin for 3 hr and subjected to immunofluorescence microscopy with anti-TOMM20 antibody . Scale bars , 10 μm . ( B ) YFP-LC3 morphologies in ( A ) were quantified . Percentages of cells harboring diffuse , punctate or accumulated YFP-LC3 are shown . The error bars represent ±SD from three independent replicates . Over 100 cells were counted in each well . ( C ) The indicated proteins were transiently overexpressed ( OE ) in TBC1D15/17 DKO cells stably expressing YFP-LC3 and mCherry-Parkin . The cells were treated with valinomycin for 3 hr followed by immunofluorescence microscopy with anti-HA antibody . Scale bars , 10 μm . ( D ) YFP-LC3 morphologies in ( C ) were quantified . The error bars represent ±SD from three independent replicates . Over 50 cells were counted in each well . DOI: http://dx . doi . org/10 . 7554/eLife . 01612 . 020 The human genome encodes at least 40 different TBC domain-containing proteins . According to phylogenetic and domain analysis of human and mouse TBC proteins ( Itoh et al . , 2006 ) , TBC1D17 has a high similarity to the primary protein sequence of TBC1D15 . Specifically , the region outside the TBC domain comprising 193-312 aa of TBC1D17 is 35% identical ( 55% positive ) to the 201-333 aa region of TBC1D15 that contains the Fis1-interacting region ( data not shown ) , suggesting that TBC1D17 may also interact with Fis1 . To test this idea , we expressed HA-tagged TBC1D17 ( HA-TBC1D17 ) with or without Fis1 overexpression in HeLa cells . Overexpressed HA-TBC1D17 alone was cytosolic ( Figure 6C ) . However , after co-overexpression with Fis1 , HA-TBC1D17 localized to mitochondria ( Figure 6C ) , indicating that like TBC1D15 , TBC1D17 has an ability to bind Fis1 . Similar results were obtained by co-immunoprecipitation with YFP-Fis1 ( Figure 6D ) . Although we were unable to detect endogenous TBC1D17 because no antibody is available , these results suggest that TBC1D17 may also be targeted to mitochondria via an interaction with Fis1 . By co-immunoprecipitation , we found that TBC1D15 and TBC1D17 can form homo- and hetero-dimers ( Figure 6E ) . Furthermore , the TBC1D15 Δ221-250 mutant that lacks a Fis1-binding region translocates to mitochondria when co-overexpressed with Fis1 and WT TBC1D15 or WT TBC1D17 ( Figure 6F ) , indicating that TBC1D17 can recruit the mutant TBC1D15 to Fis1 through dimerization . These results are also interesting in light of evidence that Fis1 also forms dimers ( Jofuku et al . , 2005; Lees et al . , 2012 ) suggesting a new model of Fis1/TBC topology ( Figure 6G ) . To examine the functional importance of TBC1D17 in mitophagy , we generated TBC1D17 single KO and TBC1D15/TBC1D17 double KO ( DKO ) HCT116 cells by the TALEN methodology ( Figure 6—figure supplement 1 ) . Mitochondrial and peroxisomal morphologies in these cells were indistinguishable from those in WT cells ( Figure 6—figure supplement 2A , B ) . TBC1D17−/− cells displayed LC3 accumulation during mitophagy similar to , but less severe than that of TBC1D15−/− cells ( Figure 6—figure supplement 3A , B ) . However , the degree of LC3 accumulation in TBC1D15/TBC1D17 DKO cells was not further increased as compared to that of TBC1D15 single KO . Therefore , in HCT116 cells , TBC1D15 appears to be dominant for contouring autophagosome morphogenesis . TBC1D15 and TBC1D17 may have different cell type expression profiles or different regulatory mechanisms . Overexpression of either TBC1D15 or TBC1D17 can rescue the LC3 accumulation in TBC1D15/17DKO cells , whereas the TBC1D15 Δ221-250 mutant cannot ( Figure 6—figure supplement 3C , D ) . Thus , Fis1 binding is required for TBC1D15 to govern isolation membrane biogenesis during mitophagy . The absence of TBC1D15 causes the accumulation and/or tubulation of the autophagosome marker , LC3 upon Parkin activation of mitophagy . This would be consistent with Rab-GAP activity of TBC1D15 switching a Rab protein to an inactive form to counterbalance Parkin-induced Rab activation . We hypothesized that the localization of such a Rab-GAP to mitochondria could inhibit Rab activity on one side of the isolation membrane to help contour the membrane expansion to surround the cargo . TBC1D15 may coordinate mitochondrial attachment to pre-autophagosomal membranes with asymmetric inhibition of Rab proteins to guide autophagosomes to tightly engulf mitochondria . Consistent with this hypothesis , several TBC family proteins such as TBC1D25/OATL1 and TBC1D2B have been reported to directly bind Atg8 family proteins ( Itoh et al . , 2011; Popovic et al . , 2012 ) and TBC1D15 was identified by proteomics to bind Atg8 family members ( Behrends et al . , 2010 ) . To test this model , we assessed Atg8 family protein interaction with TBC1D15 and sought to identify novel LC3-interacting region ( LIR ) motifs . In mammals , there are at least six different ATG8 homologues including LC3A , LC3B , LC3C , GABARAP , GABARAPL1 and GABARAPL2 . Both LC3 and GABARAP subfamilies are found on autophagosomal membranes ( Kabeya et al . , 2004 ) . Similar to YFP-LC3B ( Figure 2A ) , YFP-GABARAPL1 also forms massively expanded structures in FIS1−/− and in TBC1D15−/− cells after Parkin translocation induced by valinomycin treatment ( Figure 7A ) . We prepared recombinant GST-tagged Atg8 family proteins and subjected them to in vitro binding assays with cell extracts overexpressing YFP-TBC1D15 . While GST alone does not bind YFP-TBC1D15 , LC3 , and GABARAP subfamily members bind in moderate and high amounts to YFP-TBC1D15 , respectively ( Figure 7B ) . Fis1 cannot bind GABARAPL1 , confirming that the binding between TBC1D15 and GABARAPL1 is not mediated indirectly through Fis1 ( Figure 7F ) . In general , Atg8 homologues bind through their LDS ( LIR docking site ) to an LIR motif of a substrate ( Noda et al . , 2008 ) . We performed more detailed analyses of the mechanism of GABARAPL1 binding to TBC1D15 because the GABARAP subfamily proteins display greater binding than LC3 subfamily members ( Figure 7B ) . We made a GABARAPL1 LDS mutant by replacing Y49 and L50 with alanines . Although equal amounts of GST-tagged GABARAPL1 WT and LDS mutant were expressed and immunoprecipitated , the LDS mutant failed to bind YFP-TBC1D15 ( Figure 7C ) , indicating that the binding occurs through LIR–LDS interactions . This result was confirmed in binding assays of HA-TBC1D15 and YFP-GABARAPL1 expressed in cultured cells ( Figure 7D ) . 10 . 7554/eLife . 01612 . 021Figure 7 . TBC1D15 binds ATG8 family proteins . ( A ) The indicated cells transiently expressing YFP-GABARAPL1 ( green ) and mCherry-Parkin ( red ) were treated with valinomycin for 3 hr . Scale bars , 10 μm . ( B ) YFP-TBC1D15 overexpressed in HEK293 cells was subjected to binding assays with GST-fused proteins ( GABA , L1 , and L2 represent GABARAP , GABARAPL1 , and GABARAPL2 , respectively ) . 5% input and bound fractions were analyzed by immunoblotting with anti-GFP antibody ( upper panel ) . Coomassie brilliant blue ( CBB ) staining shows GST-fusion proteins in bound fractions ( lower panel ) . ( C ) Binding assay carried out as in ( B ) with GST-GABARAPL1 WT ( GST-WT ) or its Y49A/L50A mutant ( GST-YL ) . Immunoblotting with anti-GFP antibody ( upper panel ) and CBB staining ( lower panel ) are shown . ( D ) Cell extracts from HEK293 overexpressed HA-TBC1D15 and YFP , YFP-GABARAPL1 ( YFP-WT ) , or its Y49AL50A mutant ( YFP-YL ) were subjected to pull down assays with GFP-Trap . 5% input and bound fractions were analyzed by immunoblotting with anti-HA ( upper panel ) and anti-GFP ( lower panel ) antibodies . ( E–H ) The indicated YFP-tagged TBC1D15 full-length , truncated , or point-mutant protein or YFP-Fis1 overexpressed in HEK293 cells were subjected to binding assays with recombinant GST-GABARAPL1 . 5% input and bound fractions were analyzed by immunoblotting with anti-GFP antibody ( upper panel ) and CBB staining ( lower panel ) . ( I ) Summary of binding abilities of truncated or point-mutated TBC1D15 constructs . – , + , ++ , and +++ indicates binding of recombinant GST-GABARAPL1 to less than 1% , 1–5% , 5–10% , and over 10% , respectively of the total YFP-TBC1D15 fragment . Yellow boxes indicate YFP tags . ( J ) YFP-LC3 and mCherry-Parkin stably expressing TBC1D15−/− cells in the presence of HA-tagged TBC1D15 WT or F280A mutant were treated with valinomycin for 3 hr . Cells were subjected to immunofluorescence microscopy with anti-HA antibody . Scale bars , 10 μm . ( K ) The YFP-LC3 morphology of cells in ( J ) was quantified . The error bars represent ±SD from three independent replicates . Over 50 cells were counted in each well . DOI: http://dx . doi . org/10 . 7554/eLife . 01612 . 02110 . 7554/eLife . 01612 . 022Figure 7—figure supplement 1 . Schematic representation of potential LIR motifs of TBC1D15 . The amino acid position ( Pos . ) of the conserved aromatic residue in the potential LIR motifs in human TBC1D15 is indicated . Red , acidic residues; Blue , basic residues . DOI: http://dx . doi . org/10 . 7554/eLife . 01612 . 02210 . 7554/eLife . 01612 . 023Figure 7—figure supplement 2 . TBC1D17 has the ability to bind Atg8 homologues . ( A ) Protein sequence comparison of the LIR motif in TBC1D15 with that of TBC1D17 . Underlined sequence denotes consensus LIR motif . Red , acidic residues; Blue , basic residues . ( B ) Overexpressed YFP-TBC1D17 was subjected to binding assays with GST-fused Atg8 proteins ( GABA , L1 , and L2 represent GABARAP , GABARAPL1 , and GABARAPL2 , respectively ) . 5% input and bound fractions were analyzed by immunoblotting with anti-GFP antibody ( upper panel ) and coomassie brilliant blue ( CBB ) staining ( lower panel ) . ( C ) Binding assays of YFP-TBC1D17 were carried out with GST-fused GABARAPL1 WT ( GST-WT ) or its Y49AL50A mutant ( GST-YL ) . Immunoblotting with anti-GFP antibody ( upper panel ) and CBB staining ( lower panel ) are shown . ( D ) Extracts from cells overexpressing both HA-TBC1D17 and YFP , YFP-GABARAPL1 ( YFP-L1 ) , its Y49AL50A mutant ( YL ) , YFP-LC3B or its F52A/L53A mutant ( FL ) were subjected to pull down assays with GFP-Trap . 5% input and bound fractions were analyzed by immunoblotting with anti-HA ( upper panel ) and anti-GFP ( lower panel ) antibodies . ( E ) YFP-TBC1D17 WT and F261A mutant overexpressed in HEK293 cells were subjected to a binding assay with recombinant GST-LC3B . 5% input and bound fractions were analyzed by immunoblotting with anti-GFP antibody ( upper panel ) and CBB staining ( lower panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01612 . 023 To explore the LDS–LIR interactions in more detail , we assessed potential LIR motifs in TBC1D15 . The core of the LIR motif consists of the sequence ( W/F/Y ) XX ( L/I/V ) , where X is any amino acid and with acidic amino acid residues often found near the LIR motif ( Alemu et al . , 2012 ) . Based on this consensus motif , we identified 20 different LIR candidates in the TBC1D15 protein sequence ( Figure 7—figure supplement 1 ) . To narrow down candidates , we made and tested TBC1D15 truncations by in vitro binding assays using GST-tagged GABARAPL1 . TBC1D15 was first divided into four parts according to the domain structure; the N-terminus ( 1–200 ) , a region including the Fis1-interacting site ( 201-333 ) , the TBC domain ( 334–557 ) , and a C-terminal region ( 558–674 ) , all of which were tagged with YFP at the N-terminus for detection by immunoblotting . Transiently overexpressed TBC1D15 domains in HEK293 cells subjected to in vitro binding assays revealed that only the 201–333 region retained the affinity for GABARAPL1 ( Figure 7E ) . Similar results were obtained when we made longer constructs containing this region also indicating that polypeptides within the 201–333 aa region could efficiently bind GABARAPL1 ( Figure 7F ) . Five different LIR candidates were found in the 201–333 aa region . By mutating the aromatic amino acid in each of these five LIR candidates to alanine , we identified 280FEVI283 of TBC1D15 as the LIR motif responsible for full length TBC1D15 binding to GABARAPL1 ( Figure 7G , H ) . A summary of TBC1D15 binding to GABARAPL1 is shown in Figure 7I . To examine whether Atg8 binding to TBC1D15 is required to orchestrate autophagosome formation , the TBC1D15 F280A mutant was introduced into TBC1D15−/− cells . The F280A mutation resides outside the Fis1-binding region ( 221–250 aa of TBC1D15 ) ( Onoue et al . , 2013 ) and does not affect the mitochondrial targeting of TBC1D15 ( data not shown ) . The LC3 membrane accumulation phenotype of the TBC1D15−/− cells was only partially rescued by the TBC1D15 F280A mutant ( Figure 7J , K ) relative to the complete rescue seen with WT TBC1D15 , indicating that the interaction between TBC1D15 and Atg8 family proteins is important for autophagosome morphogenesis during Parkin-mediated mitophagy . Interestingly , amino acids 261–264 of TBC1D17 are identical to the sequence of the LIR motif in TBC1D15 ( FEVI ) ( Figure 7—figure supplement 2A ) . To test whether TBC1D17 also has the ability to bind Atg8 family proteins , in vitro binding assays were performed . Both LC3 and GABARAP subfamilies pulled down YFP-TBC1D17 ( Figure 7—figure supplement 2B ) , and this interaction was mediated by the LDS domain of ATG8 family proteins ( Figure 7—figure supplement 2C , D ) with less discrimination between GABARAP and LC3 subfamily members than TBC1D15 ( Figure 7B ) . Furthermore , the LIR mutation F261A prevented binding of TBC1D17 to GST-LC3B ( Figure 7—figure supplement 2E ) . These results indicate that both TBC1D15 and TBC1D17 have the ability to bind ATG8 family proteins via a LIR–LDS interaction . As Rab-GAPs , TBC family proteins play important roles by inactivating Rab proteins and different Rab proteins are thought to be regulated by different TBC family members . Although little is known about the pairing between various Rab proteins and TBC proteins , in vitro GTPase assays ( Zhang et al . , 2005 ) and pull-down assays in cells ( Peralta et al . , 2010 ) suggest that TBC1D15 may function as a Rab-GAP for Rab7 . Rab7 is required for endosome maturation and transport from the late endosome to the lysosome ( Bento et al . , 2013 ) . However , Rab7 also has been reported to be involved in the biogenesis of autophagosomes ( Gutierrez et al . , 2004; Jager et al . , 2004 ) . We hypothesized that if TBC1D15 bound to Fis1 on mitochondria normally inhibits Rab7 during mitophagy through Rab-GAP activity , the excessive LC3-labeled tubulation observed in TBC1D15−/− cells ( as well as FIS1−/− cells ) might be a consequence of excessive or unregulated Rab7 activity . To test this hypothesis , we knocked down Rab7 by siRNA . We found 2 of 4 different siRNAs tested could knock down endogenous Rab7 expression efficiently ( Figure 8A ) . These two Rab7 siRNAs were able to completely suppress the abnormal LC3 accumulation and tubulation in both FIS1−/− or TBC1D15−/− cells ( Figure 8B , C , Figure 8—figure supplement 1A and data not shown ) . We investigated Rab7 localization in WT and TBC1D15−/− cells with or without valinomycin treatment . As we could not detect endogenous Rab7 by immunofluoresence microscopy , we utilized N-terminally 2HA- or YFP-tagged Rab7 . In WT or TBC1D15−/− cells stably expressing YFP-LC3 , mCherry-Parkin and 2HA-Rab7 , Rab7 predominantly localizes on lysosomes with a weak ER signal both in WT and TBC1D15−/− cells under basal conditions ( Figure 8—figure supplement 1B ) . However , after 3 hr valinomycin treatment of WT cells , 2HA-Rab7 colocalized with some of the YFP-LC3 in WT cells in dot or crescent-shaped structures that represent isolation membranes , indicating that Rab7 is targeted to isolation membranes upon stimulation of mitophagy ( Figure 8D , WT valinomycin ) . Interestingly , Rab7 was also found on YFP-LC3 positive tubules in TBC1D15−/− cells ( Figure 8D , TBC1D15−/− valinomycin ) , strongly supporting the hypothesis that Rab7 functions in isolation membrane expansion during mitophagy . 10 . 7554/eLife . 01612 . 024Figure 8 . Rab7 is involved in autophagosome fusion during mitophagy . ( A ) Total cell lysates from HCT116 treated with control ( NTC ) or RAB7A siRNA were analyzed by immunoblotting . ( B ) The indicated cells stably expressing YFP-LC3 ( green ) and mCherry-Parkin were treated with control ( NTC ) or Rab7_#5 siRNA . After 3 hr valinomycin treatment , cells were analyzed by immunofluorescence microscopy with anti-TOMM20 antibody ( red ) . Z-stacks of confocal images are shown . Magnified images are also shown . Scale bars , 10 μm . ( C ) YFP-LC3 morphologies of cells in ( B ) were quantified . Percentages of cells harboring diffuse , punctuate or accumulated/tubulated YFP-LC3 are shown . Data and error bars were obtained from at least 50 cells in each of three independent replicates . ( D ) The indicated cells stably expressing YFP-LC3 ( green ) , mCherry-Parkin , and 2HA-Rab7 ( Red ) were treated with or without valinomycin for 3 hr and analyzed by immunofluorescence microscopy with anti-HA antibody . Magnified images are also shown . Scale bars , 10 μm . ( E ) YFP-LC3 and mCherry-Parkin stably expressing TBC1D15−/− cells in the presence of HA-tagged TBC1D15 WT or the D397A mutant were treated with valinomycin for 3 hr . Cells were subjected to immunofluorescence microscopy with anti-HA antibody . Scale bars , 10 μm . ( F ) The YFP-LC3 morphology of cells in ( E ) was quantified . The error bars represent ±SD from three independent replicates . Over 50 cells were counted in each replicate . DOI: http://dx . doi . org/10 . 7554/eLife . 01612 . 02410 . 7554/eLife . 01612 . 025Figure 8—figure supplement 1 . Rab7 is involved in LC3 accumulation of FIS1−/− and TBC1D15−/− cells . ( A ) Low magnification images of cells stably expressing YFP-LC3 and mCherry-Parkin treated with valinomycin for 3 hr . Cells were subjected to immunofluorescence microscopy with anti-TOMM20 antibody . Z-stacks of confocal images are shown . Scale bars , 20 μm . ( B ) The indicated cells stably expressing YFP-Rab7 and mCherry-Parkin were analyzed by immunostaining with anti-GFP and anti-LAMP2 antibodies . Scale bars , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01612 . 02510 . 7554/eLife . 01612 . 026Figure 8—figure supplement 2 . TBC1D17 GAP activity is important for LC3 accumulation through Rab7 . ( A ) HA-tagged TBC1D17 WT or the R381A mutant was transiently overexpressed ( OE ) in TBC1D15/17 DKO cells stably expressing YFP-LC3 and mCherry-Parkin . The cells treated with valinomycin for 3 hr were subjected to immunofluorescence microscopy with anti-HA antibody . Scale bars , 10 μm . ( B ) The YFP-LC3 morphology of cells in ( A ) was quantified . The error bars represent ±SD from three independent replicates . Over 50 cells were counted in each replicate . ( C ) TBC1D17−/− cells stably expressing YFP-LC3 ( green ) and mCherry-Parkin were treated with control ( NTC ) and Rab7_#5 siRNAs for 48 hr . The cells treated with valinomycin for 3 hr were then subjected to immunofluorescence microscopy with anti-TOMM20 antibody ( red ) . Z-stacks of confocal images are shown . Scale bars , 20 μm . ( D ) The YFP-LC3 morphology of cells in ( C ) was quantified . The percentages of cells harboring diffuse , punctate or accumulated/tubulated YFP-LC3 were quantified . The error bars represent ±SD from three independent replicates . Over 50 cells were counted in each of three replicate . DOI: http://dx . doi . org/10 . 7554/eLife . 01612 . 026 We next asked whether the activity of TBC1D15 was dependent on Rab-GAP activity . The mutant HA-tagged TBC1D15 D397A , which lacks TBC domain GAP activity ( Rak et al . , 2000 ) , was compared with WT TBC1D15 for the ability to rescue TBC1D15−/− cells from excessive LC3 accumulation during Parkin-mediated mitophagy . Although HA-TBC1D15 D397A localizes on mitochondria ( data not shown ) , it fails to rescue the accumulation of LC3 in TBC1D15−/− cells as WT TBC1D15 does ( Figure 8E , F ) . Furthermore , overexpression of TBC1D17 catalytic mutant R381A could not rescue the LC3 accumulation in the TBC1D15/TBC1D17 DKO cells ( Figure 8—figure supplement 2A , B ) . Although many different Rab proteins such as Rab5 , Rab8 , Rab21 and Rab35 have been suggested to be the targets of TBC1D17 ( Itoh et al . , 2006; Fuchs et al . , 2007; Vaibhava et al . , 2012 ) , Rab7 siRNA could suppress YFP-LC3 accumulation in TBC1D17−/− cells ( Figure 8—figure supplement 2C , D ) . These results support the model that Rab-GAP activity inhibits Rab7 to mitigate excessive autophagosome membrane expansion during mitophagy and , through Fis1 binding to mitochondria and LC3 family member binding on isolation membranes , tailors the autophagosomal membrane expansion to its cargo . In this study , we identify Rab7 , TBC1D15 and TBC1D17 , and Fis1 as mitophagy effectors functioning downstream of Parkin to regulate autophagosome morphogenesis . Mitochondrial damage is initially signaled to Parkin through the kinase PINK1 ( Youle and Narendra , 2011 ) . As PINK1 accumulates on the outer membrane of depolarized mitochondria , it recruits Parkin and activates Parkin ubiquitin ligase activity ( Iguchi et al . , 2013; Lazarou et al . , 2013; Zheng and Hunter , 2013 ) . Parkin then modifies numerous mitochondrial outer membrane proteins with K48- and K63-linked ubiquitin chains . This ubiquitin signal appears to recruit ULK1 , Atg9 and LC3 to the mitochondrion independently of one another to induce mitophagy ( Itakura et al . , 2012 ) . Our results also indicate that Rab7 is likely activated by a Rab GEF to promote autophagosomal membrane growth and microtubule associated trafficking and that TBC1D15/17 Rab GAP activity is required to temper Rab7 activity to tailor autophagosomal membrane expansion to surround mitochondrial aggregates . Without TBC1D15 , or with TBC1D15 lacking Rab GAP activity , LC3/GABARAP-labeled isolation membranes accumulate excessively and without cargo orientation sending long tubules away from mitochondria along microtubule tracks . This suggests that Rab GAP activity localized on the cargo—facing only one side of the expanding LC3 positive isolation membranes—may inhibit Rab7 activity and thereby inhibit membrane expansion proximal to the cargo forcing the growing membrane to tightly wrap odd shaped cargo . Alternatively , perhaps LC3-decorated preautophagosomal structures traffic along microtubules toward Parkin-ubiquitinated mitochondria and fuse together around mitochondrial aggregates where TBC1D15 bound to mitochondria inhibits Rab7 to mediate the release of LC3-bound membranes from microtubules as they contact the mitochondrial cargo ( Pankiv et al . , 2010 ) . TBC1D15−/− cells display normal LC3 accumulation during starvation suggesting that these membrane expansion steps are specific for mitophagy or other forms of selective autophagy . Xenophagy , for example , also requires Rab proteins for sequestering invading bacteria in autophagosome-like vacuoles ( Yamaguchi et al . , 2009 ) , consistent with the idea that isolation membrane expansion is mediated by Rab activity . Parkin also may activate Rab proteins as a step in xenophagy ( Manzanillo et al . , 2013 ) . We identify an LIR domain in TBC1D15 , conserved in TBC1D17 , that is also required to restrain isolation membrane biogenesis to the surface of the mitochondria . Proteins that contain both an LIR domain and a ubiquitin-associated domain such as p62 ( Bjorkoy et al . , 2005 ) , optineurin ( Wild et al . , 2011 ) , NBR1 ( Kirkin et al . , 2009 ) , and NDP52 ( Thurston et al . , 2009 ) can bind to LC3 and to polyubiquitin chains to recruit ubiquitinated cargo into nascent isolation membranes . TBC1D15 and TBC1D17 are also LIR domain-containing LC3/GABARAP-binding proteins . However , in contrast to the ubiquitin-binding LIR domain proteins , TBC1D15 and TBC1D17 appear to constitutively bind to mitochondria through Fis1 and thus , unless somehow regulated , would not appear to initiate the recruitment of LC3 to mitochondria . More likely , TBC1D15 and TBC1D17 use LC3/GABARAP binding in the context of Rab GAP activity to orient growing isolation membranes to the surface of the cargo . Interestingly , TBC1D15 and TBC1D17 appear to heterodimerize and homodimerize , which may increase their affinity for LC3/GABARAP by multivalent binding . Fis1 also forms dimers consistent with its role as the TBC1D15/17 receptor . The self association of TBC1D15 and TBC1D17 is shared by other ubiquitin-binding LIR domain proteins such as p62 ( Birgisdottir et al . , 2013 ) . We previously reported that FIS1−/− cells display no detectable mitochondrial morphology phenotype ( Otera et al . , 2010 ) . However , it is well established that yeast Fis1 binds Dnm1 through Mdv1/Caf4 and regulates mitochondrial fission ( Okamoto and Shaw , 2005 ) . Furthermore , Fis1 loss in C . elegans causes LC3 accumulation and has been linked to stress-induced mitochondrial fission ( Shen et al . , 2014 ) . Interestingly , Dnm1 activity in yeast ( Mao et al . , 2013 ) , and Drp1 activity in mammalian cells is required for mitophagy ( Twig et al . , 2008; Tanaka et al . , 2010; Frank et al . , 2012 ) suggesting a link with Fis1 in mitophagy . In contrast to human Fis1 , yeast Fis1 has an N-terminal extension that folds back into a pocket used by other tetratricopeptide motif proteins to bind ligands ( Suzuki et al . , 2003 , 2005 ) . It will be important to determine how TBC1D15 interacts with human Fis1 and if the Rab GAP-binding activity of human Fis1 is conserved in yeast . Fis1 also localizes to peroxisomes and would be predicted to recruit TBC1D15/17 to peroxisomes as well , suggesting that Fis1 and TBC1D15/17 may also play a role in pexophagy . Human LC3B was cloned into BglII/EcoRI sites of pEYFP-C1 ( Clontech , Mountain View , CA ) to make YFP-LC3 , which was then cloned into BamHI/NotI sites of pCHAC/IRES ( Allele Biotechnology , San Diego , CA ) to make a retrovirus plasmid pCHAC/YFP-LC3B-IRES-MCS2 . To generate a retrovirus mCherry-Parkin plasmid ( pBMNz/mCherry-Parkin ) , the gene for N-terminally mCherry-tagged Parkin ( Narendra et al . , 2008 ) was cloned into BamHI/NotI sites of pBMN-Z vector ( Addgene plasmid 1734 ) . 2×HA-tagged RAB7A gene from DsRed-rab7 WT ( Addgene plasmid 12661 ) was subcloned into pBABE-puro vector ( Addgene plasmid 1764 ) using BamHI to generate pBABE-puro/2HA-Rab7 . YFP-GABARAPL1 was generated by inserting a GABARAPL1 gene into BglII/EcoRI sites of pEYFP-C1 vector . The Y49AL50A mutation in GABARAPL1 and the F52AL53A mutation in LC3B were introduced by primer-based PCR mutagenesis . For GST expression , pGEX-KG vector was used . The following plasmids were also used in this study: pMRX-IP GFP-ULK1 and pMRXs-puro GFP-DFCP1 ( kind gifts from Dr Noboru Mizushima ) , pMXs-IP GFP-Atg14 ( Addgene plasmid 38264 ) To generate YFP-TBC1D15 and YFP-TBC1D17 plasmids , TBC1D15 and TBC1D17 genes were amplified from HA-TBC1D15 plasmid ( a kind gift from Dr Naotada Ishihara ) and from POTB7/TBC1D17 ( purchased from Thermo Scientific #MHS6278-202827046 ) by PCR and inserted into the XhoI/KpnI sites of pEYFP-C1 plasmid . YFP-TBC1D15-truncated mutants were generated by the same method with appropriate primer pairs . Alanine mutants were generated by primer-based PCR mutagenesis . Construction of HA-TBC1D15 WT and HA-TBC1D17 was done as described in Onoue et al . ( 2013 ) . R381A mutation in HA-TBC1D17 , F280A or D397A mutations in HA-TBC1D15 and deletion of 221–250 aa in HA-TBC1D15 were achieved by primer-based PCR mutagenesis . These HA-TBC1D15 genes were then subcloned into the BamHI/EcoRI sites of pBMN-z . 3xFLAG-Fis1 was generated by cloning 3×FLAG and FIS1 genes into pBABE-puro . The following antibodies were used for immunoblotting: rabbit anti-GFP ( A-11122; Invitrogen , Grand Island , NY ) , mouse anti-HA ( clone 16B12; Covance , Berkeley , CA ) , mouse anti-α-Tubulin ( Invitrogen clone B-5-1-2 ) , mouse anti-Actin ( clone AC-40; Sigma , St . Louis , MO ) , rabbit anti-TOMM20 ( sc-11415; Santa Cruz Biotechnology , Inc . , Dallas , TX ) , rabbit anti-PINK1 ( BC100-494; Novus Biologicals , Littleton , CO ) , mouse anti-TIMM23 ( clone 32; BD Biosciences , San Jose , CA ) , rabbit anti-LC3B ( L7543; Sigma ) , mouse anti-HSP60 ( clone LK-1; Stressgen , Victoria , Canada ) , rabbit anti-Mfn1 ( generated as described previously in the study by Karbowski et al . , 2007 ) , mouse anti-Rab7 ( Rab7-117; Abcam , Cambridge , MA ) , rabbit anti-TBC1D15 ( described previously in the study by Onoue et al . , 2013 ) , rabbit anti-Fis1 ( ALEXIS Biochemicals , San Diego , CA ) , rabbit anti-Mff ( described previously in the study by Gandre-Babbe and van der Bliek , 2008 ) , mouse anti-Drp1 ( clone 8/DLP1; BD Biosciences ) , mouse anti-MTCOXII ( clone 12C4F12; Abcam ) , and mouse anti-HA ( clone 16B12; COVANCE ) . The following antibodies were used for immunostaining: rabbit anti-TOMM20 ( sc-11415; Santa Cruz Biotechnology , Inc . ) , mouse anti-HA ( clone 16B12; COVANCE ) , mouse anti-FLAG M2 ( Agilent Technologies , Cedar Creek , TX ) , rabbit anti-Fis1 ( ALEXIS Biochemicals ) , mouse anti-Cytochrome c ( clone 6H2 . B4; BD Biosciences ) , rabbit anti-PMP70 ( 71-8300; Invitrogen ) , rabbit anti-GFP antibody ( A-11122; Invitrogen ) , mouse anti-α-Tubulin ( Invitrogen clone B-5-1-2 ) , mouse anti-LAMP2 antibody ( sc-18822; Santa Cruz Biotechnology , Inc . ) , mouse anti-GM130 antibody ( clone 35/GM130; BD Biosciences ) , mouse anti-KDEL antibody ( clone 10C3; Stressgen ) , and rabbit anti-Atg16L1 antibody ( a kind gift from Dr Noboru Mizushima ) . F-Actin was stained with Alexa Fluor 647 Phalloidin ( invitrogen ) . RAB7A siRNA oligos were purchased from QIAGEN ( Valencia , CA ) . The target sequences are as follows: RAB7_#5 , CACGTAGGCCTTCAACACAAT , RAB7_#6 , CTGCTGCGTTCTGGTATTTGA , RAB7_#2 , TCCCGTTAGATCAGCATTCTA , RAB7_#4 , TAGATCAGCATTCTACTACAA . Nontargeting control siRNA and PINK1 siRNA were described previously ( Lazarou et al . , 2013 ) . HeLa , HEK293 , and HEK293T cells were cultured in Dulbecco’s modified Eagle’s medium ( DMEM ) supplemented with 10% ( vol/vol ) fetal calf serum , 10 mM HEPES buffer , 1 mM sodium pyruvate , 1 mM glutamine , and nonessential amino acids . HCT116 cells were cultured in McCoy’s 5A medium supplemented with 10% ( vol/vol ) fetal calf serum , 1 mM glutamine , and nonessential amino acids . The cells were cultured at 37°C in a 5% CO2 incubator . FIS1−/− HCT116 cell was reported previously ( Otera et al . , 2010 ) . For transient transfection of plasmids , Fugene HD ( Promega , Madison , WI ) transfection reagent was used according to the manufacture’s instruction . For gene knock-down by siRNA , the cells were transfected with nontargeting control or RAB7A siRNA ( at a final concentration of 12 . 5 nM ) or PINK1 siRNA ( at a final concentration of 6 nM ) with Lipofectamine RNAiMAX ( Invitrogen ) according to the manufacture’s instruction . After 24 hr , the medium was changed to fresh medium and the cells were grown for further 24 hr before analysis . Stable cell lines were established by recombinant retrovirus infection as follows . Vector particles were produced in HEK293T cells grown in poly-lysine-coated plates by cotransfection with Gag-Pol , VSV-G and a retrovirus plasmid ( pCHAC/YFP-LC3B-IRES-MCS2 , pBMNz/mCherry-Parkin , pBABE-puro/2HA-Rab7 , pBABE-puro/YFP-Rab7 , pBABE-puro/3FLAG-Fis1 , pMRX-IP GFP-ULK1 , pMRXs-puro GFP-DFCP1 , pMXs-IP GFP-Atg14 , pBMN-z/HA-TBC1D15 or its mutants ) . After 12 hr of transfection , the medium was changed to a fresh medium and the cells were further incubated for 24 hr . The viral supernatants were then infected into HCT116 cells with 8 μg/ml polybrene ( Sigma ) . Amino acid starvation was induced by washing cells twice with starvation buffer ( 20 mM HEPES pH 7 . 4 , 140 mM NaCl , 1 mM CaCl2 , 1 mM MgCl2 , 5 mM glucose ) followed by incubation with starvation buffer containing 1% ( wt/vol ) BSA . Valinomycin ( Sigma ) was used at a final concentration of 10 μM . When cells were treated with valinomycin more than 6 hr , 10 μM Q-VD-OPH ( SM Biochemicals , Anaheim , CA ) was added to block apoptopic cell death . Nocodazole ( Calbiochem , Darmstadt , Germany ) was used at a final concentration of 10 μM . The MFF−/− cell line constructed by TALENs is described in Shen et al . , 2014 and the DRP1−/− cell line constructed by TALENs is described in Wang et al . ( unpublished ) . The TBC1D15−/− cell line was generated by TALEN targeting the following site: 5′- GCC CTG TTG TTC AAA GGA GAG Aac cgg tAT CAC TGG AAG AAT GGA CTA AGA ACA TTG -3′ ( underlined sequences and lower-case sequences indicate the target sites for left and right TALEs and AgeI restriction enzyme site , respectively ) . The targeting site for knock out of TBC1D17 gene was 5′- CTT CCC CCA ACA CAG TGC CGT CTC cct agg TGC AGA GCC CAG CTG CCC CCA GG -3′ ( AvrII localized at the junction between intron and exon5 was used for restriction enzyme site ) . 16-mer left and right TALEs were assembled according to Huang et al . ( 2011 ) and cloned into a final TALEN vector modified from Miller et al . ( 2011 ) . 0 . 8 μg of left and right TALEN constructs were cotransfected with 0 . 4 μg pEYFP-C1 vector in HCT116 cells and the cells were grown for 2 days . YFP-positive cells were sorted by FACS and plated into 96-well plates . Genomic DNA was isolated from single colonies , PCR amplification of the target site was performed , and assayed by restriction enzyme digestion . Finally , the gene knock out clones were confirmed by DNA sequencing and/or immunoblotting . Cells grown on 2-well coverglass chamber slides were fixed with 4% paraformaldehyde in PBS for 25 min at room temperature , permeabilized with 0 . 015% ( vol/vol ) TX-100 in PBS for 15 min , and preincubated with 5% ( wt/vol ) BSA for 30 min . The fixed cells were incubated with primary antibodies and appropriate secondary antibodies ( Alexa Fluor 488 , 594 or 647 goat anti-rabbit or anti-mouse IgG from Invitrogen ) for immunostaining . The images of the cells were captured using an inverted confocal microscope ( LSM510 Meta , Carl Zeiss ) with a 63×/1 . 4 NA oil differential interference contrast Plan-Apochromat objective lens . For image analysis , Volocity ( PerkinElmer ) and/or Photoshop ( Adobe ) software were used . For preparation of total cell lysate , cells grown in a six-well plate were washed twice with PBS and solubilized with 2% CHAPS buffer ( 25 mM HEPES-KOH pH 7 . 5 , 300 mM NaCl , 2% ( wt/vol ) CHAPS , protease inhibitor cocktail ( Roche , Indianapolis , IN ) on ice for 30 min and then protein concentrations were determined . Proteins precipitated with trichroloacetic acid were lysed with NuPAGE LDS sample buffer ( Invitrogen ) supplemented with 80 mM dithiothreitol . The appropriate amounts of proteins were applied and separated on 4–12% Bis-Tris SDS-PAGE ( Invitrogen ) with MES or MOPS SDS running buffer ( Invitrogen ) . After transfer to PVDF membrane , blocking and incubation with primary antibodies , proteins were detected using horseradish peroxidase-coupled secondary antibodies ( GE Healthcare Life Sciences , Piscataway , NJ ) and ECL Plus or ECL Prime western blotting detection reagents ( GE Healthcare Life Sciences ) . GST alone or GST-tagged Atg8 family proteins were overexpressed in Escherichia coli strain BL21 ( DE3 ) from the plasmids , which were a gift from Dr Terje Johansen ( Pankiv et al . , 2007 ) . Transformants were grown in LB medium supplemented with 100 mg/L ampicillin at 37°C and the protein expression was induced by the addition of 1 mM isopropyl β-D-thiogalactopyranoside for 24 hr at 16°C . Cells were harvested by centrifugation , washed with PBS , and kept at −80°C before use . The cell pellets were lysed with B-PER bacterial extraction reagent ( Pierce , Rockford , IL ) supplemented with 100 μg/ml lysozyme and 5 units/ml DNaseI and incubated at room temperature for 10 min . After sonication , the cell debris and insoluble fraction were removed by centrifugation ( 20 , 000×g , 15 min , 4°C ) . The resulting supernatant was incubated with Glutathione Sepharose 4 Fast Flow ( GE Healthcare Life Sciences ) equilibrated with PBS for 1 hr at 4°C . GST-bound sepharose was washed three times with PBS-containing 500 mM NaCl . YFP-tagged TBC1D15 and its derivatives were overexpressed in HEK293 cells grown in a six-well plate . The cells were washed with PBS twice , and solubilized with 0 . 5% TX-100 buffer ( 50 mM HEPES pH 7 . 5 , 150 mM NaCl , 1 mM EDTA , 1 mM EGTA , 8% [vol/vol] glycerol , 25 mM NaF , 0 . 5 [vol/vol] Triton X-100 , protease inhibitor cocktail [Roche] ) for 20 min on ice . Insoluble proteins were removed by centrifugation ( 20 , 000×g , 15 min , 4°C ) . The resulting supernatants were mixed with GST protein-bound sepharose in TBC-binding buffer ( 50 mM HEPES pH 7 . 5 , 150 mM NaCl , 1 mM EDTA , 1 mM EGTA , 8% [vol/vol] glycerol , 25 mM NaF ) for 3 hr at 4°C . The sepharose slurry was washed with 0 . 1% TX-100 buffer three times and proteins were eluted with NuPAGE LDS sample buffer ( Invitrogen ) containing 80 mM dithiothreitol . The in vitro binding assay using GFP-Trap was performed as follows . YFP-tagged proteins and interacting candidates were co-overexpressed in HEK293 cells grown in six-well plates by transient transfection with Lipofectamine LTX ( Invitrogen ) . After 24 hr of transfection , the cells were washed with PBS twice and solubilized with 0 . 5% TX-100 buffer for 20 min on ice . The resulting lysates were clarified by centrifugation and mixed with equilibrated GFP-Trap resin ( Chromotek , Planegg-Martinsried , Germany ) in TBC-binding buffer for 1 hr at 4°C . Proteins were washed with 0 . 1% TX-100 buffer three times before elution with LDS sample buffer containing 80 mM dithiothreitol . HCT116 WT and TBC1D15−/− cells stably expressing YFP-LC3 and mCherry-Parkin were treated with Valinomycin for 3 hr and fixed for 30 min with 4% paraformaldehyde and 0 . 1% glutaraldehyde in PBS . The fixed cells were washed four times with PBS , followed by permeabilization for 40 min with 0 . 1% Saponin and 5% goat serum in PBS . The cells were then incubated for 1 hr with mouse anti-GFP antibody ( Invitrogen clone 3E6 ) , followed by 1 hr with nanogold-conjugated anti-mouse IgG antibody ( Nanoprobes ) and further processing as described ( Tanner et al . , 1996 ) . Thin sections ( ∼80 nm ) were counter stained with uranyl acetate and lead citrate . The sections were examined with a JEOL 200 CX transmission electron microscope . Images were collected with a digital CCD camera ( AMT XR-100; Danvers , MA ) .	Parkinson disease is a common degenerative brain disorder that causes tremors , muscle stiffening , and slowing down of movement . Scientists believe that these symptoms are caused by a progressive loss of brain cells called dopaminergic neurons , which help regulate movement . Most cases have no obvious genetic cause , but around 15% of people with the disease have a close relative who also has the disease , and mutations in the genes encoding two proteins—PINK1 and Parkin—have been identified as prime suspects in familial Parkinson disease . These proteins help to eliminate damaged mitochondria from cells . In addition to producing the energy that cells need to function , mitochondria also help to trigger cell death . Pesticides and other chemicals linked to non-familial cases of Parkinson disease also damage mitochondria . Taken together , this evidence suggests that the accumulation of damaged mitochondria may contribute to the excessive loss of dopaminergic neurons that is seen in both forms of the disease . Yamano et al . provide new details on the ways that autophagosomes—structures that help cells to recycle nutrients and remove debris—destroy mitochondria . Previous studies have shown that when a mitochondrion is damaged , PINK1 sends a signal to Parkin , which then helps to recruit the proteins that are needed to form an autophagosome around the damaged mitochondrion . However , the identity of the proteins that guide the formation of the autophagosome remained a mystery . Yamano et al . have now identified two of these proteins and helped to explain their specific roles in the assembly of autophagosomes . The two proteins , which are called TBC1D15 and TBC1D17 , are both GAP proteins , which are well known for their role in deactivating enzymes called RAB GTPases . Yamano et al . show that TBC1D15 binds to the damaged mitochondrion and also to the autophagosome as it grows around the mitochondrion . TBC1D15 also inhibits the action of an enzyme called Rab7 to prevent excessive growth of the autophagosome . TBC1D17 has a similar role . The work of Yamano et al . indicates that Parkin activates Rab7 , perhaps by placing chains of a protein called ubiquitin on mitochondria , which would mean that an unexpected new step in this pathway remains to be discovered . Understanding how Parkin activates Rab7 could help identify new targets for drugs that might treat Parkinson disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2014	Mitochondrial Rab GAPs govern autophagosome biogenesis during mitophagy
The climbing fiber input to Purkinje cells acts as a teaching signal by triggering a massive influx of dendritic calcium that marks the occurrence of instructive stimuli during cerebellar learning . Here , we challenge the view that these calcium spikes are all-or-none and only signal whether the instructive stimulus has occurred , without providing parametric information about its features . We imaged ensembles of Purkinje cell dendrites in awake mice and measured their calcium responses to periocular airpuffs that serve as instructive stimuli during cerebellar-dependent eyeblink conditioning . Information about airpuff duration and pressure was encoded probabilistically across repeated trials , and in two additional signals in single trials: the synchrony of calcium spikes in the Purkinje cell population , and the amplitude of the calcium spikes , which was modulated by a non-climbing fiber pathway . These results indicate that calcium-based teaching signals in Purkinje cells contain analog information that encodes the strength of instructive stimuli trial-by-trial . The climbing fiber ( CF ) input to Purkinje cells ( PCs ) plays a key role in theories of cerebellar learning ( Houk et al . , 1996; Ito , 2000 ) by providing a teaching signal that sounds the alarm when an unexpected sensory event is encountered ( Simpson et al . , 1996; De Zeeuw et al . , 1998; Ito , 2013 ) . Support for this hypothesis comes from studies of Pavlovian eyeblink conditioning ( Medina et al . , 2000 ) , a simple associative task in which subjects learn to blink to an initially neutral cue if it is repeatedly paired with a blink-eliciting instructive stimulus , such as a periocular airpuff . Previous work has shown that some CFs fire a burst of spikes when the unexpected periocular airpuff is delivered ( Sears and Steinmetz , 1991; Nicholson and Freeman , 2003 ) , and that this signal is sufficient for conditioning ( Mauk et al . , 1986; Thompson et al . , 1998 ) . All of these studies indicate that CFs are an important source of instructive signals to the cerebellum . However , we do not understand how CF signals encode even the most basic stimulus features such as the strength of the periocular airpuff , which has a major impact on the magnitude and the rate of learning ( Spence , 1953; Smith , 1968 ) . CFs have peculiar physiological properties that have inspired a number of hypotheses about the underlying neural code ( De Zeeuw et al . , 2011; Najafi and Medina , 2013 ) . CFs fire bursts spontaneously at ∼1 Hz ( Thach , 1968 ) , a low rate that is barely modulated during sensory stimulation ( Simpson et al . , 1996 ) . Indeed , CFs are often described as binary because they respond by either not firing at all or firing a single burst regardless of how strongly they are stimulated ( Crill , 1970; Gibson et al . , 2004 ) . Another peculiarity is that in the adult cerebellum , each PC is innervated by only one CF ( Simpson et al . , 1996 ) . The CF-PC synapse is one of the most powerful and reliable in the brain ( Schmolesky et al . , 2002; Ohtsuki et al . , 2009 ) : each time the CF fires a burst , it evokes in PCs both a burst of sodium spikes in the soma known as a complex spike ( Eccles et al . , 1966; Thach , 1968 ) , and a massive calcium-based spike in the dendrite ( Llinás and Sugimori , 1980 ) . Based on these observations , it has been suggested that analog information may be encoded probabilistically , in the total number of bursts generated by an individual CF across many repeated presentations of the same stimulus ( Fujita , 1982; Kenyon et al . , 1998 ) . Others have pointed out that sensory events synchronize CFs ( Llinás and Sasaki , 1989; Lou and Bloedel , 1992; Wylie et al . , 1995; Ozden et al . , 2009; Schultz et al . , 2009; Wise et al . , 2010 ) , which raises the possibility that stimulus information may be available in the precise timing of CF inputs ( Schweighofer et al . , 2004; Van Der Giessen et al . , 2008 ) , or in the level of co-activation in the CF population ( Ghosh et al . , 2011; Tokuda et al . , 2013 ) . Recent findings suggest an additional possibility: analog information , like the strength of an instructive stimulus , might be encoded post-synaptically by modulating the size of the PC response to individual sensory-driven CF bursts ( Maruta et al . , 2007; Najafi et al . , 2014; Yang and Lisberger , 2014 ) . We have imaged PC dendrites of awake mice to investigate how CF-triggered calcium events encode information about the strength of a periocular airpuff stimulus . Because each PC receives input from a single CF ( Simpson et al . , 1996 ) , we were able to discriminate the post-synaptic calcium events corresponding to individual pre-synaptic CF bursts and analyze their amplitude , timing and probability . Two-photon imaging also allowed us to analyze ensembles of PC dendrites whose CF inputs were co-activated by the periocular airpuff . Thus our experiments provide a unique opportunity to evaluate a variety of calcium-based codes in PCs , both at the individual dendrite and population levels . As in previous reports ( Sullivan et al . , 2005; Ozden et al . , 2008 ) , PC dendrites in our experiments appeared as parasagittally aligned , tube-like structures ( Figure 1A ) , in which large calcium transients ( Figure 1B , circles ) occurred spontaneously and in response to periocular airpuff stimuli ( Figure 1B , triangles ) . We have previously shown that these calcium transients are triggered in each individual PC dendrite by activation of its one-and-only climbing fiber ( CF ) input , which also evokes a complex spike in the PC somata ( Ozden et al . , 2008 ) . Hereafter , we will use the term ‘calcium event’ to refer to these CF-triggered calcium transients . 10 . 7554/eLife . 03663 . 003Figure 1 . Imaging climbing fiber-triggered calcium transients in Purkinje cell dendrites . ( A ) Field of view of an example experiment including 15 dendrites . ( B ) Example fluorescence traces of some of the dendrites in ( A ) . Triangles indicate periocular airpuff stimuli of different strengths . Circles mark CF-triggered calcium events . ( C ) Box plot showing frequency of spontaneous calcium events across all dendrites . ( D ) Mean ΔF/F trace of spontaneous calcium events for all dendrites ( gray lines; mean: black ) . ( E ) Pearson correlation coefficient of calcium events in pairs of dendrites as a function of the mediolateral separation ( black: real data; gray: shuffled-frame control data ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03663 . 003 A number of features confirmed the CF origin of the calcium events in our experiments . First , they occurred spontaneously at about 1 Hz ( 0 . 4–1 . 4 Hz; median 0 . 7 Hz; Figure 1C ) , which is similar to the characteristic spontaneous firing rate of CFs reported previously in awake animals ( Thach , 1968 ) . Second , they had a fast rise ( ∼10 ms Figure 1D ) , as observed for CF-triggered signals in other calcium imaging studies ( Miyakawa et al . , 1992; Eilers et al . , 1995; Schmidt et al . , 2003 ) , and a slower decay t1/2 of 74 ± 13 ms ( mean ± SD ) , which is in the midrange of previously observed kinetics using synthetic indicators ( decay t1/2 = 25–170 ms; [Sullivan et al . , 2005; Sarkisov and Wang , 2008; Ozden et al . , 2009; Kitamura and Häusser , 2011] ) . Third , the probability of observing two spontaneous calcium events at the same time was highest for adjacent PC dendrites , and decreased rapidly as the mediolateral distance between dendrites increased ( Figure 1E , black ) . This finding is consistent with previous studies demonstrating the prevalence of synchronous CF input to neighboring PCs in the same parasagittal microzone ( Bell and Kawasaki , 1972; Sasaki et al . , 1989; Ozden et al . , 2009; Schultz et al . , 2009 ) . We imaged 101 sites on the surface of cerebellar cortex in 22 mice , including paravermal locations in lobules V , VI , and more lateral locations in simplex ( Figure 2A , B ) . We were less frequently able to image the most medial parts of simplex due to the high density of blood vessels in that area . Consistent with the location of trigeminal CFs reported in previous studies ( Miles and Wiesendanger , 1975a , 1975b; Manni and Petrosini , 2004 ) , we found that many dendrites in the paravermal regions of lobules V and VI responded to periocular airpuff stimulation with a CF-triggered calcium event ( Figure 2C , D ) . In contrast , dendrites on the surface of lobule simplex were mostly unresponsive , which is expected given the deep location of periocularly-related CF zones in this lobule ( Hesslow , 1994; Mostofi et al . , 2010; Heiney et al . , 2014 ) . 10 . 7554/eLife . 03663 . 004Figure 2 . Location of imaged spots . ( A ) Dorsal view of an exposed cerebellum showing all imaged spots ( colored dots ) . Colors indicate the fraction of periocular-responsive dendrites in each spot . ( B ) Spots examined for response laterality are shown . Colors indicate the fraction of ipsi-specific dendrites in each spot . IC: Inferior Colliculus . IV–V , VI , VII , Simplex , and CrusI/II: cerebellar lobules . pf: primary fissure . Dashed line: midline . ( C ) Top: example ΔF/F traces from an ipsi-specific spot in response to ipsilateral ( black , left ) and contralateral ( gray , right ) airpuff stimuli . Bottom: each row shows PSTH of a dendrite in the example ipsi-specific spot in response to ipsi- and contra-lateral stimuli ( left , right , respectively ) . ( D ) Same as ( C ) , but an example bilateral spot is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 03663 . 004 As observed in electrophysiological work ( Hesslow , 1994; Mostofi et al . , 2010 ) , we found two types of periocular PC dendrites that could be classified according to the receptive field properties of their CF inputs: some dendrites responded with a CF-triggered calcium event only after ipsilateral periocular airpuffs ( Figure 2B , C ) , while others had bilateral CF receptive fields and responded after ipsi- and contralateral periocular airpuffs ( Figure 2B , D ) . In all the analyses presented below , we only evaluated the data for ipsilateral airpuffs . Our results were the same for both types of dendrites . Previous studies indicate that the likelihood of eliciting a response in an individual CF is proportional to stimulus intensity ( Eccles et al . , 1972b; Bosman et al . , 2010 ) . Thus , we first examined if individual CFs encode information about the strength of the periocular airpuff probabilistically , across repeated presentations of the stimulus . The raster plots of all the PC dendrites in our entire duration dataset demonstrate that CF-triggered calcium events occurred more reliably as the duration of the airpuff was increased ( Figure 3A bottom; top: average ) . On average across all the PC dendrites , we found that the probability of calcium events increased gradually with increasing airpuff duration ( Figure 3B , left; two-way ANOVA: F[496] = 238 . 5 , p < 0 . 0001; Tukey's HSD , p < 0 . 01 for all pairwise comparisons ) , and pressure ( Figure 3B , right; two-way ANOVA: F[238] = 58 . 93 , p < 0 . 0001; Tukey's HSD , p < 0 . 001 for all pairwise comparisons ) . 10 . 7554/eLife . 03663 . 005Figure 3 . Calcium-event probability encodes stimulus strength . ( A ) Bottom: raster plots represent all trials of the duration dataset . Dots indicate calcium events . White and gray shades mark different experiments . For each experiment , all the trials corresponding to an individual dendrite are arranged consecutively . Number of dendrites imaged in each experiment is indicated on the left . Top: PSTHs , corresponding to the raster plots , indicate calcium event frequency at each time point . ( B ) Calcium-event probability for the spontaneous ( sp ) and airpuff-evoked conditions ( d1–d4 , p1–p2 ) ( black: individual dendrites; red: mean ± SEM; left: duration data; right: pressure data ) . ( C ) Top to bottom: five dendrite categories based on how calcium-event probability varies with airpuff duration . For each category , calcium-event probability of individual dendrites ( left: gray lines; right: rows of heatmaps ) and their average ( left , black lines ) is shown . Triangles ( left ) : The threshold of airpuff duration for evoking calcium events . Colors ( right ) : calcium-event probability . d1–d4: different airpuff durations . p1–p2: different airpuff pressures . DOI: http://dx . doi . org/10 . 7554/eLife . 03663 . 005 The graded increase of calcium event probability with airpuff duration , which is shown averaged across all dendrites in Figure 3B , was evident in 58% of the individual dendrites ( Figure 3C , top row ) , in which the probability and the stimulus duration increased with the same rank order . The remaining 42% of the dendrites were relatively unresponsive to all airpuffs below a certain threshold ( Figure 3C , arrowheads ) , and responded with similar probability for all durations higher than the threshold ( Figure 3C , rows 2–5 ) . These latter CFs may encode duration information about stimulus strength in a binary manner , by providing a signal that tells the post-synaptic PC if the periocular airpuff lasted longer than a certain threshold . In summary , the response probability of individual CFs showed monotonic ( i . e . , non-decreasing ) dependence on stimulus duration , with a dependency that ranged from graded to threshold-like . In agreement with previous electrophysiological reports of sensory-driven CFs ( Ekerot et al . , 1987; Kobayashi et al . , 1998 ) , we found that calcium events were evoked in PC dendrites at a wide range of latencies relative to the onset of the airpuff stimulus ( onset latency: ∼25–150 ms; Figure 4A , D ) . Increasing the duration of the periocular airpuff resulted in progressively more long-latency calcium events ( Figure 4A–C left; 67 . 1 ± 1 . 5 , 71 . 8 ± 1 . 8 , 78 . 8 ± 1 . 4 , and 80 . 9 ± 1 . 4 ms for d1–d4 respectively; two-way ANOVA: F[393] = 11 . 85 , p < 0 . 0001; Tukey's HSD: p < 0 . 05 except for d1–d2 comparison and d3–d4 comparison ) . The temporal jitter , as quantified by median absolute deviation from median latency ( Figure 4C , right ) , was also reduced ( 23 . 3 ± 1 . 2 , 21 . 1 ± 1 . 0 , 19 . 3 ± 0 . 7 , and 18 . 0 ± 0 . 8 ms for d1–d4 respectively; two-way ANOVA: F[393] = 5 . 42 , p < 0 . 01; Tukey's HSD: p < 0 . 05 for d1–d3 , d1–d4 , d2–d4 comparisons ) . The analysis shown in Figure 4B confirmed that long-duration airpuffs evoked significantly more calcium events than short-duration airpuffs in the interval 75–150 ms after the periocular stimulation ( Figure 4B; Kolmogorov–Smirnov test , p < 0 . 0001 ) but not in the interval 0–75 ms ( Figure 4B; Kolmogorov–Smirnov test , p = 0 . 8 ) . In summary , responses to airpuffs of increased duration are characterized by a higher likelihood of calcium events as the stimulus continues over time . 10 . 7554/eLife . 03663 . 006Figure 4 . Calcium-event latency is modulated by stimulus strength . ( A ) Onset-latency distributions of calcium events for duration data . ( B ) Each panel corresponds to a particular latency interval ( indicated in the title ) and compares for each dendrite ( dots ) the number of events evoked by two different durations of airpuff ( y-axis: longer duration; x-axis: shorter duration; blue dot: mean; dashed: unity line ) . ( C ) Median onset latency ( left ) and variability ( median of absolute deviation from median ) of onset latency ( right ) . Circles: average across dendrites; Error bars: SEM . ( D–F ) Same as ( A–C ) , but for pressure data . DOI: http://dx . doi . org/10 . 7554/eLife . 03663 . 006 To test whether CF timing can provide a code for stimulus strength under conditions of constant stimulus duration , we varied the pressure of the airpuff . This led to a modest reduction in onset latency ( 81 . 3 ± 2 . 4 vs 79 . 6 ± 3 . 4 ms for p1 and p2 respectively ) that was not significant ( Figure 4D–F left; two-way ANOVA: F[134] = 0 . 98 , p = 0 . 3 ) , and did not alter temporal jitter ( Figure 4F right; two-way ANOVA: F[134] = 0 . 13 , p = 0 . 7 ) . Higher-pressure airpuffs evoked more calcium events throughout the analysis window , both in the 0–75 ms interval ( Figure 4E; Kolmogorov–Smirnov test , p < 0 . 01 ) , and in the 75–150 ms interval ( Figure 4E; Kolmogorov–Smirnov test , p < 0 . 001 ) . These findings indicate that the onset latency of an individual CF input cannot be used to encode the pressure of the airpuff . However , additional information about stimulus strength may be available in the timing of the CF population ( population co-activation ) . This question is addressed next . Previous work has demonstrated that groups of CFs converging on the same zone of cerebellar cortex become synchronized in response to sensory stimulation ( Lou and Bloedel , 1992; Ozden et al . , 2009; Schultz et al . , 2009; Wise et al . , 2010; Ghosh et al . , 2011 ) . We examined if the level of co-activation in the CF population provides information about the strength of the periocular airpuff stimulus . In 16 experiments in which we were able to image at least six PC dendrites simultaneously ( Figure 5 ) , we found that the number of synchronized calcium events in a 150 ms window after stimulus onset increased in response to airpuffs of longer durations ( Figure 5A , B , top: two-way ANOVA , F[4394] = 199 . 75 , p < 0 . 0001; Tukey's HSD , p < 0 . 05 for all pairwise comparisons ) , and higher pressures ( Figure 5B , bottom: two-way ANOVA , F[2209] = 106 . 56 , p < 0 . 0001; Tukey's HSD , p < 0 . 0001 for all pairwise comparisons ) . 10 . 7554/eLife . 03663 . 007Figure 5 . Population coding of stimulus strength . ( A ) Bottom: % coactive dendrites at different time points for all trials of the duration data . Colors indicate % coactivation . Top: PSTHs correspond to heatmaps at the bottom and show the average coactivation across all trials at each time point surrounding the stimulus . ( B ) Cumulative distribution of % coactive dendrites across all trials for the spontaneous ( sp ) and airpuff-evoked conditions ( top: duration data; bottom: pressure data ) . ( C ) Measured and independent joint probabilities are shown for each dendrite pair ( gray; dashed: unity line ) for different airpuff durations . ( D ) % extra synchrony ( measured minus independent joint probability ) averaged across all dendrite pairs ( error bars: SEM; sp: spontaneous; d1–d4: different airpuff durations . p1–p2: different airpuff pressures ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03663 . 007 Since stronger airpuffs increase the probability of calcium events in individual PC dendrites ( Figure 3 ) , it is possible that the gradual increase in the number of synchronized calcium events ( Figure 5A , B ) simply reflects an increase in probability in a group of otherwise independent dendrites . To assess this possibility , we measured the joint probability for every pair of dendrites in each one of our experiments , that is the probability of observing a CF-triggered calcium event in both dendrites within the same 150 ms time window , and compared it to the joint probability expected for independent dendrites ( Pij = Pi × Pj , where Pi and Pj represent the calcium-event probability of dendrites i and j in that same time window ) . We found that joint probability deviated significantly from the independence assumption for all airpuff strengths ( Figure 5C; two sample t test , p < 0 . 001 ) . We call this deviation ‘extra synchrony’ because it represents the synchrony beyond that expected solely from the calcium-event probability of two independent dendrites . Figure 5D shows that there was a gradual boost in the amount of extra synchrony as the airpuff duration or pressure was increased ( Duration data: two-way ANOVA: F[4361] = 113 . 49 , p < 0 . 0001; Tukey's HSD: p < 0 . 0001 , except for d3–d4 comparison . Pressure data: two-way ANOVA: F[2136] = 10 . 27 , p < 0 . 0001; Tukey's HSD: p < 0 . 05 , except for p1–p2 comparison ) . In other words , the degree to which CFs are dependent on each other scales up smoothly with their overall response probability . These results suggest that CF co-activation may be controlled upstream , perhaps in the inferior olive , in a way that provides information about stimulus strength at the level of the PC population . We have recently shown that the amplitude of CF-triggered calcium events is enhanced during sensory stimulation ( Najafi et al . , 2014 ) . Here , we examined if the magnitude of this sensory-driven enhancement is graded according to periocular airpuff strength . Compared to the average fluorescence traces for spontaneous calcium events ( Figure 6A , ‘sp’ ) , the average fluorescence traces for sensory-driven calcium events revealed a gradual enhancement as the duration or the pressure of the airpuff was increased ( Figure 6A top; only duration data is displayed ) . Note that the fluorescence trace of each individual dendrite was normalized to the peak value of its mean spontaneous calcium event ( ‘Materials and methods’ ) . However , calcium events occurred with variable latency after the periocular airpuff , and for this reason the peaks of the average fluorescence traces in Figure 6A are substantially lower than ‘1’ ( for comparison purposes , the ‘sp’ trace is plotted with the same temporal jitter as the d1–d4 traces ) . 10 . 7554/eLife . 03663 . 008Figure 6 . Stimulus strength is represented in the size of calcium events and size of non-CF signals . ( A ) Top: mean ΔF/F trace of calcium events across all dendrites for the spontaneous ( ‘sp’ , green ) and airpuff-evoked conditions ( d1–d4: different airpuff durations; shades: SEM ) . Bottom: mean size of calcium events ( ‘ΔF/F-integral’ ) shown for each dendrite ( black dots ) . Left: duration data: Right: pressure data . ΔF/F-integral values are normalized to the mean size of spontaneous events ( dashed line ) . Red: mean ± SEM . ( B ) Same as ( A ) , but for the non-CF signal . ( C ) Each panel corresponds to a duration of airpuff , and compares ΔF/F traces of calcium-event enhancement ( i . e . , evoked minus spontaneous event; black ) and non-CF signal ( green ) in response to that particular airpuff duration ( lines: mean across dendrites; shades: SEM ) . ( D ) Mean size of calcium-event enhancement is compared with mean size of non-CF signal for different airpuff durations ( d1–d4; circles: average across dendrites; bars: SEM; dashed: unity line; ΔF/F-integral values are normalized to the mean size of spontaneous events . Arrowhead marks the longest duration airpuff , for which supralinearity is evident ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03663 . 008 To quantify the gradual enhancement in Figure 6A ( top ) , we measured the size of each individual calcium event by computing the integral of its fluorescence trace over a 100 ms time window after the peak ( ‘ΔF/F-integral’ ) , and normalizing this value to the average ΔF/F-integral of all the spontaneous calcium events of the corresponding PC dendrite . Note that we only examined the fluorescence traces of individual calcium events and excluded trials in which the periocular stimulation resulted in two or more calcium events separated from each other by less than 100 ms ( this occurred in <2% of trials and did not affect our results ) . This analysis confirmed that periocular airpuffs of longer duration and higher pressure evoked progressively larger calcium events ( Figure 6A , bottom; Duration data: two-way ANOVA: F[496] = 80 . 74 , p < 0 . 0001; Tukey's HSD: p < 0 . 05 except for d2–d3 comparison . Pressure data: two-way ANOVA: F[238] = 62 . 88 , p < 0 . 0001; Tukey's HSD: p < 0 . 01 for all pairwise comparisons ) . Thus , the calcium elevation evoked in PC dendrite after activation of its CF input provides information about the strength of peripheral stimulation . What neural mechanisms may contribute to the gradual enhancement of the fluorescence traces in Figure 6A ? In addition to triggering calcium events in the PC dendrite , sensory stimulation also elicits a smaller calcium response that has a non-CF origin ( ‘non-CF signal’; [Najafi et al . , 2014] ) . We found that the rise time ( 73 ± 8 , 114 ± 7 , 130 ± 7 , 148 ± 7 ms for d1–d4 respectively; two-way ANOVA: F[375] = 12 , p < 0 . 0001; Tukey's HSD: p < 0 . 05 except for d2–d3 comparison ) and the size of this non-CF signal were graded ( Figure 6B , top ) , with ΔF/F-integral becoming progressively larger as the stimulus duration or pressure was increased ( Figure 6B bottom; Duration data: two-way ANOVA: F[496] = 122 . 61 , p < 0 . 0001; Tukey's HSD: p < 0 . 0001 , except for d3–d4 comparison . Pressure data: two-way ANOVA: F[238] = 71 . 85 , p < 0 . 0001; Tukey's HSD: p < 0 . 0001 for all pairwise comparisons ) . These properties are consistent with the calcium responses driven by activation of parallel fiber inputs to PCs ( Finch and Augustine , 1998; Takechi et al . , 1998 ) which are known to increase progressively with stimulus duration ( Gandolfi et al . , 2014 ) . For the duration dataset , we had enough trials to make a direct comparison between the size of the non-CF signal and the size of the enhancement of the calcium event , which was obtained by taking the difference between the average fluorescence traces for spontaneous and stimulus-evoked calcium events . For short-duration airpuffs ( Figure 6C , d1 and d2 ) , the average non-CF signal resembled the average enhancement trace . However , for the longest-duration airpuffs ( Figure 6C , d4 ) , the non-CF signal was significantly smaller than the enhancement trace . These results are consistent with a model in which calcium events evoked by relatively weak sensory stimulation comprise a constant CF-triggered signal that adds linearly with a non-CF signal graded according to the strength of the stimulus ( Figure 6D , d1 , d2 , d3 ) . The supralinear response evoked by very strong airpuffs ( Figure 6D , d4 , arrowhead ) could be explained if these stimuli were effective in triggering more spikes in each individual CF burst ( Maruta et al . , 2007 ) . The response of a PC to sensory-driven activation of its CF input was originally described as an ‘all-or-nothing’ event ( Eccles et al . , 1966 ) . Such a binary response cannot provide analog information about the graded features of an instructive stimulus in a single trial . However , current models of cerebellar function have pointed out that an individual PC could still extract analog information from its CF input by reading out a probabilistic code: by taking into account the reliability ( Fujita , 1982; Kenyon et al . , 1998 ) , or the temporal precision ( Schweighofer et al . , 2004; Van Der Giessen et al . , 2008; Tokuda et al . , 2013 ) , with which the all-or-nothing CF input is activated across many repeated presentations of the instructive stimulus . We found that many individual PCs represent information about airpuff duration and pressure in a manner that is consistent with the probabilistic coding hypothesis . Higher-pressure ( or longer-lasting ) airpuffs evoked CF-triggered calcium events in PC dendrites in a larger fraction of trials than weaker ( or shorter ) airpuffs . This type of probabilistic coding may be particularly useful for regulating the efficacy of instructive signals during cerebellar tasks that are learned over many training trials . Every time the CF input fires , it triggers a wide range of synaptic changes in the cerebellar cortex ( Schmolesky et al . , 2002; Ohtsuki et al . , 2009; Gao et al . , 2012 ) . Thus , the total amount of plasticity induced in the cerebellar cortex during training is likely to depend on the reliability with which CFs are activated by the repeated presentations of the instructive stimulus . Our observation that the latency of CF-triggered events becomes progressively longer and less variable as a function of stimulus duration may also have implications for cerebellar learning . As is the case for neurons other brain regions ( Dan and Poo , 2004 ) , plasticity in PCs is strongly influenced by the relative timing of synaptic inputs . For example , parallel fiber ( PF ) synapses activated just before the CF input get weaker , whereas those activated just after the CF input get stronger ( Piochon et al . , 2012 ) . Therefore , we predict that the total amount of plasticity induced in a PC over the course of multiple training trials will vary as a function of the number of times the instructive stimulus is able to activate the CF within the same small window of time . We have shown that the amplitude of sensory-driven calcium events in a PC dendrite is graded and provides analog information about stimulus strength in single trials , that is more dendritic calcium if CF input fires in response to a strong periocular airpuff , less calcium for weak airpuffs . There are a number of non-mutually exclusive mechanisms that could contribute to this graded regulation of calcium events . Our results provide considerable support for the possibility that the amplitude of calcium events in PC dendrites may be modulated by sensory-driven activation of a non-CF input . In support of this hypothesis we found that periocular stimulation activated CF and non-CF inputs converging on the same PC dendrite , and that activation of the non-CF input by itself was sufficient to cause a small dendritic calcium response that was graded according to stimulus strength . Although the source of the non-CF signal is unknown , we note that the excitatory parallel fiber ( PF ) input satisfies two conditions necessary to play such a role: ( 1 ) PF and CF inputs with similar receptive fields often converge on the same PC ( Eccles et al . , 1972a; Eccles , 1973 ) , and ( 2 ) stimulation of PFs generates graded calcium responses in PC dendrites ( Eilers et al . , 1995; Gandolfi et al . , 2014 ) , that have similar kinetics to the non-CF signal ( Finch and Augustine , 1998; Takechi et al . , 1998; Wang et al . , 2000 ) and can boost the amplitude of CF-triggered dendritic calcium events ( Wang et al . , 2000 ) . For the longest duration airpuffs , the amplitude of the non-CF signal was not large enough to fully account for the sensory-driven enhacement of the calcium events . This finding suggests that in addition to the non-CF signal , other mechanisms may also contribute to the modulation of calcium signals in PC dendrites during sensory stimulation . Recent studies have shown that the number of spikes in the presynaptic CF burst varies systematically depending on experimental conditions both in vitro ( Mathy et al . , 2009 ) and in vivo ( Maruta et al . , 2007; Bazzigaluppi et al . , 2012 ) , and that having just one extra spike in the CF burst can cause a substantial enhancement of the postsynaptic calcium response ( Mathy et al . , 2009 ) . These observations raise the possibility that the graded modulation of calcium events in our experiments could be driven in part by small increases in the number of spikes of the CF burst ( Najafi and Medina , 2013; Yang and Lisberger , 2014 ) , especially for the longest stimulus durations . The discovery of graded calcium signals in PC dendrites has important implications for theories of cerebellar learning . Dendritic calcium is responsible for triggering a variety of mechanisms of cellular plasticity that cause short- and long-term modifications in the strength of PC synapses ( Gao et al . , 2012 ) . These plasticity mechanisms are tightly regulated and can be differentially engaged depending on the precise amplitude and duration of the calcium signal in the PC dendrite . For example , small differences in dendritic calcium can influence how much PF synapses will change in vitro ( Coesmans et al . , 2004; Tanaka et al . , 2007 ) , and can also determine the direction of learning-related changes in the firing of PCs recorded in vivo during eyeblink conditioning ( Rasmussen et al . , 2013 ) . Thus , we predict that the amount and direction of plasticity induced in a PC during cerebellar learning will vary trial-by-trial depending on the dendritic calcium response triggered by each presentation of the instructive stimulus . This type of trial-by-trial regulation of CF-related instructive signals has been recently demonstrated in monkeys learning a cerebellar-dependent eye movement task ( Yang and Lisberger , 2014 ) . Previous studies have reported that CFs converging on the same parasagittal strip of cerebellar cortex fire synchronously to signal the occurrence of an unexpected stimulus ( Llinás and Sasaki , 1989; Lou and Bloedel , 1992; Ozden et al . , 2009; Schultz et al . , 2009; Wise et al . , 2010 ) . Our results confirm and extend these findings in two ways: First , we demonstrated that the level of synchronization is regulated by the strength of the stimulus , that is a larger fraction of the local CF population was activated after strong periocular airpuffs than after weak airpuffs . Second , we found that stimulus-related differences in the level of synchronization are partly driven by a mechanism that boosts synchrony beyond what is expected for independent CFs . We do not know the source of this extra synchrony or the mechanisms that modulate it during sensory stimulation . However , we note that electrical coupling of cells in the inferior olive ( IO ) plays an important role in synchronizing CF activity ( De Zeeuw et al . , 1997; Blenkinsop and Lang , 2006; Van Der Giessen et al . , 2008 ) , and that the coupling coefficient can be dynamically modulated up and down via stimulus-related activation of synaptic inputs to the IO ( Llinás and Sasaki , 1989; Lang , 2002; Lefler et al . , 2014; De Gruijl et al . , 2014b ) . Regulation of CF synchrony could be used to control the efficacy of plasticity signals ( De Gruijl et al . , 2012; Tokuda et al . , 2013 ) . For example , sensory events that activate many CFs simultaneously may be particularly effective for triggering NMDA-dependent plasticity in molecular layer interneurons of the cerebellar cortex ( Duguid and Smart , 2004 ) , via synchronized spillover of glutamate from multiple CF release sites ( Szapiro and Barbour , 2007; Mathews et al . , 2012 ) . Another possibility is that CF synchrony could be used to set the strength of the plasticity signals sent by inhibitory PCs to downstream cells of the deep cerebellar nuclei ( DCN; Otis et al . , 2012 ) , which are the final output of the cerebellum . This hypothesis is consistent with recent work showing that DCN cells exhibit rebound firing immediately after being released from the hyperpolarization caused by CF-driven synchronization of the PC population ( Hoebeek et al . , 2010; Bengtsson et al . , 2011 ) , and that this rebound can serve as a trigger for plasticity ( Pugh and Raman , 2006 ) . Release from hyperpolarization may also be sufficient to trigger calcium-based excitable events in dendrites of the DCN ( Schneider et al . , 2013 ) , an effect that may be enhanced by direct monosynaptic excitation of DCN neurons by CF collaterals firing at the same time ( Llinás and Muhlethaler , 1988 ) . In addition to playing a role in the modulation of plasticity signals in the cerebellum , sensory-driven regulation of CF synchrony could also have a more direct and immediate effect on motor control ( Lang et al . , 1999; Kitazawa and Wolpert , 2005; Llinás , 2011 ) . Increased CF synchrony is observed at the onset of movement ( Welsh et al . , 1995; Ghosh et al . , 2011; Ozden et al . , 2012 ) , and the level of synchrony has been shown to correlate with the timing and the velocity of spinocerebellar reflexes after sensory perturbations ( De Gruijl et al . , 2014a ) . In our experiments , a higher level of CF synchrony after a strong periocular airpuff might serve to enhance the reflex response by generating a faster or longer duration eyeblink movement ( Evinger et al . , 1991 ) . CFs are thought to play the role of teachers ( Simpson et al . , 1996; De Zeeuw et al . , 1998; Ito , 2013 ) , providing the instructive signals that trigger the mechanisms of plasticity necessary for cerebellar learning ( Gao et al . , 2012 ) . To be useful , CFs must do more than alert about the occurrence of an unexpected event such as the presentation of a periocular airpuff in the early stages of eyeblink conditioning ( Sears and Steinmetz , 1991; Nicholson and Freeman , 2003 ) ; they must also provide analog information to indicate how unexpected the event was , that is , how much stronger or weaker the airpuff was compared with expectations . Our experiments demonstrate that CFs and non-CF pathways together encode information about stimulus strength in PC dendrites when the presentation of the stimulus cannot be predicted . It will be important to understand whether and how the different calcium-based codes we have uncovered may be modulated during cerebellar learning , as the stimulus becomes predictable . Experimental procedures were approved by the Princeton University Institutional Animal Care and Use Committee and performed in accordance with the animal welfare guidelines of the National Institutes of Health . The details of the animal preparation have been described previously ( Najafi et al . , 2014 ) . Briefly , C57BL/6J mice ( female , 8–14 weeks ) were deeply anesthetized by inhalation of isoflurane ( 3–5% induction; 0 . 5–1 . 5% maintenance ) . A small area of the cerebellum was exposed ( diameter: 3 mm ) , and a Kwik-Sil plug , pre-molded on a coverslip , was secured over the dura using a two-piece , stainless steel headplate . Throughout the surgery , sterile saline was used to keep the dura wet . Animals’ body temperature was monitored and maintained near 37°C . Analgesics ( Meloxicam , 5 mg/kg , subcutaneous ) were injected . At the end of the surgery , anesthesia was removed and mice were returned to their cage for recovery . Two-photon calcium imaging was performed the day after surgery . Mice were anesthetized with isoflurane ( 3–5% induction; 0 . 5–1 . 5% maintenance ) . Kwik-Sil plug was removed and calcium indicator ( Oregon Green 488 BAPTA-1/AM , Invitrogen , Carlsbad , CA ) was injected 150–200 µm below dura , by applying brief positive pressure through a glass pipette ( Ozden et al . , 2012 ) . A new Kwik-Sil plug was used , anesthesia was removed , and the animal was transferred and mounted on a cylindrical treadmill integrated with the imaging apparatus . Calcium imaging was performed on awake animals using a custom-built two-photon laser scanning microscope ( Sullivan et al . , 2005 ) . Fluorescence movies ( 32 × 128-pixel; 64 ms/frame ) were recorded using ScanImage software ( Pologruto et al . , 2003 ) . Animals were monitored throughout the experiment with a camera . PC dendrites were identified from the imaging movies using independent component analysis ( Ozden et al . , 2012 ) . The fluorescence trace ( ΔF/F ) of each dendrite was computed , frame-by-frame , as ( F-Fb ) /Fb , where F is the mean fluorescence intensity of the pixels contributing to a dendrite . Fb is the baseline defined as the lowest eighth percentile of the fluorescence values within a 1-s window . Calcium transients were identified as CF-evoked calcium events using a two-step process ( Ozden et al . , 2012 ) : first , the kinetic properties of the ΔF/F signal had to match a template . Second , the peak amplitude of the transient had to be larger than a predefined threshold . A pressure injector system ( Toohey Spritzer ) connected to a 25 gage needle was used to deliver airpuffs to the animal's ipsilateral eye ( inter-trial interval: 4 s; 35 trials per airpuff condition per experiment ) . Two different sets of experiments were performed . In one set , ‘duration’ experiments , the airpuff pressure was kept the same ( 30 psi ) and four different durations of airpuffs were applied ( 8 , 15 , 30 , 45 ms ) . In another set , ‘pressure’ experiments , the airpuff duration was constant ( 30 ms ) , and two different airpuff pressures ( 10 , 50 psi ) were applied . The ΔF/F traces presented in all figures were normalized . Normalization was done for each dendrite separately , by dividing the ΔF/F trace of the dendrite by the peak value of its mean spontaneous calcium event . Calcium events with a peak in the interval 50–200 ms after the periocular stimulus were considered airpuff-evoked . The onset latency of calcium events ( Figure 4 ) was computed manually by inspecting each airpuff-evoked calcium event . Coactivation ( Figure 5 ) was computed for each trial separately by dividing the number of dendrites with an airpuff-evoked calcium event , by the total number of responsive dendrites in the field of view . The measured joint probability for a pair of dendrites ( Figure 5C ) was calculated as the fraction of trials in which both dendrites in the pair exhibited airpuff-evoked calcium events . The independent joint probability was computed by multiplying the calcium-event probability of the two dendrites ( Pij = Pi × Pj ) . The d1–d4 traces in Figure 6A were computed by averaging the normalized ΔF/F signals of all the individual dendrites in trials with a calcium event . To provide a fair comparison , the ‘sp’ trace was computed the same way , after substituting the ΔF/F signal of all the stimulus-evoked calcium events in a given dendrite with the normalized ΔF/F signal corresponding to the average spontaneous calcium event for that dendrite . Thus , the temporal jitter in the onset latency of the individual calcium events comprising the ‘sp’ trace was the same as in the d1–d4 conditions . The size of individual calcium events ( ΔF/F-integral , Figure 6 ) was computed by taking the integral of the normalized ΔF/F signal over the interval [t t + 100 ms] , where t is the time point at which the peak of the event occurs . The non-CF signal ( Figure 6B , C ) was analyzed in trials without any calcium events within 50–200 ms of the airpuff stimulus . The size of the non-CF signal was measured in a similar way as the calcium events , by taking the integral of the normalized ΔF/F signal over the interval [t t + 100 ms] , where t is a time point selected randomly within 50–200 ms of the stimulus . Error bars in figures indicate SEM . Values in text are mean ± SEM unless otherwise specified . Tests of significance were two-tailed .	A region of the brain known as the cerebellum plays a key role in learning how to anticipate an event . For example , if you know that a puff of air is going to be directed at your eye , it's a good idea to close it in advance . However , how much you need to close it depends on how strong that puff of air is . A very strong puff might require closing the eye completely to protect it . In contrast , it is probably better to only partially close the eye if you know a lighter puff of air is coming , so that you can still see . Extensive research has focused on how neurons in and around the cerebellum work together to achieve this goal . When an event—such as a puff of air—occurs , signals are sent to large neurons in the cerebellum , called Purkinje cells , by ‘climbing fibers’ . However , climbing fibers were thought to be able to respond in only two ways: either they fire in a single burst to signal that an event has occurred , or they don't fire . It was therefore unclear how the finer details of the event ( for example , the strength of the puff of air ) are transmitted to the cerebellum . Najafi et al . imaged the level of calcium in the cerebellum of mice , as this indicates how active the neurons are . When a puff of air was directed at the eyes of the mice , Najafi et al . saw that the size of the response of the Purkinje cells corresponded with how big the puff of air was . Najafi et al . show that the size of this response , which is based mostly on input from the climbing fibers , is also influenced by input from an additional unknown source . These findings show that Purkinje cells of the cerebellum receive detailed information about the nature of an event , such as a puff of air . What remains to be seen is whether the cerebellum uses this information to learn the correct response , that is how hard to blink to avoid the expected puff .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2014	Coding of stimulus strength via analog calcium signals in Purkinje cell dendrites of awake mice
The unfolded protein response ( UPR ) is a cellular homeostatic circuit regulating protein synthesis and processing in the ER by three ER-to-nucleus signaling pathways . One pathway is triggered by the inositol-requiring enzyme 1 ( IRE1 ) , which splices the X-box binding protein 1 ( Xbp1 ) mRNA , thereby enabling expression of XBP1s . Another UPR pathway activates the activating transcription factor 6 ( ATF6 ) . Here we show that murine cytomegalovirus ( MCMV ) , a prototypic β-herpesvirus , harnesses the UPR to regulate its own life cycle . MCMV activates the IRE1-XBP1 pathway early post infection to relieve repression by XBP1u , the product of the unspliced Xbp1 mRNA . XBP1u inhibits viral gene expression and replication by blocking the activation of the viral major immediate-early promoter by XBP1s and ATF6 . These findings reveal a redundant function of XBP1s and ATF6 as activators of the viral life cycle , and an unexpected role of XBP1u as a potent repressor of both XBP1s and ATF6-mediated activation . The endoplasmic reticulum ( ER ) is responsible for synthesis , posttranslational modification , and folding of a substantial portion of cellular proteins . When protein synthesis is increased or ER function is compromised , the folding capacity of the ER may get out of balance , leading to an accumulation of unfolded or misfolded proteins in the ER . To alleviate ER stress and restore homeostasis , the cell activates three ER-to-nucleus signaling pathways , collectively called the unfolded protein response ( UPR ) , which lead to a reduced protein synthesis and an increased expression of folding chaperones and ER-associated degradation ( ERAD ) factors ( Walter and Ron , 2011 ) . Subsequently , ER folding capacity increases and terminally misfolded protein species are exported from the ER and targeted for proteasomal degradation ( Christianson and Ye , 2014 ) . In mammalian cells , the UPR comprises three main signaling pathways named after the initiating ER stress sensors: PERK ( PKR-like ER kinase ) , ATF6 ( activating transcription factor 6 ) , and IRE1 ( inositol-requiring enzyme 1 ) ( Walter and Ron , 2011 ) . Upon activation by ER stress , PERK phosphorylates the translation initiation factor eIF2α , which leads to a massive attenuation of protein synthesis and an immediate reduction of the protein load in the secretory system . However , phosphorylated eIF2α selectively supports the translation of selected cellular proteins such as the transcription factor ATF4 , which activates a negative feedback loop resulting in dephosphorylation of eIF2α ( Novoa et al . , 2001 ) . Upon activation by ER stress , ATF6 travels to the Golgi , where it undergoes intramembrane proteolysis . This process liberates its cytosolic N-terminus , the basic leucine zipper ( bZIP ) transcription factor ATF6 ( N ) , and enables it to travel to the nucleus , where it activates the transcription of chaperone genes as well as of the gene encoding XBP1 ( Lee et al . , 2002 ) . The third sensor , IRE1 ( also known as ER-to-nucleus signaling 1 , ERN1 ) , is an ER transmembrane protein kinase that oligomerizes upon accumulation of unfolded proteins in the ER lumen . Oligomerization and auto-transphosphorylation activate the RNase function of IRE1 , which mediates an unconventional splicing of the XBP1 mRNA in the cytosol ( Calfon et al . , 2002; Lee et al . , 2002; Yoshida et al . , 2001 ) . Removal of the 26-nt intron from the XBP1 mRNA leads to a frame shift and expression of transcription factor XBP1s , comprising an N-terminal basic leucine zipper ( bZIP ) domain followed by a C-terminal transcription activation domain . By contrast , the unspliced XBP1 mRNA encodes XBP1u , which lacks the transcription activation domain but contains in its C-terminal part a hydrophobic patch and a translational pausing region required for the recruitment of its own mRNA to the ER membrane ( Kanda et al . , 2016; Yanagitani et al . , 2011 ) . XBP1u is rapidly degraded and has a short half-life ( Tirosh et al . , 2006 ) . It can interact with XBP1s and ATF6 ( N ) and target them for proteasomal degradation . Therefore , XBP1u is thought to act as a negative regulator involved in fine-tuning the UPR ( Tirosh et al . , 2006; Yoshida et al . , 2006; Yoshida et al . , 2009 ) . Moreover , XBP1u affects autophagy by interacting with transcription factor FOXO1 ( Zhao et al . , 2013 ) . Apart from mediating XBP1 mRNA splicing , IRE1 can also cleave ER-associated mRNA molecules that contain a specific recognition motif ( Moore and Hollien , 2015 ) . This process , which leads to mRNA degradation , is called regulated IRE1-dependent mRNA decay ( RIDD ) . However , the importance of RIDD in different cellular processes such as lipid metabolism , antigen presentation , and apoptosis remains incompletely understood ( Maurel et al . , 2014 ) . During viral replication large quantities of viral proteins must be synthesized . Folding , maturation , and posttranslational modification of secreted and transmembrane proteins take place in the ER and require a plethora of chaperones , foldases , and glycosylating enzymes . While properly folded proteins are transported to the Golgi , unfolded or misfolded proteins are retained in the ER and exported to the cytosol for proteasomal degradation via the ERAD pathway ( Smith et al . , 2011 ) . However , the high levels of viral envelope glycoproteins that are being synthesized particularly during the late phase of the viral life cycle can overwhelm the folding and processing capacity of the ER and cause accumulation of unfolded and misfolded proteins in the ER ( Zhang and Wang , 2012 ) . Cytomegaloviruses ( CMVs ) are prototypic members of the β subfamily of the Herpesviridae . Their large double-stranded DNA genomes contain at least 165 protein-coding ORFs ( Dolan et al . , 2004 ) end encode an even larger number of polypeptides ( Stern-Ginossar et al . , 2012 ) . Through millions of years of co-evolution with their respective hosts , the CMVs have acquired the ability to moderate immune recognition and modulate cellular stress responses to their own benefit ( Alwine , 2008; Mocarski , 2002 ) . Considering the important role of the UPR in controlling cell fitness , it is hardly surprising that the CMVs have evolved means to modify the UPR . For instance , human and murine CMV ( HCMV and MCMV ) induce PERK activation , but limit eIF2α phosphorylation . By doing this the CMVs prevent a global protein synthesis shutoff but allow eIF2α phosphorylation-dependent activation of transcription factor ATF4 ( Isler et al . , 2005; Qian et al . , 2012 ) . The CMVs also increase expression of the ER chaperone BiP to facilitate protein folding and virion assembly ( Buchkovich et al . , 2008; Buchkovich et al . , 2010; Qian et al . , 2012 ) , and HCMV uses PERK to induce lipogenesis by activating the cleavage of sterol regulatory element binding protein 1 ( Yu et al . , 2013 ) . We have previously shown that both , MCMV and HCMV , downregulate IRE1 levels and inhibit IRE1 signaling at late times post infection . This downregulation is mediated by the viral proteins M50 and UL50 , respectively ( Stahl et al . , 2013 ) . However , a real-time transcriptional profiling study has revealed that cellular ER stress response transcripts are upregulated as early 5–6 hr after MCMV infection ( Marcinowski et al . , 2012 ) . Here , we show that MCMV transiently activates the IRE1-XBP1 pathway at early times post infection to relieve repression of viral gene expression and replication by XBP1u . When IRE1-mediated XBP1 mRNA splicing is inhibited , XBP1u blocks the activation of the viral major immediate-early promoter ( MIEP ) by XBP1s and ATF6 ( N ) . Thus , MCMV exploits UPR signaling to boost the activity of its most important promoter . Moreover , these findings reveal a redundant function of XBP1s and ATF6 as activators of viral gene expression and replication , and an unexpected role of XBP1u as a potent repressor of both XBP1s and ATF6-mediated activation . Previous studies have shown that MCMV inhibits IRE1-XBP1 signaling at late times ( ≥24 hr ) post infection ( Qian et al . , 2012; Stahl et al . , 2013 ) . However , cellular ER stress response transcripts were shown to be upregulated at 5–6 hr after MCMV infection ( Marcinowski et al . , 2012 ) , suggesting that UPR signaling is activated at early times post infection . Thus , we decided to analyze whether MCMV activates the IRE1-XPB1 signaling pathway within the first few hours after infection . To do this , we infected mouse embryonic fibroblasts ( MEFs ) with MCMV and quantified spliced and unspliced XBP1 transcripts by qRT-PCR . We detected a short and transient increase of XBP1 splicing between 5 and 7 hr post infection ( hpi ) ( Figure 1A ) . This increase was massively reduced when cells were infected with UV-inactivated MCMV ( Figure 1A ) , suggesting that XBP1 splicing was not caused by viral attachment and entry into cells but required viral gene expression . The transient activation of the IRE1-XBP1 pathway was confirmed by immunoblot detection of phosphorylated IRE1 and XBP1s ( Figure 1B ) . MCMV-induced XBP1 splicing was suppressed by cycloheximide ( CHX , a translation inhibitor ) , but not by phosphonoacetic acid ( PAA ) , an inhibitor of viral DNA replication and late gene expression ( Figure 1C ) . These results suggested that the transient activation of the IRE1-XBP1 pathway is caused by viral proteins expressed at immediate-early or early times post infection . To determine whether IRE1 signaling is important for the MCMV life cycle , we used IRE1-deficient ( Ern1-/- ) cells expressing IRE1-GFP under tight control of a tetracycline-inducible promoter ( TetON-IRE1-GFP cells , Figure 2A ) for analyses of viral replication . IRE1-GFP expression was induced with different concentrations of doxycycline , and cells were infected at low or high multiplicity of infection ( MOI ) for multi-step and single-step growth curves , respectively ( Figure 2B and C ) . In both types of replication analysis , MCMV replicated to low titers when IRE1 expression was induced with very low or very high doxycycline concentrations . High MCMV titers ( ~106 infectious units per ml ) , comparable to those obtained in wildtype ( WT ) MEFs , were attained only upon moderate induction of IRE1-GFP with 5 to 10 nM doxycycline ( Figure 2B and C ) . We observed that IRE1-GFP induction with high doxycycline concentrations resulted in a significantly decreased cell viability ( Figure 2D ) , suggesting that IRE1 overexpression is detrimental to cell viability . This is consistent with the previously reported cytotoxicity of IRE1 overexpression ( Han et al . , 2009 ) . To formally exclude the possibility that strong activation of the TetON transactivator alone impairs MCMV replication , we infected TetON-expressing cells with MCMV and analyzed viral replication in the presence of different doxycycline concentrations . As expected , MCMV replication was not affected by different doxycycline concentration ( Figure 2—figure supplement 1 ) . Thus , we concluded that IRE1 is necessary for efficient replication of MCMV , but needs to be carefully regulated as too high expression levels are detrimental for cell viability and viral replication . Activated IRE1 can splice Xbp1 mRNA ( Calfon et al . , 2002; Lee et al . , 2002; Yoshida et al . , 2001 ) and can also recruit TRAF2 to activate ASK1 ( Urano et al . , 2000 ) . To test which IRE1-dependent signaling pathway is required for efficient MCMV replication , we used CRISPR/Cas9-mediated gene editing to generate knockout ( ko ) MEFs for Ern1 ( the gene encoding IRE1 ) , Xbp1 , and Traf2 . For each gene knockout , two independent cell clones were generated with different guide RNAs . The absence of the respective gene products was verified by immunoblot analysis ( Figure 3A ) . Then , the cell clones were used to assess MCMV replication . In Ern1 ko MEFs , viral replication ( Figure 3B ) and viral gene transcription ( Figure 3—figure supplement 1 ) were massively reduced as compared to WT MEFs ( Figure 3B ) , similar to the reduction seen in IRE1-GFP cells without doxycycline induction ( Figure 2B ) . By contrast , MCMV replication was virtually unimpaired in the absence of Xbp1 ( Figure 3B ) or Traf2 ( Figure 3C ) . We also analyzed the expression of a viral immediate-early ( IE1 ) , an early ( M57 ) , and a late protein ( gB ) at different times after high-MOI infection . Compared to WT MEFs , the expression of all three proteins was reduced in Ern1 ko MEFs ( Figure 3D ) , but not in Xbp1 or Traf2 ko MEFs ( Figure 3E and F ) . Next , we tested whether the RNase activity of IRE1 is required for efficient MCMV replication . To do this , WT IRE1 or an ‘RNase-dead’ IRE1-K907A mutant ( Tirasophon et al . , 2000 ) was expressed in Ern1 ko MEFs by retroviral transduction . Expression of WT and mutant IRE1 and the ability to splice Xbp1 was verified by immunoblot analysis ( Figure 4A ) . While expression of WT IRE1 restored MCMV replication to high titers , expression of IRE1-K907A did not increase MCMV titers ( Figure 4B ) , indicating that the IRE1 RNase activity is necessary for efficient MCMV replication . Our observations that the RNase activity of IRE1 is required for efficient MCMV replication , but XBP1 is not , allowed two possible explanations: ( i ) MCMV replication could benefit from RIDD , another RNase-dependent function of IRE1 . However , this possibility is difficult to verify as selective inactivation of RIDD is complicated . ( ii ) Alternatively , MCMV replication could be inhibited by XBP1u , the product of the unspliced Xbp1 mRNA , since Ern1 ko cells differ from other cells in that they express only XBP1u ( Figure 3A ) . To test the latter option , we used two experimental approaches . First , we knocked out Ire1 in Xbp1-deficient cells ( Figure 5A ) . As shown in Figure 5B , MCMV replicated to similar titers in Xbp1 ko and Xbp1/Ern1 double-knockout ( dko ) cells , indicating that the loss of IRE1 is not detrimental for MCMV replication when XBP1 is absent . Next , we used retroviral transduction to restore XBP1 expression in Xbp1-/- cells . The retroviral vectors expressed the WT Xbp1 transcript ( which can be spliced by IRE1 ) , a truncated Xbp1 transcript encoding only the DNA-binding domain ( XBP1stop ) , or an 'unspliceable' Xbp1 transcript . Upon treatment with thapsigargin , these cells expressed XBP1s , XBP1stop , or XBP1u , respectively ( Figure 5C ) . Whereas re-introduction of WT XBP1 had no detrimental effect , MCMV replication was severely impaired upon expression of the unspliceable Xbp1 transcript ( Figure 5D ) , indicating that Xbp1u reduces viral replication . A similar inhibitory effect was observed when the truncated XBP1 protein was expressed ( Figure 5D ) . XBP1u is thought to function as a negative regulator to XBP1s ( Tirosh et al . , 2006; Yoshida et al . , 2006 ) . However , the inhibition of MCMV replication by XBP1u cannot be explained solely by a repressive effect on XBP1s as a complete loss of XBP1 is not detrimental to MCMV replication in vitro ( Figure 3B ) and has only a modest effect on viral replication in vivo ( Drori et al . , 2014 ) . Thus , we hypothesized that XBP1u might impair MCMV replication by repressing the activity of additional transcription factors besides XBP1s . For several reasons ATF6 appeared to be a likely target of XBP1u: ( i ) Like XBP1s , ATF6 is a bZIP transcription factor activated by ER stress ( Yoshida et al . , 1998 ) ; ( ii ) XBP1s and ATF6 are known to synergize in the activation of numerous ER stress response genes ( Lee et al . , 2002 ) ; and ( iii ) XBP1u can interact with ATF6 and inhibit its activity ( Yoshida et al . , 2009 ) . Therefore , we tested whether the loss of both , XBP1 and ATF6 , was detrimental for MCMV replication . First , we analyzed MEFs from Atf6-/- mice and found that MCMV gene expression and replication were not impaired ( Figure 6A and B ) . When we knocked out Xbp1 in Atf6-/- cells by CRISPR/Cas9 gene editing ( Figure 6C ) , MCMV replication was substantially reduced in Atf6/Xbp1 dko cells ( Figure 6D ) , suggesting that ATF6 and XBP1s have overlapping or redundant functions and that at least one of them is necessary for efficient MCMV replication . The MCMV MIEP has a key role in viral gene expression as it drives the expression of the viral IE1 and IE3 proteins . IE3 is the major transactivator protein that activates early and late gene expression ( Lacaze et al . , 2011 ) . Thus we interrogated whether XBP1s and ATF6 can activate the MCMV MIEP . First , we searched for ACGT motifs within the MCMV MIEP . ACGT is a minimal consensus sequence contained within XBP1s and ATF6 binding motifs ( Kanemoto et al . , 2005; Wang et al . , 2000 ) . Five ACGT motifs were identified within the MCMV MIEP ( Figure 7A ) . To analyze the function of these putative transcription factor binding sites , we inserted the WT MIEP and six mutant versions ( Supplementary file 1 ) into the luciferase reporter plasmid pGL3-Basic . In the mutant MIEPs , one or all five ACGT motifs were changed to CTAG . Using a luciferase reporter assay , we measured MIEP activity in WT , Ern1 ko and Xbp1 ko fibroblasts . As shown in Figure 7B , MIEP activity was strongly reduced in Ern1 ko cells but was not significantly altered in Xbp1 ko cells . This result is consistent with the MCMV replication defect observed in Ern1 , but not Xbp1 ko cells ( Figure 3B ) . MIEP activity was not reduced in Xbp1/Ern1 dko cells compared to the parental Xbp1 ko cell ( Figure 7C ) , but was significantly reduced in Atf6/Xbp1 dko cells compared to the parental Atf6-/- cells ( Figure 7D ) . Again , the MIEP activities were consistent with the results of the viral replication kinetics ( Figures 5B and 6D ) . Thus , we concluded that activation of the MCMV MIEP correlated with viral replication in the same cells . We also found that the activity of the MIEP had a substantially reduced activity when all 5 ACGT motifs were mutated ( Figure 7B–D ) . Next , we tested whether XBP1s and ATF6 ( N ) , the active form of ATF6 , can activate the MCMV MIEP and whether XBP1u can repress it . To assess the contribution of endogenous levels of the TFs , Xbp1 ko , Atf6-/- , and Xbp1/Atf6 dko cells were used . In Xbp1 ko cells , MIEP activity was slightly increased by expression of XBP1s , but substantially reduced by XBP1u . XBP1u also antagonized XBP1s in a dose-dependent manner ( Figure 7E ) . A similar result was obtained in Atf6-/- cells: MIEP activity was slightly increased by expression of ATF6 ( N ) , but substantially reduced by XBP1u . XBP1u also antagonized ATF6 ( N ) in a dose-dependent manner ( Figure 7F ) . In Atf6/Xbp1 dko cell , expression of either XBP1s or ATF6 ( N ) was sufficient to increase MIEP activity substantially . XBP1u alone did not reduce MIEP activity in these cells , but it antagonized the activity of ATF6 ( N ) expressed by transfection ( Figure 7G ) . Taken together , these results suggest that both , XBP1s and ATF6 ( N ) , can activate the MCMV MIEP in a largely redundant fashion , and that XBP1u represses the activity of both , XBP1s and ATF6 ( N ) . As a first step to determine , which of the five ACGT motifs function as XBP1s and/or ATF6 ( N ) binding sites for promoter activation , we measured the binding of these TFs to portions of the MIEP by using a DNA-protein interaction ELISA ( DPI-ELISA , [Brand et al . , 2010; Underwood et al . , 2013] ) . Microtiter plates were coated with double-stranded oligonucleotides encoding three copies ( 27 nucleotides each ) of a putative binding motif with adjacent sequences on either side ( Supplementary file 2 ) . A known XBP1 binding motif of the ERdj4 promoter ( Kanemoto et al . , 2005 ) served as positive control and a sequence of the SV40 origin of replication as negative control . The oligonucleotides were incubated with HA-tagged XBP1s , XBP1u , or ATF6 ( N ) ( Figure 8A ) , and DNA binding was quantified by ELISA . Mutated TFs lacking the DNA-binding domain ( ΔDBD ) served as negative controls . In this DPI-ELISA , XBP1s and XBP1u showed the strongest interaction with motifs 3 and 4 , whereas ATF6 ( N ) interacted with motifs 2 and 4 ( Figure 8B ) . Next , we used the luciferase reporter assay to test which of the five motifs was required for MIEP activation by XBP1s and ATF6 ( N ) . We used five mutant MIEPs having one of the ACGT motifs changed to CTAG . These reporter plasmids were transfected into Atf6/Xbp1 dko cells , together with expression plasmids for XBP1s , XBP1u , and ATF6 ( N ) . As shown in Figure 9A , XBP1s or ATF6 ( N ) expression increased the activity of all MIEP constructs except MIEP-4-mut and MIEP-all-mut , indicating that ACGT motif 4 is necessary for MIEP activation by XBP1s and ATF6 ( N ) . We tested then whether motif 4 was also sufficient for MIEP activation by XBP1s and ATF6 ( N ) . Motif 4 was restored in MIEP-all-mut to generate a MIEP containing only ACGT motif 4 ( MIEP-4-only , Supplementary file 1 ) . Indeed , MIEP-4-only was inducible by XBP1s and ATF6 ( N ) ( Figure 9B ) , suggesting that motif 4 is sufficient to confer MIEP responsiveness to XBP1s and ATF6 . To determine whether motif 4 is also of crucial importance for MCMV replication , we mutated motif four within the MCMV genome . The resulting virus , MCMV-GFP-4-mut , was compared to the parental MCMV-GFP virus in a multi-step replication kinetic analysis ( Figure 9C ) . Indeed , viral replication was massively reduced when motif 4 within the MIEP was mutated . This result confirmed that the XBP1 and ATF6 ( N ) target motif identified in DNA-binding and luciferase reporter assays is required for efficient replication of the virus . Taken together , the results of this study show that MCMV transiently activates the IRE1-XBP1 signaling pathway in the early phase of infection in order to relieve XBP1u-mediated repression of viral gene expression and replication . The MCMV major immediate-early promoter requires XBP1s or ATF6 ( N ) binding to motif 4 for full activation , and XBP1u prevents promoter activity by both TFs . Thus , XBP1u acts as a key repressor of two UPR pathways , the IRE1-XBP1 and the ATF6 pathways , which have overlapping and redundant functions . This study shows how MCMV harnesses UPR signaling to regulate its own life cycle . The viral MIEP , the key promotor of the lytic replication cycle , contains five ACGT motifs , of which motif 4 is most important for MIEP activation by XBP1s or ATF6 ( N ) ( Figure 9 ) . The fact that MIEP activity and viral replication are virtually unaffected in the absence of either XBP1 or ATF6 but are massively reduced in the absence of both , strongly suggests that the two UPR TFs have redundant roles in the activation of the MIEP . The most important finding , however , is the dominant role of XBP1u as a repressor of viral gene expression and replication . When IRE1 is absent or its RNase activity is blocked , XBP1u dominates and prevents MIEP activation by XBP1s and ATF6 ( N ) . Thus , the transient activation of IRE1 during the early phase of MCMV infection serves the virus by increasing XBP1s and decreasing XBP1u expression , thereby relieving XPB1u-mediated repression . At present , the MCMV protein ( s ) responsible for the early activation of IRE1 remains unknown . Most of the viral envelope glycoproteins are expressed at late times post infection ( i . e . , starting 12 hpi ) and are thus unlikely to be responsible . However , it is conceivable that certain early glycoproteins are particularly potent stress inducers , either by binding and sequestering BiP ( Bakunts et al . , 2017; Vitale et al . , 2019 ) or by inducing ER calcium release . Possible candidates for the latter include the MCMV homolog of the HCMV UL37x1 protein ( Sharon-Friling et al . , 2006 ) or the viral G protein-coupled receptors , which can activate phospholipase C and the inositol trisphosphate receptor ( Boeck and Spencer , 2017; Sherrill et al . , 2009 ) . How exactly XBP1u antagonizes XBP1s and ATF6 ( N ) -mediated promoter activation , remains to be determined . Two possible mechanisms may be involved: XBP1u could interact with XBP1s and ATF6 ( N ) and target them for proteasomal degradation as previously reported ( Yoshida et al . , 2006; Yoshida et al . , 2009 ) . Additionally , XBP1u could bind directly to DNA and inhibit promoter activation by XBP1s and ATF6 ( N ) . Indeed , we show that XBP1u is fully capable of binding to the same DNA sequences that XBP1s binds to ( Figure 8 ) . Moreover , a truncated XBP1 protein , XBP1stop , which lacks the C-terminal destabilizing domain of XBP1u , had a similar repressive effect as XBP1u ( Figure 5D ) indicating that destabilization is not required for repression . Ever since its discovery , the importance of XBP1u has been the subject of controversy . In yeast , the unspliced mRNA of HAC1 , the yeast homolog of XBP1 , is posttranscriptionally silenced , and a protein product of the unspliced HAC1 mRNA has not been detected ( Rüegsegger et al . , 2001 ) . In mammalian cells , the XBP1u protein is readily detectable , but XBP1u has a short half-life ( Navon et al . , 2010; Tirosh et al . , 2006 ) . This has led to the conclusion that XBP1u is of minor importance and its function restricted to fine-tuning of the UPR ( Byrd and Brewer , 2012 ) . Hence , many review articles on the UPR barely mention XBP1u . On the other hand , XBP1u has been shown to dimerize with XBP1s and ATF6 ( N ) and destabilize them ( Yoshida et al . , 2006; Yoshida et al . , 2009 ) , suggesting a more important regulatory role . However , the physiological relevance of these findings has been called into question by some scientists because the findings were made in cells overexpressing XBP1u . The results of our study show that XBP1u plays a very important role as a repressor of XBP1s and ATF6 ( N ) -mediated activation of the MCMV MIEP , which results in a massively ( up to 100-fold ) reduced production of progeny virus ( Figures 2 , 3B and 5D ) . In IRE1-deficient cells , the Xbp1 transcript is expressed from its endogenous promoter , not from a strong heterologous promoter . Thus , the observed effects cannot be dismissed as overexpression artifacts . Moreover , the fact that MCMV replication is massively reduced in Ern1 ko cells ( Figures 2 and 3B ) , but not in Ern1/Xbp1 dko cells ( Figure 5B ) , demonstrates that XBP1u expression rather than the absence of RIDD is responsible for the observed effect . Hence , the data of our study suggest that XBP1u plays an unexpectedly important role as a repressor , at least under the conditions of viral infection . Whether XBP1u can repress expression of cellular genes in a similar fashion is an important question that needs to be answered in future studies . XBP1 binding sites have also been identified in immediate-early gene promoters of the γ-herpesviruses Epstein-Barr virus , Kaposi's sarcoma-associated herpesvirus , and murine gammaherpesvirus 68 ( MHV-68 ) , suggesting a potential role for XBP1s in transactivating these promoters during reactivation from latency ( Bhende et al . , 2007; Matar et al . , 2014; Sun and Thorley-Lawson , 2007; Wilson et al . , 2007; Yu et al . , 2007 ) . However , one study demonstrated that XBP1 was not required for MHV-68 reactivation in B cells ( Matar et al . , 2014 ) . The authors of the study speculated that the apparent independence of MHV-68 reactivation from XBP1 expression in B cells might reflect redundancy among CREB/ATF family TFs . In light of our data showing a redundancy of XBP1s and ATF6 ( N ) in the activation of the MCMV MIEP , this conjecture may well be correct . Conversely , it would be interesting to know whether XBP1 and ATF6 also regulate MCMV reactivation from latency . Unfortunately , this obvious question is difficult to address as there is no manageable cell culture system for MCMV latency and reactivation . Nonetheless , it appears likely that β and γ-herpesviruses harness UPR TFs in a similar fashion to regulate their life cycle . The following immortalized fibroblast lines were used: wildtype MEFs ( Manzl et al . , 2009 ) , 10 . 1 cells ( Harvey and Levine , 1991 ) , Ern1-/- MEFs ( Calfon et al . , 2002 ) , and Xbp1-/- MEFs ( Reimold et al . , 2000 ) . Primary Atf6+/+ and Atf6-/- MEFs ( Wu et al . , 2007 ) were kindly provided by D . Thomas Rutkowski ( University of Iowa , USA ) . They were immortalized by transduction with a retroviral vector encoding SV40 large T antigen ( pBABE-zeo largeTcDNA , Addgene ) . The Atf6+/+ vs . Atf6-/- state was verified by PCR as described ( Wu et al . , 2007 ) . TetON-MEFs were obtained by transduction of wiltype MEFs with a lentiviral vector encoding the TetON transactivator ( pLIX-402 , Addgene ) . TetON-Ern1-/- cells were generated in a similar fashion . TetON-IRE1-GFP MEFs were obtained by reconstituting TetON-Ern1-/- MEFs with GFP-tagged murine IRE1 under control of a ‘tight’ Tet-responsive element ( Clontech ) essentially as described ( Bakunts et al . , 2017; Cohen et al . , 2017 ) . The GFP tag was introduced into the juxtamembrane cytosolic linker domain of IRE1 , where such tagging had been shown before to not interfere with function of human IRE1 ( Li et al . , 2010 ) . All cells were grown under standard conditions in Dulbecco´s modified Eagle´s medium supplemented with 10% fetal calf serum or 10% fetal calf serum tetracycline-free , 100 IU/ml penicillin and 100 µg/ml streptomycin ( Sigma ) . All cells were tested regularly for mycoplasma contamination and were found to be negative . MCMV-GFP ( Brune et al . , 2001 ) was propagated and titrated on 10 . 1 fibroblasts . Viral titers were determined by using the median tissue culture infective dose ( TCID50 ) method . Virus was inactivated by 254-nm-wavelength UV irradiation with 1 J/cm2 for 30 s using a UV cross-linker ( HL-2000 HybriLinker; UVP ) . The recombinant MCMV carrying an ACGT-to-CTAG mutation of MIEP motif 4 was constructed by BAC ( bacterial artificial chromosome ) recombineering using the en passant mutagenesis protocol ( Tischer et al . , 2010 ) . A kanamycin resistant gene was PCR-amplified using primers MIEP_4_mut fwd ( 5’-GGTACTTTCCCATAGCTGATTAATGGGAAAGTACCGTTCTCGAGCCAATACCTAGCAATGGGAAGTGAAAGGGCAGtagggataacagggtaatcgattt-3’ ) and MIEP_4-mut rev ( 5’-GGGGAAAACCGGGGCGGTGTTACGTTTTGGCTGCCCTTTCACTTCCCATTGCTAGGTATTGGCTCGAGAACGGTACTgccagtgttacaaccaattaacc-3’ ) . The linear PCR product was used for homologous recombination in Escherichia coli strain GS1783 containing the MCMV-GFP BAC . The kanamycin resistance cassette was used for positive selection in the first recombination step and removed in the second step ( Tischer et al . , 2010 ) . The MCMV-GFP-4-mut BAC was examined by restriction fragment analysis and sequencing of the mutated region . Mutant and control viruses were reconstituted by BAC transfection of mouse fibroblast using GenJet transfection reagent ( SignaGen ) . XBP1-wt and XBP1-unsplicable cDNAs ( Tirosh et al . , 2006 ) were cloned in pcDNA3 . The XBP1-wt vector was used to generate by PCR an XBP1stop mutant consisting of only the N-terminal domain of XBP1 . The XBP1s cDNA was generated by isolating RNA from MEFs stimulated with tunicamycin ( Sigma ) , reverse transcription , and PCR amplification . To generate XBP1-unsplicable and XBP1-spliced plasmids lacking the DNA-binding domain ( ΔDBD ) , the complete DBD was replaced by an alternative NLS sequence as previously described ( Zhou et al . , 2011 ) . Transcripts of XBP1 proteins were HA-tagged by PCR amplification and subcloned in pMSCVpuro ( Clonetech ) , a retroviral vector plasmid . Similarly , mIRE1-wt was PCR-amplified ( without a myc tag ) from pcDNA3-mIRE1-3xmyc ( Stahl et al . , 2013 ) and subcloned in pMSCVhygro ( Clontech ) using BglII and HpaI sites . The K907A mutation was introduced by QuikChange site-directed mutagenesis ( Stratagene ) . Transient transfection was done using GenJet ( SignaGen ) or polyethylenimine ( Sigma ) . Retrovirus production using the Phoenix packaging cell line and transduction of target cells was done as described ( Swift et al . , 2001 ) . Cells transduced with MSCVpuro and MSCVhygro vectors were selected with 1 . 5 µg/ml puromycin ( Sigma ) or 50 µg/ml hygromycin B ( Roth ) , respectively . Guide RNAs ( Supplementary file 3 ) for genes of interest were designed using the online tool E-CRISP ( http://www . e-crisp . org/E-CRISP/designcrispr . html ) and inserted into the lentiviral vector pSicoR-CRISPR-puroR ( kindly provided by R . J . Lebbink , University Medical Center Utrecht , Netherlands ) . Lentiviruses were produced in 293 T cells using standard third-generation packaging vectors as described ( van Diemen et al . , 2016 ) . Lentiviruses were used to transduce MEFs in the presence of polybrene ( 5 µg/ml ) . Cells were selected with 1 . 5 µg/ml puromycin and single cell clones were obtained by limiting dilution . Whole cell lysates were obtained by lysing cells in RIPA buffer supplemented with a cOmplete , Mini protease inhibitor cocktail ( Roche ) . XBP1s , XBP1u , and ATF6 ( N ) were extracted from cells treated with thapsigargin ( Sigma ) using a nuclear extraction protocol ( Stahl et al . , 2013 ) . Insoluble material was removed by centrifugation . Protein concentrations were measured using a BCA assay ( Thermo Fisher Scientific ) . Equal protein amounts were boiled in sample buffer and subjected to SDS-PAGE and semi-dry blotting onto nitrocellulose membranes . For immunodetection , antibodies against the following epitopes were used: HA ( 16B12 , Covance ) , β-actin ( AC-74 , Sigma ) , XBP1 ( M-186 , Santa Cruz ) , IRE1 ( 14C10 , Cell Signaling ) , pIRE1 ( NB100-2323 , Novus Biologicals ) , TRAF2 ( Cell Signaling ) , and GFP ( Roche ) . Antibodies against MCMV IE1 ( CROMA101 ) , M57 , and M55/gB ( SN1 . 07 ) were provided by Stipan Jonjic ( University of Rijeka , Croatia ) . Secondary antibodies coupled to horseradish peroxidase ( HRP ) were purchased from Dako . Total RNA was isolated from MEFs using an innuPREP RNA Mini Kit ( Analytik-Jena ) . Reverse transcription and cDNA synthesis was carried out with 2 µg RNA using 200 U RevertAid H Minus Reverse Transcriptase , 100 pmol Oligo ( dT ) 18 , and 20 U RNase inhibitor ( Thermo Fisher Scientific ) . Quantitative real-time PCR reactions employing SYBR Green were run in a 7900HT Fast Real-Time PCR System ( Applied Biosystems ) . The following primers were used: 5′-GAGTCCGCAGCAGGTG-3′ and 5′-GTGTCAGAGTCCATGGGA-3′ for Xbp1s , 5′-GTGTCAGAGTCCATGGGA-3′ and 5′-GTGTCAGAGTCCATGGGA-3′ for Xbp1u , and 5′-CCCACTCTTCCACCTTCGATG-3′ and 5′-GTCCACCACCCTGTTGCTGTAG-3′ for Gapdh . Primers for the amplification of IE1 ( M123 ) , E1 ( M112 ) , gB ( M55 ) , and M37 transcripts have been described previously ( Chapa et al . , 2013; Chapa et al . , 2014 ) . Reactions were performed under the following conditions: 45 cycles of 3 s at 95°C and 30 s at 60°C . Three replicates were analyzed for each condition , and the relative amounts of mRNAs were calculated from the comparative threshold cycle ( Ct ) values by using Gapdh as reference . Cells were seeded in 6-well plates and infected by an MOI of 3 or 0 . 1 for single or multi-step replication kinetics , respectively . Six hpi the medium was exchanged to remove the inoculum . Supernatant samples were harvested at different times post infection , and viral titers were determined on 10 . 1 fibroblasts using the TCID50 method . The firefly luciferase reporter vector pGL3-Basic , the renilla luciferase control vector pGL4 . 73 , and the Dual-Glo Luciferase assay system were purchased from Promega . WT and mutant versions of the MCMV MIE promotor ( Supplementary file 1 ) were synthesized by Integrated DNA Technologies and cloned in pGL3-Basic . Cells were transfect in 6-well dishes using GenJet with pGL3-MIEP ( 0 . 5 µg ) , pGL4 . 73 ( 0 . 05 µg ) and transcription factor expression plasmids . The total amount of DNA was kept constant at 3 µg by filling up with empty pcDNA3 vector . On the following day , the medium was exchanged and cells were incubated for another 24 hr . Cells were harvested in lysis buffer and luciferase activity was measured using a Dual-Glo Luciferase assay and a luminescence plate reader ( FLUOstar-Omega , BMG Labtech ) . Firefly and Renilla luciferase activities were evaluated for each sample . At least three biological replicates were used for each condition . The DPI-ELISA was performed essentially as described in detail elsewhere ( Brand et al . , 2010; Underwood et al . , 2013 ) . Terminally biotinylated dsDNA oligonucleotides containing 3 copies of putative TF binding sites ( Supplementary file 2 ) were purchased from Eurofins . Oligonucleotides were adsorbed to streptavidin-coated 96-well microtiter plates and incubated with nuclear extracts ( 10 µg protein ) from transfected HEK 293A cells expressing the HA-tagged XBP1 or ATF6 ( N ) proteins . Nuclear extracts were obtained as described ( Stahl et al . , 2013 ) . TF binding to DNA was quantified by ELISA using an anti-HA antibody , an HRP-coupled secondary antibody , and an ABTS substrate ( Roche ) . Absorbance at 405 nm wavelength was measured using a FLUOstar-Omega plate reader . At least three biological replicates were used for each condition . TetON-IRE1-GFP cells were seeded in 96-well plates at a density of 1000 cells per well . Cells were treated with 0 , 10 or 25 nM doxycycline at the time of seeding . Medium was replenished 1 and 3 days after seeding . Cell viability was determined on day 3 and day 5 by measuring intracellular ATP levels with a Cell Titer-Glo Luminescent Cell Viability Assay kit ( Promega ) and a FLUOstar Omega luminometer ( BMG Labtech ) . The values were normalized to non-stimulated cells . All statistical analyses were performed with GraphPad Prism 5 . 0 software . One-way ANOVA followed by Bonferroni´s post hoc test was used for the analysis of the luciferase reporter assays , the cell viability assay , and the DPI-ELISA .	Cells survive by making many different proteins that each carry out specific tasks . To work correctly , each protein must be made and then folded into the right shape . Cells carefully monitor protein folding because unfolded proteins can compromise their viability . A protein called XBP1 is important in controlling how cells respond to unfolded proteins . Normally , cells contain a form of this protein called XBP1u , while increasing numbers of unfolded proteins trigger production of a form called XBP1s . The change from one form to the other is activated by a protein called IRE1 . Viruses often manipulate stress responses like the unfolded protein response to help take control of the cell and produce more copies of the virus . Murine cytomegalovirus , which is known as MCMV for short , is a herpes-like virus that infects mice; it stops IRE1 activation and XBP1s production during the later stages of infection . However , research had shown that the unfolded protein response was triggered for a short time at an early stage of infection with MCMV , and it was unclear why this might be . Hinte et al . studied the effect of MCMV on cells grown in the laboratory . The experiments showed that a small dose of cell stress , namely activating the unfolded protein response briefly during early infection , helps to activate genes from the virus that allow it to take over the cell . Together , XBP1s and another protein called ATF6 help to switch on the viral genes . The virus also triggers IRE1 helping to reduce the levels of XBP1u , which could slow down the infection . Later , suppressing the unfolded protein response allows copies of the virus to be made faster to help spread the infection . These findings reveal new details of how viruses precisely manipulate their host cells at different stages of infection . These insights could lead to new ways to manage or prevent viral infections .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "microbiology", "and", "infectious", "disease" ]	2020	Repression of viral gene expression and replication by the unfolded protein response effector XBP1u
Primary microcephaly ( MCPH ) associated proteins CDK5RAP2 , CEP152 , WDR62 and CEP63 colocalize at the centrosome . We found that they interact to promote centriole duplication and form a hierarchy in which each is required to localize another to the centrosome , with CDK5RAP2 at the apex , and CEP152 , WDR62 and CEP63 at sequentially lower positions . MCPH proteins interact with distinct centriolar satellite proteins; CDK5RAP2 interacts with SPAG5 and CEP72 , CEP152 with CEP131 , WDR62 with MOONRAKER , and CEP63 with CEP90 and CCDC14 . These satellite proteins localize their cognate MCPH interactors to centrosomes and also promote centriole duplication . Consistent with a role for satellites in microcephaly , homozygous mutations in one satellite gene , CEP90 , may cause MCPH . The satellite proteins , with the exception of CCDC14 , and MCPH proteins promote centriole duplication by recruiting CDK2 to the centrosome . Thus , centriolar satellites build a MCPH complex critical for human neurodevelopment that promotes CDK2 centrosomal localization and centriole duplication . At the heart of the centrosome , the principle microtubule-organizing center of a mammalian cell , are two centrioles . During S phase , centrioles are duplicated through the coordinated assembly of centriolar complexes near the proximal ends of the preexisting centrioles . Centriole duplication is coupled to the cell cycle preventing under- or over-duplication , either of which can disrupt spindle formation and chromosomal segregation ( Hinchcliffe et al . , 1999; Lacey et al . , 1999; Meraldi et al . , 1999; Vidwans et al . , 1999; Haase et al . , 2001 ) . Within centrosomes , a pair of centrioles are surrounded by pericentriolar material and centriolar satellites , 70–100 nm diameter electron dense structures that participate in microtubule-based transport ( Kubo et al . , 1999; Dammermann and Merdes , 2002; Kodani et al . , 2010 ) . Centriolar satellites contain PCM1 , thought to function in the delivery of proteins to the centrosome ( Kim et al . , 2008; Kodani et al . , 2010 ) . Consistent with this hypothesis , depletion of PCM1 reduces the centrosomal localization of select proteins ( Dammermann and Merdes , 2002; Kim et al . , 2004 , 2008; Kodani et al . , 2010 ) . Centriolar satellite proteins also promote ciliogenesis and centriole duplication through only partially elucidated mechanisms ( Kim et al . , 2004; Sedjai et al . , 2010; Kim and Rhee , 2011; Kim et al . , 2012; Stowe et al . , 2012; Firat-Karalar et al . , 2014; Klinger et al . , 2014 ) . In addition to chromosomal missegregation , altered centriole biogenesis is associated with human developmental growth disorders , such as primordial dwarfism and MCPH ( Mochida and Walsh , 2001 , 2004; Woods et al . , 2005; Griffith et al . , 2008; Rauch et al . , 2008; Thornton and Woods , 2009 ) . Many of the genes mutated in MCPH encode centrosomal proteins ( Bond et al . , 2005; Zhong et al . , 2005; Kumar et al . , 2009; Guernsey et al . , 2010; Nicholas et al . , 2010; Yu et al . , 2010; Sir et al . , 2011; Lin et al . , 2013 ) . The association of MCPH proteins with the centrosome is evolutionarily conserved with Caenorhabditis elegans ( Delattre et al . , 2006; Strnad and Gonczy , 2008 ) . Three MCPH proteins , CEP152 , CEP135 and STIL , interact with and promote the centrosomal localization of SAS4 ( also known as CPAP or CENPJ ) ( Strnad and Gonczy , 2008; Cizmecioglu et al . , 2010; Dzhindzhev et al . , 2010; Sir et al . , 2011; Brown et al . , 2013; Lin et al . , 2013 ) . Failure to recruit SAS4 can attenuate centriole elongation and duplication ( Schmidt et al . , 2009; Comartin et al . , 2013; Lin et al . , 2013 ) . Together , these observations have raised the possibility that CEP152 , CEP135 and STIL promote the recruitment of proteins to the centrosome to facilitate centriole duplication . However , how these MCPH-associated proteins localize to the centrosome and how they promote centriole duplication have remained largely elusive . Apart from CEP152 , CEP135 , STIL and SAS4 , the protein products of other MCPH-associated genes , including WDR62 , CDK5RAP2 and CEP63 , participate in centriole biogenesis and function ( Barrera et al . , 2010; Nicholas et al . , 2010; Yu et al . , 2010; Sir et al . , 2011 ) . Whether and , if so , how these proteins function together are unclear . We tested the hypothesis that these MCPH-associated proteins biochemically interact and cooperate in a shared mechanism of centriole biogenesis . To test this hypothesis , we identified interactors of each MCPH-associated protein and found that the MCPH proteins CDK5RAP2 , CEP152 , WDR62 and CEP63 physically associate with each other . Moreover , they form a hierarchy in which each is required to localize another to the centrosome , and that this stepwise assembly at the centrosome is essential to promote centriole duplication . In addition to interacting with each other , the MCPH-associated proteins CDK5RAP2 , CEP152 , WDR62 and CEP63 each interacts with a cognate centriolar satellite proteins . Their associated centriolar satellite partners are required for the localization of the interacting MCPH-associated protein to the centrosome . Consistent with a role in building the MCPH protein complex at the centrosome , centriolar satellites , like their MCPH-associated proteins , are necessary for centriole duplication to occur efficiently . Thus , paralleling the hierarchy of MCPH-associated proteins , there is a hierarchy of satellite proteins , each of which participates in the centriolar localization of an MCPH-associated protein . We found that a homozygous , missense mutation affecting one of these centriolar satellite components , CEP90 , is associated with microcephaly , further validating the functional connection between centriolar satellites and the function of the previously defined MCPH-associated proteins . How do the centrosomal MCPH-associated proteins promote centriole duplication ? We found that both the MCPH-associated proteins and their centriolar satellite partner proteins are required for the centrosomal localization of CDK2 , a cyclin-dependent kinase with established roles in both cell cycle progression and centriole duplication ( Hinchcliffe et al . , 1999; Meraldi et al . , 1999 ) . Thus , MCPH-associated proteins and centriolar satellites cooperate to ensure the stepwise recruitment to the centrosome of additional MCPH-assocated proteins , culminating in bringing CDK2 to the centrosome and duplicating the centrioles . To determine which centrosomal proteins CDK5RAP2 , CEP152 , WDR62 and CEP63 associate with , we immunoprecipitated endogenous CDK5RAP2 , CEP152 , WDR62 , and CEP63 from HeLa cells and identified co-precipitating proteins by LC-MS/MS . c-Myc , a non-centrosomal protein , served as a negative control in our immunoprecipitation experiments . While we detected a number of interacting proteins , we focused our study on proteins previously reported to localize to the centrosome and implicated in centriole duplication . Specifically , we used the data from Jakobsen et al . and Balestra et al . to identify likely centrosomal proteins and those required for centriole biogenesis , respectively ( Jakobsen et al . , 2011; Balestra et al . , 2013 ) . The mass spectrometry confirmed previously identified interactions , such as the association of CEP152 with CDK5RAP2 and CEP63 ( Sir et al . , 2011; Brown et al . , 2013; Lukinavicius et al . , 2013; Firat-Karalar et al . , 2014 ) , and suggested that CDK5RAP2 , CEP152 , WDR62 and CEP63 may interact with each other ( Supplementary file 1 ) . Consistent with this possibility , co-immunoprecipitation confirmed that CDK5RAP2 , CEP152 , WDR62 and CEP63 interact with each other , but not with an unrelated centriole component , CP110 ( Chen et al . , 2002 , Figure 1A ) . 10 . 7554/eLife . 07519 . 003Figure 1 . CDK5RAP2 , CEP152 , WDR62 , and CEP63 interact at the centrosome and localize to centrioles in a stepwise manner . ( A ) We immunoprecipitated endogenous CDK5RAP2 , CEP152 , WDR62 , and CEP63 from HeLa total cell lysates . Precipitation and co-precipitation were detected using antibodies specific to CDK5RAP2 , CEP152 , WDR62 , and CEP63 . The centriole component CP110 served as a negative control . ( B ) Sucrose gradient fractions of HeLa cell lysates were analyzed by immunoblot with antibodies to CP110 , CDK5RAP2 , CEP152 , WDR62 , and CEP63 . ( C ) Immunofluorescence of S phase scrambled control ( SC ) , CDK5RAP2 , CEP152 , WDR62 , and CEP63 siRNA-treated HeLa cells co-stained for Centrin ( ‘c’ , green ) to visualize centrioles , CDK5RAP2 ( red ) , CEP152 ( red ) , WDR62 ( red ) , and CEP63 ( red ) , and nuclei ( DAPI , blue ) . The inset shows magnified images of the boxed region . ( D ) Our findings indicate that CDK5RAP2 , recruits CEP152 to the centrosome , which in turn recruits WDR62 and CEP63 . Scale bars indicate 5 μm for all images . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 00310 . 7554/eLife . 07519 . 004Figure 1—figure supplement 1 . CDK5RAP2 , CEP152 , WDR62 and CEP63 are required for centriole duplication . ( A ) SC , CDK5RAP2 #1 , or CDK5RAP2 #2 siRNA-treated S phase HeLa cells were analyzed by immunofluorescence with Centrin ( ‘c’ , green ) and CDK5RAP2 ( red ) . ( B ) Immunoblotting of SC , CDK5RAP2 #1 , or CDK5RAP2 #2 siRNA transfected HeLa cell total cell lysate analyzed with an antibody to CDK5RAP2 . ( C ) Immunofluorescence images of S phase HeLa cells transfected with SC , CEP152 #1 , or CEP152 #2 siRNA and co-stained with Centrin ( ‘c’ , green ) and CEP152 ( red ) . ( D ) Total cell lysate of SC , CEP152 #1 , or CEP152 #2 siRNA-treated HeLa cells was analyzed by immunoblot with an antibody to CEP152 . ( E ) S phase HeLa cells transfected with SC , WDR62 #1 , or WDR62 #2 siRNA were co-stained with Centrin ( ‘c’ , green ) and WDR62 ( red ) . ( F ) Unboiled total cell lysate of SC , WDR62 #1 , or WDR62 #2 siRNA-treated HeLa cells was analyzed by immunoblot with an antibody to WDR62 . ( G ) Immunofluorescence analysis of SC , CEP63 #1 , or CEP63 #2 siRNA transfected HeLa cells co-stained with Centrin ( ‘c’ , green ) and CEP63 ( red ) . ( H ) Immunoblot of total cell lysate of HeLa cells transfected with SC , CEP63 #1 , or CEP63 #2 siRNA and analyzed with an antibody to CEP63 . ( I ) Quantification of S phase SC , CDK5RAP2 #1 , CDK5RAP2 #2 , CEP152 #1 , CEP152 #2 , WDR62 #1 , WDR62 #2 , CEP63 #1 , or CEP63 #2 siRNA-treated HeLa cells with four centrioles . S phase cells were identified by CyclinA immunostaining . For all quantifications at least 100 cells were counted per experiment ( n = 3 ) , p < 0 . 005 ( paired t-test ) . Actin served as a loading control for all immunoblot analyses . Scale bars indicate 5 μm for all images . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 00410 . 7554/eLife . 07519 . 005Figure 1—figure supplement 2 . CDK5RAP2/CEP215 promotes centriole duplication and centrosome organization . ( A ) S phase SC , CEP215 282 , CEP215 283 siRNA-treated HeLa cells co-stained with Centrin ( ‘c’ , green ) and CDK5RAP2 ( red ) . ( B ) Quantification of S phase SC , CEP215 282 , CEP215 283 siRNA transfected HeLa cells with four centrioles . For all quantifications at least 100 cells were counted per experiment ( n = 3 ) , p < 0 . 005 ( paired t-test ) . ( C ) Immunofluorescence images of S phase HeLa cells transfected with SC , CEP215 282 , CEP215 283 siRNA and co-stained with Centrin ( ‘c’ , green ) and CEP152 ( red ) , WDR62 ( red ) , or CEP63 ( red ) . Scale bars indicate 5 μm for all images . ( D ) Immunoblot of total cell lysate of HeLa cells transfected with SC , CEP215 282 , CEP215 283 siRNA and analyzed with antibodies to CDK5RAP2 , CEP152 , WDR62 , and CEP63 . Actin served as a loading control . ( E ) Quantification of the mean fluorescence intensities ±s . d . of CDK5RAP2 , CEP152 , WDR62 and CEP63 in SC , CEP215 282 , CEP215 283 siRNA-treated cells expressed as the mean percentage ±s . d . of the fluorescence intensities of SC cells . For fluorescence quantifications 30 cells were analyzed per experiment ( n = 3 ) . p < 0 . 005 ( paired t-test ) statistically significant differences are denoted by an asterisk . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 00510 . 7554/eLife . 07519 . 006Figure 1—figure supplement 3 . Centriole duplication defects in CDK5RAP2 , CEP152 , WDR62 and CEP63-depleted mitotic cells . ( A ) SC , CDK5RAP2 , CEP152 , WDR62 and CEP63 siRNA treated mitotic HeLa cells were co-stained for Centrin ( ‘c’ , green ) and CDK5RAP2 ( red ) , CEP152 ( red ) , WDR62 ( red ) , or CEP63 ( red ) . Scale bars indicate 5 μm for all images . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 00610 . 7554/eLife . 07519 . 007Figure 1—figure supplement 4 . CDK5RAP2 , CEP152 , WDR62 and CEP63 form a ring near parental centrioles . ( A ) HeLa cells were co-stained with antibodies to SAS6 ( green ) , CDK5RAP2 ( red ) , CEP152 ( red ) , WDR62 ( red ) , and CEP63 ( red ) and imaged using three-dimensional structured illumination microscopy ( SIM ) . Scale bars indicate 200 nm for all images . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 00710 . 7554/eLife . 07519 . 008Figure 1—figure supplement 5 . CDK5RAP2 , CEP152 , WDR62 and CEP63 do not control the stability of their binding partners . ( A ) Total cell lysate of SC , CDK5RAP2 , CEP152 , WDR62 and CEP63 siRNA transfected HeLa cells were analyzed by immunoblotting with antibodies to CDK5RAP2 , CEP152 , WDR62 and CEP63 . Actin served as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 00810 . 7554/eLife . 07519 . 009Figure 1—figure supplement 6 . CDK5RAP2 , CEP152 , WDR62 and CEP63 localize in a hierarchical manner at the centrosome . ( A ) Quantification of CDK5RAP2 , CEP152 , WDR62 and CEP63 in SC , CDK5RAP2 , CEP152 , WDR62 , CEP63-depleted cells expressed as the mean percentage ±s . d . of the fluorescence intensities of SC cells . For all quantifications 30 cells were analyzed per experiment ( n = 3 ) . p < 0 . 005 ( paired t-test ) statistically significant differences from the SC control cells are denoted by an asterisk . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 00910 . 7554/eLife . 07519 . 010Figure 1—figure supplement 7 . CEP63-depletion using a previous published siRNA destabilizes MCPH proteins . ( A ) HeLa cells in S phase transfected with SC or CEP63 Nicola Brown ( NB ) siRNA were co-stained with Centrin ( ‘c’ , green ) and CEP63 ( red ) . ( B ) Quantification of the mean percentage of SC and CEP63 NB siRNA treated cells in S phase with four centrioles . ( C ) S phase HeLa cells transfected with SC and CEP63 NB siRNA were co-stained with Centrin ( ‘c’ , green ) and CDK5RAP2 ( red ) , CEP152 ( red ) , or CEP63 ( red ) . ( D ) Quantification of the mean fluorescence intensities ±s . d . of CDK5RAP2 , CEP152 , WDR62 and CEP63 in SC and CEP63 NB siRNA treated cells expressed as the mean percentage ±s . d . of the fluorescence intensities of SC cells . For all quantifications 30 cells were analyzed per experiment ( n = 3 ) . p < 0 . 005 ( paired t-test ) statistically significant differences from SC controls are denoted by an asterisk . ( E ) Total cell lysates of SC and CEP63 NB siRNA treated cellswere analyzed by Western blot with antibodies to CEP63 , CDK5RAP2 , CEP152 , and WDR62 . Actin served as a loading control . Asterisk indicates the specific band . ( F ) Quantification of the mean intensities ±s . d . of CEP63 , CDK5RAP2 , CEP152 , and WDR62 in SC and CEP63 NB siRNA-treated cells expressed as the mean percentage ±s . d . of the intensities of SC samples . Actin served as a loading control and to normalize samples ( n = 3 ) . Asterisk denotes statistical significance compared to controls where p < 0 . 005 ( paired t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 010 To confirm the centrosomal enrichment of MCPH-associated proteins , we fractionated centrosomes from HeLa cells by sucrose gradient centrifugation . CDK5RAP2 , CEP152 , WDR62 and CEP63 co-fractionated with the centrosomal protein CP110 ( Figure 1B ) . Taken together , these findings suggest that CDK5RAP2 , CEP152 , WDR62 , and CEP63 interact with each other at the centrosome . As MCPH-associated proteins localize to the centrosome , we investigated whether the interaction of CDK5RAP2 , CEP152 , WDR62 and CEP63 reflects a shared centrosomal function . Consistent with previous studies , depletion of CEP152 or CEP63 disrupted centriole duplication ( Figure 1—figure supplement 1C , G; Brown et al . , 2013 ) . In these , and other similar experiments described below , we observed a mixture of cells in S phase ( as determined by nuclear Cyclin A localization ) with two and three centrioles , instead of four centrioles as observed in scrambled control siRNA ( SC ) -transfected cells: the images presented in the figures are of cells in which three centrioles are observed . Similar to the depletion of CEP152 and CEP63 , knockdown of CDK5RAP2 or WDR62 , inhibited centriole duplication ( Figure 1—figure supplement 1A , E ) . In contrast , mouse embryonic fibroblasts ( MEFs ) expressing a fusion between a truncated Cdk5rap2 and βGEO or decreased levels of Wdr62 exhibit centriole overaccumulation ( Barrera et al . , 2010; Chen et al . , 2014 ) , suggesting that efficient depletion of CDK5RAP2 and WDR62 has different consequences than truncations or partial reduction ( Figure 1—figure supplement 1B , F ) . To confirm that CDK5RAP2 ( also called CEP215 ) is required for centriole duplication , we repeated CDK5RAP2 knockdown using two previously published siRNAs ( Graser et al . , 2007 ) . Both of these CDK5RAP2 siRNAs also reduced the percentage of S phase cells with two pairs of centrioles ( Figure 1—figure supplement 2A , B ) . Instead of four centrioles , the majority of S phase CDK5RAP2 , CEP152 , WDR62 or CEP63-depleted cells possessed only two or three centrioles ( Figure 1C , depicting instances of three centrioles ) . A reduced number of centrioles in S phase could be caused by either a failure or delay in centriole duplication . Therefore , we also examined the number of centrioles in mitotic SC , CDK5RAP2 , CEP152 , WDR62 and CEP63-depleted cells ( Figure 1—figure supplement 3A ) . Depleting MCPH-associated proteins also caused a significant decrease in the number of centrioles in mitosis , suggesting that these MCPH-associated proteins have a shared function in promoting centriole duplication . As depletion of each of these four MCPH-associated proteins , CDK5RAP2 , CEP152 , WDR62 and CEP63 , has similar effects on centriole duplication , we examined whether they co-localize . We discovered that , like CDK5RAP2 , CEP152 , and CEP63 , WDR62 localizes to the interphase centrosome ( Figure 1C ) . To more precisely identify where WDR62 localizes at the centrosome , we used structured illumination microscopy . Also like CDK5RAP2 , CEP152 , and CEP63 , WDR62 is present in a toroidal structure with a diameter of 400–500 nm adjacent to SAS6 , a component of the proximal end of procentrioles ( Figure 1—figure supplement 4A , Strnad et al . , 2007; Sir et al . , 2011 ) . We found that the appearance of CDK5RAP2 differed depending on whether it was imaged using SIM or diffraction limited microscopy , which may be attributable to the limitations of SIM in imaging low contrast samples such as the CDK5RAP2 outside of the toroidal structure . Given their indistinguishable sub-centrosomal localization , we hypothesized that these MCPH proteins participate in each other's localization . Surprisingly , we found that there is a hierarchy by which they localize to the centrosome . CEP152 , WDR62 , and CEP63 failed to localize to the centrosome in the absence of CDK5RAP2 ( Figure 1C , and quantitated in Figure 1—figure supplements 2E , 7A ) . As this localization dependency differed from that reported by Firat-Karalar et al . we confirmed that the centrosomal localization of CEP152 depends on CDK5RAP2 using four non-overlapping siRNAs ( Figure 1C and Figure 1—figure supplement 2C ) . Conversely , depletion of CEP152 , WDR62 or CEP63 had no effect on CDK5RAP2 localization , indicating that CDK5RAP2 has a unique role in localizing other MCPH-associated protein to centrosomes ( Figure 1C , and quantitated in Figure 1—figure supplement 6A ) . The stability of CEP152 , WDR62 and CEP63 was unaltered in CDK5RAP2-depleted cells , indicating that their mislocalization was not due to degradation ( Figure 1—figure supplements 3D , 6A ) . CEP152 has previously been shown to be essential for the centrosomal localization of CEP63 ( Brown et al . , 2013; Lukinavicius et al . , 2013 ) . We confirmed this finding and extended it by finding that depletion of CEP152 also prevented WDR62 from localizing to the centrosome ( Figure 1C , and quantitated in Figure 1—figure supplement 6A ) . Consistent with a hierarchy of MCPH-assocated proteins , only CEP63 failed to localize to the centrosome in the absence of WDR62 , while CDK5RAP2 , CEP152 and WDR62 all properly localized to centrosomes in the absence of CEP63 ( Figure 1C , and quantitated in Figure 1—figure supplement 6A ) . Similar to the depletion of CDK5RAP2 , the protein levels of all four proteins were unaltered in these MCPH protein knockdown cells ( Figure 1—figure supplement 5A ) . Thus , CEP152 , WDR62 and CEP63 have progressively more restricted roles in MCPH protein localization to centrosomes than does CDK5RAP2 ( Figure 1D ) . Previous studies have reported that CEP152 and CEP63 depend on each other to localize to the centrosome ( Sir et al . , 2011; Brown et al . , 2013; Lukinavicius et al . , 2013 ) . In support of these findings , we confirmed that CEP63 failed to localize to the centrosome in CEP152-depleted cells with no detectable changes in protein stability ( Figure 1C and Figure 1—figure supplement 5A ) . Contrary to previous findings , we did not detect a decrease of centrosomal CEP152 in CEP63-depleted cells ( Figure 1C and Figure 1—figure supplement 6A ) . A CEP63 siRNA described by Brown et al . ( CEP63 NB ) efficiently depleted CEP63 and disrupted centriole duplication ( Figure 1—figure supplement 7A , B ) . Surprisingly , CEP63 NB also reduced the centrosomal localization of CDK5RAP2 , CEP152 and WDR62 ( Figure 1—figure supplement 7C , D ) . CEP63 NB also reduced the protein levels of CDK5RAP2 , CEP152 and WDR62 ( Figure 1—figure supplement 7E , F ) . Whereas CEP63 NB destabilized CEP152 , the other CEP63 siRNAs used in this study did not , suggesting that depletion of CEP63 may not necessarily lead to CEP152 destabilization and reduced centrosomal CEP152 . Taken together , these results suggest that the centrosomal localization of CEP152 , WDR62 and CEP63 depends on CDK5RAP2 , localization of WDR62 and CEP63 depends on CEP152 , and localization of CEP63 depends on WDR62 , indicating a hierarchical localization scheme . As loss of any of these MCPH proteins disrupts centriole duplication or stability , we propose that the ordered accumulation of MCPH proteins at the centrosome is critical for centriole duplication or stability . In addition to MCPH-associated proteins , CDK5RAP2 , CEP152 , WDR62 and CEP63 copurified with other proteins ( Supplementary file 1 ) . Some of these co-purifying proteins , including CEP72 , CEP131 , CEP90 , KIAA0753 and CCDC14 , localize to centriolar satellites ( Oshimori et al . , 2009; Jakobsen et al . , 2011; Kim and Rhee , 2011; Hall et al . , 2013; Firat-Karalar et al . , 2014 ) . MCPH copurifying proteins also included SPAG5 ( also called ASTRIN ) , a mitotic spindle component , which has not been previously reported to localize to interphase centrosomes ( Gruber et al . , 2002; Thein et al . , 2007 ) . CEP72 , SPAG5 CEP131 , CEP90 , and CCDC14 all colocalize with PCM1 at centriolar satellites throughout the cell cycle ( Figure 2A ) . 10 . 7554/eLife . 07519 . 011Figure 2 . CEP72 , SPAG5 , CEP131 , MNR , CEP90 and CCDC14 are centriolar satellite components . ( A ) Immunofluorescence microscopy of asynchronously growing HeLa cells co-stained with antibodies to Centrin ( ‘c’ , white ) , PCM1 ( green ) , CEP72 ( red ) , SPAG5 ( red ) , CEP131 ( red ) , MNR ( red ) , CEP90 ( red ) , CCDC14 ( red ) . Cell cycle stage was indicated by the number of Centrin ( ‘c’ , green ) foci and DNA condensation ( Hoechst , blue ) . The insets show magnified images of the boxed regions . Scale bars indicate 5 μm for all images . ( B ) HeLa total cell lysates were subjected to immunoprecipitation of PCM1 and c-Myc ( IgG ) , which served as a negative control throughout . Precipitating proteins were subjected to immunoblotting for PCM1 , CEP72 , SPAG5 , CEP131 , MNR , CEP90 , and CCDC14 . ( C–H ) Reciprocal immunoprecpitation of PCM1 and copurified proteins . Endogenous proteins were immunoprecipitated with antibodies to CEP72 , SPAG5 , CEP131 , MNR , CEP90 , CCDC14 , and a negative control IgG . Complexes were immunoblotted with antibodies to PCM1 , CEP72 , SPAG5 , CEP131 , MNR , CEP90 , and CCDC14 . Scale bars indicate 5 μm for all images . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 01110 . 7554/eLife . 07519 . 012Figure 2—figure supplement 1 . CDK5RAP2 , CEP152 , WDR62 and CEP63 , but not γ-tubulin , localize to the centrosome in a microtubule-dependent manner . ( A ) HeLa cells treated with DMSO or 17 μM nocodazole were analyzed by immunofluorescence for PCM1 ( ‘p’ , green ) , CDK5RAP2 ( red ) , CEP152 ( red ) , WDR62 ( red ) , CEP63 ( red ) and γ-tubulin ( red ) . Scale bars indicate 5 μm for all images . ( B ) Quantification of the relative mean fluorescence intensities of CDK5RAP2 , CEP152 , WDR62 , CEP63 and γ-tubulin at the centrosome ( gray ) and acentrosomal ( as defined by colocalization with the centriolar satellite component PCM1; blue , red , orange , green and teal ) in DMSO or nocodazole-treated cells . For all quantifications , 15 cells were analyzed per experiment ( n = 3 ) . p < 0 . 005 ( paired t-test ) statistically significant differences from DMSO-treated controls are denoted by an asterisk . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 012 As KIAA0753 similarly localized to satellites , we refer to it as MOONRAKER ( MNR ) , a satellite from the Ian Fleming novel of the same name ( Figure 2A ) . In contrast to PCM1 and other centriolar satellite proteins , SPAG5 also localizes to the microtubules of the mitotic spindle . The spindle localization of SPAG5 may reflect its ability to directly bind microtubules and organize the spindle during mitosis ( Gruber et al . , 2002; Fitzgerald et al . , 2006; Yuan et al . , 2009; Schmidt et al . , 2010 ) . As CEP72 , SPAG5 , CEP131 , MNR , CEP90 and CCDC14 localize to centriolar satellites throughout the cell cycle , similar to PCM1 ( Figure 2A ) , we used co-immunoprecipitation of endogenous proteins to investigate whether they interact with the core centriolar satellite component PCM1 . We confirmed previous reports that CEP72 , CEP131 , MNR , CEP90 and CCDC14 associate with PCM1 ( Figure 2B , Kim and Rhee , 2011; Staples et al . , 2012; Stowe et al . , 2012; Hall et al . , 2013; Firat-Karalar et al . , 2014 ) . In addition , we identified SPAG5 as a PCM1 interacting protein ( Figure 2B ) . Reciprocal immunoprecipitations of these new satellite proteins confirmed that they interact with PCM1 , but not with another unrelated centrosomal protein , CP110 ( Figure 2C–H ) . The colocalization with PCM1 and the physical association with PCM1 indicate that SPAG5 , like CEP72 , CEP131 , MNR , CEP90 and CCDC14 , is a centriolar satellite component . Previous studies have shown that centriolar satellites bring cargo proteins to centrosomes in a way that depends on intact microtubules ( Dammermann and Merdes , 2002; Kodani et al . , 2010 ) . After brief treatment with nocodazole to depolymerize microtubules , PCM1 forms large cytoplasmic aggregates together with satellite cargo proteins ( Dammermann and Merdes , 2002 ) . Like satellite cargo proteins , microtubule depolymerization caused a portion of CDK5RAP2 , CEP152 , WDR62 and CEP63 to relocalize to PCM1 aggregates ( Figure 2—figure supplement 1A , B ) . In contrast , γ-tubulin did not accumulate at the nocodazole-induced PCM1 aggregates , suggesting that not all centrosomal proteins move from centriolar satellites to the centrosome in a microtubule-dependent manner ( Figure 2—figure supplement 1A , B ) . The accumulation of four MCPH-associated proteins in PCM1 aggregates upon nocodazole treatment raised the possibility that they are cargoes that centriolar satellites help transport to the centrosome . Tests of this hypothesis are described in the following four sections . Mass spectrometry of CDK5RAP2 coprecipitating proteins identified the centriolar satellite proteins , CEP72 and SPAG5 ( Supplementary file 1 ) . We confirmed that the endogenous CDK5RAP2 protein interacts with CEP72 and SPAG5 by reciprocal immunoprecipitation ( Figure 3A , B ) . As CDK5RAP2 is part of a complex with other MCPH-associated proteins , we examined whether CEP72 and SPAG5 also interacted with CEP152 , WDR62 and CEP63 . Interestingly , immunoprecipitation of CEP72 and SPAG5 did not co-precipitate CEP152 , WDR62 or CEP63 , suggesting that CEP72 and SPAG5 interact with CDK5RAP2 , but not the other MCPH-associated proteins ( Figure 3—figure supplement 1A ) . In addition to interacting with CDK5RAP2 , we found that CEP72 and SPAG5 immunoprecipitated each other ( Figure 3C ) . 10 . 7554/eLife . 07519 . 013Figure 3 . Centriolar satellite components CEP72 and SPAG5 stabilize and localize CDK5RAP2 to the centrosome to promote centriole duplication . ( A ) HeLa total cell lysates were subjected to immunoprecipitation of CEP72 , SPAG5 , and a negative control . Precipitating proteins were subjected to immunoblotting for CEP72 , SPAG5 and CDK5RAP2 . ( B ) The interactions between CEP72 and SPAG5 with CDK5RAP2 were validated by immunoprecipitation using an antibody to CDK5RAP2 . Endogenous CDK5RAP2 was immunoprecipitated and complexes were immunoblotted with antibodies to CEP72 , SPAG5 and CDK5RAP2 . ( C ) Immunoprecipitated CEP72 , SPAG5 , and a negative control were immunoblotted using antibodies to CEP72 , SPAG5 and a negative control . ( D ) HeLa cells in S phase transfected with SC , CEP72 #1 or SPAG5 siRNA were co-stained for CDK5RAP2 ( red ) and Centrin ( ‘c’ , green ) to visualize centrioles . ( E ) Total cell lysates of SC , CEP72 #1 , and SPAG5 siRNA transfected HeLa cells were analyzed by immunoblotting with antibodies to CEP72 , SPAG5 , and CDK5RAP2 . Actin served as a loading control . ( F ) SC , CEP72 #1 , CEP72 #2 siRNA transfected HeLa cells were co-stained with CEP72 ( red ) and Centrin ( ‘c’ , green ) to visualize centrioles . ( G ) Percentage of S phase cells with four Centrin foci in SC , CEP72 #1 , CEP72 #2 siRNA treated HeLa cells . Error bars represent ±s . d . throughout . ( H ) S phase HeLa cells transfected with SC or SPAG5 siRNA co-stained with Centrin ( ‘c’ , green ) and SPAG5 ( red ) . ( I ) Quantification of S phase SC and SPAG5-depleted HeLa cells with four Centrin foci . ( J ) Our findings suggest that CEP72 and SPAG5 are required to localize CDK5RAP2 to the centrosome , which itself recruits CEP152 , WDR62 and CEP63 . Centrosome localized MCPH proteins are represented in red and centriolar satellites in blue . For all quantifications at least 100 cells were counted per experiment ( n = 3 ) , p < 0 . 005 ( paired t-test ) for SC vs CEP72 #1 , #2 and SPAG5 siRNA transfected cells . Scale bars indicate 5 μm for all images . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 01310 . 7554/eLife . 07519 . 014Figure 3—figure supplement 1 . CEP72 and SPAG5 interact with CDK5RAP2 but not CEP152 , WDR62 or CEP63 . ( A ) HeLa total cell lysates were immunoprecipitated with antibodies to CEP72 and SPAG5 . Co-precipitating proteins were probed with antibodies to CEP72 , SPAG5 , CDK5RAP2 , CEP152 , WDR62 , and CEP63 . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 01410 . 7554/eLife . 07519 . 015Figure 3—figure supplement 2 . CEP72 , SPAG5 and CDK5RAP2 promote centriole duplication and centrosome organization in U2OS cells . ( A ) SC , CDK5RAP2 , CEP72 and SPAG5-depleted S phase U2OS cells were co-stained with Centrin ( ‘c’ , green ) , and CDK5RAP2 ( red ) , SPAG5 ( red ) or CEP72 ( red ) . ( B ) Total cell lysate from SC , CEP72 and SPAG5 siRNA-treated U2OS cells were analyzed by western blotting using antibodies to CDK5RAP2 , CEP72 , and SPAG5 . Actin served as a loading control . ( C ) S phase SC , CDK5RAP2 , CEP72 , and SPAG5 siRNA-treated U2OS cells were co-stained with Centrin ( ‘c’ , green ) , CDK5RAP2 ( red ) , CEP152 ( red ) , WDR62 ( red ) , and CEP63 ( red ) . Scale bars indicate 5 μm for all images . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 01510 . 7554/eLife . 07519 . 016Figure 3—figure supplement 3 . CEP72 and SPAG5 are required to localize CDK5RAP2 , CEP152 , WDR62 and CEP63 to the centrosome . ( A ) S phase SC , CEP72 #1 , and SPAG5 siRNA-treated HeLa cells were co-stained with Centrin ( ‘c’ , green ) , CDK5RAP2 ( red ) , CEP152 ( red ) , WDR62 ( red ) , and CEP63 ( red ) . Scale bar indicates 5 μm . ( B ) Quantification of the intensities of centrosomal CDK5RAP2 , CEP152 , WDR62 and CEP63 in SC , SPAG5 and CEP72 siRNA-treated cells expressed as the mean percentage ±s . d . of the intensities of SC samples . For fluorescence quantifications 30 cells were analyzed per experiment ( n = 3 ) . p < 0 . 005 ( paired t-test ) statistically significant differences are denoted by an asterisk . ( C ) Total cell lysates of SC SPAG5 and CEP72 siRNA treated cells were analyzed by immunoblot with antibodies to CDK5RAP2 , CEP152 , WDR62 and CEP63 . Actin served as a loading control . Asterisk indicates the specific band . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 016 To begin to test our hypothesis that centriolar satellite components help MCPH-associated proteins localize to the centrosome , we assessed whether CEP72 and SPAG5 were required to localize CDK5RAP2 to the centrosome . In HeLa cells depleted of either CEP72 or SPAG5 , there was a sharp decrease in the localization of CDK5RAP2 at the centrosome ( Figure 3D ) . We also found that the protein levels of CDK5RAP2 were reduced in CEP72 and SPAG5-depleted cells ( Figure 3E ) . Interestingly , in addition to destabilizing CDK5RAP2 , the stability of CEP72 and SPAG5 was reduced in SPAG5 and CEP72-depleted cells , respectively ( Figure 3E ) . As in HeLa cells , CEP72 and SPAG5 were essential for stabilizing CDK5RAP2 and localizing it to the centrosome in U2OS cells ( Figure 3—figure supplement 2A–C ) . These findings suggest that the centriolar satellite proteins CEP72 and SPAG5 interact with CDK5RAP2 , stabilize each other and CDK5RAP2 , and promote the localization of CDK5RAP2 to the centrosome . Previous studies have shown that CEP72 and SPAG5 are each required for centrosome maturation during formation of the mitotic spindle ( Oshimori et al . , 2009; Schmidt et al . , 2010 ) . Given that CEP72 and SPAG5 are required for CDK5RAP2 centrosomal localization , we hypothesized that , during interphase , CEP72 and SPAG5 may be important for centriole biogenesis . Therefore , we examined whether depletion of either CEP72 or SPAG5 phenocopies the effects of CDK5RAP2 depletion on centriole duplication . Similar to the loss of CDK5RAP2 , greater than 78% of CEP72-depleted and 83% of SPAG5-depleted S phase cells had fewer than four centrioles , indicating that these proteins may function together with CDK5RAP2 to promote centriole duplication or stability ( Figure 3F–I ) . Because CEP72 and SPAG5 are required for the centrosomal localization of CDK5RAP2 , and CDK5RAP2 is required for the centrosomal localization of other MCPH-associated proteins CEP152 , WDR62 and CEP63 , we examined whether the loss of SPAG5 or CEP72 altered the localization of the CDK5RAP2-dependent MCPH proteins . Similar to the depletion of CDK5RAP2 , depletion of SPAG5 or CEP72 caused CEP152 , WDR62 and CEP63 localization to become significantly decreased at S phase centrosomes of both HeLa and U2OS cells ( Figure 3—figure supplements 2C , 3A–C ) . Thus , the centriolar satellite components SPAG5 and CEP72 are required for the centrosomal localization of CDK5RAP2 and its dependent proteins and the promotion of centriole duplication ( Figure 3J ) . Mass spectrometry of CEP152 coprecipitating proteins suggested that CEP152 interacts with CEP131 , a centriolar satellite protein involved in genomic stress responses and ciliogenesis ( Supplementary file 1B , Staples et al . , 2012; Hall et al . , 2013; Villumsen et al . , 2013 ) . We confirmed that CEP152 and CEP131 interact by reciprocally co-immunprecipitating the endogenous proteins ( Figure 4A ) . We hypothesized that the relationship between CEP152 and its satellite interactor , CEP131 , might parallel the relationship we had identifed for CDK5RAP2 and its centriolar satellite interactors , CEP72 and SPAG5 . Therefore , we examined whether CEP131 was required for the centrosomal localization of CEP152 . Interestingly , CEP152 was greatly reduced at the centrioles in CEP131-depleted cells ( Figure 4B and Figure 4—figure supplement 1A ) . Unlike the relationship between CDK5RAP2 , CEP72 and SPAG5 , The loss of CEP152 from the centrosome of CEP131-depleted cells was not due to destabilization , as CEP152 protein levels were unchanged by CEP131 knockdown ( Figure 4C ) . Similar to the siRNA-mediated depletion , we observed disrupted centrosomal localization of CEP152 in mouse embryonic fibroblasts ( MEFs ) derived from embryonic day 14 . 5 Cep131gt/gt embryos ( Figure 4—figure supplement 2A , Hall et al . , 2013 ) . 10 . 7554/eLife . 07519 . 017Figure 4 . Centriolar satellite component CEP131 interacts with and localizes CEP152 to the centrosome to promote centriole duplication . ( A ) HeLa total cell lysates were subjected to immunoprecipitation of CEP131 , CEP152 and a negative control . Precipitating proteins were subjected to immunoblotting for CEP131 and CEP152 . ( B ) HeLa cells in S phase transfected with SC or CEP131 #1 siRNA were co-stained for CEP152 ( red ) and Centrin ( ‘c’ , green ) . ( C ) Total cell lysates of SC and CEP131 #1 siRNA transfected HeLa cells were analyzed by immunoblotting with antibodies to CEP131 and CEP152 . Actin served as a loading control . ( D ) HeLa cells in S phase transfected with SC , CEP131 #1 , or CEP131 #2 siRNA were co-stained for CEP131 ( red ) and Centrin ( ‘c’ , green ) to visualize centrioles . ( E ) Total cell lysates of SC , CEP131 #1 , CEP131 #2 siRNA transfected HeLa cells were analyzed by immunoblotting for Cep131 . α-tubulin served as a loading control . 20 μg of protein lysate was loaded per lane . ( F ) Quantification of the mean percentage of SC , CEP131 #1 , or CEP131 #2 siRNA transfected cells in S phase with four centrioles . ( G ) SC and CEP131-depleted S phase cells were co-stained with Centrin ( ‘c’ , green ) , CDK5RAP2 ( red ) , CEP152 ( red ) , WDR62 ( red ) , and CEP63 ( red ) . ( H ) Schematic indicating that CEP131 is required to localize CEP152 , WDR62 and CEP63 to the centrosome . For all quantifications at least 100 cells were counted per experiment ( n = 3 ) , p < 0 . 005 ( paired t-test ) . Scale bars indicate 5 μm for all images . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 01710 . 7554/eLife . 07519 . 018Figure 4—figure supplement 1 . Cep131gt/gt MEFs exhibit centriole duplication and centrosome organizational defects . ( A ) WT and two Cep131gt/gt MEF cell lines were analyzed by immunofluorescence with antibodies to Centrin ( ‘c’ , green ) and γ-tubulin ( red ) . ( B ) Quantification of the number of centrioles in asynchronous populations of WT and two Cep131gt/gt MEFs . ( C ) Immunofluorescence of WT and Cep131gt/gt MEFs co-stained with Centrin ( ‘c’ , green ) and CEP152 ( red ) . For all quantifications at least 100 cells were counted per experiment ( n = 3 ) . p < 0 . 005 ( paired t-test ) compared to WT was denoted with an asterisk . Scale bars indicate 5 μm for all images . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 01810 . 7554/eLife . 07519 . 019Figure 4—figure supplement 2 . CEP131 is required for the centrosomal localization of CEP152 , WDR62 and CEP63 . ( A ) Quantification of the mean fluorescence intensities ±s . d . of CDK5RAP2 , CEP152 , WDR62 and CEP63 in SC or CEP131-depleted cells expressed as the mean percentage ±s . d . of the fluorescence intensities of SC cells . For all quantifications 30 cells were analyzed per experiment ( n = 3 ) . p < 0 . 005 ( paired t-test ) statistically significant differences are denoted by an asterisk . ( B ) Total cell lysate from SC and CEP131 siRNA-treated HeLa cells were analyzed by immunoblotting using antibodies to CEP131 , CDK5RAP2 , CEP152 , WDR62 , and CEP63 . Actin served as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 01910 . 7554/eLife . 07519 . 020Figure 4—figure supplement 3 . CEP152 associates with CDK5RAP2 and WDR62 in a CEP131-dependent manner . ( A ) Total cell lysate from SC and CEP131-depleted HeLa cells were subjected to immunoprecipitation using an antibody to CEP152 . Co-precipitating proteins were analyzed by immunoblotting for CDK5RAP2 and WDR62 . SC and CEP131 siRNA transfected HeLa cell total cell lysates were analyzed by immunoblotting with antibodies to CDK5RAP2 , CEP152 , WDR62 and CEP131 . Actin served as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 020 To investigate whether the failure of CEP152 to localize to centrosomes in CEP131-depleted cells compromises the interaction of CEP152 with its interacting MCPH-associated proteins , we immunoprecipitated endogenous CEP152 from control and CEP131-depleted cells and examined whether CDK5RAP2 and WDR62 co-precipitated . Although depletion of CEP131 had no effect on CDK5RAP2 , CEP152 or WDR62 stability , CEP131 was required for CEP152 to co-precipitate either CDK5RAP2 or WDR62 ( Figure 4—figure supplement 3 ) . The finding that the centriolar satellite-mediated centrosomal localization of CEP152 is critical for its interaction with its MCPH-associated partner proteins is consistent with the idea that the MCPH-associated proteins interact specifically at the centrosome . As CEP152 is required for centriole duplication , we examined whether its centriolar satellite interactor , CEP131 , is also required for centriole duplication . Depletion of CEP131 in HeLa cells and examination of Centrin and CEP131 revealed that greater than 84% of CEP131-depleted cells in S phase failed to properly duplicate their centrioles ( Figure 4D , F ) . Early passage Cep131gt/gt MEFs displayed similar defects in centriole duplication ( Figure 4—figure supplement 2B , C ) . As depletion of CEP131 disrupted the centrosomal accumulation of CEP152 and CEP152 is required for the centrosomal localization of WDR62 and CEP63 , we assessed whether the centrosomal localization of MCPH-associated proteins depends upon CEP131 . The depletion of CEP131 had no effect on the localization of CDK5RAP2 , consistent with our finding that CEP152 is not required for CDK5RAP2 localization ( Figure 4G ) . In addition to CEP152 , depletion of CEP131 abrogated the centrosomal localization , but not the stability , of the CEP152-dependent MCPH-associated proteins , WDR62 and CEP63 ( Figure 4G and quantitated in Figure 4—figure supplement 2A , B ) . Together , these findings demonstrate that the centriolar satellite protein CEP131 is critical for bringing CEP152 to the centrosome , thereby allowing CEP152 to participate in centriole duplication ( Figure 4H ) . Given the requirement of WDR62 for centriole duplication , we analyzed WDR62 co-precipitating proteins by mass spectrometry and identified MNR ( also called KIAA0753 ) , recently described as a centriolar satellite component involved in centriole duplication ( Jakobsen et al . , 2011; Firat-Karalar et al . , 2014 ) . We confirmed the interaction between MNR and WDR62 by reciprocal co-immunoprecipitation ( Figure 5A ) . Immunoblotting and immunofluorescence revealed a loss of MNR in siRNA-treated lysates and cells ( Figure 5B , C ) . We also confirmed that MNR is required to promote centriole duplication by depleting the protein and assessing centriole duplication and centrosome organization ( Firat-Karalar et al . , 2014 ) . Depletion of MNR revealed that more than 73% of S phase cells had fewer than four centrioles ( Figure 5D ) , a phenotype similar to that caused by the depletion of WDR62 or CEP63 . 10 . 7554/eLife . 07519 . 021Figure 5 . Satellite component MNR promotes centriole duplication by localizing WDR62 to the centrosome . ( A ) We immunoprecipitated endogenous WDR62 and MNR from HeLa total cell lysates . Co-precipitation was detected using antibodies specific to WDR62 and MNR . ( B ) SC , MNR #1 , and MNR #2 siRNA transfected S-phase HeLa cells were co-stained with MNR ( red ) and Centrin ( ‘c’ , green ) . ( C ) Total cell lysate of SC , MNR #1 , MNR #2 siRNA transfected HeLa cells were analyzed by immunoblotting with antibodies to MNR . The asterisk marks the band specific to MNR , which sits below a non-specific band . Actin served as a loading control . ( D ) Percentage of S-phase SC , MNR #1 , MNR #2 siRNA treated HeLa cells with four Centrin foci . ( E ) S-phase SC and MNR siRNA-transfected HeLa cells were co-stained for Centrin ( ‘c’ , green ) and CEP63 ( red ) . ( F ) SC and MNR siRNA treated S-phase HeLa cells were co-stained for WDR62 ( red ) and Centrin ( ‘c’ , green ) . ( G ) Total cell lysates of SC and MNR-depleted cells were analyzed by immunoblot with antibodies to WDR62 and MNR . Actin served as a loading control . 20 μg of protein lysate was loaded per lane . ( H ) SC and MNR #1-depleted S-phase cells were co-stained with Centrin ( ‘c’ , green ) , CDK5RAP2 ( red ) , CEP152 ( red ) , and WDR62 ( red ) . ( I ) Our findings indicate that MNR localizes WDR62 to the centrosome , which in turn recruits CEP63 . For all quantifications at least 100 cells were counted per experiment ( n = 3 ) , p < 0 . 005 ( paired t-test ) . Scale bars indicate 5 μm for all images . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 02110 . 7554/eLife . 07519 . 022Figure 5—figure supplement 1 . MNR is required to localize WDR62 and CEP63 to the centrosome . ( A ) Quantification of the mean fluorescence intensities of centrosomal CDK5RAP2 , CEP152 , WDR62 and CEP63 in SC or MNR-depleted cells expressed as the mean percentage ±s . d . of the fluorescence intensities of the centrosomal signal of SC cells . For all quantifications 30 cells were analyzed per experiment ( n = 3 ) . p < 0 . 005 ( paired t-test ) statistically significant differences are denoted by an asterisk . ( B ) SC and MNR siRNA transfected HeLa cell total cell lysates were analyzed by immunoblotting with antibodies to MNR , CDK5RAP2 , CEP152 , WDR62 and CEP63 . Actin served as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 02210 . 7554/eLife . 07519 . 023Figure 5—figure supplement 2 . WDR62 interacts with CEP152 and CEP63 in a MNR-dependent manner . ( A ) We immunoprecipitated endogenous CEP63 from SC and MNR-depleted HeLa total cell lysates . Precipitation and co-precipitation were detected using antibodies specific to CEP152 , WDR62 , and CEP63 . Total cell lysates from SC and MNR depleted cells were analyzed by immunoblotting with antibodies to CEP152 , WDR62 , CEP63 , and MNR . Actin served as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 023 Although overexpressed MNR interacts with CEP63 and MNR is essential for CEP63 localization to centrosomes ( Figure 5E and Figure 5—figure supplement 1A , Firat-Karalar et al . , 2014 ) , we did not detect an interaction between endogenous MNR and CEP63 ( data not included ) . As we had found that , instead , MNR interacts with WDR62 ( Figure 5A ) , and WDR62 is required to localize CEP63 to the centrosome , we predicted that MNR would be required to localize WDR62 to centrosomes . Indeed , WDR62 was absent from centrosomes in MNR-depleted cells , indicating MNR is required for the centrosomal localization of WDR62 ( Figure 5F ) . As a centriolar satellite protein , CEP131 , is required for CEP152 to localize to the centrosome and interact with other MCPH-associated proteins , we investigated whether , similarly , WDR62 required its cognate centriolar satellite protein to interact with the MCPH-associated proteins CEP152 and CEP63 at the centrosome . Therefore , we immunoprecipitated endogenous WDR62 from control and MNR-depleted cells , and found that , in the absence of MNR , WDR62 no longer co-precipitated CEP152 or CEP63 ( Figure 5—figure supplement 2A ) , suggesting that assembling WDR62 together with other MCPH-associated proteins requires its centriolar satellite-mediated localization to the centrosome . Similar to the relationship between CEP131 and CEP152 , depletion of MNR did not destabilize its associated MCPH protein , WDR62 ( Figure 5G ) . Consistent with a specific role in WDR62 localization , depletion of MNR did not affect CDK5RAP2 and CEP152 localization to the centrosome or its protein stability ( Figure 5H and quantified in Figure 5—figure supplement 1A , B ) . Thus , loss of MNR phenocopies loss of WDR62 , suggesting that this centriolar satellite component has a specific role in localizing WDR62 to the centrosome , which is subsequently requlred for the centrosomal localization of CEP63 and centriole duplication ( Figure 5I ) . Cooperation between the MCPH-associated proteins culminates in CEP63 localization to centrosomes ( Figure 1 ) . As we had done with the other MCPH proteins , we subjected CEP63 and its co-immunoprecipitated proteins to analysis by mass spectrometry . Similar to other MCPH proteins , we identified a centriolar satellite component , CEP90 , as a candidate interactor of CEP63 ( Supplementary file 1D ) . Reciprocal immunprecipitation of endogenous proteins confirmed the interaction of CEP63 and CEP90 ( Figure 6A ) . 10 . 7554/eLife . 07519 . 024Figure 6 . CEP90 encodes a centriolar satellite and MCPH-associated protein required to localize CEP63 to the centrosome to promote centriole duplication . ( A ) HeLa total cell lysate was subjected to endogenous immunoprecipitation using antibodies to CEP63 , CEP90 and a negative control . Precipitating proteins were analyzed by immunoblotting for CEP63 and CEP90 . ( B ) Immunofluorescence images of SC and CEP90 #2 siRNA-treated HeLa cells co-stained for CEP63 ( red ) and Centrin ( ‘c’ , green ) . ( C ) Total cell lysates of SC or CEP90 #2 siRNA transfected HeLa cells were analyzed by immunoblotting with antibodies to CEP63 , CEP90 and Actin , which served as a loading control . Asterisk indicates the specific band . ( D ) S phase SC , CEP90 #1 , and CEP90 #2 siRNA-treated cells were analyzed by immunofluorescence with antibodies to CEP90 ( red ) and Centrin ( ‘c’ , green ) . ( E ) Total cell lysates of SC , CEP90 #1 , and CEP90 #2 siRNA transfected HeLa cells were analyzed by immunoblotting for CEP90 . α-tubulin served as a loading control . ( F ) Mean percentage of S phase SC , CEP90 #1 , and CEP90 #2 siRNA-treated HeLa cells with four Centrin foci . At least 100 cells were counted per experiment ( n = 3 ) , p < 0 . 005 ( paired t-test ) . ( G ) SC and CEP90 #2-depleted S phase cells were co-stained with Centrin ( ‘c’ , green ) , CDK5RAP2 ( red ) , CEP152 ( red ) , WDR62 ( red ) , and CEP63 ( red ) . Scale bars indicate 5 μm for all images . ( H ) Schematic indicating that CEP90 is required for the centrosomal localization of CEP63 . ( I ) Diagram of a simplified pedigree . Filled squares indicate individuals with MCPH . Spontaneous abortions ( SAB ) , stillbirths ( SB ) , and individuals of unknown gender ( diamonds ) are also indicated . ( J ) Sanger sequencing of the affected patients and their unaffected mother confirms the presence of a guanine to cytosine mutation leading to a charge reversing glutamic acid to glutamine substitution . This variant was identified by whole exome sequencing of the affected individuals and the unaffected mother . ( K ) We immunoprecipitated the FLAG tags of the wild-type and E89Q forms of CEP90 and blotted for endogenous CEP63 and FLAG . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 02410 . 7554/eLife . 07519 . 025Figure 6—figure supplement 1 . CEP90 is required to localize CEP63 to the centrosome . ( A ) The centrosomal fluorescence intensity of CDK5RAP2 , CEP152 , WDR62 and CEP63 were quantified in SC or CEP90-depleted HeLa cells . Quantification is expressed as mean percentage ±s . d . of the fluorescence intensities of SC cells . For all quantifications 30 cells were analyzed per experiment ( n = 3 ) . p < 0 . 005 ( paired t-test ) statistically significant differences are denoted by an asterisk . ( B ) Total cell lysate from SC and CEP90-depleted cells were subjected to immunoblotting using antibodies to CEP90 , CDK5RAP2 , CEP152 , WDR62 , and CEP63 . Actin served as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 02510 . 7554/eLife . 07519 . 026Figure 6—figure supplement 2 . CEP63 interacts with WDR62 in a CEP90-dependent manner . ( A ) HeLa total cell lysate from SC and CEP90-depleted cells were subjected to endogenous immunoprecipitation using an antibody to CEP63 and a negative control . Precipitating proteins were analyzed by immunoblotting for CEP63 and WDR62 . Total cell lysates from SC and CEP90 siRNA transfected HeLa cells were analyzed by immunoblotting with antibodies to WDR62 , CEP63 , and CEP90 . Actin served as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 02610 . 7554/eLife . 07519 . 027Figure 6—figure supplement 3 . CEP90 human genetics . ( A ) Schematic diagram showing linkage and homozygosity mapping in MCPH-affected individuals MC-701 and MC-702 . Regions labeled with red brackets denote homozygosity in MC-701 . Blue brackets denote homozygosity in MC-702 . The purple oval highlights the region of shared homozygosity between the two patients . ( B ) 57 Mb block of overlapping homozygosity on chromosome 13 , circled in ( A ) . ( C ) Calculation of logarithmic odds ( LOD ) score over the block of overlapping homozygosity ( g . chr13:30975286-88096410 in hg19 coordinates ) yields a parametric , multipoint LOD score of 1 . 8057 . ( D ) Phylogenetic analysis of CEP90 ( using BLAST mutual best match ) identified CEP90 orthologs in unicellular organisms . Analysis of the amino-terminal portion of these proteins confirms that CEP90 is ancient and that the variant affects a highly conserved residue ( E89 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 027 We assessed whether CEP90 participated in the centrosomal localization of CEP63 . Indeed , CEP63 was absent from the centrosome in CEP90-depleted cells ( Figure 6B and quantified in Figure 6—figure supplement 1A ) , suggesting that CEP63 localizes to centrioles in a CEP90-dependent manner . The levels of CEP63 were unchanged upon depletion of CEP90 , indicating that CEP90 is essential for CEP63 centrosomal localization , but not its stability ( Figure 6C ) . Thus , each of the interacting MCPH-associated proteins has an associated centriolar satellite component that it requires for localization to centrosomes . Given the requirement for other centriolar satellite proteins in assembling their cognate MCPH-associated proteins into a larger complex , we examined whether CEP90 is essential for CEP63 to associate with an interacting MCPH-associated protein , WDR62 . We immunoprecipitated endogenous CEP63 from control and CEP90-depleted cells , and discovered that CEP90 is essential for CEP63 to co-precipitate WDR62 ( Figure 6—figure supplement 2A ) . These results are consistent with the idea that centriolar satellite components bring select MCPH-associated proteins to the centrosome , and the MCPH-associated proteins form a complex specifically at the centrosome . Previous studies have implicated CEP90 in ciliogenesis and mitotic spindle pole integrity ( Kim and Rhee , 2011; Kim et al . , 2012 ) . Whether it has additional centrosomal functions has been unclear . We hypothesized that , like CEP63 , CEP90 may also participate in centriole duplication . To investigate the role of CEP90 in centriole biogenesis , we depleted CEP90 by siRNA and examined whether centriole duplication is abrogated . Immunofluorescence analysis demonstrated that greater than 78% of S phase CEP90-depleted cells had fewer than four centrioles ( Figure 6D–F ) . These findings were similar to the centriole duplication defect we detected in CEP63-depleted cells , suggesting that CEP90 and CEP63 function together to promote centriole duplication . As CEP63 is not required for the centrosomal localization of the other MCPH proteins CDK5RAP2 , CEP152 and WDR62 , we predicted that CEP90 would similarly be dispensable for their localization . Immunofluorescence and quantitative analysis of HeLa cells treated with CEP90 siRNA revealed that the centrosomal localization and protein stability of CDK5RAP2 , CEP152 , and WDR62 were unaltered ( Figure 6G and quantified in Figure 6—figure supplement 1A , B ) . Collectively , these results reveal that CEP63 localizes to the centrosome in a manner dependent on the centriolar satellite protein CEP90 and the MCPH proteins CDK5RAP2 , CEP152 and WDR62 as well as the other centriolar satellite proteins CEP72 , SPAG5 , CEP131 and MNR ( Figure 6H ) . Our finding that centriolar satellite proteins participate in the cellular functions of established MCPH-associated proteins raised the possibility that mutations in genes encoding centriolar satellite components might also cause MCPH . Examination of whole exome sequence from >300 families with MCPH identified a homozygous missense variant in CEP90 in a consanguineous Pakistani family with two affected male siblings . The two affected boys ( MC-701 and MC-702 ) were born to parents who were first cousins and who had a total of 12 pregnancies ( Figure 6I ) . In addition to their two affected sons , they had four unaffected children , five miscarriages and one stillbirth at 6 months of gestation with multiple congenital anomalies . Both affected boys presented with a similar syndrome of microcephaly , intellectual disability , spasticity , and dysmorphisms . At birth , MC-701 measured 2 . 98 kg ( 18th percentile ) for weight , 50 cm ( 48th percentile ) for length and 33 cm ( 10th percentile ) for head circumference . Global developmental delay was noted at 5 months of age; he toe walked at 3 years , later developed tetraplegia , and is wheelchair-bound . He exhibits spasticity especially in the lower extremities with scissoring and Achilles contractures , hyperreflexia , ankle clonus and a positive Babinksi sign . He suffers from severe intellectual disability and does not speak . His dysmorphic features include hypertelorism , large ears , broad nose , anteverted nares , thick lips and a large mouth , as well as pectus carinatum and kyphoscoliosis . His brother , MC-702 , has a similar phenotype with a head circumference of 48 . 5 cm ( 4 standard deviations below the mean ) at 14 years of age . Brain magnetic resonance imaging noted normal gross morphology with atrophy . Whole exome analysis suggested that CEP90 is one of two strong candidates to contribute to the phenotype observed in this family . Both affected boys possess a homozygous missense alteration in the second coding exon of the CEP90 gene ( chr13:73366597G>C [hg19] ) , resulting in p . E89Q ( Figure 6J ) . The observed variant lies in the midst of a 57 Mb stretch of homozygosity on chr13 implicated by linkage and homozygosity mapping to likely harbor the disease-causing allele ( Figure 6—figure supplement 3A–C ) . CEP90 p . E89Q has not been detected in dbSNP138 , the 1000 Genomes Project , the ESP6500 databases , or >1000 exomes from 467 Middle Eastern families . Moreover , no other polymorphisms affecting E89 have been reported in these databases . E89 is highly conserved , including 9/9 vertebrate species including Xenopus and zebrafish as well as much more distantly related species ( Figure 6—figure supplement 3D ) . The nonconservative p . E89Q alteration is strongly predicted by SIFT and Polyphen2 to be pathogenic ( SIFT and Polyphen2 scores of 0 and 0 . 99 , respectively ) . Both affected boys also have a homozygous intronic variant in AP4S1 ( rs185246578 , NM_001254729:exon5:c . 295-3C>A ) , a gene in which truncating mutations have been reported to cause a phenotype that overlaps that described in this family ( Hardies et al . , 2015 ) . The variant detected in this family ( rs185246578 ) is of uncertain significance , as it is an intronic SNP that was previously reported by the 1000 Genomes Project , is not strongly predicted to be pathogenic by splice prediction algorithms , and lies upstream of exon 5 of 6 , which is skipped in one of the known AP4S1 isoforms . To test whether the observed CEP90 mutation affects protein function , we engineered cDNAs corresponding to wild type and E89Q alleles and expressed the constructs in HeLa cells for 6 hr . We then assessed whether the mutant form of CEP90 was capable of interacting with its binding partners , PCM1 and CEP63 by co-immuoprecipitation . Interestingly , CEP90 E89Q interacted efficiently with PCM1 , but did not bind CEP63 ( Figure 6K ) . Thus , consistent with a role for CEP90 in the localization of the MCPH protein CEP63 , our data suggests that mutations in CEP90 itself may contribute to abnormal human brain development . In addition to CEP90 , our mass spectrometric analysis of CEP63 co-immunoprecipitating proteins identified CCDC14 , another centriolar satellite component . It has previously been demonstrated that tagged versions of CEP63 bind CCDC14 ( Camargo et al . , 2007; Firat-Karalar et al . , 2014 ) . We confirmed that endogenous CEP63 and CCDC14 interact using reciprocal coimmunoprecipitations ( Figure 7A ) . Immunoblot analysis demonstrated that CCDC14 siRNAs efficiently depleted its target protein ( Figure 7B ) . Unlike depleting CEP90 , depleting CCDC14 produced supernumerary Centrin foci in S phase U2OS or HeLa cells ( Figure 7C and data not shown ) . 10 . 7554/eLife . 07519 . 028Figure 7 . Satellite component CCDC14 suppresses the formation of supernumerary Centrin foci by limiting the centrosomal accumulation of CEP63 and activity of CDK2 . ( A ) We immunoprecipitated endogenous CCDC14 and CEP63 from HeLa total cell lysates . Efficient precipitation and co-precipitation were detected using antibodies specific to CCDC14 and CEP63 . ( B ) S phase SC , CCDC14 #1 , and CCDC14 #2 siRNA transfected U2OS cells were analyzed by immunofluorescence with antibodies to CCDC14 ( red ) and Centrin ( ‘c’ , green ) . ( C ) Total cell lysates of SC , CCDC14 #1 , and CCDC14 #2 siRNA-treated U2OS cells were analyzed by immunoblotting for CCDC14 . α-tubulin served as a loading control . ( D ) S phase SC and CCDC14 #2 siRNA-treated U2OS cells were co-stained for Centrin ( ‘c’ , green ) and CEP63 ( red ) . ( E ) Total cell lysate of SC , CCDC14 #2 siRNA transfected U2OS cells were analyzed by immunoblotting with antibodies to CCDC14 and CEP63 . Actin served as a loading control . ( F ) U2OS cells transfected with SC siRNA , or siRNA targeting CCDC14 alone , or CCDC14 siRNA in combination with siRNAs targeting CDK5RAP2 , CEP152 , WDR62 , or CEP63 in S phase were co-stained with Centrin ( ‘c’ , green ) and CCDC14 ( red ) . ( G ) Quantification of S phase SC , CCDC14 , or CCDC14 and CDK5RAP2 , CEP152 , WDR62 , or CEP63 siRNA-treated U2OS cells with fewer than four centrioles . S phase cells were identified by Cyclin A immunostaining . For all quantifications at least 100 cells were counted per experiment ( n = 3 ) , p < 0 . 005 ( paired t-test ) . ( H ) S phase SC and CCDC14 #2 siRNA transfected U2OS cells were co-stained with Centrin ( ‘c’ , green ) and CDK2 ( red ) . ( I ) S phase SC and CCDC14 #2 siRNA transfected U2OS cells treated with DMSO or roscovitine were analyzed by immofluorescence with antibodies to CCDC14 ( red ) and Centrin ( ‘c’ , green ) . Scale bars indicate 5 μm for all images . ( J ) Percentage of S phase SC , CCDC14 #2 siRNA-transfected U2OS cells treated with DMSO or roscovitine with greater than four Centrin foci . ( K ) CDK5RAP2 delivery to the centrosome requires the centriolar satellite proteins CEP72 and SPAG5 . In a manner dependent on CDK5RAP2 , CEP131 localizes CEP152 to the centrosome , MNR localizes WDR62 , and CEP90 localizes CEP63 . The centriolar satellite , CCDC14 binds CEP63 and removes it from the centrosome to limit centriole duplication by limiting the localization and activity of CDK2 . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 02810 . 7554/eLife . 07519 . 029Figure 7—figure supplement 1 . CCDC14 limits the formation of Centrin positive foci that do not recruit the PCM component γ-tubulin . ( A ) Immunofluorescence images of SC and CCDC14 #2 siRNA transfected U2OS cells co-stained for Centrin ( ‘c’ , green ) , γ-tubulin ( red ) , and CP110 ( red ) , to assess association with pericentrosomal matrix and organized centriole distal ends . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 02910 . 7554/eLife . 07519 . 030Figure 7—figure supplement 2 . CCDC14 limits the centrosomal accumulation of CEP63 but does not alter CDK5RAP2 , CEP152 , or WDR62 localization . ( A ) S phase SC and CCDC14 #2 siRNA transfected U2OS cells were co-stained with Centrin ( ‘c’ , green ) , CDK5RAP2 ( red ) , CEP152 ( red ) , WDR62 ( red ) , and CEP63 ( red ) . ( B ) Quantification of the centrosomal fluorescence intensities of CDK5RAP2 , CEP152 , WDR62 and CEP63 in SC or CCDC14-depleted U2OS cells expressed as the mean percentage ±s . d . of the fluorescence intensities of SC cells . For all quantifications 30 cells were analyzed per experiment ( n = 3 ) . p < 0 . 005 ( paired t-test ) statistically significant differences are denoted by an asterisk . ( C ) Total cell lysate from SC and CCDC14-depleted HeLa cells was subjected to immunoblotting using antibodies to CCDC14 , CDK5RAP2 , CEP152 , WDR62 and CEP63 . Actin served as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 03010 . 7554/eLife . 07519 . 031Figure 7—figure supplement 3 . Co-depletion of CCDC14 and MCPH-associated proteins . ( A ) Total cell lysate of SC , CCDC14 , or CCDC14 plus CDK5RAP2 , CEP152 , WDR62 , or CEP63 siRNA transfected U2OS cells were analyzed by immunoblotting with antibodies to CCDC14 , CDK5RAP2 , CEP152 , WDR62 , and CEP63 . Actin served as a loading control . Asterisk indicates the specific band . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 03110 . 7554/eLife . 07519 . 032Figure 7—figure supplement 4 . CEP63 and CDK2 interact . ( A ) HeLa total cell lysate was subjected to immunoprecipitation of using antibodies to endogenous CDK2 , CEP63 and a negative control , FLAG . Precipitating proteins were analyzed by immunoblotting for CEP63 and CDK2 . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 03210 . 7554/eLife . 07519 . 033Figure 7—figure supplement 5 . CDK2 localizes to the centrosome in a CDK5RAP2 , CEP152 , WDR62 , CEP63 , SPAG5 , CEP72 , CEP131 , MNR , and CEP90-dependent manner . ( A ) S phase SC , CDK5RAP2 , CEP152 , WDR62 , CEP63 , SPAG5 , CEP72 , CEP131 , MNR , and CEP90 siRNA transfected U2OS cells were analyzed by immunofluorescence with antibodies to Centrin ( ‘c’ , green ) and CDK2 ( red ) . ( B ) Quantification of the mean fluorescence intensities ±s . d . of CDK2 in SC , CDK5RAP2 , CEP152 , WDR62 , CEP63 , SPAG5 , CEP72 , CEP131 , MNR , CEP90 and CCDC14 siRNA treated cells expressed as the mean percentage ±s . d . of the fluorescence intensities of SC cells . For all quantifications 30 cells were analyzed per experiment ( n = 3 ) . p < 0 . 005 ( paired t-test ) statistically significant differences from SC controls are denoted by an asterisk . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 03310 . 7554/eLife . 07519 . 034Figure 7—figure supplement 6 . CDK5RAP2 , CEP152 , WDR62 , CEP63 , SPAG5 , CEP72 , CEP131 , MNR , CEP90 and CCDC14 do not control the stability of CDK2 . ( A ) Total cell lysate of SC , CDK5RAP2 , CEP152 , WDR62 , CEP63 , SPAG5 , CEP72 , CEP131 , MNR , CEP90 and CCDC14 siRNA transfected HeLa cells were analyzed by immunoblotting with antibodies to CDK5RAP2 , CEP152 , WDR62 , CEP63 , SPAG5 , CEP72 , CEP131 , MNR , CEP90 , CCDC14 and CDK2 . Actin or α-tubulin served as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 07519 . 034 We assessed the composition of these Centrin foci in CCDC14-depleted HeLa cells by costaining with antibodies to γ-tubulin and CP110 to identify the pericentrosomal matrix and centriolar distal ends respectively . Interestingly , in CCDC14-depleted cells , Centrin foci colocalized with CP110 but only partially with γ-tubulin , suggesting that inhibiting CCDC14 cause the formation of supernumary Centrin foci not necessarily associated with pericentriolar matrices ( Figure 7—figure supplement 1A ) . A recent study demonstrated that CCDC14 restricts the centrosomal localization of CEP63 ( Firat-Karalar et al . , 2014 ) . Consistent with these results , CEP63 localized to most of the supernumerary Centrin foci in CCDC14-depleted cells ( Figure 7D ) . We confirmed by immunoblot that protein levels of CEP63 were unchanged in CCDC14-depleted cells ( Figure 7E ) . In contrast to CEP63 , CDK5RAP2 , CEP152 and WDR62 were not recruited to these Centrin-positive foci , suggesting that the role for CCDC14 in restricting the localization of CEP63 to centrioles and centriole-like structures did not extend to other MCPH proteins ( Figure 7—figure supplement 2A and quantified in Figure 7—figure supplement 2B ) . The stability of CDK5RAP2 , CEP152 , WDR62 and CEP63 was unaltered in CCDC14-depleted cells , indicating that the accumulation of CEP63 in CCDC14-depleted cells was not due to over-stabilization ( Figure 7—figure supplement 2C ) . As CEP63 depends on CDK5RAP2 , CEP152 and WDR62 to localize to centrosomes , we assessed whether these other MCPH proteins were required for the formation of supernumerary Centrin foci in CCDC14-depleted cells . Using scrambled control or CCDC14 siRNA together with siRNAs targeting CDK5RAP2 , CEP152 , WDR62 , or CEP63 , we depleted U2OS cells of CCDC14 in combination with MCPH-associated proteins ( Figure 7F ) . Co-depletion of CCDC14 and MCPH proteins was confirmed by immunoblot analysis ( Figure 7—figure supplement 3 ) . Quantitation of Centrin foci revealed that , whereas depleting CCDC14 alone caused the formation of supernumary Centrin foci , co-depleting CDK5RAP2 , CEP152 , WDR62 or CEP63 blocked the formation of supernumary Centrin foci ( Figure 7G ) . In contrast , co-depleting the MCPH-associated proteins produced a phenotype similar to depleting the MCPH-associated protein alone . These findings demonstrate that CCDC14 suppresses the formation of supernumerary Centrin foci in a manner that depends upon the MCPH-associated proteins CDK5RAP2 , CEP152 , WDR62 and CEP63 . CEP63 interacts with Cyclin dependent kinase 2 ( CDK2 ) , a cell cycle-dependent kinase that is activated at the onset of S phase and participates in centriole duplication ( Dulic et al . , 1992; Hinchcliffe et al . , 1999; Lacey et al . , 1999; Matsumoto et al . , 1999; Meraldi et al . , 1999; De Boer et al . , 2008; Loffler et al . , 2011 ) . Reciprocal immunoprecipitations of CEP63 and CDK2 confirmed the specificity of the interaction ( Figure 7—figure supplement 4A ) . To assess whether the localization of CDK2 was altered in the absence of the CEP63-interacting protein CCDC14 , we assessed the localization of CDK2 in CCDC14-depleted cells . CDK2 , which localizes in the vicinity of the centrosome , overaccumulated at the Centrin foci in CCDC14-depleted cells ( Figure 7H ) . As CDK2 activity is required for centriole duplication and its inhibition can suppress supernumerary Centrin foci induced by CEP63 overexpression ( Loffler et al . , 2011 ) , we assessed whether blocking CDK2 activity could similarly suppress the formation of supernumerary Centrin foci in CCDC14-depleted cells . We examined U2OS cells transfected with scrambled control ( SC ) or CCDC14 siRNA and treated with DMSO or the CDK2 inhibitor roscovitine . Following DMSO treatment , control cells in S phase had two pairs of centrioles while CCDC14-depleted cells had multiple Centrin foci ( Figure 7I , left , and Figure 7J ) . In contrast , CCDC14-depleted cells in S phase treated with roscovitine only contained one pair of centrioles ( Figure 7I , right , and Figure 7J ) . Together , these findings demonstrate that CCDC14 suppresses the formation of supernumerary Centrin foci by limiting the centrosomal localization of CEP63 and the activity of its binding partner CDK2 . Given the involvement of CDK2 in promoting centriole duplication , we examined whether the centrosomal localization of CDK2 was dependent on MCPH proteins and their cognate centriolar satellite interactors that promote centriole duplication . Immunostaining U2OS cells treated with siRNA to CDK5RAP2 , CEP152 , WDR62 , CEP63 , SPAG5 , CEP72 , CEP131 , MNR , or CEP90 revealed that all of these MCPH-associated genes and centriolar satellite genes promote the centrosomal localization of CDK2 ( Figure 7—figure supplement 5A and quantified in Figure 7—figure supplement 5B ) . The protein level of CDK2 was unaltered in these MCPH and centriolar satellite protein knockdown cells ( Figure 7—figure supplement 6A ) . Collectively , these results reveal that CDK2 localizes to the centrosome in a manner dependent on MCPH-associated proteins and select centriolar satellite proteins to promote centriole duplication ( Figure 7K ) . We found that the MCPH-associated proteins CDK5RAP2 , CEP152 , WDR62 and CEP63 interact with each other and function together to promote centriole duplication . These MCPH-associated proteins assemble in a step-wise hierarchical manner to form a toroidal ring at the proximal centriole . We also found that each of these MCPH-associated proteins interacts with distinct centriolar satellite components that are required for the ordered recruitment of its cognate MCPH protein to the centrosome . Therefore , the centriolar satellite proteins , like the MCPH proteins , have critical roles in promoting centriole duplication . Consistent with a role of centriolar satellites in the function of proteins previously associated with MCPH , we have found evidence that the centriolar satellite gene , CEP90 , may be mutated in rare instances of MCPH . The assembly of MCPH proteins culminates in CDK2 localization to centrosomes , a previously recognized regulator of centriole duplication ( Hinchcliffe et al . , 1999; Matsumoto et al . , 1999 ) . In contrast to the other centriolar satellites components that promote MCPH-associated protein localization to the centrosomes and centriole duplication , CCDC14 opposes this pathway by limiting the centrosomal localization of CEP63 and its interactor CDK2 . Thus , centriolar satellite proteins can have either positive or negative roles in controlling the centrosomal localization of MCPH-associated proteins and CDK2 , and are thus critical for centriole duplication ( Figure 7K ) . Several previous studies in various organisms have demonstrated how proteins can recruit each other to the centrosome to promote centriole duplication . For example , the Drosophila melanogaster homologs of CDK5RAP2 and CEP152 , Centrosomin and Asterless , interact with each other and Asterless is involved in the centrosomal localization of Centrosomin ( Conduit et al . , 2010 ) . In C . elegans , centriole duplication is coordinated by SPD-2 ( CEP192 ) recruitment of ZYG-1 ( PLK4 ) which , in turn , localizes SAS-5 ( STIL ) and SAS-6 to the centrosome , culminating in the recruitment of SAS-4 ( Delattre et al . , 2006 ) . Our results indicate that ordered recruitment of MCPH-associated proteins to the centrosome extends beyond CDK5RAP2 and CEP152 to include WDR62 and CEP63 . Moreover , the recruitment of centriole duplication proteins to the centrosome occurs in a step-wise fashion , with CDK5RAP2 being required for the recruitment of CEP152 , which recruits WDR62 , which in turn recruits CEP63 ( Figure 7J ) . While our mass spectrometry analysis of CEP152 detected interactions with CDK5RAP2 , WDR62 and CEP131 , we did not detect some previously reported interactors of CEP152 , including CEP192 , SAS4 , CENTROBIN and PLK4 ( Cizmecioglu et al . , 2010; Hatch et al . , 2010; Kim et al . , 2013; Sonnen et al . , 2013; Firat-Karalar et al . , 2014; Gudi et al . , 2014 ) . Immunoprecipitation of endogenous CEP152 may be less efficient than overexpressed CEP152 , or the use of different cell lines or lysis conditions may account for the observed differences . Conversely , our in solution mass spectrometry preparations allowed us to detect an interaction between CEP152 and WDR62 . These interaction data helped reveal that WDR62 localizes to the centrosome in a CEP152-dependent manner . Interestingly , although CEP63 has been reported to be required for the localization of CEP152 to the centrosome , we detected only a minor decrease in the centrosomal accumulation of CEP152 in CEP63-depleted cells 48 hr post-transfection ( Sir et al . , 2011; Brown et al . , 2013; Lukinavicius et al . , 2013 ) . The observed difference in dependence of localization is attributable to the loss of CEP152 protein stability upon knockdown of CEP63 using the siRNA sequence used by Brown et al . The additional CEP63 siRNAs used in this study efficiently deplete CEP63 but do not affect CEP152 protein levels ( Figure 1—figure supplement 5 ) , suggesting that CEP152 centrosomal localization does not critically depend on CEP63 . The MCPH-associated protein CEP63 interacts with CDK2 , a kinase which in addition to its well known roles in cell cycle progression , is also involved in centriole duplication ( Hinchcliffe et al . , 1999; Matsumoto et al . , 1999; Loffler et al . , 2011 ) . Consistent with a key role for CEP63 in recruiting CDK2 , the other MCPH-associated proteins CDK5RAP2 , CEP152 and WDR62 required for CEP63 localization to the centrosome are also required for CDK2 localization to the centrosome and centriole duplication . While the recruitment of CDK2 to the centrosome by MCPH-assoicated proteins can account for how these proteins participate in centriole duplication , our results do not exclude roles for the MCPH proteins in recruiting other effectors of centriole duplication to the centrosome . We also found that the centriolar satellite components SPAG5 , CEP72 , CEP131 , MNR , and CEP90 participate in centriole duplication by interacting with and localizing MCPH-associated proteins to the centrosome . Different satellite proteins interact with different MCPH proteins: CEP72 and SPAG5 interact with CDK5RAP2 , CEP131 with CEP152 , MNR with WDR62 , and CEP90 with CEP63 . By bringing MCPH-associated proteins to the centrosome , the centriolar satellite proteins are required to localize proteins to the centrosome even when no biochemical interaction was detected . For example , CEP72 and SPAG5 are required for all four MCPH-assocated proteins to localize to the centrosome , but they interact specifically with CDK5RAP2 . While it is possible that the satellite proteins assemble the MCPH protein complex in the cytoplasm prior to its delivery to the centrosome , the most parsimonious model to explain these results is that centriolar satellite proteins bring individual interacting MCPH-associated proteins to the centrosome where the MCPH complex is assembled , and that this recruitment and assembly is required for the subsequent localization of other MCPH-associated proteins farther down the hierarchy . To extend our example , by bringing CDK5RAP2 to the centrosome , the centriolar satellite proteins CEP72 and SPAG5 are required for the centrosomal localization of the other three MCPH proteins despite not interacting with them biochemically . Thus , the centriolar satellite proteins form a hierarchy that parallels the MCPH protein hierarchy , with distinct satellite components interacting with and being required for the centrosomal localization of different MCPH-associated proteins . Inhibiting centriolar satellite protein function not only prevents the interacting MCPH-associated protein from localizing to the centrosome , but it also prevents that MCPH protein from interacting with the other MCPH-associated proteins , indicating that it is critical for MCPH complex formation . These findings suggest that centriolar satellite proteins may bind their cognate MCPH-associated protein outside of the centrosome and deliver it to the centrosome where it can complex with other MCPH-associated proteins . Previous studies revealed that centriolar satellites transport cargo along microtubules in a dynein-dynactin dependent manner to deliver proteins to the centrosome ( Kubo et al . , 1999; Kodani et al . , 2010 ) . Our finding that MCPH-associated proteins colocalize with the centriolar satellite protein PCM1 upon acute depolymerization of microtubules provides further evidence that the function of centriolar satellites is to transport proteins to the centrosome in a microtubule-dependent manner ( Figure 2—figure supplement 1 , Dammermann and Merdes , 2002; Kodani et al . , 2010 ) . Thus , a critical function of centriolar satellite proteins may be to bring MCPH-associated proteins to the centrosome , thereby accumulating CDK2 and promoting centriole duplication . Recent evidence implicating centriolar satellites in various cellular processes such as cellular stress response , ciliogenesis and protein degradation , reveals that centriolar satellite function extends beyond promoting centriole duplication ( Kim et al . , 2012; Stowe et al . , 2012; Hall et al . , 2013; Villumsen et al . , 2013 ) . We identified a homozygous missense variant in the centriolar satellite gene , CEP90 , in two boys with microcephaly and intellectual disability . The charge-reversing mutation affects a glutamate residue conserved back to protists and eliminates the interaction of CEP90 with the MCPH protein CEP63 , but not its interaction with PCM1 . Therefore , we propose that hypomorphic mutations in CEP90 are a rare cause of human microcephaly , and that defects in the function of centriolar satellites can cause the human disease microcephaly . It will be interesting to determine whether mutations in genes encoding other centriolar satellite components are also causes of microcephaly . Depletion of CEP90 reduces recruitment of CEP63 and CDK2 to the centrosome and abrogates centriole duplication . We hypothesize that microcephaly-associated mutation of CEP90 causes defective centriole duplication , similar to depletion of CEP90 or of MCPH proteins such as CDK5RAP2 , CEP152 , WDR62 or CEP63 . Loss of function mutations in the mouse orthologs of the MCPH genes MCPH1 , STIL , CDK5RAP2 , and ASPM alter spindle orientation and promote the loss of neural progenitors ( Izraeli et al . , 1999; Barrera et al . , 2010; Lizarraga et al . , 2010; Pulvers et al . , 2010; Gruber et al . , 2011 ) . It seems likely that centriole duplication is critical for spindle orientation and the regulation of neuronal progenitor number , but why centriole duplication is preferentially required for brain development remains unclear . In contrast to the other MCPH protein-interacting satellite components , CCDC14 restrains supernumerary Centrin foci formation by limiting the centrosomal accumulation of its interacting protein CEP63 ( Firat-Karalar et al . , 2014 ) . We found that CCDC14 is also essential for limiting the centrosomal localization and activity of CDK2 . As several satellite proteins are required for MCPH protein localization to the centrosome , but CCDC14 limits the centrosomal localization of CEP63 and CDK2 , centriolar satellites may control the trafficking of proteins both to and away from the centrosome . In summary , we have found that diverse centriolar satellite proteins control the localization of cognate MCPH-associated proteins to the centriole , where they assemble in a hierarchical step-wise manner , culminating in the recruitment CDK2 to the centrosome and centriole duplication . These findings help elucidate how satellites and ordered assembly coordinately build the complex centrosomal structure and how MCPH-associated proteins participate in the fundamental cell biological process of centriole duplication . Polyclonal guinea pig antibodies to human CEP131 ( aa . 1053–1083 ) and CCDC14 ( aa . 194–211 ) were developed by Pierce Technologies ( Thermo Fisher Scientific Inc . , Rockford , IL ) . Antibodies obtained for this study are the following: anti-Centrin1/2 20H5 ( Millipore , Billerica , MA ) , anti-CDK5RAP2 ( ABE236 and 06–0398 , Millipore , Billerica , MA ) , anti-CEP63 ( Millipore , Billerica , MA , Proteintech , Chicago , IL and Thermo Fisher Scientific , Waltham , MA ) , anti-MNR ( Novus Biologicals , Littleton , CO and Sigma , St . Louis , MO ) , anti-CCDC14 ( GeneTex Inc . , Irvine , CA and Abcam , Cambridge , MA ) , anti-CEP131 ( Sigma Prestige , St . Louis , MO and Abcam , Cambridge , MA ) , anti-CP110 , anti-CEP72 , anti-SPAG5 , anti-CEP90 , anti-Centrin1 ( Proteintech Group , Chicago , IL ) , anti-PCM1 ( H-262 and D-19 , Santa Cruz Biotechnologies , Dallas , TX ) , anti-CDK2 ( M2 , Santa Cruz Biotechnology , Dallas , TX; 22060-1-AP , Proteintech Group , Chicago , IL ) , anti-WDR62 and anti-CEP152 ( Bethyl Labs , Rockford , IL ) , anti-Actin , anti-α-tubulin and anti-γ-tubulin ( Sigma , St . Louis , IL ) , anti-SAS6 and anti-c-Myc ( Santa Cruz Biotechnology , Dallas , TX ) . Alexa-conjugated secondary antibodies and Hoechst 33342 were obtained from Molecular Probes ( Life Technologies , Grand Island , NY ) . DyLight and HRP-conjugated secondary antibodies were obtained from Jackson ImmunoResearch Laboratories ( West Grove , PA ) and Cell Signaling ( Grand Island , NY ) . HeLa and U2OS cells ( UCSF tissue culture facility ) were cultured in Advanced DMEM ( Invitrogen , Grand Island , NY ) supplemented with 2% FBS ( Invitrogen , Grand Island , NY ) and Glutamax-I ( Invitrogen , Grand Island , NY ) . Lyophilized roscovitine ( Millipore , Billerica , MA ) was resuspended in DMSO . HeLa cells were treated with 17 μM of nocodazole for 45 min to depolymerize microtubules . To inhibit CDK2 function , U2OS cells were incubated with 2 μM of roscovitine for 24 hr . Cells were transfected with siRNA using Oligofectamine ( Invitrogen , Grand Island , NY ) according to the manufacturer's instructions and analyzed 48 hr later . Sequences for the siRNA oligonucleotides used in this study are described in Supplementary file 2 . To deplete CEP215 and CEP63 using previously published sequences , HeLa cells were grown in DMEM ( Invitrogen , Grand Island , NY ) supplemented with 10% FBS , and Glutamax-I . Asynchronous HeLa cells were incubated on ice for 5 min in chilled Ca++ and Mg++ free Dulbecco's PBS ( DPBS , Invitrogen , Grand Island , NY ) , harvested using a cell scraper and lysed on ice for 10 min in lysis buffer ( 50 mM Tris-HCL pH7 . 4 , 266 mM NaCl , 2 . 27 mM KCl , 1 . 25 mM KH2PO4 , 6 . 8 mM Na2HPO4-7H2O and 1% NP-40 ) supplemented with protease and phosphatase inhibitors ( Calbiochem , Billerica , MA and Thermo Fisher Scientific Inc . , Waltham , MA ) . Lysates were clarified ( 13 , 000 r . p . m . , 4°C , 10 min ) and the protein concentrations were determined using the Bradford assay ( Bio-Rad , Hercules , CA ) . For each immunoprecipitation reaction analyzed by mass spectrometry , 5 mg of total lysate was incubated with 10 μg of antibody for 2 hr and then incubated with protein G-sepharose ( GE Healthcare Life Sciences , Pittsburgh , PA ) for an additional 1 . 5 hr at 4°C . Immunocomplexes were washed three times in lysis buffer , once in detergent free lysis buffer and subsequently incubated with 0 . 2 M Glycine pH2 . 5 on ice for 10 min and quenched with 1 M Tris pH9 . For smaller scaled immunoprecipitation reactions , 500 µg of total lysate was incubated with 2 µg of antibody for 2 hr and then incubated with protein G-Sepharose for an additional hour at 4°C . Complexes were washed three times in lysis buffer and subsequently boiled in 2× Laemmli reducing buffer . To detect WDR62 , samples were incubated in 2× Laemmli reducing buffer , but were not boiled . Samples were separated on 4–15% gradient TGX precast gels ( Bio-Rad , Hercules , CA ) , transferred onto nitrocellulose ( Whattman , Pittsburgh , PA ) and then subjected to immunoblot analysis using ECL Lightening Plus ( Perkin–Elmer , Waltham , MA ) . For quantifications , samples were blotted with fluorescently labeled secondary antibodies and analyzed on a LiCOR Odyssey scanner . Specific protein signals were normalized to Actin . Centrosomes from asynchronously growing HeLa cells were isolated as previously described ( Bobinnec et al . , 1998 ) . HeLa cells were treated with 2 μM nocodazole and 1 μg/μl of cytochalasin D for 1 hr to disrupt the microtubule and actin cytoskeleton respectively . Cells were collected and lysed in lysis buffer ( 1 mM HEPES , pH 7 . 2 , 0 . 5% NP-40 , 0 . 5 mM MgCl2 , and 0 . 1% β-mercaptoethanol ) and clarified at 2500×g for 10 min . The resulting supernatant was layered atop a 60% sucrose cushion above a discontinuous sucrose gradient ( 70 , 50 and 40% sucrose ) and centrifuged at 40 , 000×g for 1 hr . Fractions were collected from the bottom and analyzed by Western blot . To visualize centrosome and centriolar satellite proteins , cells were fixed in chilled methanol for 3 min . Following fixation; cells were incubated in blocking buffer ( 2 . 5% BSA , 0 . 1% Triton-X100 , 0 . 03% NaN3 in DPBS ) overnight at 4°C . Primary and secondary antibodies were diluted in blocking buffer and incubated with cells at room temperature for 1 hr . To detect CDK2 using anti-CDK2 ( M2 , Santa Cruz Biotechnology ) , cells were extracted with 0 . 1% Triton X-100 in a buffer consisting of 50 mM piperazine-1 , 4-bis ( 2-ethanesulphonic acid ) at pH 7 . 4 , 5 mM MgCl2 , and 5 mM EDTA prior to fixation in chilled methanol for 3 min . Alternatively , cells were fixed in ice cold methanol:acetone ( 1:1 ) for 7 min to detect CDK2 using anti-CDK2 ( 22060-1-AP , Proteintech Group , Chicago , IL ) . Coverslips were mounted using Gelvatol mounting media and imaged with an inverted Axio Observer D1 ( Zeiss , Thornwood , NY ) , image processing was completed with Adobe Photoshop . For fluorescence quantifications , specified sized regions of interest were selected and quantified using Fiji . The signal of the MCPH proteins were normalized to Centrin signal in HeLa or U2OS cells . To analyze proteins by LC-MS/MS , immunocomplexes were digested with trypsin ( Promega , Madison , WI ) overnight at 37°C , denatured in 2 M urea , 10 mM NH4HCO3 , 2 mM DTT for 30 min at 60°C , and alkylated with 2 mM iodoacetamide for 45 min at room temperature . Samples were analyzed by LC-MS/MS with a Thermo Scientific Velos Pro ion trap mass spectrometry system equipped with a Proxeon Easy nLC 1000 ultra high pressure liquid chromatography and autosampler system . Peptide data was matched to protein sequences by the Protein Prospector algorithm and searched against the SwissProt Human protein sequence database . Subjects were identified and evaluated in a clinical setting for medical history , cognitive impairment and physical abnormalities . Those with histories suggestive of hereditary disorders of brain development , epilepsy , and/or cognition were offered research enrollment . After obtaining written informed consent from participants or their legal guardian , phenotypic information and peripheral blood samples were collected for research purposes . The informed consent for this genetic study included permission for publication . Ethical review and approval was obtained from the Committee on Clinical Investigations at Beth Israel Deaconess Medical Center ( current protocol 2001-P-000758 ) . The study adheres to the state and federal regulations governing the conduct of human subject research ( 45 CFR Part 46 and 21 CFR Parts 50 and 56 ) and the ethical principles set forth in the Belmont Report . Patient DNA samples were subject to genome-wide SNP genotyping using Illumina 660W-Quad BeadChip ( 2 . 6M markers ) . Linkage analyses were performed using Allegro , MERLIN and homozygosity analysis was performed using custom scripts as previously described ( Yu et al . , 2010 ) . Whole exome sequencing libraries were generated using Agilent SureSelect capture kits , and sequenced on Illumina HiSeq machines to a mean read depth of 170× . Alignment , variant calling , and annotation were performed as previously described ( Yu et al . , 2013 ) . 3X-FLAG tagged human CEP90 cDNA was obtained from Genecopoeia ( EX-T9157-M12 ) . The single–base pair mutation in CEP90 was introduced using the QuickChange site-directed mutagenesis kit ( Agilent Technologies ) with the primer pair: Forward: 5′-cttacaaagattgaagaattggacgagaaacttaatgatgcacttca-3′; Reverse: 5′-tgaagtgcatcattaagtttctcgtccaattcttcaatctttgtaag-3′ .	When a cell divides , the chromosomes that contain the genetic blueprint for the cell must be replicated and shared between the two new cells . A structure called the centrosome organizes the cellular machinery that separates the chromosome copies during cell division . At the center of each centrosome are two cylindrical microtubule-based structures called centrioles . Mutations in certain proteins that interact with the centrosome cause a neurodevelopmental disorder called primary microcephaly . People born with microcephaly have unusually small heads and brains . As a result , they may have difficulties with mental tasks . Scientists do not know exactly how these ‘microcephaly-associated’ proteins normally interact with the centrosomes or what they do at the centrosomes , so it is difficult to work out what goes wrong in people with microcephaly . One idea is that the proteins help to duplicate the centrioles before a cell divides . If this duplication does not occur , a cell cannot divide properly; so , people with mutations that interfere with centriole duplication cannot grow enough brain cells . Now , Kodani et al . have examined how these microcephaly-associated proteins work with ‘satellite’ proteins that congregate near the centrosome to duplicate centrioles . The satellite proteins help to recruit four microcephaly-associated proteins to the centrosome , where they are built into a ring . The microcephaly-associated proteins congregate at the centrosome in a particular order , with each protein recruiting the next one in the sequence . Once all four are in place near the centrosome , an enzyme that helps to duplicate the centrioles joins them . Further experiments suggest that mutations that affect one of the satellite proteins—known as CEP90—may cause microcephaly . Future analysis of how microcephaly-associated genes work may reveal the cell biological mechanisms by which centrioles participate in brain development .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2015	Centriolar satellites assemble centrosomal microcephaly proteins to recruit CDK2 and promote centriole duplication
In many excitatory synapses , mobile zinc is found within glutamatergic vesicles and is coreleased with glutamate . Ex vivo studies established that synaptically released ( synaptic ) zinc inhibits excitatory neurotransmission at lower frequencies of synaptic activity but enhances steady state synaptic responses during higher frequencies of activity . However , it remains unknown how synaptic zinc affects neuronal processing in vivo . Here , we imaged the sound-evoked neuronal activity of the primary auditory cortex in awake mice . We discovered that synaptic zinc enhanced the gain of sound-evoked responses in CaMKII-expressing principal neurons , but it reduced the gain of parvalbumin- and somatostatin-expressing interneurons . This modulation was sound intensity-dependent and , in part , NMDA receptor-independent . By establishing a previously unknown link between synaptic zinc and gain control of auditory cortical processing , our findings advance understanding about cortical synaptic mechanisms and create a new framework for approaching and interpreting the role of the auditory cortex in sound processing . As a constituent for nearly 3000 proteins , zinc plays a key role in protein structure , enzymatic catalysis , and cellular regulation ( Vallee , 1988 ) . Although the chemistry and biology of zinc metalloproteins have historically dominated the field of zinc biology , there is a growing appreciation for a signaling role of free , mobile zinc found in secretory tissues such as the prostate , pancreas , and especially the brain ( Frederickson et al . , 2005; Kelleher et al . , 2011 ) . In many brain areas , including the neocortex , limbic structures and auditory brainstem , high concentrations of synaptic zinc are found in glutamatergic excitatory presynaptic vesicles ( Frederickson et al . , 2005 ) . In the hippocampus , more than half of presynaptic glutamatergic contain synaptic zinc ( Sindreu et al . , 2003 ) , attesting to zinc’s abundance and importance in synaptic function . Ex vivo studies established that synaptic zinc is an inhibitory neuromodulator of AMPA receptors , NMDA receptors , and vesicular release probability at lower frequencies of synaptic activity ( Kalappa et al . , 2015; Anderson et al . , 2015; Vergnano et al . , 2014; Pan et al . , 2011; Vogt et al . , 2000; Perez-Rosello et al . , 2013 ) . In contrast , during higher frequencies of synaptic activity and during enhanced vesicular release probability , synaptic zinc inhibits synaptic responses during the first few stimuli but enhances steady state responses during subsequent stimuli ( Kalappa and Tzounopoulos , 2017 ) . However , it is unknown whether and how zinc affects neuronal processing in vivo . To answer this question , we investigated the role of synaptic zinc in shaping the sound-evoked activity in cortical neurons of awake mice . Namely , we investigated whether synaptic zinc modulates the relationship between sound intensity and the amplitude of neuronal evoked responses – termed gain modulation . Gain modulation controls the dynamic range of neuronal responses to sounds and is a fundamental feature of sound processing enabling adaptation to changes in sound stimulus statistics , as well as central compensation to peripheral damage ( Wen et al . , 2012; Dean et al . , 2005; Watkins and Barbour , 2008; Rabinowitz et al . , 2011; Natan et al . , 2017; Chambers et al . , 2016 ) . We used widefield transcranial imaging of the genetically-encoded calcium indicator GCaMP6 to identify the effects of synaptic zinc on populations of specific neuronal types in the auditory cortex , and two-photon imaging to interrogate the effects of zinc on individual layer 2/3 neurons . Our results highlight synaptic zinc as a novel modulator of cortical responses to sound . We began our exploration by visualizing sound-evoked responses in the auditory cortex of awake mice . To image and quantify the cortical sound-evoked responses , we used adeno-associated virus ( AAV ) driven by the synapsin promoter to express the genetically-encoded calcium indicator GCaMP6s in auditory cortical neurons ( AAV-Syn-GCaMP6s , ( Chen et al . , 2013 ) ; Materials and methods ) . Eleven to 24 days following stereotaxic viral injections into the posterior temporal cortex , we performed in vivo calcium imaging in head-fixed unanesthetized mice ( Materials and methods ) . To locate the primary auditory cortex ( A1 ) , we presented low frequency tones ( 5 or 6 kHz , 40–60 dB SPL ) and imaged the sound-evoked changes in transcranial GCaMP6s fluorescence ( Figure 1a , Materials and methods ) . Due to the mirror-like reversal of tonotopic gradients between A1 and the anterior auditory field ( AAF ) ( Guo et al . , 2012; Joshi et al . , 2015 ) , these sounds activated two discrete regions of the auditory cortex corresponding to the low frequency regions of A1 and the AAF ( Figure 1a bottom ) . These results are consistent with previous studies mapping auditory cortical fields in the mouse ( Issa et al . , 2014; Kato et al . , 2015; Guo et al . , 2012 ) . After locating A1 , we performed a small craniotomy adjacent to A1 and inserted a thin glass micropipette containing the extracellular , high-affinity , fast , zinc-specific chelator ZX1 ( Pan et al . , 2011; Anderson et al . , 2015 ) ( Figure 1a top right , Materials and methods ) . Then , we quantified the sound-evoked responses of A1 neuronal populations to 400 msec long , 12 kHz tones of 30–80 dB SPL – an intensity span that encompasses sounds ranging from a quiet room to a busy city street . To assess the effect of zinc on A1 responses , we infused ZX1 in the auditory cortex and verified the intracortical diffusion of the chelator into A1 by visualizing the spread of the extracellular red fluorescent dye Alexa-594 , co-infused with ZX1 ( Figure 1—figure supplement 1a–c ) . We observed that the response to an 80 dB SPL tone was significantly enhanced after ZX1 application ( Figure 1b , c ) ; whereas , the response to a 50 dB SPL tone was not affected ( Figure 1d , e ) . Overall , ZX1 , but not vehicle containing ACSF and Alexa-594 , enhanced the amplitude of the fluorescence response for sounds of ≥70 dB SPL ( Figure 1f and Figure 1—figure supplement 1d , e ) . These results indicate that endogenous extracellular zinc inhibits the gain of sound-evoked A1 responses . Does zinc scale by division or shift by subtraction the gain of sound-evoked A1 responses ? To answer this question , we plotted the normalized evoked responses before vs . after ZX1 infusion . By fitting a regression line to this function , we found a slope steeper than unity and no change in the y-intercept ( Figure 1g–i ) , indicating that extracellular zinc signaling exerts divisive gain control of sound-evoked A1 responses . To track the origin of the extracellular zinc modulating A1 gain control , we performed similar experiments in ZnT3 KO mice , which lack the vesicular zinc transporter ZnT3 and synaptic zinc ( Cole et al . , 1999 ) . Compared to WT mice , ZnT3 KO mice showed larger responses to sounds of 80 dB SPL ( Figure 1—figure supplement 1f ) , suggesting that synaptic zinc inhibits the gain of A1 responses . Moreover , zinc chelation in KO mice had no effect on the gain of A1 responses ( Figure 1j–m , Materials and methods ) , suggesting that , in WT mice , ZX1 enhances A1 gain via chelation of extracellular , ZnT3-dependent zinc . Consistent with the high selectivity of ZX1 over calcium and other biologically relevant metal ions such as magnesium ( Pan et al . , 2011 ) , the lack of ZX1 effects on the sound-evoked responses in ZnT3 KO mice further validate that the observed effects of ZX1 on the sound evoked responses of WT mice are due to the chelation of synaptic zinc – and not to the potential non-specific effects of ZX1 on calcium or other metal ions . Finally , because extracellular tonic , not synaptically-evoked , zinc levels in brain slices are ZnT3-independent ( Anderson et al . , 2015 ) , our results suggest that synaptically released zinc modulates the gain of sound-evoked responses in A1 . We next investigated how synaptic zinc affects the gain of populations of specific classes of auditory cortical neurons . We recorded transcranial sound-evoked responses from populations of principal excitatory neurons , by utilizing an AAV driven by the calcium/calmodulin-dependent protein kinase 2 ( CaMKII ) promoter to express GCaMP6f in principal neurons ( Figure 2a–c and Figure 2—figure supplement 1a; AAV-CaMKII-GCaMP6f , [Pakan et al . , 2016] ) . ZX1 reduced the amplitude of sound-evoked responses of principal neurons to sounds of 60–80 dB SPL ( Figure 2a , b ) . Linear regression analysis showed that the ZX1-induced reduction in the gain was due to a slope shallower than unity ( Figure 2c ) , indicating that extracellular zinc signaling exerts multiplicative gain control of the sound-evoked A1 responses in principal neurons . In contrast to the enhancing effects of ZX1 on responses to sounds of 70–80 dB SPL in all neurons ( Figure 1f ) , the divisive effect of ZX1 on responses to sounds of 60–80 dB SPL in principal neurons suggests that the effects of synaptic zinc are intensity- and cell type-specific . To track the origin of the extracellular zinc increasing the gain of principal neurons , we studied the effects of ZX1 in ZnT3 KO mice , which had been injected with AAV-CaMKII-GCaMP6f . ZX1 had no effect on the sound-evoked responses of principal neurons in ZnT3 KO mice , indicating that synaptic zinc increases the gain of principal neurons ( Figure 2—figure supplement 1b , c ) . Whereas synaptic zinc is , mostly , an inhibitor of glutamatergic neurotransmission ( Perez-Rosello et al . , 2013; Vergnano et al . , 2014; Anderson et al . , 2015; Kalappa et al . , 2015 ) , zinc chelation decreased the sound-evoked activity of principal neurons . To explain this result , we hypothesized that the effects of zinc chelation are circuit-dependent . Since the sound-evoked responses of principal neurons in A1 are suppressed by inhibitory inputs from parvalbumin-expressing interneurons ( PV neurons; Li et al . , 2013; Seybold et al . , 2015; Phillips and Hasenstaub , 2016; Resnik and Polley , 2017 ) , a ZX1-induced increase of the sound-evoked responses of PV neurons would be consistent with the ZX1-induced decrease of the sound-evoked responses observed in principal neurons . To image and quantify the sound-evoked responses of PV neurons , we injected AAV expressing Cre-dependent GCaMP6f ( AAV-Flex-GCaMP6f , Materials and methods ) into the auditory cortex of PV-Cre mice ( Figure 2—figure supplement 1a , Materials and methods ) . ZX1 increased the sound-evoked responses of PV neurons to all tested sound intensities ( 30–80 dB SPL , Figure 2d , e ) . Linear regression analysis showed that the ZX1-induced increase in response gain was due to a slope steeper than unity ( Figure 2f ) , indicating that extracellular zinc exerts divisive gain control of sound-evoked responses in PV neurons . In PV neurons , the enhancing effect of ZX1 on sound-evoked responses was largest for sounds of 60–80 dB SPL ( Figure 2—figure supplement 1d ) – the intensities where principal neurons showed a ZX1-induced reduction in their responses ( Figure 2b ) . Although our approach involving the bulk application of ZX1 is not conclusive for distinguishing cell- ( direct ) vs . circuit-dependent effects ( see Discussion for details ) , this coupling between the ZX1-induced enhancement in the gain of PV neurons ( Figure 2ef ) and the ZX1-induced reduction in the gain of principal neurons ( Figure 2bc ) is consistent with the notion that synaptic zinc increases , at least in part , the gain of principal neurons by decreasing the gain of PV neurons . We next focused on the effect of zinc chelation on the somatostatin-expressing interneurons ( SOM neurons ) , which inhibit PV neurons ( Harris and Shepherd , 2015; Tremblay et al . , 2016 ) . We selectively expressed GCaMP6f in SOM neurons by injecting AAV-Flex-GCaMP6f into SOM-Cre mice ( Figure 2—figure supplement 1a , Methods ) . ZX1 increased the sound-evoked responses of SOM neurons to sounds of lower intensities ( 30–60 dB SPL , Figure 2gh , Figure 2—figure supplement 1d ) , and increased the gain of these neurons through a positive shift in the y-intercept ( Figure 2i ) . These results support that extracellular zinc exerts subtractive gain control of sound-evoked A1 responses in SOM neurons . The smaller ZX1-induced enhancement of the sound-evoked responses of PV neurons to lower intensity sounds ( 30–40 dB SPL ) is consistent with the ZX1-induced increases of the sound-evoked responses of SOM neurons to the same sound intensities ( Figure 2—figure supplement 1d ) . Together , our results from SOM , PV and principal neurons are consistent with a scheme where the zinc-mediated inhibition of the sound-evoked responses in SOM neurons contributes to the sound intensity dependence of the zinc-mediated inhibition in PV neurons , which , in turn , contributes to the zinc-mediated increase in the sound-evoked responses of principal neurons ( see Discussion for more details ) . We next investigated the effect of zinc chelation on the vasoactive intestinal polypeptide-expressing interneurons ( VIP neurons ) , which inhibit SOM and PV neurons ( Harris and Shepherd , 2015; Tremblay et al . , 2016; Pi et al . , 2013 ) . We selectively expressed GCaMP6f in VIP neurons by injecting AAV-Flex-GCaMP6f into VIP-Cre mice ( Figure 2—figure supplement 1a , Materials and methods ) . ZX1 increased the responses of VIP neurons to sounds of 60–80 dB SPL but did not affect their overall response gain ( Figure 2j–l ) . The ZX1-induced enhancements of the responses of VIP neurons to sounds of 60–80 dB SPL are consistent with the lack of ZX1 effects on the responses of SOM neurons to sounds of the same intensity ( Figure 2—figure supplement 1d ) . Although the lack of ZX1 effects on the gain of VIP neurons may reflect either a circuit-based cancellation of opposing effects or no direct effect ( see Discussion for more details ) , together , our results show that synaptic zinc is a novel modulator of cortical sound processing – a modulator that increases the gain of principal neurons , but reduces the gain of PV and SOM neurons . We next focused on the molecular targets ( i . e . , zinc-interacting proteins ) that contribute to the effects of synaptic zinc on cortical sound-evoked activity . Because nanomolar levels of zinc inhibit NMDA receptors ( NMDARs; Paoletti et al . , 1997; Hansen et al . , 2014 ) , we tested whether NMDARs contribute to the observed zinc-mediated gain modulation in A1 . To answer this question , we measured the effects of ZX1 on the neuronal sound-evoked responses after pharmacological blockade of NMDARs with a selective NMDAR antagonist ( D-2-Amino-5-phosphonopentanoic acid , APV , infusion , Materials and methods ) . In the presence of APV ( control ) , ZX1 increased the responses of principal neurons to sounds of 70–80 dB SPL ( Figure 3ab ) , which is a reversal of the observed ZX1-induced reduction in the sound-evoked responses in the absence of APV ( Figure 2ab ) . Moreover , APV enhanced the sound-evoked responses of principal neurons ( Figure 3—figure supplement 1ab ) , suggesting that APV disinhibits principal neurons . This disinhibition may also disrupt the coupling between the ZX1-induced enhancement in the gain of PV neurons ( Figure 2e , f ) and the ZX1-induced reduction in the gain of principal neurons ( Figure 2b , c ) . Consistent with this hypothesis , in the presence of APV , zinc chelation caused an increase in the sound-evoked responses of PV neurons to sound intensities of 30–80 dB SPL ( Figure 3a , c ) . These results suggest that the disinhibition of principal neurons , in the presence of APV , unmasked the NMDAR-independent inhibitory effects of zinc on the sound-evoked responses of principal neurons , which are likely direct and non-circuit-dependent . However , similarly to ZX1 application , our approach involving the bulk application of APV is not conclusive for distinguishing cell autonomous vs . circuit-dependent effects ( see Discussion for more details ) . The ZX1-induced increases in the sound-evoked responses of PV neurons in the presence of APV ( Figure 3c ) were not different from the increases observed in the absence of APV ( Figure 2f ) , suggesting that the effects of zinc on the responses of PV neurons are , at least in part , NMDAR-independent . Next , we investigated whether NMDARs contribute to the effects of synaptic zinc on the sound-evoked responses of SOM neurons . APV reduced the sound-evoked responses of SOM neurons ( Figure 3—figure supplement 1ef ) and abolished the effects of zinc chelation on the sound-evoked responses of SOM neurons ( Figure 3d , e ) . Although potential circuit effects due to NMDAR-dependent and NMDAR-independent effects that cancel each other out can’t be ruled out ( see Discussion for more details ) , our results are consistent with the notion that the effects of synaptic zinc on the sound-evoked responses of SOM neurons are NMDAR-dependent . We next tested whether NMDARs contribute to the effects of synaptic zinc on the sound-evoked responses of VIP neurons . APV enhanced the sound-evoked responses of VIP neurons ( Figure 3—figure supplement 1gh ) and abolished the effects of ZX1 on the sound-evoked responses of VIP neurons ( Figure 3df ) , suggesting that synaptic zinc inhibits the sound-evoked responses of VIP neurons , likely , via NMDAR-dependent mechanisms . However , similarly to the limitations in the interpretation of our results in SOM neurons due to potential circuit-based cancelation , NMDAR-independent mechanisms can’t be excluded . Overall , our results highlight that synaptic zinc modulates non-NMDAR targets to inhibit the gain of sound-evoked responses of principal and PV neurons; and suggest that NMDAR-mediated signaling is involved in the zinc-mediated gain control of SOM and VIP neurons . Transcranial imaging reflects a sound-evoked population fluorescent signal arising from neurons residing in different cortical layers , as well as from different neuronal compartments ( e . g . somata vs . dendrites ) . To explore the layer and sub-cellular specificity of the zinc-mediated effects on gain control , we next performed two-photon imaging of individual A1 L2/3 neurons in awake mice . In these experiments , we investigated the effects of ZX1 on the sound evoked responses from individual L2/3 neuronal somata ( Methods ) . We began these experiments by injecting AAV-CaMKII-GCaMP6f into A1 to record responses from principal neurons ( Materials and methods ) . After locating A1 as shown in Figure 1 , we performed a craniotomy and subsequent two-photon imaging to obtain sound-evoked responses from the somata of individual principal neurons ( Figure 4a , Materials and methods ) . We then infused ZX1 and were able to locate the same group of neurons to remeasure their sound-evoked responses ( Figure 4a ) . ZX1 reduced the responses to sounds of 70 dB SPL ( Figure 4b , c ) , and linear regression analysis showed that this reduction in gain was divisive ( Figure 4d ) , which is consistent with our results obtained with transcranial imaging ( Figure 2c ) . These results show that synaptic zinc exerts multiplicative gain control of sound-evoked responses in individual L2/3 principal neurons . We next measured the effects of ZX1 on the responses of individual L2/3 PV neurons . We injected AAV-Flex-GCaMP6f into PV-Cre mice and performed 2-photon imaging ( Figure 4e ) . We measured the sound-evoked responses from the somata of individual L2/3 PV neurons , infused ZX1 , and remeasured the responses of the same neurons ( Figure 4e ) . PV neurons showed increased sound-evoked responses to sounds of 50 and 70 dB SPL ( Figure 4f , g ) and a multiplicative increase in their gain ( Figure 4h ) , suggesting that synaptic zinc exerts divisive gain control of sound-evoked responses in individual L2/3 PV neurons . In individual L2/3 PV neurons , in contrast to our transcranial measurements , we did not observe any increases in their somatic responses to sounds of 30 dB SPL . Since calcium transients in neuronal dendrites can be tuned to different stimulus features compared to somata ( Chen et al . , 2013 ) , and can generate action potentials independently from the soma ( Golding and Spruston , 1998 ) , this discrepancy may reflect the differential zinc modulation of sound-evoked responses in dendritic vs . somatic domains of L2/3 PV neurons . It could also indicate that the sound-evoked responses of non-L2/3 PV neurons may contribute to the effects synaptic zinc to sounds of 30 dB SPL observed with transcranial imaging . Together , these results show that synaptic zinc signaling decreases the gain of sound-evoked responses of individual L2/3 PV neurons in A1 . We performed similar two-photon experiments in individual L2/3 SOM and VIP neurons ( Figure 5 ) and the results we obtained were , overall , consistent with the results we observed with transcranial imaging: ZX1 caused an additive gain increase in L2/3 SOM neurons ( Figure 5a–d ) and no significant changes in the gain of L2/3 VIP neurons ( Figure 5e–h ) . Whereas with transcranial imaging we observed ZX1-induced increases in the responses of VIP neurons to sounds of 70 dB SPL , we did not observe such an increase in individual L2/3 VIP neurons . This discrepancy may reflect the differential zinc modulation of sound-evoked responses in dendritic vs . somatic domains of L2/3 VIP neurons and/or may suggest that non-L2/3 VIP neurons are responsible for the enhancement we observed with transcranial imaging . Together , our two-photon experiments on individual L2/3 neurons support the cell-specific effect of synaptic zinc on A1 neurons: synaptic zinc increases the gain of principal neurons , but decreases the gain of PV and SOM interneurons . Our results show that synaptic zinc modulates the gain of sound-evoked auditory cortical responses . In the auditory system , as sound statistics of the auditory environment change , auditory neurons adjust their gain – the dynamic range of their responses – to match the stimulus statistics . In the auditory nerve , inferior colliculus , and auditory cortex , neurons adapt their responses to match the mean sound intensity levels ( Wen et al . , 2012; Dean et al . , 2005; Watkins and Barbour , 2008 ) . In addition to adaptation to the mean sound intensity levels , auditory cortical response gain is modulated by higher-level stimulus statistics , such as spectrotemporal contrast , reflecting tone distributions with identical mean intensity but different widths , and stimulus temporal correlation ( Rabinowitz et al . , 2011; Natan et al . , 2017 ) . Moreover , increases in cortical gain restore sound-evoked responses following near complete cochlear denervation , thus providing central compensation to peripheral damage ( Chambers et al . , 2016 ) . Therefore , gain modulation provides a mechanism for generating sensory representations that are robust to noise , thus maximizing sensitivity to changes in stimulus intensity across variable auditory contexts . Moreover , gain modulation maintains the robustness of central responses after peripheral damage . We propose , although not addressed in this manuscript , that sound-dependent changes in zinc signaling may contribute to these mechanisms . But , is there any evidence for activity-dependent changes in cortical zinc signaling ? Sensory experience alters synaptic zinc levels in primary somatosensory and visual cortex ( Brown and Dyck , 2002; Dyck et al . , 2003 ) ; however , the physiological roles of this plasticity in sensory processing remain unknown . Based on our results , we propose that sound-dependent changes in zinc signaling , which have been demonstrated in the dorsal cochlear nucleus ( Kalappa et al . , 2015 ) , may contribute to the activity-dependent changes in cortical gain . Such a contribution , which will be the focus of future experiments , would provide a role for zinc signaling in the adaptation of neuronal input-output functions to changing sound statistics , as well as in the preservation of central responses after peripheral damage . Synaptic zinc is present in many cortical areas ( Danscher and Stoltenberg , 2005 ) and dynamic gain modulation is a fundamental mechanism across many brain regions affecting sensory computations , attention , multisensory integration and value estimation ( Somers et al . , 1995; Reynolds and Heeger , 2009; Ohshiro et al . , 2011; Louie et al . , 2013 ) . We therefore propose , although not addressed in this manuscript , that activity-dependent modification of synaptic zinc signaling may be a fundamental mechanism capable of mediating dynamic gain modulation in response to context , history , training , or injury across sensory modalities . Gain modulation is mediated by selective increases or decreases in cortical inhibition ( Li et al . , 2013; Olsen et al . , 2012; Wilson et al . , 2012; Phillips and Hasenstaub , 2016 ) . Recent studies , including optogenetic approaches to increase the activity of PV and SOM neurons , have suggested that PV neurons provide multiplicative/divisive gain control of principal neurons , whereas SOM neurons provide additive/subtractive gain control of principal neurons ( Atallah et al . , 2012; Li et al . , 2014; Wilson et al . , 2012; Moore and Wehr , 2013 ) . However , activation and inactivation of PV and SOM neurons revealed asymmetric effects on neuronal gain in auditory cortex ( Phillips and Hasenstaub , 2016 ) . Here , we observed that zinc caused a multiplication in the gain of principal neurons , which is consistent with the observed zinc-mediated division in the gain of PV neurons . Therefore , our results are consistent with the role of PV neurons in providing dynamic gain control in A1 and add zinc as a player in the gain control of L2/3 cortical circuits . The lack of ZX1 effects on the A1 neuronal responses to specific sound intensities may reflect circuit-based cancelation of opposing ZX1 effects . Although the bulk application of ZX1 limits our ability to distinguish cell- from circuit-specific effects , the lack of ZX1 effect on the sound-evoked responses of principal and VIP neurons to lower sound intensities ( 30–50 dB SPL ) , and on the responses of SOM neurons to higher intensities ( 70–80 dB SPL ) suggests that the disruption of extracellular zinc signaling does not affect the balance of excitation and inhibition in these cases ( Wehr and Zador , 2003 ) . In principal neurons , it is interesting that this balance begins to break at sound intensities of ≥60 dB SPL , which is the intensity range where we observed the ZX1-induced increases in the sound-evoked responses of VIP neurons . Since VIP neurons provide inhibition to both SOM and PV neurons ( Pi et al . , 2013; Harris and Shepherd , 2015 ) , these findings suggest that synaptic zinc modulates the VIP-mediated disinhibition of SOM and PV neurons , which , in turn , shape the gain of principal neurons . This scheme , and our previously stated suggestions on the hierarchical contribution of the zinc-mediated effects on SOM , PV and principal neuronal responses to sound ( Results ) , are consistent with the overall sequential hodology of the three main inhibitory cortical cell classes ( Harris and Shepherd , 2015 ) . Importantly , the effects of ZX1 over a wide range of sound intensities ( 30–80 dB SPL ) reflect the ethological relevance of synaptic zinc signaling in modulating cortical responses to sound . Blockade of NMDARs with APV reduced the sound-evoked responses of PV neurons ( Figure 3—figure supplement 1cd ) and increased the sound-evoked responses of principal neurons ( Figure 3—figure supplement 1ab ) . Similarly , in the prefrontal cortex pharmacological blockade of NMDARs reduced the spontaneous firing rates of PV neurons and increased the spontaneous firing rates of principal neurons ( Homayoun and Moghaddam , 2007 ) . Moreover , genetic deletion of NMDARs selectively from PV neurons increased the spontaneous firing rates of principal neurons ( Carlén et al . , 2012 ) , suggesting that NMDA receptors expressed by PV-neurons contribute to PV neuron-mediated inhibition of principal neurons . Finally , PV neurons show high expression of GluN2A containing NMDAR subunits ( Xi et al . , 2009 ) , which are exceptionally sensitive to low nanomolar zinc levels ( Paoletti et al . , 1997; Hansen et al . , 2014 ) , thus rendering excitatory inputs to PV neurons very zinc-sensitive . Together , these results are consistent with the view that NMDAR signaling and its modulation by zinc support and modulate the ability of PV neurons to inhibit principal neurons . However , it is unclear whether NMDARs expressed by PV neurons are responsible for these effects , for previous studies showed relatively weak NMDA receptor-mediated excitation of PV neurons ( Goldberg et al . , 2003; Rotaru et al . , 2011 ) , suggesting that NMDARs are not a strong source of excitation for PV neurons . In this context , it is interesting that blockade of NMDARs increased the sound-evoked responses of VIP neurons ( Figure 3—figure supplement 1gh ) , which provide inhibition to PV neurons ( Tremblay et al . , 2016 ) . This suggests that NMDAR signaling may serve to increase the sound evoked-responses of PV neurons via a combination of direct excitation of PV neurons and reduced VIP-mediated sound-evoked inhibition to these cells . Future studies will be required to explore these possibilities . The responses of PV neurons to sound are the most sensitive to zinc modulation , both in amplitude and sound intensity span ( Figure 2 and Figure 2—figure supplement 1d ) . It is interesting to note that plasticity in auditory cortex PV neurons during the first days following auditory nerve damage predicts the eventual recovery of cortical sound processing , which is achieved via increased A1 gain modulation weeks later ( Resnik and Polley , 2017 ) . The mechanisms of this PV neuronal plasticity remain unknown; however , activity-dependent changes in zinc levels may contribute to this plasticity – future experiments will be needed to test this hypothesis . We observed both NMDAR-dependent and NMDAR-independent zinc-mediated effects on A1 responses to sound . This is consistent with in vitro studies showing that synaptic zinc inhibits , albeit at different concentrations , NMDARs ( Paoletti et al . , 1997; Hansen et al . , 2014 ) , AMPA receptors ( Kalappa et al . , 2015; Kalappa and Tzounopoulos , 2017 ) , and reduces vesicular release probability via endocannabinoid signaling ( Perez-Rosello et al . , 2013; Kalappa and Tzounopoulos , 2017 ) . Although previous studies have reported potentiation , inhibition and no effects of zinc on AMPARs ( McAllister and Dyck , 2017 ) , the vast majority of these did not employ ZX1 , which , due to its improved kinetic and zinc binding properties , is the most appropriate chelator for investigating the effects of fast , transient elevations of synaptic zinc on synaptic targets ( Pan et al . , 2011; Anderson et al . , 2015; Kalappa et al . , 2015 ) . However , ZX1 had no effects on synaptic AMPA currents in mossy fiber-CA3 synapses , suggesting that the effects of zinc on synaptic AMPA responses are subunit- and/or synapse-specific . Together , our results suggest that synaptic zinc shapes the gain of L2/3 neurons via multiple zinc-interacting targets , including NMDARs . In the presence of APV , the lack of ZX1 effects on the sound-evoked responses of SOM and VIP neurons may be due to potential circuit- or to direct but multiple target-based cancelation of ZX1 effects , and it therefore does not necessarily support total NMDAR-dependence . Despite these limitations in the interpretation of the APV experiments showing negative results , our APV experiments showing enhancing effects of ZX1 on the sound-evoked responses of principal and PV neurons , suggest that synaptic zinc exerts its effects on these neurons , at least partially , in an NMDAR-independent manner . Nonetheless , future experiments investigating the effect of sound-evoked zinc on subthreshold conductances mediating the sound-evoked response of different types of neurons in A1; in vitro experiments on the effect of synaptic zinc on NMDAR and AMPAR EPSCs in A1 synapses; as well as the use of optogenetic approaches are needed to further resolve the detailed molecular mechanisms via which synaptic zinc shapes the sound-evoked responses of A1 L2/3 neurons . Our findings are mostly specific to cells and circuits in L2/3 . In the neocortex , synaptic zinc containing terminals are found predominantly in L1 , 2/3 and 5 , with moderate presence in L6 and light presence in L4 ( McAllister and Dyck , 2017 ) . This distribution together with the differential distribution of the different classes of interneurons – with VIP neurons more predominant in the superficial layers , SOM neurons more predominant in the deeper layers , and PV neurons present throughout the cortex ( Tremblay et al . , 2016 ) – make it unclear whether our findings apply to the deeper layers of cortex . Future studies will be required to address the effects of synaptic zinc on the gain of neurons and circuits located outside of L2/3 . Together , our results and previous findings on the role of zinc in modulating glutamatergic neurotransmission demonstrate that synaptic zinc fine-tunes excitatory synaptic transmission and sensory processing . Although this fine-tuning does not impair baseline neurotransmission or basic sensory and sensorimotor functions , such as hearing thresholds and prepulse inhibition ( Cole et al . , 2001; Thackray et al . , 2017 ) , it has profound consequences on synaptic plasticity ( Pan et al . , 2011 ) , sound-evoked cortical processing and more intricate , yet ethologically crucial sensory processing tasks , such as gain modulation . We propose that zinc is a novel ‘knob’ in the brain that tunes cortical gain . All procedures were approved by the Institutional Animal Care and Use Committee at the University of Pittsburgh . Male and female ZnT3 ( Slc30a3 ) KO and WT mice ( Jackson , Bar Harbor , ME ) , were used for experiments shown in Figure 1a–m , Figure 1—figure supplement 1a–c , f , and Figure 2—figure supplement 1bc . Male and female ICR mice ( Harlan , Indianapolis , IN ) were used for experiments shown in Figure 2a–c , Figure 3a , b , Figure 4a–d , Figure 1—figure supplement 1a–f , Figure 2—figure supplement 1a , d ( AAV-CaMKII-GCaMP6f ) , and Figure 3—figure supplement 1ab . PV ( Pvalb ) -Cre mice ( Jackson , Bar Harbor , ME ) were used for experiments shown in Figure 2d–f , Figure 3a–c , Figure 4e–h , Figure 2—figure supplement 1a–d , and Figure 3—figure supplement 1cd . SOM ( Sst ) -Cre mice ( Jackson , Bar Harbor , ME ) were used for experiments shown in Figure 2g–i , Figure 3d , e , Figure 5a–d , Figure 2—figure supplement 1a–d , and Figure 3—figure supplement 1ef . VIP-Cre mice ( Jackson , Bar Harbor , ME ) were used for experiments shown in Figure 2j–l , Figure 3d , f , Figure 5e–h , Figure 2—figure supplement 1a , d , and Figure 3—figure supplement 1gh . Mice between postnatal day ( P ) 19 and P36 were anesthetized with inhaled isoflurane ( induction: 3% in oxygen , maintenance: 1 . 5% in oxygen ) and secured in a stereotaxic frame ( Kopf , Tujunga , CA ) . Core body temperature was maintained at ~37°C with a heating pad and eyes were protected with ophthalmic ointment . Lidocaine ( 1% ) was injected under the scalp and an incision was made into the skin at the midline to expose the skull . Using a 27-gauge needle as a scalpel , a small craniotomy ( ~0 . 4 mm diameter ) was made over the temporal cortex ( ~4 mm lateral to lambda ) . A glass micropipette , containing AAV vectors , driven by synapsin promoter for neuron-specific expression , was inserted into the cortex 1 mm past the surface of the skull with a micromanipulator ( Kopf ) . We used AAV9 . CaMKII . GCaMP6fast . WPRE . SV40 ( GCaMP6f ) and AAV9 . CAG . Flex . GCaMP6f . WPRE . SV40 for two-photon imaging and AAV9 . Syn . GCaMP6slow . WPRE . SV40 ( GCaMP6s ) , AAV9 . Syn . GCaMP6f . WPRE . SV40 , AAV9 . CaMKII . GCaMP6f . WPRE . SV40 , and AAV9 . CAG . Flex . GCaMP6f . WPRE . SV40 for transcranial imaging experiments ( titer 5e12 – 5e13 genome copies/mL , Penn Vector Core; [Chen et al . , 2013] ) . The glass micropipette was backfilled with mineral oil and connected to a 5 μL glass syringe ( Hamilton , Reno , NV ) . We used a syringe pump ( World Precision Instruments , Sarasota , FL ) to inject 200–400 nL of this solution over the course of 5 min . The pipette was left in place for 2 min after the end of the injection . The pipette was then removed and the scalp of the mouse was closed with cyanoacrylate adhesive . Mice were fed a diet containing the non-steroidal anti-inflammatory drug carprofen ( Medigel , Westbrook , ME ) for 24 hr prior to and 48 hr after surgery . Mice were monitored for signs of postoperative stress and pain . 11–24 days after AAV injections , mice were prepared for in vivo calcium imaging . Mice were anesthetized with inhaled isoflurane ( induction: 3% in oxygen , maintenance: 1 . 5% in oxygen ) and positioned into a head holder . Core body temperature was maintained at ~37°C with a heating pad and eyes were protected with ophthalmic ointment . Lidocaine ( 1% ) was injected under the scalp and an incision ( ~1 . 5 cm long ) was made into the skin over the right temporal cortex . The head of the mouse was rotated ~45 degrees in the coronal plane to align the pial surface of the right temporal cortex with the imaging plane of the upright microscope optics . The skull of the mouse was secured to the head holder using dental acrylic ( Lang , Wheeling , IL ) and cyanoacrylate adhesive . A tube ( the barrel of a 25 mL syringe or an SM1 tube from Thorlabs , Newton , NJ ) was placed around the animal’s body to reduce movement , and the mouse received an injection of the sedative chlorprothixene ( 0 . 36 mg/kg intramuscular ) to reduce animal movement during in vivo imaging ( Chen et al . , 2013; Kato et al . , 2015 ) . A dental acrylic reservoir was created to hold warm artificial cerebrospinal fluid ( ACSF ) over the exposed skull . In preparing the ACSF , we removed contaminating zinc by incubating with Chelex 100 resin ( Biorad , Hercules , CA ) for 1 hr . Subsequently , we removed the Chelex by vacuum filtration , and added high purity calcium and magnesium salts ( 99 . 995% purity; Sigma-Aldrich , St . Louis , MO ) . The solution contained in millimolar: 130 NaCl , 3 KCl , 2 . 4 CaCl2 , 1 . 3 MgCl2 , 20 NaHCO3 , 3 Hepes , and 10 D-glucose , pH = 7 . 25–7 . 35 , ∼300 mOsm . For better optical access of the auditory cortex , we injected lidocaine-epinephrine ( 2% lidocaine , 1/100 , 000 weight/volume epinephrine ) into the temporal muscle and retracted a small portion of the muscle from the skull . Mice were then positioned under the microscope objective in a sound- and light-attenuation chamber containing the microscope and a calibrated speaker ( ES1 , Tucker-Davis Davis Technologies , Alachua , FL ) . Acoustic stimuli were calibrated with ¼ inch microphone ( Brüel and Kjær , Nærum , Denmark ) placed at the location of the animal’s ear within the chamber . We performed transcranial imaging to locate A1 in each mouse . We removed the isoflurane from the oxygen flowing to the animal and began imaging sound-evoked responses at least ten minutes later ( Issa et al . , 2014 ) . Sounds were delivered from a free-field speaker 10 cm from the left ear of the animal ( ES1 speaker , ED1 driver , Tucker-Davis Technologies ) , controlled by a digital to analog converter with an output rate of 250 kHz ( USB-6229 , National Instruments , Austin , TX ) . We used ephus ( Suter et al . , 2010 ) to generate the sound waveforms and synchronize the sound delivery and image acquisition hardware . We presented 50 or 60 dB SPL , 5 or 6 kHz tones to the animal while illuminating the skull with a blue LED ( nominal wavelength of 490 nm , M490L2 , Thorlabs ) . We imaged the change in green GCaMP6 emission with epifluorescence optics ( eGFP filter set , U-N41017 , Olympus , Center Valley , PA ) and a 4X objective ( Olympus ) using a cooled CCD camera ( Rolera , Q-Imaging , Surrey , BC , Canada ) . Images were acquired at a resolution of 174 × 130 pixels ( using 4X spatial binning , each pixel covered an area of = 171 . 1 μm2 of the image ) at a frame rate of 20 Hz . After locating A1 in each animal ( see analysis section below ) , we reanesthetized the mouse with isoflurane and performed a small craniotomy ( 0 . 4–1 mm2 ) adjacent to the location of A1 ( the edge of craniotomy was ~0 . 5 mm medial and ~0 . 5 mm anterior to A1 ) . Using a micromanipulator ( Siskiyou , Grants Pass , OR ) , we inserted a glass micropipette backfilled with mineral oil and connected to a 5 μL glass syringe into the cortex as above . The pipette contained ACSF including 100 μM of ZX1 and 50 μM Alexa-594 . Once the pipette was inserted into the cortex , we removed the isoflurane . After ten minutes of recovery from isoflurane , we presented sound stimuli ( 12 kHz tones , 30–80 dB SPL , 0 . 4 s duration , 10 msec ramps ) while measuring the changes in GCaMP6 fluorescence . After recording the responses to different sounds ( 3 to 5 presentations of each sound level ) , we began to infuse the ZX1 solution into the cortex at a rate of 30 nL/min . We monitored the intracortical spread of the ZX1 solution with red epifluorescence optics ( excitation: FF01-543/22 , emission: FF01-593/40 , dichroic: Di02-R561 , Semrock , Rochester , NY ) and a green LED ( nominal wavelength of 530 nm , M530L2 , Thorlabs ) for transcranial illumination . After ten to twenty minutes , we observed strong transcranial red fluorescence throughout A1 indicating that the ZX1 solution was present in the cortex . At this point we reduced the pump speed to 9 nL/min and remeasured the sound-evoked responses . For experiments in which we infused 1 mM APV and then 1 mM APV and 100 μM ZX1 into the cortex , we constructed two-barrel infusion pipettes by hot-gluing 2 infusion pipettes together so that their tips were within ~200 μm of each other . Each pipette was backfilled with mineral oil and connected to a 5 μL glass syringe as above . One pipette contained ACSF with APV and 10 μM Alexa-594 , and the other contained ACSF with APV , ZX1 , and 50 μM Alexa-594 . We inserted the tips of both pipettes into the small craniotomy as above . We measured control sound-evoked responses , infused APV and remeasured these responses , and then infused APV and ZX1 and measured these responses a third time . We monitored the diffusion of each solution in the cortex by the increase in red transcranial fluorescence as above . We followed a similar approach for the sequential infusion of vehicle and ZX1 in Figure 1—figure supplement 1d , e . Mice were euthanized at the end of the recording session . To localize A1 , we used 50 or 60 dB SPL , 5 or 6 kHz tones and we normalized the sound-evoked change in fluorescence after sound presentation ( ΔF ) to the baseline fluorescence ( F ) , where F is the average fluorescence of 1 s preceding the sound onset ( for each pixel in the movie ) . We applied a two-dimensional , low-pass Butterworth filter to each frame of the ΔF/F movie , and then created an image consisting of a temporal average of 10 consecutive frames ( 0 . 5 s ) ; the temporal average started at the end of the sound stimulus . This image indicated two sound-responsive regions corresponding to the low frequency tonotopic areas of A1 and the AAF ( Figure 1 ) . A region of interest ( ROI , 150–200 μm x 150–200 μm ) over A1 was then used to quantify the sound-evoked responses to 12 kHz sounds . We averaged the fluorescent intensity from all pixels in the ROI for each frame and normalized the ΔF to the F of the ROI to yield ΔF/F responses . ΔF/F responses from 3 to 5 presentations of the same sound level and frequency were averaged . Response amplitude was the peak ( 50 msec window ) of the transcranial response that occurred within one second of the sound onset . To quantify of the effects of ZX1 on the type of gain change ( multiplicative/divisive and additive/subtractive ) the response amplitudes from each mouse ( in control and ZX1 ) were normalized to the largest response in control . The normalized response to each sound in ZX1 was then plotted against the corresponding control response , and the slope and y-intercept were quantified with a regression line fit through the data . For 2-photon imaging in awake mice , we followed the same steps as above to locate A1 , but created a larger craniotomy ( ~1 mm2 ) over A1 for improved optical access , and inserted the pipette containing ZX1 into the cortex at the edge of this craniotomy . Mode-locked infrared laser light ( 940 nm , 100–200 mW intensity at the back focal plane of the objective , MaiTai HP , Newport , Santa Clara , CA ) was delivered through a galvanometer-based scanning 2-photon microscope ( Scientifica , Uckfield , UK ) controlled with scanimage ( Pologruto et al . , 2003 ) , using a 40X , 0 . 8 NA objective ( Olympus ) with motorized stage and focus controls . We imaged green and red fluorescence simultaneously with 2 photomultiplier tubes using red and green emission filters ( FF01-593/40 , FF03-525/50 , Semrock ) and a dichroic splitter ( Di02-R561 , Semrock ) . We acquired movies at a frame rate of 5 Hz over an area of 145 μm x 145 μm and at a resolution of 256 × 256 pixels . We imaged neurons in L2/3 at an average depth of 197 μm ± 34 μm from pia , the range represents standard deviation . After identifying A1 L2/3 neurons responding to sounds , we presented different levels ( 30 , 50 , 70 dB SPL ) of 12 kHz tones ( 500 msec duration , 20 msec ramps ) while monitoring the changes in GCaMP6f fluorescence . We recorded neuronal activity in ten-second long movies and presented sound stimuli 4 s after the start of each movie . We presented different sound stimuli every 30 s . After obtaining movies of responses to different sound stimuli , we began to infuse ZX1 . Once ZX1 diffused in A1 , we remeasured the responses of the same neurons to the same sounds . Mice were euthanized at the end of the recording session . To quantify the neuronal responses to sounds we identified neurons that were present in the field of view before and after ZX1 infusion and targeted only these cells for analysis . Using FluoroSNNAP software ( Patel et al . , 2015 ) , we selected ROIs within the soma of each L2/3 neuron from the temporal average of each movie ( 50 frames , ten-sec long ) . The pixels in each ROI from each frame were averaged and converted into ΔF/F as above . We then averaged the fluorescent response for 4–7 presentations of the same sound intensity and frequency for each neuron . Sound-evoked responses were measured for 1 s after of the sound onset and were defined as responses if the sound-evoked increases in ΔF/F were larger than the mean +3 standard deviations of the baseline fluorescence measured prior to the sound onset . The response was quantified as the integral of the fluorescence during this 1 s period . For neurons that responded to 12 kHz tones , we quantified the type of gain change ( multiplicative/divisive and additive/subtractive ) following zinc chelation by plotting the sound-evoked response amplitudes in ZX1 against the corresponding response amplitudes in control conditions . We quantified the slope and y-intercept of this relationship for each neuron with a major axis regression ( Phillips and Hasenstaub , 2016 ) . To image GCaMP6f expressing neurons in brain slices , mice that had undergone AAV injections were deeply anesthetized with isoflurane and decapitated . Brains were quickly removed and sectioned , with a vibratome ( Leica , Buffalo Grove , IL ) , into 300 μm acute slices containing the temporal cortex . We prepared the slices in a solution containing ( in mM ) : 2 . 5 KCl , 1 . 25 NaH2PO4 , 25 NaHCO3 , 0 . 5 CaCl2 , 7 , MgCl2 , 7 dextrose , 205 sucrose , 1 . 3 ascorbic acid , and 3 sodium pyruvate ( pH 7 . 35 , bubbled with 95% O2/5% CO2 ) . The slices were then transferred and incubated at 36°C in ACSF ( see above ) for 30 min . Subsequently , slices were transferred to a solution containing 4% paraformaldehyde ( Electron Microscopy Sciences , Hattfield , PA ) in 0 . 01M phosphate buffered saline ( PBS ) and incubated at 4°C overnight . Following this fixation , slices were rinsed five times in PBS and mounted on glass slides . We acquired fluorescent images of GCaMP6f expressing neurons in cortical slices with 2-photon microscopy ( see above ) . Experiments with WT and ZnT3 KO were blinded . Analysis was performed with MATLAB ( Mathworks , Natick , MA ) and QuickCalcs ( Graphpad , La Jolla , CA ) . Group data are presented as mean ± standard error of the mean . Pairwise comparisons between groups were performed with the Student’s paired t-test , t-test or one sample t-test ( for normally distributed data ) or the Wilcoxon signed-rank or rank sum tests ( for non-normally distributed data ) . Normality of the distribution of data was assessed with the Lilliefors test . Significance is defined as p<0 . 05 . The sample size for experiments was chosen to be consistent with similar studies in the field , such as in ( Kato et al . , 2015; Phillips and Hasenstaub , 2016 ) .	Many people find it easy to follow a conversation while on a busy city street , but this seemingly simple task requires sophisticated processing of sounds . The brain must accurately distinguish speech sounds from background noise , even though the volumes and pitches of those sounds overlap . To make this possible , neurons that process sounds continually adjust the relationship between the volume of a sound and the size of their response . This helps the brain to distinguish more precisely between different sounds , but how this works remains unclear . Zinc ions form part of almost 3 , 000 different enzymes and regulatory proteins , and also help neurons to communicate with one another at junctions called synapses . Changes to the amount of zinc ions at the synapses have been seen in disorders including depression and Alzheimer’s disease . By imaging the brains of mice , Anderson , Kumar et al . now show that zinc ions affect how the healthy brain processes sounds . Treating the mice with a substance that temporarily mops up zinc ions changed how neurons responded to sounds of different volumes . This revealed that zinc ions cause excitatory neurons , which activate neighboring cells , to increase their responses to sounds . Conversely , zinc ions cause inhibitory neurons , which reduce the activity of other cells , to decrease their responses to sounds . The overall effect is to change the balance of excitatory and inhibitory activity in areas of the brain that process sound . Anderson , Kumar et al . propose that these changes make it easier for the brain to process and distinguish different sounds as the environment changes from quiet to loud and vice versa . As well as revealing a role for zinc ions in normal hearing , these findings may help us to understand disorders such as tinnitus and auditory neuropathies ( conditions where the nerve that carries signals from the ear to the brain is damaged , leading to hearing loss ) . Both tinnitus and auditory neuropathies involve changes in the brain’s ability to increase or decrease its responses to sounds with particular characteristics – processes that may involve the activity of zinc ions .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Cell-specific gain modulation by synaptically released zinc in cortical circuits of audition
Only a subset of cancer patients respond to T-cell checkpoint inhibitors , highlighting the need for alternative immunotherapeutics . We performed CRISPR-Cas9 screens in a leukemia cell line to identify perturbations that enhance natural killer effector functions . Our screens defined critical components of the tumor-immune synapse and highlighted the importance of cancer cell interferon-γ signaling in modulating NK activity . Surprisingly , disrupting the ubiquitin ligase substrate adaptor DCAF15 strongly sensitized cancer cells to NK-mediated clearance . DCAF15 disruption induced an inflamed state in leukemic cells , including increased expression of lymphocyte costimulatory molecules . Proteomic and biochemical analysis revealed that cohesin complex members were endogenous client substrates of DCAF15 . Genetic disruption of DCAF15 was phenocopied by treatment with indisulam , an anticancer drug that functions through DCAF15 engagement . In AML patients , reduced DCAF15 expression was associated with improved survival . These findings suggest that DCAF15 inhibition may have useful immunomodulatory properties in the treatment of myeloid neoplasms . Major advances in tumor control have recently been achieved by targeting immune inhibitory signaling pathways . Treatment with ‘checkpoint inhibitors , ’ antibodies targeting PD1 , PD-L1 , or CTLA4 , lead to durable responses across a wide range of indications , but only in a subset of patients . Treatment response is positively correlated with tumor mutational burden and infiltration of CD8+ effector T cells , which recognize tumor cells via peptides bound to major histocompatibility complex class I ( MHC-I ) molecules , suggesting that checkpoint inhibitors work best at clearing highly immunogenic cancers with repressed T cell responses ( Snyder et al . , 2014; Tumeh et al . , 2014; Mariathasan et al . , 2018 ) . Substantial efforts are being made to extend the benefits of immunotherapy to additional patients , including combining checkpoint inhibitors with other therapies , drugging additional lymphocyte-suppressive pathways , and boosting the activity of other arms of the immune system ( Galon and Bruni , 2019 ) . Resistance to therapy has long been a major problem in cancer treatment . Drugs targeting tumor growth pathways can profoundly reduce tumor burden , but resistance invariably arises , driven by the substantial genetic and phenotypic heterogeneity present within human tumors ( Easwaran et al . , 2014 ) . Recent clinical and experimental data have similarly highlighted the ability of cancer cells to escape checkpoint inhibitor-induced immune control . B2M and JAK1/2 mutations have been identified in melanoma patients with acquired resistance to checkpoint inhibitors ( Zaretsky et al . , 2016; Sade-Feldman et al . , 2017 ) . These mutations impair recognition of the tumor by the adaptive immune system , either by directly disrupting antigen presentation or by rendering the cells insensitive to IFNγ , an important inducer of MHC-I expression . Functional genetic screens using T cell-cancer cell cocultures have highlighted similar mechanisms of resistance in vitro ( Kearney et al . , 2018; Pan et al . , 2018; Manguso et al . , 2017; Patel et al . , 2017 ) . Even treatment-naïve tumors can be highly immuno-edited , presenting with IFNγ pathway mutations , reduced MHC-I expression and loss of the peptide sequences that can serve as antigens ( Dunn et al . , 2004; Rodig et al . , 2018; Gao et al . , 2016; McGranahan et al . , 2017 ) . Together , these findings highlight a critical need for therapies that can either increase MHC expression or work in a MHC-independent fashion . Anti-tumor immunity is not solely mediated by the adaptive immune compartment . Innate immune cells , most notably natural killer ( NK ) cells , can have both direct tumoricidal activity and also help to fully elaborate long-lasting anti-tumor responses ( Marcus et al . , 2014; Moynihan et al . , 2016; Lopez-Soto et al . , 2017; Chiossone et al . , 2018 ) . NK cells are cytotoxic lymphocytes capable of mounting rapid responses to damaged , infected , or stressed cells , including cancer cells . T and NK cells share effector functions , releasing cytokines and exocytosing lytic granules upon activation to kill target cells . However , NK activation status is controlled by the integrated signals from germline-encoded NK-activating and -inhibiting receptors ( aNKRs/iNKRs ) . Generally , iNKR ligands are expressed by normal and healthy cells , whereas aNKR ligands are upregulated after DNA damage or viral insult ( Chiossone et al . , 2018; Ljunggren and Malmberg , 2007 ) . MHC-I molecules provide a potent inhibitory signal sensed by NK cells , enabling the innate immune system to respond productively to MHC-deficient cells . As a result , there is considerable interest in amplifying NK responses to cancers , as well as developing NK-based cell therapies ( Lopez-Soto et al . , 2017; Chiossone et al . , 2018; Ljunggren and Malmberg , 2007 ) . Here , we performed genetic screens in an MHC-deficient leukemic cell line to systematically identify modulators of NK-mediated anti-cancer immunity . These screens , unexpectedly , revealed the potential therapeutic utility of targeting the cullin4-RING E3 ubiquitin ligase ( CRL4 ) substrate adaptor DCAF15 in myeloid malignancies . Disruption of DCAF15 strongly sensitized cancer cells to NK-mediated killing , resulting from increased cancer cell expression of lymphocyte costimulatory molecules . Proteomic experiments revealed that DCAF15 interacted with and promoted the ubiqitination of the cohesin complex members . Treatment with indisulam , an anticancer drug that modulates DCAF15 function , reduced interaction with cohesin members and mimicked DCAF15 loss-of-function immunophenotypes . We performed genome-scale CRISPR screens in K562 cells to identify perturbations that modulate NK-92-mediated killing ( Figure 1A ) . K562 human chronic myelogenous leukemia cells are a NK-sensitive cancer cell line that weakly expresses MHC-I . For screening purposes , a clonal isolate of K562 cells expressing high levels of spCas9 was generated and validated ( Figure 1—figure supplement 1A–C ) . NK-92 cells are a human lymphoma-derived cell line phenotypically similar to activated NK cells ( Klingemann et al . , 1994 ) . These cells exhibit interleukin-2 ( IL-2 ) -dependent growth , express a large number of aNKRs and few iNKRs ( Maki et al . , 2001 ) , and display potent cytolytic activity against K562 cells . In pilot experiments , labeled K562 cells were co-cultured with NK-92 cells to determine an effector-to-target ( E:T ) ratio that applied sufficient selective pressure for screening ( Figure 1—figure supplement 1D–F ) . IL-2 was removed during the co-culture to promote the eventual death of NK-92 cells , allowing the collection of genomic DNA preparations undiluted by effector cell DNA . A multi-day timeframe between the NK-92 challenge and screen readout was used to capture tumor cell fitness changes related both to the direct cytolytic activities of NK cells as well as the longer-term effects from NK-released cytokines . For the co-culture screen , cas9-expressing K562 cells were infected with a genome-scale single guide RNA ( sgRNA ) library targeting all unique coding genes and miRNAs , as well as one thousand non-targeting controls ( Supplementary file 1 ) . Seven days post-infection , cells were either grown normally or challenged with NK-92 cells at a 1:1 or 2 . 5:1 E:T ratio , reducing K562 cell counts 19-fold or 43-fold , respectively , by the end of the screen ( Figure 1B ) . Deep sequencing was used to compare changes in sgRNA abundance between the challenged and unchallenged state after 8 days of co-culture , and genes were ranked using the MAGeCK software ( Li et al . , 2014 ) . ( Figure 1C , Supplementary file 2 ) There was good agreement between the results from screens performed at the different E:T ratios ( Figure 1—figure supplement 2 ) . The screen revealed two broad classes of ‘hits’— sgRNAs targeting components of the tumor-immune synapse or components of the IFNγ signaling pathway ( Figure 1C–D ) . Disruption of ICAM1 was the top-ranked NK-92 evasion mechanism , scoring many orders of magnitude stronger than any other gene—an observation consistent with the critical role of ICAM1-LFA1 interactions in establishing initial target-lymphocyte adhesion and polarizing cytotoxic granules towards the synapse ( Marcus et al . , 2014 ) . Single guide RNAs targeting multiple other tumor-immune synapse components were also enriched after NK-92 challenge , including NCR3LG1 ( #26-ranked gene by MaGeCK score ) , the activating ligand for NKP30 on NK cells ( Brandt et al . , 2009 ) ; CD58 ( #37 ) , an adhesion molecule that binds CD2 ( Selvaraj et al . , 1987; Rölle et al . , 2016 ) ; and CD84 ( #80 ) , a SLAM-related receptor that binds homotypically to promote activation and cytokine secretion in lymphocytes ( Martin et al . , 2001; Veillette , 2006; Wang et al . , 2010 ) . Other than NECTIN2 , sgRNAs targeting NK-inhibitory surface proteins did not score prominently as NK-sensitization mechanisms , consistent with the weak MHC-I expression on K562 cells and the limited repertoire of NK inhibitory receptors expressed on NK-92 cells ( Maki et al . , 2001 ) . NECTIN2 transmits both stimulatory or inhibitory signals to NK cells , depending on whether it is bound to DNAM1 ( CD226 ) or TIGIT , respectively ( Stanietsky et al . , 2009; Bottino , 2003 ) . After ICAM1 , the top 10 highest scoring NK-92 evasion mechanisms were dominated by sgRNAs targeting the proximal components of the IFNγ signaling pathway , including STAT1 , JAK1 , IFNGR2 , JAK2 and INFGR1 ( Figure 1D ) . Consistent with the importance of cancer cell IFNγ signaling , sgRNAs targeting negative regulators of the interferon response were strongly depleted after NK-92 challenge . Disruption of the protein tyrosine phosphatases PTPN2 and PTPN1 were the #2 and #5 ranked NK-92 -sensitizing mechanisms , respectively . Presumably , these proteins suppress IFNγ-induced immunomodulation by dephosphorylating STAT and JAK proteins , as has been reported in CRISPR screens using T-cell coculture systems or syngeneic tumor models ( Pan et al . , 2018; Manguso et al . , 2017 ) . Taken together , these findings indicate that our in vitro functional genomics screens effectively revealed known components of physiologically-relevant immune synapse and cytokine pathways . Mechanisms of NK-92 sensitization identified in the screen were diverse , revealing many strongly-scoring genes not previously linked to either interferon signaling or NK cell biology ( Figure 1 ) . Most surprisingly , the top-ranked mechanism for promoting NK-92 mediated clearance was disruption of DCAF15 , an uncharacterized substrate adaptor for CRL4 ubiquitin E3 ligases . DCAF15 is a member of the large family of DDB1 and Cul4-associated factors ( DCAFs ) ( Jin et al . , 2006 ) . CRL4 complexes enable cells to mark proteins for proteosomal degradation , helping regulate intracellular protein homeostasis . As substrate adaptors for CRL4 , DCAF proteins provide specificity to the complex , determining which client proteins are ubiquitinated ( Jackson and Xiong , 2009 ) . As with most substrate adaptors , the normal client repertoire of DCAF15 is undefined , and relatively little is known about the biological function of DCAF15 . We also noted that disruption of two cohesin-related genes , STAG2 and HDAC8 , scored as NK-92 sensitization factors ( ranked #26 and #19 , respectively ) . Cohesin is a ring-shaped complex involved in chromatin replication , organization and repair , with STAG2 acting as a core complex member and HDAC8 controlling chromatin accessibility ( Uhlmann , 2016 ) . Cohesin dysregulation has cell context-specific consequences , including DNA damage and aneuploidy; in leukemic cells , cohesin mutations are thought to enforce stem cell programs by altering chromatin organization ( Mazumdar and Majeti , 2017 ) . The prominent role for IFNγ signaling in the immune response to cancer cells , both clinically and in our screens , prompted us to define more specifically which NK-92-sensitizing genes are involved in modulating the IFNγ response . MHC-I levels are highly upregulated in K562 cells after IFNγ exposure , increasing 5 . 9+ /- 0 . 98 fold after 24 hr of exposure to IFNγ . This induction was dependent on STAT1 and was nearly doubled by disrupting PTPN2 ( Figure 2A–B and Figure 2—figure supplement 1 ) . We therefore used IFNγ-induced cell surface MHC-I expression as a proxy for the strength of the interferon response . K562 cells transduced with a genome-scale CRISPR library were treated with IFNγ for 24 hr and MHC-I expression was measured by flow cytometry . The brightest 20% and dimmest 20% of cells were sorted , and deep sequencing was used to compare sgRNA abundance between the populations ( Figure 2C , Supplementary file 3 ) . As expected , cells with impaired MHC-I upregulation were highly enriched for sgRNAs targeting the IFNγ-JAK-STAT pathway ( IFNGR1/2 , JAK1/2 , STAT1 , IRF1/2 ) , as well as the antigen processing/presentation machinery ( B2M , TAP1/2 , TAPBP , PDIA3 , HLA-C/B ) ( Figure 2D ) . Conversely , disruption of PTPN2 or STAG2 induced an exuberant MHC-I response . Surprisingly , sgRNAs targeting epigenetic factors were highly enriched within the brightest MHC-I expressing cells—most prominently , members of the BCOR complex PCGF1 and KDM2B , members of the PRC2 complex EZH2 and SUZ12 , as well as factors affecting histone methylation/acetylation status . Rank-rank comparisons between the NK and MHC screens were informative in prospectively defining a core group of IFNγ response genes in K562 cells ( Figure 2E ) . Comparing sgRNAs enriched after NK-92 challenge with those causing impaired MHC-I upregulation clearly delineated the known proximal components of the IFNγ signaling pathway ( IFNGR1/2 , JAK1/2 , STAT1 ) , and highlighted several poorly characterized genes such as GSE1 , SPPL3 and NR2F2 ( Figure 2E ) . Surprisingly , comparing sgRNAs depleted after NK-92 challenge with those causing an exaggerated MHC-I response highlighted the CRL4 substrate adaptor DCAF15 most prominently , alongside the cohesin members STAG2 and HDAC8 ( Figure 2E ) . As expected , negative feedback regulators of the IFNγ pathway ( PTPN1 and PTPN2 ) were also recovered by this analysis . We focused additional studies on understanding the function of DCAF15 , given its prominence in both the NK sensitization ( #1 ranked hit at 2 . 5 E:T ratio; #12 ranked hit at 1:1 E:T ratio ) and MHC upregulation ( #13 ranked hit ) screens . To evaluate hits from the CRISPR screens , we generated individual gene knockout ( KO ) cell lines by lentiviral sgRNA expression , producing polyclonal cell lines with high levels of gene disruption ( Figure 2—figure supplement 1 and Figure 3—figure supplement 1 ) . Fluorescently-labeled control or test KO target cell lines were subjected to competitive co-culture assays in the presence of either NK-92 or primary NK effector cells , with changes in the relative ratios of target cell types measured over time by flow cytometry ( Figure 3A ) . As expected , disrupting ICAM1 in K562 cells conferred very high levels of protection against NK-92 cells ( Figure 3B–C; 19 . 7-fold enrichment ) . Disabling signaling downstream of IFNγ by disrupting STAT1 provided an intermediate level of resistance ( 2 . 45-fold enrichment ) . K562s are very sensitive to NK-mediated killing , providing a large dynamic range for detection of resistance-promoting factors , while limiting the ability of the assay to detect similarly large increases in sensitization . Nevertheless , multiple independent sgRNAs targeting DCAF15 promoted sensitization to NK-92 cells ( 1 . 6-fold depletion ) , with a similar degree of preferential killing observed for NECTIN2 or PTPN2 KO cells ( NECTIN2: 1 . 95-fold depletion; PTPN2: 1 . 4-fold depletion ) . We repeated NK-92 competitive co-culture experiments after disruption of DCAF15 , PTPN2 , STAT1 and ICAM1 in Daudi cells , a B2M-deficient B-cell lymphoma line ( Figure 3D , Figure 3—figure supplement 1D–F ) . ICAM1 KO Daudi cells were highly protected against NK-92 cell killing , whereas disruption of DCAF15 or PTPN2 led to enhanced killing . In contrast to K562 cells , STAT1 disruption in Daudi cells promoted their preferential killing . To extend these observations to primary NK cells , human peripheral NK cells were isolated from PBMCs of 6 healthy donors , activated and challenged in competitive co-cultures with various K562 KO cell genotypes ( Figure 3E and Figure 3—figure supplement 2 ) . Disruption of DCAF15 or PTPN2 promoted sensitization to primary NK cells , albeit with reduced magnitudes of effect compared to NK-92 cells ( PTPN2: 1 . 3-fold depletion; DCAF15: 1 . 15-fold depletion ) . In 3 out of 6 donors , NK cells showed increased degranulation , as measured by cell surface CD107a expression , when challenged with DCAF15 KO cells ( Figure 3F ) . ICAM1 disruption promoted resistance to NK cell attack , but only conferred partial protection ( 2 . 3-fold enrichment ) . The effect of STAT1 disruption was extremely variable , with STAT1 KO K562 cells strongly preferentially killed by primary NK cells from a subset of donors . These findings implicate DCAF15 and PTPN2 as novel modulators of NK-mediated cancer cell immunity . DCAF15 was a strong hit in both the NK sensitization and MHC upregulation screens , suggesting that DCAF15 disruption sensitizes K562 cells to NK-mediated killing by dysregulating the IFNγ response . Consistent with the screening results , polyclonal K562 cells expressing DCAF15 sgRNAs ( ‘DCAF15 KO cells’ ) displayed 2 . 45-fold higher levels of MHC-I than control knockout cells after 24 hr of IFNγ exposure , an effect comparable in magnitude to PTPN2 disruption ( Figure 4A ) . We then tested whether DCAF15 KO cells exhibited hallmarks of dysregulated JAK-STAT signaling , using PTPN2 KO cells as a positive control . We were unable to see any difference in induction of STAT1 phosphorylation after IFNγ exposure in DCAF15 KO cells , or differences in steady state levels of STAT1/2 , JAK1/2 or IFNGR1 ( Figure 4B and data not shown ) . DCAF15 KO cells appeared healthy and proliferated at a normal rate ( Figure 4—figure supplement 1A ) . As in wild-type K562 cells , long-term IFNγ exposure was neither cytotoxic nor cytostatic to DCAF15 KO cells ( Figure 4c and Chen et al . , 1986 ) . In contrast , PTPN2 KO cells showed higher levels of STAT1PY701 induction after IFNγ exposure , and their proliferative rate was temporarily reduced after transient exposure to IFNγ , or more substantially slowed down by continuous treatment with the cytokine ( Figure 4B–C ) . We explored the transcriptional and immunophenotypic response of cells to IFNγ treatment . RNA-seq and flow cytometry was performed on control , DCAF15 KO or PTPN2 KO K562 cells basally and after 24 hr of IFNγ exposure . Wild-type cells dramatically upregulated transcription of anti-viral genes and components of the antigen processing and presentation pathway after IFNγ treatment ( Figure 4—figure supplement 1B and Supplementary file 4 ) . On the cell surface , K562 cells exhibited STAT1-dependent upregulation of ICAM1 expression after IFNγ treatment , with a variety of other important NK ligands unaffected by cytokine treatment ( Figure 4—figure supplement 1C ) . Clustering analysis clearly showed that PTPN2 KO cells were transcriptionally distinct from control or DCAF15 KO cells both before and after cytokine exposure ( Figure 4D ) . In the basal state , PTPN2 KO cells were enriched for inflammation and interferon-associated Gene Ontology ( GO ) terms ( Figure 4—figure supplement 1D ) . After cytokine exposure , gene set enrichment analysis revealed that PTPN2 KO cells had exaggerated transcriptional responses to interferon and were also enriched for apoptotic GO gene categories ( Figure 4—figure supplement 1E ) . PTPN2 KO cells also showed greater IFNγ-induced MHC-I and ICAM1 cell surface expression ( Figure 2A and Figure 4—figure supplement 1C ) . These results suggest that loss of appropriate IFNγ negative feedback may both promote cell death and modulate NK cell interactions . In contrast , DCAF15 KO cells did not show substantial differences in their transcriptional response to IFNγ ( Figure 4D ) . However , DCAF15 KO cells were enriched for GO terms associated with NK-mediated cytotoxicity , antigen presentation and cell adhesion , consistent with our phenotypic characterization of these cells ( Figure 4E ) . Together , these findings indicate that while DCAF15 KO K562 cells exhibit a relatively normal response to IFNγ stimulation , they are nonetheless in an inflamed state primed to interact with cytotoxic lymphocytes . Intriguingly , differential expression analysis showed that one of the most significantly upregulated genes in DCAF15 KO K562 cells was CD80 ( Figure 4F; Q value 3 . 7e-23 , Beta value 0 . 98 ) . CD80 is an important co-stimulatory molecule for lymphocytes , regulating T cell activation and tolerance by ligation to CD28 , CTLA4 or PDL1 ( Chen and Flies , 2013 ) . During antigen-presenting cell ( APC ) activation , the upregulation of MHC molecules and CD80 provide critical antigenic and costimulatory signals to T cells ( Acuto and Michel , 2003 ) . K562 cells are an undifferentiated and multipotential CML cell line , well-studied for their ability to differentiate towards many different lineages , including APC-like states ( Lindner et al . , 2003 ) . We hypothesized that upregulation of MHC-I and CD80 in DCAF15 KO cells may reflect a broader acquisition of APC-like properties . Indeed , immunophenotyping of unstimulated DCAF15 KO cells revealed higher levels of the APC markers CD80 , CD40 as well as class I and II MHC molecules ( Figure 5A; 2 . 23-fold CD80 increase; 2 . 01-fold CD40 increase; 1 . 44-fold MHC-I increase; 1 . 22-fold MHC-II increase ) . DCAF15 KO cells did not display higher levels of the APC maturation marker CD83 ( Figure 5A ) . Expression levels of B7H6 , ICAM1 , ULBP2/5/6 , IFNGR1 , CD58 and NECTIN2 were either unaltered in DCAF15 KO cells or modestly changed in a fashion not expected to increase sensitivity to NK cells ( Figure 5—figure supplement 1A ) . Importantly , the changes to the DCAF15 KO cell immunophenotype could be rescued by constitutive expression of a sgRNA-resistant DCAF15 open-reading frame ( Figure 5—figure supplement 1B–D ) . Transducing tumors with the B7 ligands CD80 or CD86 can enhance anti-tumor immunity by enabling the tumor cells to directly deliver antigenic and costimulatory signals to T and NK cells ( Townsend and Allison , 1993; Chen et al . , 1992; Wilson et al . , 1999; Galea-Lauri et al . , 1999 ) . While best understood in the context of T cell biology , B7 ligands have been shown to promote NK activation , via CD28-dependent and -independent pathways ( Wilson et al . , 1999; Galea-Lauri et al . , 1999; Chambers et al . , 1996; Azuma et al . , 1992; Martín-Fontecha et al . , 1999 ) . We confirmed that NK-92 cells are CD28-positive , whereas we could not detect CD28 on peripheral CD3- CD56+ NK cells ( Figure 5—figure supplement 1E–F ) . K562 cells stably over-expressing wild-type CD80 were generated ( Figure 5B; 49-fold higher CD80 levels than endogenous ) . Over-expression was sufficient to increase K562 sensitivity to NK-92 mediated killing ( Figure 5C ) , whereas overexpression of a mutant form of CD80 carrying point mutations that abrogate CD28 binding ( Peach et al . , 1995 ) ( CD80Q65A , M72A ) had no effect . We next determined whether the increased CD80 expression in DCAF15 KO cells was important for their altered NK-92 sensitivity . Changes in NK-92 degranulation were measured after incubation with K562 cells pretreated with either control or CD80 blocking antibodies ( Figure 5D ) . As expected , ICAM1 KO cells triggered less NK-92 degranulation than control cells ( by 52 ± 12% ) and were not significantly affected by CD80 antagonism ( Figure 5E ) . DCAF15 KO cells showed a 17 ± 8% increased ability to trigger NK-92 cells and were approximately twice as sensitive to CD80 antagonism compared to control cells ( Figure 5E–F ) . Following CD80 antagonism , degranulation triggered by DCAF15 KO was not significantly different from untreated control cells . Taken together , these results indicate that DCAF15 disruption in K562 cells induces an APC-like immunophenotype conducive to promoting lymphocyte responses , with higher CD80 expression especially important for increased NK-92 cell triggering . Aryl sulfonamide drugs have demonstrated promising anti-cancer properties in hematological malignancies ( Assi et al . , 2018 ) . Recently , it was discovered that these agents work by binding DCAF15 and redirecting the ubiquitination activity of the CRL4-DCAF15 E3 ligase towards the essential splicing factor RBM39 ( Han et al . , 2017; Uehara et al . , 2017 ) . This mechanism of action is conceptually similar to that of the ‘IMiD’ thalidomide analogs , which promote the degradation of various lymphocyte transcription factors by engaging the CRL4-cereblon E3 ubiquitin ligase ( Ito et al . , 2010; Fischer et al . , 2014; Krönke et al . , 2014 ) . Presumably , sulfonamides and IMiDs also impair the degradation of the normal client proteins when they induce neomorphic activity of the substrate adaptor . This has not been proven , however , as it is difficult to systematically determine the normal substrate repertoire of adaptor proteins . We hypothesized that treating cells with low concentrations of the aryl sulfonamide indisulam would phenocopy DCAF15 depletion ( Figure 6A ) . Three-day dose-response experiments revealed that K562 cells were sensitive to indisulam , and that DCAF15 disruption reduced this sensitivity , consistent with previous reports ( Uehara et al . , 2017 ) ( Figure 6B ) . Dose-response experiments across a panel of 16 hematological cancer cell lines confirmed the reported positive relationship between DCAF15 mRNA expression levels and indisulam sensitivity ( Han et al . , 2017 ) ( Figure 6—figure supplement 1A; R2 = 0 . 33 , p=0 . 02 ) . We empirically determined that 100 nM indisulam treatment moderately lowered RBM39 levels while minimally affecting cell viability and proliferation over a four-day period ( Figure 6C-D and Uehara et al . , 2017 ) . Remarkably , this treatment regime was able to recapitulate the increased CD80 expression seen in K562 DCAF15 KO cells ( Figure 6E; 2 . 14-fold increase ) , and more modestly , the effects on MHC-I and CD40 expression ( 1 . 36-fold and 1 . 27-fold increase , respectively ) . CD80 upregulation was first detected 24 hr after treatment initiation and plateaued after 48 hr ( Figure 6—figure supplement 1B ) . Importantly , indisulam treatment did not further upregulate CD80 in K562 DCAF15 KO cells , suggesting that the pharmaco-modulation of CD80 levels was entirely mediated through DCAF15 ( Figure 6F ) . To extend these observations to other cell lines , a panel of hematological cancer cell lines was screened to identify those with detectable CD80 expression ( Figure 6—figure supplement 1C ) . The CML cell line KU812 expressed similar levels of CD80 as K562 , whereas the Daudi lymphoma cell line expressed significantly higher basal CD80 levels . These cells lines were subjected to similar 4 day low-dose regimes of indisulam , which only modestly affected the growth and viability of the cells ( Figure 6—figure supplement 1D ) . Both Daudi and KU812 cells up-regulated CD80 levels after indisulam treatment ( Figure 6G; 2 . 45-fold for Daudi , 1 . 71-fold for KU812 ) . Indisulam was not able to induce de novo CD80 expression in CD80-negative cell lines ( Figure 6—figure supplement 1E ) . Thus , in certain cellular contexts , aryl sulfonamides are immuno-modulatory agents that alter co-stimulatory protein levels by disrupting the normal functions of DCAF15 . Given the in vitro findings , we hypothesized that lower DCAF15 expression in myeloid malignancies could be associated with better clinical outcomes . We tested this hypothesis using publicly available acute myeloid leukemia ( AML ) datasets ( Bolouri et al . , 2018; Ley et al . , 2013 ) . In both adult and pediatric AML , lower expression of DCAF15 mRNA was associated with increased median overall survival time ( Figure 6H–I ) . The improved survival of DCAF15-low patients was not driven by a correlation between DCAF15 expression and more aggressive AML subtypes ( Figure 6—figure supplement 1F–G ) . Taken together , these findings indicate that lower DCAF15 function , achieved pharmacologically or by genetic means , is associated with favorable immunophenotypes in vitro and improved outcomes in AML patients . The normal substrate repertoire of DCAF15 is unknown . To systemically identify direct DCAF15 client proteins , we undertook proximity-based proteomic analysis of DCAF15 interaction partners ( Figure 7A ) . DCAF15 was fused to a promiscuous bacterial biotin ligase ( Roux et al . , 2012 ) ( ‘DCAF15-BioID’ ) and stably expressed in K562 cells , enabling recovery of interaction partners by streptavidin pull-down . We first confirmed that exogenous C-terminally tagged DCAF15 was able to rescue DCAF15 KO phenotypes and associate with CRL4 complex members DDB1 and CUL4A ( Figure 5—figure supplement 1B–D and Figure 7—figure supplement 1A ) . During stable DCAF15 overexpression , we observed that the basal concentration of biotin in the media ( ~3 μM ) was sufficient to induce BioID activity in the absence of exogenous ( 50 μM ) biotin supplementation ( Figure 7B ) . However , proteasome inhibition by MG132 increased accumulation of DCAF15-BioID and biotinylated species . As a control , results were compared to a GFP-BioID fusion , expected to generically biotinylate proteins . GFP-BioID accumulated much more readily than DCAF15-BioID , and its biotinylation activity was not affected by MG132 treatment . After 24 hr of biotin and MG132 treatment , biotinylated protein species were recovered under stringent denaturing conditions . Isobaric labeling and mass spectrometry were used to quantitatively compare the DCAF15 interactome to the GFP interactome ( Figure 7C and Supplementary file 5 ) . This approach clearly recovered DCAF15 and the core CRL4 complex , including DDA1 , DDB1 and CUL4A ( DCAF15: 155 . 6-fold change , p=2 . 5e-152; DDA1: 24 . 5-fold change , p=6 . 5e-139; DDB1: 4 . 86-fold change , p=8 . 1e-21; CUL4A: 3 . 8-fold change , p=8 . 5e-18 ) . Surprisingly , two of the most differentially biotinylated proteins were the cohesin complex members SMC1A and SMC3 ( SMC1: 2 . 39-fold change , p=0 . 00098; SMC3: 2 . 76-fold change , p=5e-5 ) . We confirmed the interaction between DCAF15-BioID and endogenous SMC1 and 3 by streptavidin-pulldown followed by western blotting ( Figure 7D ) . To determine whether this association with cohesin was a generic feature of CRL4 complexes or specific to the CRL4 loaded with DCAF15 , we examined the interaction partners of a different substrate adaptor . DCAF16 is a nuclear-localized CUL4 substrate adaptor , which , like DCAF15 , interacts with DDB1 despite lacking a canonical WD40 docking domain ( Jin et al . , 2006 ) . When stably expressed in K562 cells , DCAF16-BioID fusions accumulated similarly to DCAF15-BioID , interacted with DDB1 but did not biotinylate SMC proteins ( Figure 7E ) . As low concentrations of indisulam phenocopy certain aspects of DCAF15 depletion ( Figure 6 ) , we asked whether indisulam treatment would alter the interaction of DCAF15 with cohesin . DCAF15-BioID cells were pre-treated with indisulam for 72 hr prior to biotin and MG132 addition . This treatment regime lead to substantial indisulam-dependent biotinylation of RBM39 and reduced recovery of biotinylated SMC proteins ( Figure 7E ) . To test whether DCAF15 promoted SMC1 ubiquitination , we co-transfected 293 T cells with his-tagged ubiquitin , DCAF15 and SMC1A , and purified ubiquitinated species under denaturing conditions by nickel affinity chromatography . Co-expression of DCAF15 , but not GFP , led to the recovery of poly-ubiquitinated SMC1A ( Figure 7F ) . Treatment with indisulam reduced the amount of ubiquitinated SMC1A recovered . These orthogonal proteomic and biochemical assays support the notion that cohesin proteins are bona fide client ubiquitination substrates for DCAF15 , with these interactions impaired by DCAF15-engaging aryl sulfonamide drugs . Clinical and experimental data have revealed that disrupting optimal antigen presentation levels is a common mechanism by which cancer cells escape recognition by the adaptive immune system . We performed CRISPR-Cas9 screens using NK-92:K562 co-cultures to uncover perturbations that enhance natural killer mediated anti-cancer immunity . We discovered and characterized how disruption of DCAF15 or PTPN2 sensitizes a variety of cancer cell types to both NK-92 and primary NK cells . In addition , the screens clearly identified known lymphocyte adhesion factors and aNKR/iNKR ligands , such as ICAM1 , NCR3LG1 , CD58 , CD84 and NECTIN2 . Performing additional genetic screens on diverse cancer cell lines and natural killer subtypes will enable a more complete understanding of the relative importance of various aNKRs and iNKRs and should identify novel immunotherapeutic targets . There was a strong IFNγ signature within our NK-92:K562 screens . We determined that disruption of PTPN2 , a negative regulator of IFNγ signaling , consistently enhanced NK cell sensitivity . Interestingly , preventing cancer cell IFNγ signaling had a much more variable effect than removing negative feedback on the pathway , promoting resistance or sensitization to NK cells in target and effector cell dependent manner . This variability likely reflects the complexity of cancer cell IFNγ signaling , which can include MHC upregulation , growth suppression and immunomodulation . In K562 cells , PTPN2 KO rendered IFNγ treatment growth suppressive , likely through the induction of apoptosis , while also enhancing the immunomodulatory effects of the cytokine . The immunophenotypic changes include enhanced MHC-I expression , which likely inhibits full activation of NK cell cytotoxicity , especially in primary NK cells that express a broader KIR repertoire than NK-92 cells ( Maki et al . , 2001 ) . PTPN2 disruption has previously been shown to enhance activated T-cell mediated killing , as well as potentiate the effect of immunotherapy in syngeneic tumor models ( Pan et al . , 2018; Manguso et al . , 2017; Luo et al . , 2018 ) . These data suggest that targeting PTPN2 may be a generalizable strategy to sensitize tumor cells to multiple arms of the immune system . The importance of the cancer cell IFNγ response prompted us to systemically identify perturbations that modulated IFNγ signaling , as read out by cytokine-induced MHC-I upregulation . As expected , this screen clearly delineated the proximal components of the IFNγ-JAK-STAT pathway , as well as the antigen processing/presentation machinery . Surprisingly , disruption of DCAF15 , a poorly characterized substrate adaptor for the CRL4 E3 ligase , was a top scoring hit in both the MHC-I and NK screens . We therefore focused our efforts on understanding the role of DCAF15 in this context . We determined that DCAF15 KO cells did not have a grossly dysregulated response to IFNγ , but scored in the MHC-I screen due to higher unstimulated levels of MHC-I . This change reflected a broader phenotypic switch in DCAF15 KO cells reminiscent of APC activation , including the upregulation of a variety of co-stimulatory and antigen-presenting molecules ( Figure 7G ) . Higher CD80 levels in DCAF15 KO cells were especially important for increasing NK-92 cell triggering . Cross-talk between APCs and NK cells is a well-established phenomenon that mutually regulates both cell types ( Ferlazzo and Münz , 2004 ) . These interactions , which often occur at sites of inflammation or secondary lymphoid organs , can promote APC maturation or lysis in a context-dependent fashion . Further work is needed to determine whether DCAF15 plays a role in the activation of APCs and their interactions with NK cells . Our data suggest that the surface factors upregulated by DCAF15 disruption are not direct client proteins of the substrate receptor , but rather represent events secondary to altered turnover of the normal DCAF15 client protein ( s ) . We pioneered a novel approach to systemically purify substrates of DCAF family members , as conventional biochemical methods are poorly suited to recover the transient , low-affinity interactions between substrate adaptors and client proteins . The use of DCAF-BioID fusion proteins and proteasome inhibition protects labile substrates from degradation and robustly recovers biotinylated proteins under stringent , denaturing conditions . We anticipate this will be a generalizable strategy for discovering client proteins for the whole family of CUL4 CRL substrate receptors . Proteomic analysis and subsequent validation experiments showed that DCAF15 loaded into CLR4 complexes , interacted with cohesin complex members SMC1 and SMC3 , and promoted their ubiquitination . We were intrigued by these interactions given the similar CRISPR screening scoring pattern of DCAF15 to the cohesin factors STAG2 and HDAC8 ( Figure 2E ) ; the shared roles of cohesin and CRL4 E3 ligases in DNA metabolism , organization , replication and repair ( Uhlmann , 2016; O'Connell and Harper , 2007; Litwin et al . , 2018 ) ; and the ability of cohesin mutations to dysregulate hematopoietic differentiation in myeloid malignancies ( Mazumdar and Majeti , 2017 ) . Rather than globally controlling SMC protein levels , we speculate that CRL4-DCAF15 complexes ubiquitinate cohesin at specific genomic sites to regulate chromatin topology or repair ( Figure 7G ) . In this model , disruption of either STAG2 , HDAC8 or DCAF15 impair cohesin function with overlapping phenotypic consequences . Further work is needed to elaborate the control of cohesin function by DCAF15 and how this may promote APC-like differentiation . Recently , it was discovered that aryl sulfonamide drugs including indisulam are capable of binding to DCAF15 and altering CRL-DCAF15 substrate specificity towards the splicing factor RBM39 ( Han et al . , 2017; Uehara et al . , 2017 ) . The cytotoxic effects of indisulam were driven by splicing defects resulting from RBM39 degradation . Our studies confirmed indisulam-induced RBM39 degradation and the indisulam-dependent interaction between DCAF15 and RBM39 . We also discovered indisulam-induced phenotypes attributable to inhibition of DCAF15’s normal functions . Concentrations of indisulam with limited RBM39 degradation or cytotoxicity had immunomodulatory properties that phenocopied genetic DCAF15 disruption . Biochemically , the recruitment of endogenous client proteins to CRL4-DCAF15 and subsequent ubiquitination was impaired by indisulam treatment . We also determined that AML patients with naturally occurring lower levels of DCAF15 had improved overall survival . While these clinical data are preliminary in nature , they provide a rationale for drugging DCAF15 in myeloid neoplasms , achieved through judicious dosing of existing anti-cancer sulfonamides or the development of pure DCAF15 inhibitors . All cell lines were purchased from ATCC and were tested monthly for mycoplasma contamination . Cell lines other than NK-92 were maintained in RPMI supplemented with 10% FBS , 1 mM GlutaMAX and 1% antibiotic , antimycotic . NK-92 cells were grown in Myelocult H5100 ( Stem cell Technologies ) supplemented with 100 U/ml human IL-2 ( Peprotech cat#200–02 ) . IL-2 stock solution was made by reconstituting lyophilized cytokine to 10e6 U/ml in 50 mM acetic acid , 0 . 1% BSA in PBS . CRISPR screening was performed using bespoke genome-scale libraries , to be described in detail elsewhere . In brief , the sgRNA library was designed with 120 , 021 sgRNAs present , representing six guides each against 21 , 598 genes and four guides each against 1918 miRNAs , as well as 1000 non-targeting negative control guides ( Supplementary file 1 ) . sgRNAs targeting protein-coding genes were based on the Avana libraries ( Doench et al . , 2016 ) . sgRNAs targeting miRNAs were based on the design of the Gecko v2 . 0 libraries ( Sanjana et al . , 2014 ) . The protospacer library was synthesized by Twist Biosciences . The synthesized oligo library was amplified using emulsion PCR followed by purification . The vaccinia virus DNA polymerase was used to clone the protospacers into a lentiviral construct ( In-fusion , Takara ) . The lentiviral construct was linearized by Bfua1 digestion ( New England Biolabs ) . To ensure complete digestion , Bfua1 activity was stimulated by addition of 500 nM of a double-stranded oligo containing a Bfua1 site ( 5’ atagcacctgctata 3’ ) ( based on Grigaite et al . , 2002 ) . The guide RNAs were expressed from a human U6 promoter , using a modified sgRNA design ( A-U flip , longer stem-loop ) previously described ( Chen et al . , 2013 ) . An EF1a-Puro-T2A-cerulean-wpre ORF was used for selection purposes and for measuring infection rates . The library was electroporated into MegaX cells and plated across 37 × 500 cm2 LB-carbenicillin plates . The library was recovered and pooled by scraping and column-based plasmid purification ( Zymopure GigaPrep ) . 16M 293 T cells were plated onto 177 cm2 dishes ( 9x plates total ) . 24 hr later , cell media was replaced with 32mls of DMEM+10% FBS ( D-10 ) . Cells were transfected with the library by lipofection ( per plate: 158 µl lipofectamine 2000 , 8mls of Optimem , 3 . 95 μg of VSVG , 11 . 8 μg of Pax2 , 15 . 78 μg of library ) . The transfection mixture was left on the cells overnight , then changed to 25mls of D-10 . Viral supernatant was collected 48 hr later . Debris was removed by centrifugation at 200 g . Aliquots were flash-frozen and stored at −80°C . K562 cells were lentivirally infected with an EF1a-spCas9-T2A-blastR construct . Following 10 μg/ml blasticidin selection , cells were dilution cloned . Clonal isolates with high levels of cas9 expression ( as determined by western blot ) were selected for further characterization . To determine the kinetics and efficacy of gene cutting under screening conditions ( e . g . , low multiplicity of infection ( MOI ) ) of the sgRNA construct ) , cells were infected with a lentiviral construct expressing both EGFP and a sgRNA targeting GFP . Following puromycin selection , the loss of EGFP expression was monitored by flow cytometry . In other experiments , the ability to effectively deplete endogenous proteins was determined by using a series of sgRNAs targeting the mismatch repair complex and measuring protein depletion by western blot . Five hundred million spCas9-expressing K562 cells were mixed with CRISPR sgRNA library virions and 1L of media , then distributed across 34 6-well plates . Cells were spin-infected at 1900 rpm for 30 min at room-temperature with CRISPR library virus , conditions calculated to achieve a MOI of ~0 . 3 and 1000 cells per sgRNA library representation . MOI was measured by tracking the percent of the population expressing the cerulean marker found in the sgRNA library . Cells were incubated overnight in viral supernatant prior to being pooled , spun down to remove virions , and returned to spinner-flask culture ( Bell-Flo Flask , Bellco Glass Inc ) . 24 to 48 hr post-infection , 2 μg/ml puromycin selection was started for four days . Cells were maintained in log-growth phase with a minimal representation of 500 M cells . Seven days post guide-infection , NK cell challenge was initiated . For the 2 . 5:1 E:T challenge , 100M K562 cells were mixed with 250M NK-92 cells in 1L of Myelocult , and split across 25 177 cm2 dishes . For the 1:1 E:T challenge , 100M K562 cells were mixed with 100M NK-92 cells in 400mls of Myelocult , and split across 10 177 cm2 dishes . As a control , K562 cells were continuously propagated in RPMI media . 2 days after initiating the NK-92 challenge , the co-culture was switched to RPMI media . NK-92 cells , when grown in RPMI without IL-2 , are rapidly lost from the culture . Six days after the media switch , 100 M cell pellets were generated for library construction . All cell pellets were stored cryopreserved in cell-freezing media ( Gibco #12648 ) . Generation of K562 cells expressing the sgRNA library was performed as described above . On day 8 post-guide infection , 150 M cells were stimulated with 10 ng/ml IFNγ ( R and D Systems 285-IF ) . 24 hr later , cells were spun down , washed , and stained in 3mls of 1:200 anti-HLA ABC-APC ( W6/32 clone , BioLegend ) in PBS , 2% FBS for 30 min at 4°C . Approximately , 10M of the brightest 20% and dimmest 20% of cells were sorted ( BD FACSAria Fusion ) . Cell purity was determined to be >95% by re-analysis post-sorting . A replicate of this experiment was performed on day 14 post-guide infection . Genomic DNA was prepared by thawing cryopreserved cell pellets and proceeding with DNA extraction using column-based purification methods ( NucleoSpin Blood XL , Machery-Nagel ) . Protospacer libraries were generated by a two-step PCR strategy , modified from van Overbeek et al . ( 2016 ) . In the first round of PCR , 150 μg of gDNA ( equivalent to 15M K562 cells and 125-fold coverage of the library ) was used in a 7 . 5 ml PCR reaction to amplify the protospacers . This reaction was performed using a 500 nM mixture of primers containing 0–9 bp staggers , to ensure base-pair diversity during Illumina sequencing ( see Supplementary file 7 ) . The reaction was performed with Phusion master mix ( New England Biolabs ) and 3% DMSO with the following cycling conditions: one cycle X 30 s at 98°C , 21 cycles X 15 s at 98°C , 20 s at 63°C , 15 s at 72°C , one cycle X 2 min at 72°C . In the second round of PCR , 4 μl of the initial PCR product was used as the template in a 200 μl PCR reaction to make the sample compatible with Illumina chemistry and to add unique I5 and I7 barcodes to the sample . The reaction was performed with Phusion master mix ( New England Biolabs ) , 500 nM primers and 3% DMSO with the following cycling conditions: one cycle X 30 s at 98°C , 12 cycles X 15 s 98°C , 20 s at 60°C , 15 s at 72°C , one cycle X 2 min at 72°C . The library was size-selected first by a 1:1 SPRI bead selection ( AMPure XP beads , Beckman Coulter ) , quantified by high-sensitivity dsDNA Qubit ( ThermoFisher Scientific ) , and pooled . An agarose size selection step ( PippinHT , Sage Science ) was performed prior to sequencing on an Illumina Hiseq4000 . Libraries were sequenced to a depth of ~20 million reads per condition . Reads were aligned to the sgRNA library using bowtie2 . Protospacer count tables were generated from these alignments with python scripts and processed with MAGeCK . MAGeCK analysis was used to score and prioritize sgRNAs , using default settings in the algorithm ( Li et al . , 2014 ) . A subset of genes , mostly from highly related gene families , have more than six sgRNAs targeting them . As the MAGeCK scoring method tends to prioritize consistency of effect over magnitude of effect , genes with more than six guides targeting them were excluded from the analysis . MAGeCK scores were -log10 normalized , and values were plotted against FDR values . Cas9-expressing K562 or Daudi cells were transduced at a high MOI with a lentiviral sgRNA construct expressing the guide of interest , puromycin resistance and CMV promoter-driven expression of either a red or green fluorescent protein . Knockout cell populations were maintained in a polyclonal state . Gene disruption was confirmed at the protein level by flow cytometry or western blot , and by RNA-seq when antibody reagents were not available . Complete knockout in >90% of the population was routinely achieved ( Figure 2—figure supplement 1 and Figure 3—figure supplement 1 ) . Cells were counted and 0 . 25M red-labeled test cells were mixed with 0 . 25M green-labeled control cells ( expressing a sgRNA against an olfactory receptor gene ) . The mixture was either grown in RPMI or mixed with 1 . 25M NK-92 cells in 4mls Myelocult media in 6-well plates . The ratio of green-to-red cells was measured 2 to 4 days post-challenge , with the fold-change normalized to the ratio in the non-challenged state ( to control for differences in basal cell growth rate ) . PBMCs were isolated from leukocyte-enriched blood of human donors ( Stanford Blood Center ) . Donors were not genotyped or pre-screened for infectious disease markers . Natural killer cells were isolated by negative selection using magnetic columns ( Miltenyi 130-092-657 ) . Purity post-selection was routinely >97% CD56+ CD3- . Isolated NK cells were plated at 1 M/ml density in 96-well u-bottom plates and stimulated overnight with 1000 U/ml IL-2 in complete RPMI with 10% FBS . Following IL-2 activation , NK cells were counted and mixed with target cells in 96-well u-bottom plates . 100 , 000 NK cells were mixed with 40 , 000 target cells in a final volume of 200 μl for a 2 . 5:1 E:T ratio . Cells were co-cultured at 37°C for 2 hr , then stained with anti-CD56-BV421 ( 1:200 ) and anti-CD107a-APC ( 1:200 dilution ) for 30 m at 4°C . CD107a expression was assayed by flow cytometry in the 7-AAD- CD56+ RFP- GFP- cell population . Primary NK cells exhibited very little basal degranulation in the absence of target cells ( <1% CD107a+ ) . The assay was repeated with six primary donors with technical duplicates . Following IL-2 activation , NK cells were counted and mixed at a 1:1 E:T ratio with RFP-labeled target cells and GFP-labeled control cells in a 96-well u-bottom plate . 50 , 000 NK cells were combined with 25 , 000 Red and 25 , 000 Green cells in a final volume of 200 μl and continuously expanded in 96-well plates to as needed . The ratio of GFP+ to RFP+ cells was measured by flow cytometry on days 2 and 4 post-challenge , and the fold change was normalized to the ratio in the non-challenged state ( to control for differences in basal cell growth rate ) . The assay was repeated with six primary donors with technical duplicates . 0 . 5 to 1 M cells were spun down and stained with APC-conjugated antibodies for 30mins at 4°C in PBS with 2% FBS . Cells were analyzed on a BD LSRFortessa . The geometric mean fluorescence intensity of singlet , DAPI-excluding cells was measured , and normalized to the background fluorescence of that particular genotype of cells . The DCAF15 open reading frame was synthesized based off reference sequence NM_138353 , but with silent mutations designed to confer resistance to all three sgRNAs . To ensure resistance to sgRNA-directed gene cutting , silent mutations were introduced to the PAM domain and the proximal region of the protospacer ( guide #1: 5’ GCTGCACACCAAGTACCAGGTGG to GCTGCACACCAAaTAtCAaGTaG . Guide #2: 5’ TGACATCTACGTCAGCACCGTGG to TGACATCTACGTCtcCACaGTaG; Guide #3: 3’ GCAGCTTCCGGAAGAGGCGAGGG to GCAGCTTCCGGAAtaaaCGtGGt ) . The rescue construct was expressed lentivirally from an EF1a promoter , with a c-terminal 3-flag epitope tag . Rescue cells were selected with hygromycin . The rescue construct was introduced 2 weeks after the initial introduction of the DCAF15 sgRNAs , and stable cell lines generated by a week of selection in 375 μg/ml hygromycin . Expression of the construct was confirmed by lysing cells in PBS+0 . 1% NP40 ( as per Han et al . , 2017 ) and western blotting for the Flag tag . RFP-labeled target cells ( 0 , 0 . 75M , or 2 . 5M ) were resuspended in 0 . 5mls Myelocult + 5 μg/ml control ( MOPC-21 clone ) or blocking CD80 ( 2D10 clone ) antibodies . Cells were incubated for 30 mins at room temperature . 0 . 5M NK-92 cells in 0 . 5mls Myelocult were added to the well with 1:200 anti-CD107a-APC antibody . Cells were co-cultured for 4 hr at 37°C . Flow cytometry was used to measure CD107A-APC expression in the NK-92 ( RFP-negative ) population . NK-92 cells were found to display a basal level of CD107a expression in the absence of effector cells . Increases above basal staining levels were used to define NK-92 degranulation . The experiment was repeated four times to establish biological replicates . A full-length CD80 open reading frame , based on reference sequence NM_005191 . 4 , was synthesized with a c-terminal 3x Flag tag and expressed lentivirally from the EF1a promoter . Over-expressing cells were selected with hygromycin . As a control construct , a mutant version of CD80 was synthesized with Q65A and M72A mutations in the ‘V-type’ Ig-like domain . Each of these residues makes contacts with CTLA4 in published CD80-CTLA4 co-crystal structures ( Stamper et al . , 2001 ) and alanine scanning experiments have shown that these residues are required for CD80 binding to CD28 or CTLA4 ( Peach et al . , 1995 ) . See Supplementary file 6 . See Supplementary file 8 . Cells were plated at 0 . 25 M/ml in media with or without 10 ng/ml IFNγ . Every day , an aliquot of cells was counted ( Vi-CELL XR , Beckman Coulter ) , and cells were diluted in fresh media to maintain them in logarithmic growth phase . Continuously-treated cells were re-fed with fresh 10 ng/ml IFNγ every day , whereas pulse-treated cells only received 24 hr of cytokine treatment . Cells were counted , spun down and washed in PBS prior to lysis in NP40 lysis buffer ( 25 mM HEPES pH 7 . 5 , 150 mM NaCl , 1 . 5 mM MgCl2 , 0 . 5% NP40 , 10% glycerol , 2 mM DTT ) with protease and phosphatase inhibitors . Lysates were normalized by cell count and/or protein concentration . 20–40 μg of lysates were used for western blot analysis using standard procedures . Antibody binding was detected by enhanced chemiluminescence ( SuperSignal Dura , Thermo Scientific ) . Cells were seeded at 0 . 33 M/ml density . Twenty-four hours later , 2 M cells were collected . RNA was first purified by TRizol extraction and then further purified using column-based methods and polyA selection . RNAseq libraries were constructed using TruSeq Stranded mRNA Library Prep Kits ( Illumina ) and were sequenced on an Illumina Hiseq4000 machine using 150 bp paired-end reads . Transcript abundances were quantified using Salmon ( v0 . 9 . 1 ) in pseudo-alignment mode , without adapter trimming , using the Ensembl GRCh38 transcriptome ( Patro et al . , 2017 ) . Differential expression analysis was performed using Sleuth ( v0 . 29 . 0 ) ( Pimentel et al . , 2017 ) . RNA-seq analysis was executed and visualized using an in-house web-based platform . RNA sequencing data are available under accession number GEO:GSE134173 . Indisulam ( Sigma SML1225 ) was reconstituted at 10 mM in DMSO and stored in single use aliquots at −80°C . For indisulam dose-response experiments , 5000 cells were plated in 384-well plates ( Greiner ) and treated with indisulam over a 72 hr period . A 12-point dose-response was performed between 10 μM and 4 . 9 μM of drug , as dispensed by a Tecan D300e . Cell viability was measured by ATP quantification ( Cell Titer Glo , Promega ) . Dose-response measurements were fitted to a sigmoidal curve and an IC50 determined ( Prism , GraphPad Software ) . DCAF15 expression data from different cell lines was downloaded from the Cancer Cell Line Encyclopedia ( Barretina et al . , 2012 ) ( https://portals . broadinstitute . org/ccle ) . For low-dose indisulam experiments , cells were plated at 0 . 4 M/ml in media with or without 100 nM indisulam ( 1:100 , 000 dilution of stock ) . Cells were diluted every day in fresh media and drug to maintain them in logarithmic growth phase . Cells were analyzed for CD80 expression 96 hr after initiation of treatment . FPKM RNAseq quantification for patient samples from the TARGET and TCGA LAML cohorts was obtained from the NIH NCI Genomic Data Commons DATA portal ( https://portal . gdc . cancer . gov/ ) . Clinical data for that TARGET AML and TCGA LAML cohorts were obtained from the Genomic Data Commons DATA portal and the Broad Institute TCGA Genome Data Analysis Center ( http://gdac . broadinstitute . org/ ) respectively . Patients with matching clinical and transcript abundance data patients were stratified by DCAF15 expression as indicated . Survival time and vital status were defined as 'Overall Survival Time in Days' and 'Vital Status' respectively for the Target AML cohort . For the TCGA LAML study survival time for deceased patients ( 'patient . vital_status'=dead ) was defined as 'patient . days_to_death' while for living patients ( 'patient . vital_status'=alive ) 'patient . days_to_last_followup' was used for survival time . Survival analysis and Kaplan–Meier plots were generated using lifelines software for python ( https://doi . org/10 . 5281/zenodo . 2584900 ) . K562 cells were infected with lentivirus expressing DCAF15-3flag-BioID-HA-T2A-BlastR , DCAF16-3flag-BioID-HA-T2A-BlastR , or 3flag-GFP-T2A-BioID-HA-T2A-BlastR . Stable cell lines were generated by 10 μg/ml blasticidin selection . In triplicate , 20 M cells were grown in media supplemented with 50 μm biotin and 5 μm MG132 . 18 hr later , cell pellets were washed three times with ice cold PBS and lysed in mild lysis buffer ( PBS 0 . 1% NP40 + PI/PPI ) , conditions determined to maximize the solubility of over-expressed DCAF15 . 1 mg of clarified lysate was used for enrichment of biotinylated species . Lysates were mixed with 60 µl of streptavidin beads ( Pierce #88817 ) for 4 hr at 4°C in 500 μl total volume . Beads were collected on a magnetic column , washed twice with 1 ml RIPA buffer ( 25 mM HEPES-KOH pH 7 . 4 , 150 mM NaCl , 1% Triton X-100 , 0 . 5% sodium deoxycholate , 0 . 1% SDS , 1 mM EDTA ) and three times with 1 ml urea buffer ( 2M urea , 10 mM TRIS-HCl pH 8 . 0 ) . For western blotting experiments , after urea washing , beads were equilibrated in mild lysis buffer . Elution was performed by incubating the beads 15 min at 23°C , 15 min at 95°C in 30 μl 2 . 5X Laemlli buffer supplemented with 10 mM biotin and 20 mM DTT . The eluate was brought down to 1X concentration with lysis buffer prior to western blotting . We note that detection of total biotinylated species by western blot was extremely sensitive to detection conditions . Membranes were blocked for 10 min in PBS , 2 . 5%BSA , 0 . 4% Triton-X 100 . 1 ng/ml streptavidin-HRP in blocking buffer was added for 25 min at room temperature . The membrane was washed for 15 min in PBS 0 . 4% TritonX-100 prior to ECL exposure . To measure BioID activity after indisulam treatment , cells were treated for 48 hr with 0 . 1 μm indisulam , followed by 24 hr in indisulam with 50 μm biotin and 5 μm MG132 . For proteomics experiments , after the urea buffer washes , the samples were resuspended in denaturing buffer ( 8M urea , 50 mM ammonium bicarbonate pH 7 . 8 ) . Proteins were reduced with dithiothreitol ( 5 mM , RT , 30 min ) and alkylated with iodoacetamide ( 15 mM RT , 45 min in the dark ) . Excess iodoacetamide was quenched with dithiothreitol ( 5 mM , room temperature , 20 min in the dark ) . The samples were diluted to 1 M urea using 50 mM ammonium bicarbonate and then digested with trypsin ( 37°C , 16 hr ) . After protein digestion , samples were acidified with trifluoroacetic acid to a final concentration of 0 . 5% and desalted using C18 StageTips ( Rappsilber et al . , 2007 ) . Peptides were eluted with 40% acetonitrile/5% formic acid then 80% acetonitrile/5% formic acid and dried overnight under vacuum at 25°C ( Labconco CentriVap Benchtop Vacuum Concentrator , Kansas City , Mo ) . For tandem mass tag ( TMT ) labeling , dried peptides were resuspended in 50 μL 200 mM HEPES/30% anhydrous acetonitrile . TMT reagents ( 5 mg ) were dissolved in anhydrous acetonitrile ( 250 μL ) of which 10 μL was added to peptides to achieve a final acetonitrile concentration of approximately 30% ( v/v ) . Following incubation at room temperature for 1 hr , the reaction was quenched with 5% hydroxylamine/200 mM HEPES to a final concentration of 0 . 3% ( v/v ) . The TMT labeled peptides were acidified with 50 μL 1% trifluoroacetic acid and pooled prior to desalting with SepPak ( Waters ) and dried under vacuum . The pooled TMT-labeled peptides were fractionated using high pH RP-HPLC . The samples were resuspended in 5% formic acid/5% acetonitrile and fractionated over a ZORBAX extended C18 column ( Agilent , 5 μm particles , 4 . 6 mm ID and 250 mm in length ) . Peptides were separated on a 75 min linear gradient from 5% to 35% acetonitrile in 10 mM ammonium bicarbonate at a flow rate of 0 . 5 mL/min on an Agilent 1260 Infinity pump equipped with a degasser and a diode array detector ( set at 214 , 220 , and 254 nm wavelength ) from Agilent Technologies ( Waldbronn , Germany ) . The samples were fractionated into a total of 96 fractions and then consolidated into 12 as described previously ( Edwards and Haas , 2016 ) . Samples were dried down under vacuum and reconstituted in 4% acetonitrile/5% formic acid for LC-MS/MS processing . Peptides were analyzed on an Orbitrap Fusion Lumos mass spectrometer ( Thermo Fisher Scientific ) coupled to an Easy-nLC ( Thermo Fisher Scientific ) . Peptides were separated on a microcapillary column ( 100 μm internal diameter , 25 cm long , filled using Maccel C18 AQ resin , 1 . 8 μm , 120A; Sepax Technologies ) . The total LC-MS run length for each sample was 180 min comprising a 165 min gradient from 6% to 30% acetonitrile in 0 . 125% formic acid . The flow rate was 300 nL/min and the column was heated to 60°C . Data-dependent acquisition ( DDA ) mode was used for mass spectrometry data collection . A high resolution MS1 scan in the Orbitrap ( m/z range 500–1 , 200 , 60 k resolution , AGC 5 × 10^5 , max injection time 100 ms , RF for S-lens 30 ) was collected from which the top 10 precursors were selected for MS2 analysis followed by MS3 analysis . For MS2 spectra , ions were isolated using a 0 . 5 m/z window using the mass filter . The MS2 scan was performed in the quadrupole ion trap ( CID , AGC 1 × 10^4 , normalized collision energy 30% , max injection time 35 ms ) and the MS3 scan was analyzed in the Orbitrap ( HCD , 60 k resolution , max AGC 5 × 10^4 , max injection time 250 ms , normalized collision energy 50 ) . For TMT reporter ion quantification , up to six fragment ions from each MS2 spectrum were selected for MS3 analysis using synchronous precursor selection ( SPS ) . Mass spectrometry data were processed using an in-house software pipeline ( Huttlin et al . , 2010 ) . Raw files were converted to mzXML files and searched against a composite human uniprot database ( downloaded on 29th March 2017 ) containing sequences in forward and reverse orientations using the Sequest algorithm . Database searching matched MS/MS spectra with fully tryptic peptides from this composite dataset with a precursor ion tolerance of 20 ppm and a product ion tolerance of 0 . 6 Da . Carbamidomethylation of cysteine residues ( +57 . 02146 Da ) and TMT tags of peptide N-termini ( +229 . 162932 Da ) were set as static modifications . Oxidation of methionines ( +15 . 99492 Da ) was set as a variable modification . Linear discriminant analysis was used to filter peptide spectral matches to a 1% FDR ( false discovery rate ) as described previously ( Huttlin et al . , 2010 ) . Non-unique peptides that matched to multiple proteins were assigned to proteins that contained the largest number of matched redundant peptides sequences using the principle of Occam’s razor ( Huttlin et al . , 2010 ) . Quantification of TMT reporter ion intensities was performed by extracting the most intense ion within a 0 . 003 m/z window at the predicted m/z value for each reporter ion . TMT spectra were used for quantification when the sum of the signal-to-noise for all the reporter ions was greater than 200 and the isolation specificity was greater than 0 . 75 ( Ting et al . , 2011 ) . Base 2 logarithm of protein fold-changes were estimated by fitting a previously described Bayesian model ( O'Brien et al . , 2018 ) to the peptide level intensities . Protein estimates are reported as the mean of the posterior distribution for each parameter . Similarly , coefficients of variation are calculated by taking the posterior variance divided by the posterior mean . The probability of a small change ( ‘P_null’ ) was estimated as the frequency of posterior samples that fall within the interval ( −1 , 1 ) on the log2 scale . Experiments were performed as per Lu et al . ( 2014 ) . One million 293Ts were plated in 6-well dishes overnight . Cells were transfected with 500 ng of plasmids containing CMV-6his-ubiquitin , CMV-3fl-EGFP , EF1A-DCAF15-3fl and/or CMV-SMC1A-HA . The SMC1A open reading frame was synthesized based on reference sequence NM_006306 . 3 . After 36 hr , cells were treated with 2 μM indisulam and/or 10 μM MG132 . 12 hr later , replicate wells were harvested for whole-cell lysates or for Ni-NTA pulldowns . Whole-cell lysates were made by extraction in 500 μl NP40 lysis buffer . For Ni-NTA pulldowns , cells were solubilized in 700 μl of guanidine buffer ( 6M guanidine-HCL , 0 . 1M Na2HPO4/NaH2PO4 , 10 mM imidazole , 0 . 05% TWEEN 20 , pH 8 . 0 ) , run through QIAshredder columns ( Qiagen ) and briefly sonicated . Purifications of 6his-ubiquinated species were performed as described in Lu et al . ( 2014 ) , except for the use of magnetic Ni-NTA beads ( Thermo Fisher Scientific 88831 ) and the addition of 0 . 05% TWEEN 20 to wash buffers . Unless otherwise specified , data were graphed and statistically analyzed using Prism ( GraphPad Software ) . Sample size was not predetermined , No outliers were excluded . Unless otherwise noted , all data points represent biological replicates rather than technical replicates . We define ‘technical replicates’ as running an assay multiple times on the exact same sample .	The human immune system can recognize and kill cancer cells growing in the body . Certain immune cells recognize mutated proteins on the surface of cancer cells known as antigens , and this ability can be enhanced by drugs . These so-called immunotherapies can be effective to treat several cancer types , but only some patients benefit from them . This is because cancer cells often stop presenting antigens on their surface , thus hiding from the immune response . Natural killer cells are a type of immune cell that does not rely on antigen presentation to recognize cancer cells . Scientists are now trying to develop drugs to increase the effectiveness with which natural killer cells attack cancer . Pech et al . used cells from a human leukemia , a type of blood cancer , to look for proteins that made these cells more vulnerable to natural killer cells . The main experiment , in which every single protein in the cancer cells was deleted one by one , revealed that a protein called DCAF15 changes how cancer and natural killer cells interact . Leukemia cells lacking DCAF15 could be attacked by natural killer cells much more easily because the cancer cells exhibited inflammation-like symptoms that stimulated the immune response . DCAF15 is part of a family of ‘adaptors’ that that provide specificity to the cellular machinery that controls proliferation , the recycling of proteins and DNA repair . Inhibiting DCAF15 with a drug also made natural killer cells more efficient at eliminating leukemia cells . Patients with leukemia whose cancer cells make little DCAF15 protein have a better chance of survival , suggesting that this process may already be happening in some patients . Together these data indicate that targeting DCAF15 in leukemia patients may help natural killer cells attack cancer cells . Future research is needed to see if a similar process takes place in other cancer types .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cancer", "biology" ]	2019	Systematic identification of cancer cell vulnerabilities to natural killer cell-mediated immune surveillance
The rapid expansion of human activities threatens ocean-wide biodiversity . Numerous marine animal populations have declined , yet it remains unclear whether these trends are symptomatic of a chronic accumulation of global marine extinction risk . We present the first systematic analysis of threat for a globally distributed lineage of 1 , 041 chondrichthyan fishes—sharks , rays , and chimaeras . We estimate that one-quarter are threatened according to IUCN Red List criteria due to overfishing ( targeted and incidental ) . Large-bodied , shallow-water species are at greatest risk and five out of the seven most threatened families are rays . Overall chondrichthyan extinction risk is substantially higher than for most other vertebrates , and only one-third of species are considered safe . Population depletion has occurred throughout the world’s ice-free waters , but is particularly prevalent in the Indo-Pacific Biodiversity Triangle and Mediterranean Sea . Improved management of fisheries and trade is urgently needed to avoid extinctions and promote population recovery . Populations and species are the building blocks of the communities and ecosystems that sustain humanity through a wide range of services ( Mace et al . , 2005; Díaz et al . , 2006 ) . There is increasing evidence that human impacts over the past 10 millennia have profoundly and permanently altered biodiversity on land , especially of vertebrates ( Schipper et al . , 2008; Hoffmann et al . , 2010 ) . The oceans encompass some of the earth’s largest habitats and longest evolutionary history , and there is mounting concern for the increasing human influence on marine biodiversity that has occurred over the past 500 years ( Jackson , 2010 ) . So far our knowledge of ocean biodiversity change is derived mainly from retrospective analyses usually limited to biased subsamples of diversity , such as: charismatic species , commercially-important fisheries , and coral reef ecosystems ( Carpenter et al . , 2008; Collette et al . , 2011; McClenachan et al . , 2012; Ricard et al . , 2012 ) . Notwithstanding the limitations of these biased snapshots , the rapid expansion of fisheries and globalized trade are emerging as the principal drivers of coastal and ocean threat ( Polidoro et al . , 2008; Anderson et al . , 2011b; McClenachan et al . , 2012 ) . The extent and degree of the global impact of fisheries upon marine biodiversity , however , remains poorly understood and highly contentious . Recent insights from ecosystem models and fisheries stock assessments of mainly data-rich northern hemisphere seas , suggest that the status of a few of the best-studied exploited species and ecosystems may be improving ( Worm et al . , 2009 ) . However , this view is based on only 295 populations of 147 fish species and hence is far from representative of the majority of the world’s fisheries and fished species , especially in the tropics for which there are few data and often less management ( Sadovy , 2005; Newton et al . , 2007; Branch et al . , 2011; Costello et al . , 2012; Ricard et al . , 2012 ) . Overfishing and habitat degradation have profoundly altered populations of marine animals ( Hutchings , 2000; Lotze et al . , 2006; Polidoro et al . , 2012 ) , especially sharks and rays ( Stevens et al . , 2000; Simpfendorfer et al . , 2002; Dudley and Simpfendorfer , 2006; Ferretti et al . , 2010 ) . It is not clear , however , whether the population declines of globally distributed species are locally reversible or symptomatic of an erosion of resilience and chronic accumulation of global marine extinction risk ( Jackson , 2010; Neubauer et al . , 2013 ) . In response , we evaluate the scale and intensity of overfishing through a global systematic evaluation of the relative extinction risk for an entire lineage of exploited marine fishes—sharks , rays , and chimaeras ( class Chondrichthyes ) —using the Red List Categories and Criteria of the International Union for the Conservation of Nature ( IUCN ) . We go on to identify , ( i ) the life history and ecological attributes of species ( and taxonomic families ) that render them prone to extinction , and ( ii ) the geographic locations with the greatest number of species of high conservation concern . Chondrichthyans make up one of the oldest and most ecologically diverse vertebrate lineages: they arose at least 420 million years ago and rapidly radiated out to occupy the upper tiers of aquatic food webs ( Compagno , 1990; Kriwet et al . , 2008 ) . Today , this group is one of the most speciose lineages of predators on earth that play important functional roles in the top-down control of coastal and oceanic ecosystem structure and function ( Ferretti et al . , 2010; Heithaus et al . , 2012; Stevens et al . , 2000 ) . Sharks and their relatives include some of the latest maturing and slowest reproducing of all vertebrates , exhibiting the longest gestation periods and some of the highest levels of maternal investment in the animal kingdom ( Cortés , 2000 ) . The extreme life histories of many chondrichthyans result in very low population growth rates and weak density-dependent compensation in juvenile survival , rendering them intrinsically sensitive to elevated fishing mortality ( Musick , 1999b; Cortés , 2002; García et al . , 2008; Dulvy and Forrest , 2010 ) . Chondrichthyans are often caught as incidental , but are often retained as valuable bycatch of fisheries that focus on more productive teleost fish species , such as tunas or groundfishes ( Stevens et al . , 2005 ) . In many cases , fishing pressure on chondrichthyans is increasing as teleost target species become less accessible ( due to depletion or management restrictions ) and because of the high , and in some cases rising , value of their meat , fins , livers , and/or gill rakers ( Fowler et al . , 2002; Clarke et al . , 2006; Lack and Sant , 2009 ) . Fins , in particular , have become one of the most valuable seafood commodities: it is estimated that the fins of between 26 and 73 million individuals , worth US$400-550 million , are traded each year ( Clarke et al . , 2007 ) . The landings of sharks and rays , reported to the Food and Agriculture Organization of the United Nations ( FAO ) , increased steadily to a peak in 2003 and have declined by 20% since ( Figure 1A ) . True total catch , however , is likely to be 3–4 times greater than reported ( Clarke et al . , 2006; Worm et al . , 2013 ) . Most chondrichthyan catches are unregulated and often misidentified , unrecorded , aggregated , or discarded at sea , resulting in a lack of species-specific landings information ( Barker and Schluessel , 2005; Clarke et al . , 2006; Iglésias et al . , 2010; Bornatowski et al . , 2013 ) . Consequently , FAO could only be ‘hopeful’ that the catch decline is due to improved management rather than being symptomatic of worldwide overfishing ( FAO , 2010 ) . The reported chondrichthyan catch has been increasingly dominated by rays , which have made up greater than half of reported taxonomically-differentiated landings for the past four decades ( Figure 1B ) . Chondrichthyan landings were worth US$1 billion at the peak catch in 2003 , since then the value has dropped to US$800 million as catch has declined ( Musick and Musick , 2011 ) . A main driver of shark fishing is the globalized trade to meet Asian demand for shark fin soup , a traditional and usually expensive Chinese dish . This particularly lucrative trade in fins ( not only from sharks , but also of shark-like rays such as wedgefishes and sawfishes ) remains largely unregulated across the 86 countries and territories that exported >9 , 500 mt of fins to Hong Kong ( a major fin trade hub ) in 2010 ( Figure 1C ) . 10 . 7554/eLife . 00590 . 003Figure 1 . The trajectory and spatial pattern of chondrichthyan fisheries catch landings and fin exports . ( A ) The landed catch of chondrichthyans reported to the Food and Agriculture Organization of the United Nations from 1950 to 2009 up to the peak in 2003 ( black ) and subsequent decline ( red ) . ( B ) The rising contribution of rays to the taxonomically-differentiated global reported landed catch: shark landings ( light gray ) , ray landings ( black ) , log ratio [rays/sharks] , ( red ) . Log ratios >0 occur when more rays are landed than sharks . The peak catch of taxonomically-differentiated rays peaks at 289 , 353 tonnes in 2003 . ( C ) The main shark and ray fishing nations are gray-shaded according to their percent share of the total average annual chondrichthyan landings reported to FAO from 1999 to 2009 . The relative share of shark and ray fin trade exports to Hong Kong in 2010 are represented by fin size . The taxonomically-differentiated proportion excludes the ‘nei’ ( not elsewhere included ) and generic ‘sharks , rays , and chimaeras’ category . DOI: http://dx . doi . org/10 . 7554/eLife . 00590 . 003 Overall , we estimate that one-quarter of chondrichthyans are threatened worldwide , based on the observed threat level of assessed species combined with a modeled estimate of the number of Data Deficient species that are likely to be threatened . Of the 1 , 041 assessed species , 181 ( 17 . 4% ) are classified as threatened: 25 ( 2 . 4% ) are assessed as Critically Endangered ( CR ) , 43 ( 4 . 1% ) Endangered ( EN ) , and 113 ( 10 . 9% ) Vulnerable ( VU ) ( Table 1 ) . A further 132 species ( 12 . 7% ) are categorized as Near Threatened ( NT ) . Chondrichthyans have the lowest percentage ( 23 . 2% , n = 241 species ) of Least Concern ( LC ) species of all vertebrate groups , including the marine taxa assessed to date ( Hoffmann et al . , 2010 ) . Almost half ( 46 . 8% , n = 487 ) are Data Deficient ( DD ) meaning that information is insufficient to assess their status ( Table 1 ) . DD chondrichthyans are found across all habitats , but particularly on continental shelves ( 38 . 4% of 482 species in this habitat ) and deepwater slopes ( 57 . 6% , Table 2 ) . Of the 487 DD species for which we had sufficient maximum body size ( n = 396 ) and geographic distribution data ( n = 378 ) , we were able to predict that at least a further 68 DD species are likely to be threatened ( Table 3 , Supplementary file 1 ) . Accounting for the uncertainty in threat levels due to the number of DD species , we estimate that more than half face some elevated risk: at least one-quarter ( n = 249; 24% ) of chondrichthyans are threatened and well over one-quarter are Near Threatened ( Table 1 ) . Only 37% are predicted to be Least Concern ( Table 1 ) . 10 . 7554/eLife . 00590 . 004Table 1 . Observed and predicted number and percent of chondrichthyan species in IUCN Red List categoriesDOI: http://dx . doi . org/10 . 7554/eLife . 00590 . 004TaxonSpecies number ( % ) Threatened species number ( % ) CRENVUNTLCDDSkates and rays539 ( 51 . 8 ) 107 ( 19 . 9 ) 14 ( 1 . 3 ) 28 ( 2 . 7 ) 65 ( 6 . 2 ) 62 ( 6 . 0 ) 114 ( 11 . 0 ) 256 ( 24 . 6 ) Sharks465 ( 44 . 7 ) 74 ( 15 . 9 ) 11 ( 1 . 1 ) 15 ( 1 . 4 ) 48 ( 4 . 6 ) 67 ( 6 . 4 ) 115 ( 11 . 0 ) 209 ( 20 . 1 ) Chimaeras37 ( 3 . 6 ) 00003 ( 0 . 3 ) 12 ( 1 . 2 ) 22 ( 2 . 1 ) All observed1041181 ( 17 . 4 ) 25 ( 2 . 4 ) 43 ( 4 . 1 ) 113 ( 10 . 9 ) 132 ( 12 . 7 ) 241 ( 23 . 2 ) 487 ( 46 . 8 ) All predicted249 ( 23 . 9 ) –––312 ( 29 . 9 ) 389 ( 37 . 4 ) 91 ( 8 . 7 ) CR , Critically Endangered; EN , Endangered; VU , Vulnerable; NT , Near Threatened; LC , Least Concern; DD , Data Deficient . Number threatened is the sum total of the categories CR , EN and VU . Species number and number threatened are expressed as percentage of the taxon , whereas the percentage of each species in IUCN categories is expressed relative to the total number of species . 10 . 7554/eLife . 00590 . 005Table 2 . Number and percent of chondrichthyans in IUCN Red List categories by their main habitatsDOI: http://dx . doi . org/10 . 7554/eLife . 00590 . 005HabitatSpecies ( % ) Threatened ( % ) CR ( % ) EN ( % ) VU ( % ) NT ( % ) LC ( % ) DD ( % ) Coastal and continental shelf482 ( 46 . 3 ) 127 ( 26 . 3 ) 20 ( 4 . 1 ) 26 ( 5 . 4 ) 81 ( 16 . 8 ) 73 ( 15 . 1 ) 97 ( 20 . 1 ) 185 ( 38 . 4 ) Neritic and epipelagic39 ( 3 . 7 ) 17 ( 43 . 6 ) 03 ( 7 . 7 ) 14 ( 35 . 9 ) 13 ( 33 . 3 ) 5 ( 12 . 8 ) 4 ( 10 . 3 ) Deepwater479 ( 46 . 0 ) 25 ( 5 . 2 ) 2 ( 0 . 4 ) 6 ( 1 . 3 ) 17 ( 3 . 5 ) 45 ( 9 . 4 ) 133 ( 27 . 8 ) 276 ( 57 . 6 ) Mesopelagic8 ( 0 . 8 ) 000004 ( 50 . 0 ) 4 ( 50 . 0 ) Freshwater ( obligate species only ) 33 ( 3 . 2 ) 12 ( 36 . 4 ) 3 ( 9 . 1 ) 8 ( 24 . 2 ) 1 ( 3 . 0 ) 1 ( 3 . 0 ) 2 ( 6 . 1 ) 18 ( 54 . 5 ) Totals1041181 ( 17 . 4 ) 25 ( 2 . 4 ) 43 ( 4 . 1 ) 113 ( 10 . 9 ) 132 ( 12 . 7 ) 241 ( 23 . 2 ) 487 ( 46 . 8 ) CR , Critically Endangered; EN , Endangered; VU , Vulnerable; NT , Near Threatened; LC , Least Concern; DD , Data Deficient . 10 . 7554/eLife . 00590 . 006Table 3 . Summary of predictive Generalized Linear Models for life history and ecological correlates of IUCN statusDOI: http://dx . doi . org/10 . 7554/eLife . 00590 . 006ModelModel structure and hypothesisDegrees of freedom , kLog likelihoodAICcΔAICAIC weightAccuracy ( AUC ) R21∼maximum length2−227 . 47945943 . 670 . 0000 . 6780 . 1392∼ …+ minimum depth3−210 . 299426 . 711 . 340 . 0030 . 7460 . 2433∼ …+…+ depth range4−204 . 703417 . 52 . 190 . 250 . 7620 . 2764∼ …+…+…+ geographic range5−202 . 578415 . 300 . 7480 . 7720 . 298Species were scored as threatened ( CR , EN , VU ) = 1 or Least Concern ( LC ) = 0 for n = 367 marine species . AICc is the Akaike information criterion corrected for small sample sizes and ΔAIC is the change in AICc . The models are ordered by increasing complexity and decreasing AIC weight ( largest ΔAIC to lowest ) , coefficient of determination ( R2 ) , and prediction accuracy ( measured using Area Under the Curve , AUC ) . The main threats to chondrichthyans are overexploitation through targeted fisheries and incidental catches ( bycatch ) , followed by habitat loss , persecution , and climate change . While one-third of threatened sharks and rays are subject to targeted fishing , some of the most threatened species ( including sawfishes and large-bodied skates ) have declined due to incidental capture in fisheries targeting other species . Shark-like rays , especially sawfishes , wedgefishes and guitarfishes , have some of the most valuable fins and are highly threatened . Although the global fin trade is widely recognized as a major driver of shark and ray mortality , demand for meat , liver oil , and even gillrakers ( of manta and other devil rays ) also poses substantial threats . Half of the 69 high-volume or high-value sharks and rays in the global fin trade are threatened ( 53 . 6% , n = 37 ) , while low-value fins often enter trade as well , even if meat demand is the main fishery driver ( Supplementary file 2A ) . Coastal species are more exposed to the combined threats of fishing and habitat degradation than those offshore in pelagic and deepwater ecosystems . In coastal , estuarine , and riverine habitats , four principal processes of habitat degradation ( residential and commercial development , mangrove destruction , river engineering , and pollution ) jeopardize nearly one-third of threatened sharks and rays ( 29 . 8% , n = 54 of 181 , Supplementary file 2B ) . The combined effects of overexploitation and habitat degradation are most acute in freshwater , where over one-third ( 36 . 0% ) of the 90 obligate and euryhaline freshwater chondrichthyans are threatened . Their plight is exacerbated by high habitat-specificity and restricted geographic ranges ( Stevens et al . , 2005 ) . Specifically , the degradation of coastal , estuarine and riverine habitats threatened 14% of sharks and rays: through residential and commercial development ( 22 species , including river sharks Glyphis spp . ) ; mangrove destruction for shrimp farming in Southeast Asia ( 4 species , including Bleeker’s variegated stingray Himantura undulata ) ; dam construction and water control ( 8 species , including Mekong freshwater stingray Dasyatis laosensis ) , and pollution ( 20 species ) . Many freshwater sharks and rays suffer multiple threats and have narrow geographic distributions , for example the Endangered Roughnose stingray ( Pastinachus solocirostris ) that is found only in Malaysian Borneo and Indonesia ( Kalimantan , Sumatra and Java ) . Population control of sharks , in particular due to their perceived risk to people , fishing gear , and other fisheries has contributed to the threatened status of at least 12 species ( Supplementary file 2B ) . Sharks and rays are also threatened due to capture in shark control nets ( e . g . Dusky shark Carcharhinus obscurus ) , and persecution to minimise: damage to fishing nets ( e . g . Green sawfish Pristis zijsron ) ; their predation on aquacultured molluscs ( e . g . Estuary stingray Dasyatis fluviorum ) ; interference with spearfishing activity ( e . g . Grey nurse shark Carcharias taurus ) , and the risk of shark attack ( e . g . White shark Carcharodon carcharias ) . So far the threatened status of only one species has been directly linked to climate change ( New Caledonia catshark Aulohalaelurus kanakorum , Supplementary file 2B ) . the climate-sensitivity of some sharks has been recognized ( Chin et al . , 2010 ) and the status of shark and ray species will change rapidly in climate cul-de-sacs , such as the Mediterranean Sea ( Lasram et al . , 2010 ) . Elevated extinction risk in sharks and rays is a function of exposure to fishing mortality coupled with their intrinsic life history and ecological sensitivity ( Figures 2–6 ) . Most threatened chondrichthyan species are found in depths of less than 200 m , especially in the Atlantic and Indian Oceans , and the Western Central Pacific Ocean ( 79 . 6% , n = 144 of 181 , Figure 2 ) . Extinction risk is greater in larger-bodied species found in shallower waters with narrower depth distributions , after accounting for phylogenetic non-independence ( Figures 3 and 4 ) . The traits with the greatest relative importance ( >0 . 95 ) are maximum body size , minimum depth , and depth range . In comparison , geographic range ( measured as Extent of Occurrence ) has a much lower relative importance ( 0 . 79 , Figure 3 ) , and in the predictive models it improved the variance explained by 2% and the prediction accuracy by 1% ( Table 3 ) . The probability that a species is threatened increases by 1 . 2% for each 10 cm increase in maximum body length , and decreases by 10 . 3% for each 50 m deepening in the minimum depth limit of species . After accounting for maximum body size and minimum depth , species with narrower depth ranges have a 1 . 2% greater threat risk per 100 m narrowing of depth range . There is no significant interaction between depth range and minimum depth limit . Geographic range , measured as the Extent of Occurrence , varies over six orders of magnitude , between 354 km2 and 278 million km2 and is positively correlated with body size ( Spearman’s ρ = 0 . 58 ) , and hence is only marginally positively related to extinction risk over and above the effect of body size . Accounting for the body size and depth effects , the threat risk increases by only 0 . 5% for each 1 , 000 , 000 km2 increase in geographic range ( Table 4 ) . The explanatory and predictive power of our life history and geographic distribution models increased with complexity , though geographic range size contributed relatively little additional explanatory power and a high degree of uncertainty in the parameter estimate ( Tables 3 and 4 ) . The maximum variance explained was 69% ( Table 4 ) and the predictive models ( without controlling for phylogeny ) explained 30% of the variance and prediction accuracy was 77% ( Table 3 ) . 10 . 7554/eLife . 00590 . 007Figure 2 . IUCN Red List Threat status and the depth distribution of chondrichthyans in the FAO Fishing Areas of the Atlantic , Indian and Pacific Oceans , and Polar Seas . Each vertical line represents the depth range ( surface-ward minimum to the maximum reported depth ) of each species and is colored according to threat status: CR ( red ) , EN ( orange ) , VU ( yellow ) , NT ( pale green ) , LC ( green ) , and DD ( gray ) . Species are ordered left to right by increasing median depth . The depth limit of the continental shelf is indicated by the horizontal gray line at 200 m . The Polar Seas include the following FAO Fishing Areas: Antarctic–Atlantic ( Area 48 ) , Indian ( Area 58 ) , Pacific ( Area 88 ) , and the Arctic Sea ( Area 18 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00590 . 00710 . 7554/eLife . 00590 . 008Figure 2—figure supplement 1 . Map of Food and Agriculture Organization of the United Nations Fishing Areas and their codes: 18 , Arctic Sea; 21 , Atlantic , Northwest; 27 , Atlantic , Northeast; 31 , Atlantic , Western Central; 34 , Atlantic , Eastern Central; 37 , Mediterranean and Black Sea; 41 , Atlantic , Southwest; 47 , Atlantic , Southeast; 48 , Atlantic , Antarctic; 51 , Indian Ocean , Western; 57 , Indian Ocean , Eastern; 58 , Indian Ocean , Antarctic and Southern; 61 , Pacific , Northwest; 67 , Pacific , Northeast; 71 , Pacific , Western Central; 77 , Pacific , Eastern Central; 81 , Pacific , Southwest; 87 , Pacific , Southeast; and , 88 , Pacific , Antarctic . DOI: http://dx . doi . org/10 . 7554/eLife . 00590 . 00810 . 7554/eLife . 00590 . 009Figure 3 . Standardized effect sizes with 95% confidence intervals from the two best explanatory models of life histories , geographic range and extinction risk in chondrichthyans . The data were standardized by subtracting the mean and dividing by one standard deviation to allow for comparison among parameters . The relative importance is calculated as the sum of the Akaike weights of the models containing each variable . Chondrichthyans were scored as threatened ( CR , EN , VU ) = 1 or Least Concern ( LC ) = 0 for n = 367 marine species . Threat status was modeled using General Linear Mixed-effects Models , with size , depth and geography treated as fixed effects and taxonomy hierarchy as a random effect to account for phylogenetic non-independence . DOI: http://dx . doi . org/10 . 7554/eLife . 00590 . 00910 . 7554/eLife . 00590 . 010Figure 4 . Life history sensitivity , accessibility to fisheries and extinction risk . Probability that a species is threatened due to the combination of intrinsic life history sensitivity ( maximum body size , cm total length , TL ) and accessibility to fisheries which is represented as minimum depth limit , depth range , and geographic range size ( Extent of Occurrence ) . The lines represent the variation in body size-dependent risk for the upper quartile , median , and lower quartile of each range metric . The examplar species are all of similar maximum body length and the difference in risk is largely due to differences in geographic distribution . Chondrichthyans were scored as threatened ( CR , EN , VU ) = 1 or Least Concern ( LC ) = 0 for n = 366 marine species . The lines are the best fits from General Linear Mixed-effects Models , with maximum body size and geographic distribution traits treated as fixed effects and taxonomy hierarchy as a random effect to account for phylogenetic non-independence . Each vertical line in each of the ‘rugs’ represents the maximum body size and Red List status of each species: threatened ( red ) and LC ( green ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00590 . 01010 . 7554/eLife . 00590 . 011Figure 5 . Life history , habitat , and extinction risk in chondrichthyans . IUCN Red List status as a function of maximum body size ( total length , TL cm ) and accessibility to fisheries in marine chondrichthyans in three main habitats: coastal and continental shelf <200 m ( ‘Continental shelf’ ) ; neritic and oceanic pelagic <200 m ( ‘Pelagic’ ) ; and , deepwater >200 m ( ‘Deepwater’ ) , n = 367 ( threatened n = 148; Least Concern n = 219 ) . The upper and lower ‘rug’ represents the maximum body size and Red List status of each species: threatened ( upper rugs ) and Least Concern ( lower rugs ) . The lines are best fit using Generalized Linear Mixed-effects Models with 95% confidence intervals ( Table 9 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00590 . 01110 . 7554/eLife . 00590 . 012Figure 6 . Evolutionary uniqueness and taxonomic conservation priorities . Threat among marine chondrichthyan families varies with life history sensitivity ( maximum length ) and exposure to fisheries ( depth distribution ) . ( A ) Proportion of threatened data sufficient species and the richness of each taxonomic family . Colored bands indicate the significance levels of a one-tailed binomial test at p=0 . 05 , 0 . 01 , and 0 . 001 . Those families with significantly greater ( or lower ) than expected threat levels at p<0 . 05 against a null expectation that extinction risk is equal across families ( 35 . 6% ) . ( B ) The most and least threatened taxonomic families . ( C ) Average life history sensitivity and accessibility to fisheries of 56 chondrichthyan families . Significantly greater ( or lower ) risk than expected is shown in red ( green ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00590 . 01210 . 7554/eLife . 00590 . 013Figure 6—source data 1 . Number and IUCN Red List status of chondrichthyan species in IUCN Red List categories by family ( alphabetically within each order ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00590 . 01310 . 7554/eLife . 00590 . 014Table 4 . Summary of explanatory Generalized Linear Mixed-effect Models of the life history and geographic distributional correlates of IUCN statusDOI: http://dx . doi . org/10 . 7554/eLife . 00590 . 014Model structure and hypothesisDegrees of freedom , kLog likelihoodAICcΔAICAIC weightR2GLMM ( m ) of fixed effects onlyR2GLMM ( c ) of fixed and random effects∼ maximum length5−197 . 06404 . 328 . 310 . 0000 . 320 . 58∼ …+ minimum depth6−187 . 013386 . 310 . 290 . 0050 . 480 . 65∼ …+…+ depth range7−182 . 139378 . 62 . 620 . 2120 . 490 . 66∼ …+…+…+ geographic range8−179 . 785376 . 000 . 7840 . 690 . 80Species were scored as threatened ( CR , EN , VU ) = 1 or Least Concern ( LC ) = 0 for n = 367 marine species . AICc is the Akaike information criterion corrected for small sample sizes; ΔAIC is the change in AICc . The models are ordered by increasing complexity and decreasing AIC weight ( largest ΔAIC to lowest ) . R2GLMM ( m ) is the marginal R2 of the fixed effects only and R2GLMM ( c ) is the conditional R2 of the fixed and random effects . By habitat , one-quarter of coastal and continental shelf chondrichthyans ( 26 . 3% , n = 127 of 482 ) and almost half of neritic and epipelagic species ( 43 . 6% , n = 17 of 39 ) are threatened . Coastal and continental shelf and pelagic species greater than 1 m total length have a more than 50% chance of being threatened , compared to ∼12% risk for a similar-sized deepwater species ( Figure 5 ) . While deepwater chondrichthyans , due to their slow growth and lower productivity , are intrinsically more sensitive to overfishing than their shallow-water relatives ( García et al . , 2008; Simpfendorfer and Kyne , 2009 ) for a given body size they are less threatened—largely because they are inaccessible to most fisheries ( Figure 5 ) . As a result of their high exposure to coastal shallow-water fisheries and their large body size , sawfishes ( Pristidae ) are the most threatened chondrichthyan family and arguably the most threatened family of marine fishes ( Figure 6 ) . Other highly threatened families include predominantly coastal and continental shelf-dwelling rays ( wedgefishes , sleeper rays , stingrays , and guitarfishes ) , as well as angel sharks and thresher sharks; five of the seven most threatened families are rays . Least threatened families are comprised of relatively small-bodied species occurring in mesopelagic and deepwater habitats ( lanternsharks , catsharks , softnose skates , shortnose chimaeras , and kitefin sharks , Figure 6 , Figure 6—source data 1 ) . Local species richness is greatest in tropical coastal seas , particularly along the Atlantic and Western Pacific shelves ( Figure 7A ) . The greatest uncertainty , where the number of DD species is highest , is centered on four areas: ( 1 ) Caribbean Sea and Western Central Atlantic Ocean , ( 2 ) Eastern Central Atlantic Ocean , ( 3 ) Southwest Indian Ocean , and ( 4 ) the China Seas ( Figure 7B ) . The megadiverse China Seas face the triple jeopardy of high threat in shallow waters ( Figure 7CD ) , high species richness ( Figure 7A ) , and a large number of threatened endemic species ( Figure 8 ) , combined with high risk due to high uncertainty in status ( large number of DD species , Figure 7B ) . Whereas the distribution of threat in coastal and continental shelf chondrichthyans is similar to the overall threat pattern across tropical and mid-latitudes , the spatial pattern of threat varies considerably for pelagic and deepwater species . Threatened neritic and epipelagic oceanic sharks are distributed throughout the world’s oceans , but there are also at least seven threat hotspots in coastal waters: ( 1 ) Gulf of California , ( 2 ) southeast US continental shelf , ( 3 ) Patagonian Shelf , ( 4 ) West Africa and the western Mediterranean Sea , ( 5 ) southeast South Africa , ( 6 ) Australia , and ( 7 ) the China Seas ( Figure 7D ) . Hotspots of deepwater threatened chondrichthyans occur in three areas where fisheries penetrate deepest: ( 1 ) Southwest Atlantic Ocean ( southeast coast of South America ) , ( 2 ) Eastern Atlantic Ocean , spanning from Norway to Namibia and into the Mediterranean Sea , and ( 3 ) southeast Australia ( Figure 7E ) . 10 . 7554/eLife . 00590 . 015Figure 7 . Global patterns of marine chondrichthyan diversity , threat and knowledge . ( A ) Total chondrichthyan richness , ( B ) the number of Data Deficient and threat by major habitat: ( C ) coastal and continental shelf ( <200 m depth ) , ( D ) neritic and epipelagic ( <200 m depth ) , and ( E ) deepwater slope and abyssal plain ( >200 m ) habitats . Numbers expressed as the total number of species in each 23 , 322 km2 cell . The numbers are hotspots refereed to in the text . DOI: http://dx . doi . org/10 . 7554/eLife . 00590 . 01510 . 7554/eLife . 00590 . 016Figure 8 . Irreplaceability hotspots of the endemic threatened marine chondrichthyans . Endemics were defined as species with an Extent of Occurrence of <500 , 000 km2 ( n = 66 ) . Irreplaceable cells with the greatest number of small range species are shown in red , with blue cells showing areas of lower , but still significant irreplaceability . Irreplaceability is the sum of the inverse of the geographic range sizes of all threatened endemic species in the cell . A value of 0 . 1 means that on average a single cell represents one tenth of the global range of all the species present in the cell . The numbers are hotspots referred to in the text . DOI: http://dx . doi . org/10 . 7554/eLife . 00590 . 016 Spatial conservation priority can be assigned using three criteria: ( 1 ) the greatest number of threatened species ( Figure 7A ) , ( 2 ) greater than expected threat ( residuals of the relationship between total number of species and total number of threatened species per cell , Figure 9 ) , and ( 3 ) high irreplaceability—high numbers of threatened endemic species ( Figure 8 ) . Most threatened marine chondrichthyans ( n = 135 of 169 ) are distributed within , and are often endemic to ( n = 73 ) , at least seven distinct threat hotspots ( e . g . , for neritic and pelagic species Figure 7D ) . With the notable exception of the US and Australia , threat hotspots occur in the waters of the most intensive shark and ray fishing and fin-trading nations ( Figure 1C ) . Accordingly these regions should be afforded high scientific and conservation priority ( Table 5 ) . 10 . 7554/eLife . 00590 . 017Figure 9 . Spatial variation in the relative extinction risk of marine chondrichthyans . Residuals of the relationship between total number of data sufficient chondrichthyans and total number of threatened species per cell , where positive values ( orange to red ) represent cells with higher threat than expected for their richness alone . DOI: http://dx . doi . org/10 . 7554/eLife . 00590 . 01710 . 7554/eLife . 00590 . 018Table 5 . Scientific and conservation priority according to threat , knowledge and endemicity by FAO Fishing AreaDOI: http://dx . doi . org/10 . 7554/eLife . 00590 . 018FAO Fishing Area ( ranked priority ) Threatened species ( % of total , n = 181 ) Data Deficient species ( % of total , n = 487 ) Number of endemic species ( threatened endemics ) Threatened endemic species ( 1 ) Indian , Eastern67 ( 37 . 0 ) 69 ( 14 . 2 ) 58 ( 5 ) Atelomycterus baliensis , Himantura fluviatilis , Zearaja maugeana , Trygonorrhina melaleuca , Urolophus orarius ( 2 ) Pacific , Western Central76 ( 42 . 0 ) 81 ( 16 . 6 ) 51 ( 14 ) Glyphis glyphis , Aulohalaelurus kanakorum , Hemitriakis leucoperiptera , Brachaelurus colcloughi , Hemiscyllium hallstromi , H . strahani , Himantura hortlei , H . lobistoma , Pastinachus solocirostris , Aptychotrema timorensis , Rhinobatos jimbaranensis , Rhynchobatus sp . nov . A , Rhynchobatus springeri , Urolophus javanicus ( 3 ) Pacific , Northwest48 ( 26 . 5 ) 116 ( 23 . 8 ) 80 ( 6 ) Benthobatis yangi , Narke japonica , Raja pulchra , Squatina formosa , S . japonica , S . nebulosa ( 4 ) Indian , Western61 ( 33 . 7 ) 104 ( 21 . 4 ) 62 ( 8 ) Carcharhinus leiodon , Haploblepharus kistnasamyi , H . favus , H . punctatus , Pseudoginglymostoma brevicaudatum , Electrolux addisoni , Dipturus crosnieri , Okamejei pita ( 5 ) Atlantic , Western Central32 ( 17 . 7 ) 81 ( 16 . 6 ) 62 ( 4 ) Diplobatis colombiensis , D . guamachensis , D . ommata , D . pictus ( 6 ) Pacific , Southwest34 ( 18 . 8 ) 49 ( 10 . 1 ) 28 ( 7 ) Atlantic , Southwest52 ( 28 . 7 ) 52 ( 10 . 7 ) 37 ( 19 ) Galeus mincaronei , Schroederichthys saurisqualus , Mustelus fasciatus , M . schmitti , Atlantoraja castelnaui , A . cyclophora , A . platana , Rioraja agassizii , Sympterygia acuta , Benthobatis kreffti , Dipturus mennii , Gurgesiella dorsalifera , Rhinobatos horkelii , Zapteryx brevirostris , Rhinoptera brasiliensis , Squatina argentina , S . guggenheim , S . occulta , S . punctata ( 8 ) Atlantic , Southeast37 ( 20 . 4 ) 51 ( 10 . 5 ) 13 ( 9 ) Atlantic , Eastern Central42 ( 23 . 2 ) 44 ( 9 . 0 ) 6 ( 10 ) Pacific , Southeast26 ( 14 . 4 ) 67 ( 13 . 8 ) 32 ( 3 ) Mustelus whitneyi , Triakis acutipinna , T . maculata ( 11 ) Pacific , Eastern Central20 ( 11 . 0 ) 52 ( 10 . 7 ) 19 ( 2 ) Urotrygon reticulata , U . simulatrix ( 12 ) Atlantic , Northeast33 ( 18 . 2 ) 23 ( 4 . 7 ) 8 ( 13 ) Atlantic , Northwest22 ( 12 . 2 ) 17 ( 3 . 5 ) 3 ( 1 ) Malacoraja senta ( 14 ) Mediterranean & Black Sea34 ( 18 . 8 ) 16 ( 3 . 3 ) 3 ( 1 ) Leucoraja melitensis ( 15 ) Pacific , Northeast9 ( 5 . 0 ) 11 ( 2 . 3 ) 0 ( 16 ) Indian , Antarctic1 ( 0 . 6 ) 4 ( 0 . 8 ) 2 ( 17 ) Atlantic , Antarctic1 ( 0 . 6 ) 4 ( 0 . 8 ) 2 ( 18 ) Pacific , Antarctic03 ( 0 . 6 ) 0 ( 19 ) Arctic Sea000Endemics were defined as those species found only within a single FAO Fishing Area . FAO Fishing Areas were ranked according to greatest species richness , percent threatened species , percent Data Deficient species , number of endemic species and number of threatened endemic species . The greatest number of threatened species coincides with the greatest richness ( Figure 7A vs 7C–E ) ; by controlling for species richness we can reveal the magnitude of threat in the pelagic ocean and two coastal hotspots that have a greater than expected level of threat: the Indo-Pacific Biodiversity Triangle and the Red Sea . Throughout much of the pelagic ocean , threat is greater than expected based on species richness alone , species richness is low ( n = 30 ) and a high percentage ( 86% ) are threatened ( n = 16 ) or Near Threatened ( n = 10 ) . Only four are of Least Concern ( Salmon shark Lamna ditropis , Goblin shark Mitsukurina owstoni , Longnose pygmy Shark Heteroscymnoides marleyi , and Largetooth cookiecutter shark Isistius plutodus ) ( Figure 9 ) . The Indo-Pacific Biodiversity Triangle , particularly the Gulf of Thailand , and the islands of Sumatra , Java , Borneo , and Sulawesi , is a hotspot of greatest residual threat especially for coastal sharks and rays with 76 threatened species ( Figure 9 ) . Indeed , the Gulf of Thailand large marine ecosystem has the highest threat density with 48 threatened chondrichthyans in an area of 0 . 36 million km2 . The Red Sea residual threat hotspot has 29 threatened pelagic and coastal species ( Figure 9 ) . There are 15 irreplaceable marine hotspots that harbor all 66 threatened endemic species ( Figure 8; Supplementary file 2C ) . In a world of limited funding , conservation priorities are often based on immediacy of extinction , the value of biodiversity and conservation opportunity ( Marris , 2007 ) . In this study , we provide the first estimates of the threat status and hence risk of extinction of chondrichthyans . Our systematic global assessment of the status of this lineage that includes many iconic predators reveals a risky combination of high threat ( 17% observed and 23 . 9% estimated ) , low safety ( Least Concern , 23% observed and >37% estimated ) , and high uncertainty in their threat status ( Data Deficient , 46% observed and 8 . 7% estimated ) . Over half of species are predicted to be threatened or Near Threatened ( n = 561 , 53 . 9% , Table 1 ) . While no species has been driven to global extinction—as far as we know—at least 28 populations of sawfishes , skates , and angel sharks are locally or regionally extinct ( Dulvy et al . , 2003; Dulvy and Forrest , 2010 ) . Several shark species have not been seen for many decades . The Critically Endangered Pondicherry shark ( Carcharhinus hemiodon ) is known only from 20 museum specimens that were captured in the heavily-fished inshore waters of Southeast Asia: it has not been seen since 1979 ( Cavanagh et al . , 2003 ) . The now ironically-named and Critically Endangered Common skate ( Dipturus batis ) and Common angel shark ( Squatina squatina ) are regionally extinct from much of their former geographic range in European waters ( Cavanagh and Gibson , 2007; Gibson et al . , 2008; Iglésias et al . , 2010 ) . The Largetooth sawfish ( Pristis pristis ) and Smalltooth sawfish ( Pristis pectinata ) are possibly extinct throughout much of the Eastern Atlantic , particularly in West Africa ( Robillard and Séret , 2006; Harrison and Dulvy , 2014 ) . Our analysis provides an unprecedented understanding of how many chondrichthyan species are actually or likely to be threatened . A very high percentage of species are DD ( 46% , 487 species ) ; that is one of the highest rates of Data Deficiency of any taxon to date ( Hoffmann et al . , 2010 ) . This high level of uncertainty in status further elevates risk and presents a key challenge for future assessment efforts . We outline a first step through our estimation that 68 DD species are likely to be threatened based on their life histories and distribution . Numerous studies have retrospectively explained extinction risk , but few have made a priori predictions of risk ( Dulvy and Reynolds , 2002; Davidson et al . , 2012 ) . Across many taxa , extinction risk has been shown to be a function of an extrinsic driver or threat ( Jennings et al . , 1998; Davies et al . , 2006 ) and the corresponding life history and ecological traits: large body size ( low intrinsic rate of population increase , high trophic level ) , small geographic range size , and ecological specialization . Maximum body size is an essential predictor of threat status , we presume because of the close relationship between body size and the intrinsic rate of population increase in sharks and rays ( Smith et al . , 1998; Frisk et al . , 2001; Hutchings et al . , 2012 ) . Though we note that this proximate link may be mediated ultimately through the time-related traits of growth and mortality ( Barnett et al . , 2013; Juan-Jordá et al . , 2013 ) . Our novel contribution is to show that depth-related geographic traits are more important for explaining risk than geographic range per se . The shallowness of species ( minimum depth limit ) and the narrowness of their depth range are important risk factors ( Figure 3 ) . We hypothesize that this is so because shallower species are more accessible to fishing gears and those with narrower depth ranges have lower likelihood that a proportion of the species distribution remains beyond fishing activity . For example , the Endangered Barndoor skate ( Dipturus laevis ) was eliminated throughout much of its geographic range and depth distribution due to bycatch in trawl fisheries , yet may have rebounded because a previously unknown deepwater population component lay beyond the reach of most fisheries ( Dulvy , 2000; Kulka et al . , 2002; COSEWIC , 2010 ) . We find that geographic range ( measured as Extent of Occurrence ) is largely unrelated to extinction risk . This is in marked contrast to extinction risk patterns on land ( Jones et al . , 2003; Cardillo et al . , 2005; Anderson et al . , 2011a ) and in the marine fossil record ( Harnik et al . , 2012a , 2012b ) , where small geographic range size is the principal correlate of extinction risk . We suggest that this is because fishing activity is now widespread throughout the world’s oceans ( Swartz et al . , 2010 ) , and even species with the largest ranges are exposed and often entirely encompassed by the footprint of fishing activity . By contrast , with a few exceptions ( mainly eastern Atlantic slopes , Figure 7E ) , fishing has a narrow depth penetration and hence species found at greater depths can still find refuge from exploitation ( Morato et al . , 2006; Lam and Sadovy de Mitcheson , 2010 ) . The status of chondrichthyans is arguably among the worst reported for any major vertebrate lineage considered thus far , apart from amphibians ( Stuart et al . , 2004; Hoffmann et al . , 2010 ) . The percentage and absolute number of threatened amphibians is high ( >30% are threatened ) , but a greater percentage are Least Concern ( 38% ) , and uncertainty of status is lower ( 32% DD ) than for chondrichthyans . Our discovery of the high level of threat in freshwater chondrichthyans ( 36% ) is consistent with the emerging picture of the intense and unmanaged extinction risk faced by many freshwater and estuarine species ( Darwall et al . , 2011 ) . Our threat estimate is comparable to other marine biodiversity status assessments , but our findings caution that ‘global’ fisheries assessments may be underestimating risk . The IUCN Global Marine Species Assessment is not yet complete , but reveals varying threat levels among taxa and regions ( Polidoro et al . , 2008 , 2012 ) . The only synoptic summary to-date focused on charismatic Indo-Pacific coral reef ecosystem species . Of the 1 , 568 IUCN-assessed marine vertebrates and invertebrates , 16% ( range: 12–34% among families ) were threatened ( McClenachan et al . , 2012 ) . This is a conservative estimate of marine threat level because although they may be more intrinsically sensitive to extinction drivers , charismatic species are more likely to garner awareness of their status and support for monitoring and conservation ( McClenachan et al . , 2012 ) . The predicted level of chondrichthyan threat ( >24% ) is distinctly greater than that provided by global fisheries risk assessments . These studies provide modeled estimates of the percentage of collapsed bony fish ( teleost ) stocks in both data-poor unassessed fisheries ( 18% , Costello et al . , 2012 ) and data-rich fisheries ( 7–13% , Branch et al . , 2011 ) . This could be because teleosts are generally more resilient than elasmobranchs ( Hutchings et al . , 2012 ) , but in addition we caution that analyses of biased geographic and taxonomic samples may be underestimating risk of collapse in global fisheries , particularly for species with less-resilient life histories . Our work relies on consensus assessments by more than 300 scientists . However , given the uncertainty in some of the underlying data that inform our understanding of threat status , such as fisheries catch landings data , it is worth considering whether these uncertainties mean our assessments are downplaying the true risk . While there are methods of propagating uncertainty through the IUCN Red List Assessments ( Akcakaya et al . , 2000 ) , in our experience this approach was uninformative for even the best-studied species , because it generated confidence intervals that spanned all IUCN Categories . Instead it is worth considering whether our estimates of threat are consistent with independent quantitative estimates of status . The Mediterranean Red List Assessment workshop in 2005 prompted subsequent quantitative analyses of catch landings , research trawl surveys , and sightings data . Quantitative trends could be estimated for five species suggesting they had declined by 96% to >99 . 9% relative to their former abundance suggesting they would meet the highest IUCN Threat category of Critically Endangered ( Ferretti et al . , 2008 ) . By comparison the earlier IUCN regional assessment for these species , while suggesting they were all threatened , was more conservative for two of the five species: Hammerhead sharks ( Sphyrna spp . ) —Critically Endangered , Porbeagle shark ( Lamna nasus ) —Critically Endangered , Shortfin mako ( Isurus oxyrinchus ) —Critically Endangered , Blue shark ( Prionace glauca ) —Vulnerable , and Thresher shark ( Alopias vulpinus ) —Vulnerable . We can also make a complementary comparison to a recent analysis of the status of 112 shark and ray fisheries ( Costello et al . , 2012 ) . The median biomass relative to the biomass at Maximum Sustainable Yield ( B/BMSY ) of these 112 shark and ray fisheries was 0 . 37 , making them the most overfished groups of any of the world’s unassessed fisheries . Assuming BMSY occurs at 0 . 3 to 0 . 5 of unexploited biomass then the median biomass of shark and ray fisheries had declined by between 81% and 89% by 2009 . These biomass declines would be sufficient to qualify all of these 112 shark and ray fisheries for the Endangered IUCN category if they occurred within a three-generation time span . By comparison our results are considerably more conservative . Empirical analyses show that an IUCN threatened category listing is triggered only once teleost fishes ( with far higher density-dependent compensation ) have been fished down to below BMSY ( Dulvy et al . , 2005; Porszt et al . , 2012 ) . Hence , our findings are consistent with only around one-quarter of chondrichthyan species having been fished down below the BMSY target reference point . While there may be concern that expert assessments may overstate declines and threat , it is more likely that our conservative consensus-based approach has understated declines and risk in sharks and rays . For marine species , predicting absolute risk of extinction remains highly uncertain because , even with adequate evidence of severe decline , in many instances the absolute population size remains large ( Mace , 2004 ) . There remains considerable uncertainty as to the relationship between census and effective population size ( Reynolds et al . , 2005 ) . Therefore , Red List categorization of chondrichthyans should be interpreted as a comparative measure of relative extinction risk , in recognition that unmanaged steep declines , even of large populations , may ultimately lead to ecosystem perturbations and eventually biological extinction . The Red List serves to raise red flags calling for conservation action , sooner rather than later , while there is a still chance of recovery and of forestalling permanent biodiversity loss . Despite more than two decades of rising awareness of chondrichthyan population declines and collapses , there is still no global mechanism to ensure financing , implementation and enforcement of chondrichthyan fishery management plans that is likely to rebuild populations to levels where they would no longer be threatened ( Lack and Sant , 2009; Techera and Klein , 2011 ) . This management shortfall is particularly problematic given the large geographic range of many species . Threat increased only slightly when geographic range is measured as the Extent of Occurrence; however , geographic range becomes increasingly important when it is measured as the number of countries ( legal jurisdictions ) spanned by each species . The proportion of species that are threatened increases markedly with geographic size measured by number of Exclusive Economic Zones ( EEZs ) spanned; one-quarter of threatened species span the EEZs of 18 or more countries ( Figure 10 ) . Hence , their large geographic ranges do not confer safety , but instead exacerbates risk because sharks and rays require coherent , effective international management . 10 . 7554/eLife . 00590 . 019Figure 10 . Elevated threat in chondrichthyans with the largest geographic ranges , spanning the greatest number of national jurisdictions . Frequency distribution of number of jurisdictions spanned by all chondrichthyans ( black , n = 1 , 041 ) and threatened species only ( red , n = 174 ) , for ( A ) country EEZs , and ( B ) the overrepresentation of threatened species spanning a large number of country EEZs , shown by the log ratio of proportion of threatened species over the proportion of all species . The proportion of threatened species is greater than the proportion of all species where the log ratio = 0 , which corresponds to range spans of 16 and more countries . DOI: http://dx . doi . org/10 . 7554/eLife . 00590 . 019 With a few exceptions ( e . g . , Australia and USA ) , many governments still lack the resources , expertise , and political will necessary to effectively conserve the vast majority of shark and rays , and indeed many other exploited organisms ( Veitch et al . , 2012 ) . More than 50 sharks are included in Annex I ( Highly Migratory Species ) of the 1982 Law of the Sea Convention , implemented on the high seas under the 1992 Fish Stocks Agreement , but currently only a handful enjoy species-specific protections under the world’s Regional Fishery Management Organizations ( Table 6 ) , and many of these have yet to be implemented domestically . The Migratory Sharks Memorandum of Understanding ( MoU ) adopted by the Parties to the Convention on Migratory Species ( CMS ) so far only covers seven sharks , yet there may be more than 150 chondrichthyans that regularly migrate across national boundaries ( Fowler , 2012 ) . To date , only one of the United Nations Environment Programme’s Regional Seas Conventions , the Barcelona Convention for the Conservation of the Mediterranean Sea , includes chondrichthyan fishes and only a few of its Parties have taken concrete domestic action to implement these listings . Despite two decades of effort , only ten sharks and rays had been listed by the Convention on International Trade in Endangered Species ( CITES ) up to 2013 ( Vincent et al . , 2014 ) . A further seven species of shark and ray were listed by CITES in 2013—the next challenge is to ensure effective implementation of these trade regulations ( Mundy-Taylor and Crook , 2013 ) . OSPAR ( the Convention for the Protection of the marine Environment of the North-East Atlantic ) lists many threatened shark and ray species , but its remit excludes fisheries issues . Many chondrichthyans qualify for listing under CITES , CMS , and various regional seas conventions , and should be formally considered for such action as a complement to action by Regional Fisheries Management Organizations ( RFMOs ) ( Table 6 ) . 10 . 7554/eLife . 00590 . 020Table 6 . Progress toward regional and international RFMO management measures for sharks and raysDOI: http://dx . doi . org/10 . 7554/eLife . 00590 . 0201 . Bans on ‘finning’ ( the removal of a shark’s fins and discarding the carcass at sea ) through most RFMOs ( Fowler and Séret , 2010 ) ;2 . North East Atlantic Fisheries Commission ( NEAFC ) bans on directed fishing for species not actually targeted within the relevant area ( Spiny dogfish [Squalus acanthias] , Basking shark [Cetorhinus maximus] , Porbeagle shark [Lamna nasus] ) ( NEAFC , 2009 ) ;3 . Convention on the Conservation of Antarctic Marine Living Resources bans on ‘directed’ fishing for skates and sharks and bycatch limits for skates ( CCMLR , 2011 ) ;4 . A Northwest Atlantic Fisheries Organization ( NAFO ) skate quota ( note: this has consistently been set higher than the level advised by scientists since its establishment in 2004 ) ( NAFO , 2011 ) ;5 . International Commission for the Conservation of Atlantic Tunas ( ICCAT ) bans on retention , transshipment , storage , landing , and sale of Bigeye Thresher ( Alopias superciliosus ) , and Oceanic whitetip shark ( Carcharhinus longimanus ) , and partial bans ( developing countries excepted under certain circumstances ) on retention , transshipment , storage , landing , and sale of most hammerheads ( Sphyrna spp . ) , and retention , transshipment , storage , and landing ( but not sale ) of Silky shark ( Carcharhinus falciformis ) ( Kyne et al . , 2012 ) ;6 . An Inter-American Tropical Tuna Commission ( IATTC ) ban on retention , transshipment , storage , landing , and sale of Oceanic whitetip sharks ( IATTC , 2011 ) ;7 . An Indian Ocean Tuna Commission ( IOTC ) ban on retention , transshipment , storage , landing , and sale of thresher sharks-with exceptionally low compliance and reportedly low effectiveness ( IOTC , 2011 ) ; and , 8 . A Western and Central Pacific Fisheries Commission ban on retention , transshipment , storage , and landing ( but not sale ) of Oceanic whitetip sharks ( Clarke et al . , 2013 ) . Bans on ‘finning’ ( slicing off a shark’s fins and discarding the body at sea ) are the most widespread shark conservation measures . While these prohibitions , particularly those that require fins to remain attached through landing , can enhance monitoring and compliance , they have not significantly reduced shark mortality or risk to threatened species ( Clarke et al . , 2013 ) . Steep declines and the high threat levels in migratory oceanic pelagic sharks suggest raising the priority of improved management of catch and trade through concerted actions by national governments working through RFMOs as well as CITES , and CMS ( Table 7 ) . 10 . 7554/eLife . 00590 . 021Table 7 . Management recommendations: the following actions would contribute to rebuilding threatened chondrichthyan populations and properly managing associated fisheriesDOI: http://dx . doi . org/10 . 7554/eLife . 00590 . 021Fishing nations and regional fisheries management organizations ( RFMOs ) are urged to: 1 . Implement , as a matter of priority , scientific advice for protecting habitat and/or preventing overfishing of chondrichthyan populations; 2 . Draft and implement Plans of Action pursuant to the International Plan Of Action ( IPOA–Sharks ) , which include , wherever possible , binding , science-based management measures for chondrichthyans and their essential habitats; 3 . Significantly increase observer coverage , monitoring , and enforcement in fisheries taking chondrichthyans; 4 . Require the collection and accessibility of species-specific chondrichthyan fisheries data , including discards , and penalize non-compliance; 5 . Conduct population assessments for chondrichthyans; 6 . Implement and enforce chondrichthyan fishing limits in accordance with scientific advice; when sustainable catch levels are uncertain , set limits based on the precautionary approach; 7 . Strictly protect chondrichthyans deemed exceptionally vulnerable through Ecological Risk Assessments and those classified by IUCN as Critically Endangered or Endangered; 8 . Prohibit the removal of shark fins while onboard fishing vessels and thereby require the landing of sharks with fins naturally attached; and , 9 . Promote research on gear modifications , fishing methods , and habitat identification aimed at mitigating chondrichthyan bycatch and discard mortality . National governments are urged to: 10 . Propose and work to secure RFMO management measures based on scientific advice and the precautionary approach; 11 . Promptly and accurately report species-specific chondrichthyan landings to relevant national and international authorities; 12 . Take unilateral action to implement domestic management for fisheries taking chondrichthyans , including precautionary limits and/or protective status where necessary , particularly for species classified by IUCN as Vulnerable , Endangered or Critically Endangered , and encourage similar actions by other Range States; 13 . Adopt bilateral fishery management agreements for shared chondrichthyan populations; 14 . Ensure active membership in Convention on International Trade in Endangered Species ( CITES ) , Convention for the Conservation of Migratory Species ( CMS ) , RFMOs , and other relevant regional and international agreements; 15 . Fully implement and enforce CITES chondrichthyan listings based on solid non-detriment findings , if trade in listed species is allowed; 16 . Propose and support the listing of additional threatened chondrichthyan species under CITES and CMS and other relevant wildlife conventions; 17 . Collaborate on regional agreements and the CMS migratory shark Memorandum of Understanding ( CMS , 2010 ) , with a focus on securing concrete conservation actions; and , 18 . Strictly enforce chondrichthyan fishing and protection measures and impose meaningful penalties for violations . A high proportion of catch landings come from nations with a large number of threatened chondrichthyans and less-than-comprehensive chondrichthyan fishery management plans . Future research is required to down-scale these global Red List assessments and analyses to provide country-by-country diagnoses of the link between specific fisheries and specific threats to populations of more broadly distributed species ( Wallace et al . , 2010 ) . Such information could be used to focus fisheries management and conservation interventions that are tailored to specific problems . There is no systematic global monitoring of shark and ray populations and the national fisheries catch landings statistics provide invaluable data for tracking fisheries trends in unmanaged fisheries ( Newton et al . , 2007; Worm et al . , 2013 ) . However , the surveillance power of such data could be greatly improved if collected at greater taxonomic resolution . While there have been continual improvements , catches are under-reported ( Clarke et al . , 2006 ) , and for those that are reported only around one-third is reported at the species level ( Fischer et al . , 2012 ) . To complement improved catch landings data , we recommend the development of repeat regional assessments of the Red List Status of chondrichthyans to provide an early warning of adverse changes in status and to detect and monitor the success of management initiatives and interventions . Aggregate Red List Threat indices for chondrichthyans , like those available for mammals , birds , amphibians , and hard corals ( Carpenter et al . , 2008 ) would provide one of the few global scale indicators of progress toward international biodiversity goals ( Walpole et al . , 2009; Butchart et al . , 2010 ) . Our global status assessment of sharks and rays reveals the principal causes and severity of global marine biodiversity loss , and the threat level they face exposes a serious shortfall in the conservation management of commercially-exploited aquatic species ( McClenachan et al . , 2012 ) . Chondrichthyans have slipped through the jurisdictional cracks of traditional national and international management authorities . Rather than accept that many chondrichthyans will inevitably be driven to economic , ecological , or biological extinction , we warn that dramatic changes in the enforcement and implementation of the conservation and management of threatened chondrichthyans are urgently needed to ensure a healthy future for these iconic fishes and the ecosystems they support . We applied the Red List Categories and Criteria developed by the International Union for Conservation of Nature ( IUCN ) ( IUCN , 2004 ) to 1 , 041 species at 17 workshops involving more than 300 experts who incorporated all available information on distribution , catch , abundance , population trends , habitat use , life histories , threats , and conservation measures . Some 105 chondrichthyan fish species had been assessed and published in the 2000 Red List of Threatened Species prior to the initiation of the Global Shark Red List Assessment ( GSRLA ) . These assessments were undertaken by correspondence and through discussions at four workshops ( 1996—London , UK , and Brisbane , Australia; 1997—Noumea , New Caledonia , and 1999—Pennsylvania , USA ) . These assessments applied earlier versions of the IUCN Red List Criteria and , where possible , were subsequently reviewed and updated according to version 3 . 1 Categories and Criteria during the GSRLA . The IUCN Shark Specialist Group ( SSG ) subsequently held a series of 13 regional and thematic Red List workshops in nine countries around the world ( Table 8 ) . Prior to the workshops , each participant was asked to select species for assessment based on their expertise and research areas . Where possible , experts carried out research and preparatory work in advance , thus enabling more synthesis to be achieved during each workshop . SSG Red List-trained personnel facilitated discussion and consensus sessions , and coordinated the production of global Red List Assessments for species in each region . For species that had previously been assessed , participants provided updated information and assisted in revised assessments . Experts completed assessments for some wide-ranging , globally distributed species over the course of several workshops . In total , 302 national , regional , and international experts from 64 countries participated in the GSRLA workshops and the production of assessments . All Red List Assessments were based on the collective knowledge and pooled data from dedicated experts across the world , ensuring global consultation and consensus to achieve the best assessment for each species with the knowledge and resources available ( ‘Acknowledgements’ ) . Any species assessments not completed during the workshops were finalized through subsequent correspondence among experts . The SSG evaluated the status of all described chondrichthyan species that are considered to be taxonomically valid up to August 2011 ( see “Systematics , missing species and species coverage” below ) . Experts compiled peer-reviewed Red List documentation for each species , including data on: systematics , population trends , geographic range , habitat preferences , ecology , life history , threats , and conservation measures . The SSG assessed all species using the IUCN Red List Categories and Criteria version 3 . 1 ( IUCN , 2001 ) . The categories and their standard abbreviations are: Critically Endangered ( CR ) , Endangered ( EN ) , Vulnerable ( VU ) , Near Threatened ( NT ) , Least Concern ( LC ) , and Data Deficient ( DD ) . Experts further coded each species according to the IUCN Habitats , Threats and Conservation Actions Authority Files , enabling analysis of their habitat preferences , major threats and conservation action requirements . SSG Program staff entered all data into the main data fields in the IUCN Species Information Service Data Entry Module ( SIS DEM ) and subsequently transferred these data into the IUCN Species Information Service ( SIS ) in 2009 . The SSG and the GMSA created ArcGIS distribution maps as polygons describing the geographical range of each chondrichthyan depending on the individual species’ point location and depth information . Pelagic species distribution maps were digitized by hand from the original map sources . For spatial analyses , we merged all species maps into a single shapefile . We mapped species using a hexagonal grid composed of individual units ( cells ) that retain their shape and area ( ∼23 , 322 km2 ) throughout the globe . Specifically , we used the geodesic discrete global grid system , defined on an icosahedron and projected to the sphere using the inverse Icosahedral Snyder Equal Area ( ISEA ) ( Sahr et al . , 2003 ) . A row of cells near longitude 180°E/W was excluded , as these interfered with the spatial analyses ( Hoffmann et al . , 2010 ) . Because of the way the marine species range maps are buffered , the map polygons are likely to extrapolate beyond known distributions , especially for any shallow-water , coastal species , hence not only will range size itself likely be an overestimate , but so will the number of hexagons . We excluded obligate freshwater species from the final analysis as their distribution maps have yet to be completed . The maps of the numbers of threatened species represent the sum of species that have been globally assessed as threatened , in IUCN Red List categories VU , EN or CR , existing in each ∼23 , 322 km2 cell . We caution that this should not be interpreted to mean that species existing within that grid cell are necessarily threatened in this specific location , rather that this location included species that are threatened , on average , throughout their Extent of Occurrence . The number of threatened species was positively related to the species richness of cells ( F1 , 14 , 846 = 1 . 5 e5 , p<0 . 001 , r2 = 0 . 91 ) . To remove this first-order effect and reveal those cells with greater and lower than expected extinction risk , we calculated the residuals of a linear regression of the number of threatened species on the number of non-DD species ( referred to as data sufficient species ) . Cells with positive residuals were mapped to show areas of greater than expected extinction risk compared to cells with equal or negative residuals . Hexagonal cell information was converted to point features and smoothed across neighboring cells using ordinary kriging using a spherical model in the Spatial Analyst package of ArcView . Such smoothing can occasionally lead to contouring artefacts , such as the yellow wedge west of southern Africa in Figure 7D , and we caution against over-interpreting marginal categorization changes . We identified hotspots of threatened endemic chondrichthyans to guide conservation priorities . To describe the potential cost of losing unique chondrichthyan faunas , we calculated irreplaceability scores for each cell . Irreplaceability scores were calculated for each species as the reciprocal of its area of occupancy measured as the number of cells occupied . For example , for a species with an Extent of Occurrence spanning 100 hexagons , each hexagon in its range would have an irreplaceability 1/100 or 0 . 01 in each of the 100 hexagons of its Extent of Occurrence . The irreplaceability of each cell was calculated by averaging log10 transformed irreplaceability scores of each species in each cell . Averaging irreplaceability scores controls for varying species richness across cells . We calculated irreplaceability both for all chondrichthyans and for threatened species only . Irreplaceability was also calculated using only endemic threatened species , whereby endemicity was defined as species having an Extent of Occurrence of <50 , 000 , 100 , 000 , 250 , 000 or 500 , 000 km2 . Different definitions of endemicity gave similar patterns of irreplaceability and we present the results of only the largest-scale definition of endemicity . Hence the irreplaceability of threatened species and particularly the threatened endemic chondrichthyans represents those locations or ‘hotspots’ ( Myers et al . , 2000 ) at greatest risk of losing the most unique chondrichthyan biodiversity . We extracted chondrichthyan landings reported to FAO by 146 countries and territories from a total of 128 countries ( as some chondrichthyan fishing nations are overseas territories , unincorporated territories , or British Crown Dependencies ) from FishStat ( FAO , 2011 ) . We categorized landings into 153 groupings , comprised of 128 species-specific categories ( e . g . , angular roughshark , piked dogfish , porbeagle , Patagonian skate , plownose chimaera , small-eyed ray , etc ) and 25 broader nei ( nei = not elsewhere included ) groupings ( e . g . , such as various sharks nei , threshers sharks nei , ratfishes nei , raja rays nei ) . For each country , all chondrichthyan landings in metric tonnes ( t ) were averaged over the decade 2000–2009 . Landings reported as ‘<0 . 5’ were assigned a value of 0 . 5 t . Missing data reported as ‘ . ’ were assigned a zero . Total annual chondrichthyan landings are underestimated as data are not reported for 1 , 522 out of a total count of 13 , 990 entries in the dataset . Therefore , 11% of chondrichthyan landings reported to the FAO over the 10-year period are ‘data unavailable , unobtainable’ . We mapped FAO chondrichthyan landings as the national percent share of the average total landings from 2000 to 2009 . For the analysis of landings over time we removed the aggregate category ‘sharks , rays , skates , etc’ and all nine of the FAO chimaera reporting categories . The ‘sharks , rays , skates , etc’ FAO reported category comprised 15 , 684 , 456 tonnes of the reported catch from all countries during 1950–2009 , which is a total of 45% of the total reported catch for this time period . However , the proportion of catch in this category has declined from around 50% of global catch to around 35% , presumably due to better reporting of ray catch and as sharks have declined or come under stronger protection ( Figure 1 ) . The nine chimaera categories make up a small fraction of the global catch , 249 , 404 . 5 tonnes from 1950 to 2009 , representing 0 . 72% of the total catch . Hong Kong has long served as one of the world’s largest entry ports for the global shark fin trade . While fins are increasingly being exported to Mainland China where species-specific trade data is more difficult to obtain , each year ( from 1996 to 2001 ) Hong Kong handled around half of all fin imports ( Clarke et al . , 2006 ) . Data on shark fin exports to Hong Kong were requested directly from the Hong Kong Census and Statistics Department ( Hong Kong Special Administrative Region Government , 2011 ) . We mapped exports to Hong Kong as the proportion of the summed total weight of the four categories of shark fin exported to Hong Kong in 2010: ( 1 ) shark fins ( with or without skin ) , with cartilage , dried , whether or not salted but not smoked ( trade code: 3055950 ) , ( 2 ) shark fins ( with or without skin ) , without cartilage , dried , whether or not salted but not smoked ( 3055930 ) , ( 3 ) shark fins ( with or without skin ) , without cartilage , salted or in brine , but not dried , or smoked ( 3056940 ) , and ( 4 ) shark fins ( with or without skin ) , with cartilage , salted or in brine , but not dried or smoked ( 3056930 ) . We could not correct the difference in weight due to product type . To identify the threat classification of the chondrichthyan species in the fin trade , we included records of the most numerous species used in the Hong Kong fin trade as well as those species with the most-valued fins ( Clarke et al . , 2006 , 2007; Clarke , 2008 ) .	Ocean ecosystems are under pressure from overfishing , climate change , habitat destruction and pollution . These pressures have led to documented declines of some fishes in some places , such as those living in coral reefs and on the high seas . However , it is not clear whether these population declines are isolated one-off examples or , instead , if they are sufficiently widespread to risk the extinction of large numbers of species . Most fishes have a skeleton that is made of bone , but sharks and rays have a skeleton that is made of cartilage . A total of 1 , 041 species has such a skeleton and they are collectively known as the Chondrichthyes . To find out how well these fish are faring , Dulvy et al . worked with more than 300 scientists around the world to assess the conservation status of all 1 , 041 species . Based on this , Dulvy et al . estimate that one in four of these species are threatened with extinction , mainly as a result of overfishing . Moreover , just 389 species ( 37 . 4% of the total ) are considered to be safe , which is the lowest fraction of safe species among all vertebrate groups studied to date . The largest sharks and rays are in the most peril , especially those living in shallow waters that are accessible to fisheries . A particular problem is the ‘fin trade’: the fins of sharks and shark-like rays are a delicacy in some Asian countries , and more than half of the chondrichthyans that enter the fin trade are under threat . Whether targeted or caught by boats fishing for other species , sharks and rays are used to supply a market that is largely unmonitored and unregulated . Habitat degradation and loss also pose considerable threats , particularly for freshwater sharks and rays . Dulvy et al . identified three main hotspots where the biodiversity of sharks and rays was particularly seriously threatened—the Indo-Pacific Biodiversity Triangle , Red Sea , and the Mediterranean Sea—and argue that national and international action is needed to protect them from overfishing .	[ "Abstract", "Introduction", "Results", "Discussion", "Methods" ]	[ "ecology" ]	2014	Extinction risk and conservation of the world’s sharks and rays
The spliceosome is a complex machine composed of small nuclear ribonucleoproteins ( snRNPs ) and accessory proteins that excises introns from pre-mRNAs . After assembly the spliceosome is activated for catalysis by rearrangement of subunits to form an active site . How this rearrangement is coordinated is not well-understood . During activation , U4 must be released to allow U6 conformational change , while Prp19 complex ( NTC ) recruitment is essential for stabilizing the active site . We used multi-wavelength colocalization single molecule spectroscopy to directly observe the key events in Saccharomyces cerevisiae spliceosome activation . Following binding of the U4/U6 . U5 tri-snRNP , the spliceosome either reverses assembly by discarding tri-snRNP or proceeds to activation by irreversible U4 loss . The major pathway for NTC recruitment occurs after U4 release . ATP stimulates both the competing U4 release and tri-snRNP discard processes . The data reveal the activation mechanism and show that overall splicing efficiency may be maintained through repeated rounds of disassembly and tri-snRNP reassociation . The spliceosome is one of the most dynamic molecular machines inside the cell . Removal of introns from precursors to mRNAs ( pre-mRNAs ) involves the coordinated action of 5 small nuclear RNAs ( snRNAs ) and >100 proteins ( Wahl et al . , 2009; Hoskins and Moore 2012 ) . Some of these proteins along with the snRNAs assemble into small nuclear ribonucleoprotein particles ( the U1 , U2 , U4 , U5 , and U6 snRNPs ) that work together with other accessory proteins to catalyze splicing . Experiments in vitro ( Hoskins et al . , 2011; Ruby and Abelson 1988; Konarska and Sharp 1986; Cheng and Abelson 1987 ) and in cells ( Tardiff and Rosbash 2006 ) indicate that spliceosomes are unlikely to exist as preformed complexes . Instead , spliceosomes are built from their snRNP and accessory protein components on pre-mRNAs , carry out splicing , and then are disassembled after each reaction ( Wahl et al . , 2009 ) . Consequently , the overall process can be described as sequential progression through distinct stages of spliceosome assembly , formation of the active site ( called activation ) , catalysis , disassembly , and component recycling . A number of biochemical and genetic experiments have elucidated the splicing factors present at each stage ( Wahl et al . , 2009; Fabrizio et al . , 2009 ) , as well as characteristic interactions between snRNA , pre-mRNA , and protein components ( Brow 2002; Wahl et al . , 2009 ) . The U1 and U2 snRNPs identify the 5' splice site ( SS ) and branch site ( BS ) , respectively , during early stages of spliceosome assembly . While U1 binding is ATP-independent , U2 base pairing with the intron to form the pre-spliceosome or A complex typically requires ATP hydrolysis ( Figure 1A ) . A pre-formed complex of U4 , U5 , and U6 ( the U4/U6 . U5 tri-snRNP ) then joins A complex to form B complex . While B complex contains dozens of proteins and five snRNAs , it is not capable of mediating either of the two chemical steps of splicing ( 5' SS cleavage and exon ligation ) since the spliceosomal components are not yet rearranged into a configuration capable of catalysis . This rearrangement encompasses two stages during the activation process . In the first stage , the Prp19-associated complex ( NTC ) , the final major spliceosomal subcomplex , joins and U1 and U4 are expelled to form the Bact spliceosome . In the second stage , Bact is further remodeled to the B* complex ( Lardelli et al . , 2010; Wlodaver and Staley 2014; Liu and Cheng 2012 ) . Single molecule FRET ( smFRET ) experiments suggest that it is only in B* complex that the 5' SS and BS become juxtaposed , a necessary prerequisite for formation of a spliceosome competent for 5' SS cleavage ( Crawford et al . , 2013; Krishnan et al . , 2013 ) . These catalytically activated B* spliceosomes then progress further through stages of exon ligation , mRNA product release , and finally disassembly of the lariat intron-containing product complex . 10 . 7554/eLife . 14166 . 003Figure 1 . Cartoon of major steps in spliceosome assembly and activation and impact of 2 mM ( dark green ) and 50 µM ( light green ) ATP concentrations on U1 and NTC interactions with pre-mRNA . ( A ) Spliceosome assembly and activation . snRNPs interact with the pre-mRNA to form the A complex , containing U1 and U2 associated with the 5' SS and BS , respectively . The U4/U6 . U5 tri-snRNP is then recruited to form B complex . At 2 mM ATP , activation can proceed and result in release of U1 and U4 and acquisition of the NTC to form Bact . Subsequent steps then lead to splicing . At 50 µM ATP , the first step of activation is inhibited . ( B ) Schematic of a two-color CoSMoS experiment for observing U1 binding dynamics . The U1 snRNP contained two green-excited ( Dy549 ) fluorophores attached to two different proteins while the pre-mRNA was immobilized to the slide surface and contained a single Alexa 488 or Alexa 647 fluorophore . ( C ) Time record of the number of U1 fluorescence spots relative to the number of surface-tethered pre-mRNA molecules at 2 mM and 50 µM ATP . ( D ) Example fluorescence intensity records showing binding of U1 to individual pre-mRNA molecules at 2 mM and 50 µM ATP . ( E ) Probability density histogram of dwell times for U1 at 2 mM and 50 µM ATP ( N = 535 events on 166 pre-mRNAs at 50 µM ATP ) . Results at 2 mM ATP were originally reported by Hoskins et al . and those data sets were used to generate this new figure for comparison ( Hoskins et al . , 2011 ) . Lines represent fits of the distributions of dwell times to multi-exponential equations . ( F ) Schematic of a two-color CoSMoS experiment for observing NTC binding dynamics . ( G ) Time record of the number NTC fluorescence spots relative to the number of surface-tethered pre-mRNA molecules at 2 mM and 50 µM ATP . ( H ) Example fluorescence intensity records showing binding of the NTC to individual pre-mRNA molecules at 2 mM and 50 µM ATP . ( I ) Probability density histogram of dwell times for the NTC at 2 mM and 50 µM ATP ( N = 234 events on 169 pre-mRNAs at 50 µM ATP ) . Results at 2 mM ATP were originally reported by Hoskins et al . and those data sets were used to generate this new figure for comparison ( Hoskins et al . , 2011 ) . Lines represent fits of the distributions of dwell times to multi-exponential equations . Parameters for fits in ( E ) and ( I ) are shown in Figure 1— source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 14166 . 003 10 . 7554/eLife . 14166 . 004Figure 1—source data 1 . Fit parameters describing the distribution of dwell times observed for the U1 and NTC subcomplexes . DOI: http://dx . doi . org/10 . 7554/eLife . 14166 . 004 10 . 7554/eLife . 14166 . 005Figure 1—figure supplement 1 . Additional examples ( to supplement those shown in Figure 1D ) of fluorescence intensity traces showing individual U1-SNAP subcomplexes co-localizing with surface-tethered pre-mRNA in the presence of 2 mM ( A–E , dark green ) or 50 µM ( F–J , light green ) ATP . DOI: http://dx . doi . org/10 . 7554/eLife . 14166 . 00510 . 7554/eLife . 14166 . 006Figure 1—figure supplement 2 . Additional examples ( to supplement those shown in Figure 1H ) of fluorescence intensity traces showing individual NTC-SNAP subcomplexes co-localizing with surface-tethered pre-mRNA in the presence of 2 mM ( A–E , dark green ) or 50 µM ( F–J , light green ) ATP . DOI: http://dx . doi . org/10 . 7554/eLife . 14166 . 006 Despite the central importance of the B-to-Bact transition , key mechanistic questions remain unanswered . For example , the order of U4 loss and NTC association is uncertain and alternative models have been proposed ( Fabrizio et al . , 2009; Chan et al . , 2003; Tarn et al . , 1993 ) . It is also unclear whether U4 snRNP loss is irreversible or whether it can rejoin an activated spliceosome to fix incorrectly assembled complexes by possible proofreading or spliceosomal discard pathways . Single-molecule fluorescence microscopy techniques have proven valuable in elucidating the kinetic mechanisms of several parts of the splicing process ( Hoskins et al . , 2011; Abelson et al . , 2010; Shcherbakova et al . , 2013; Crawford et al . , 2013; Krishnan et al . , 2013; Cherny et al . , 2010; Karunatilaka and Rueda 2013; Guo et al . , 2009 ) . To investigate the mechanism of spliceosomal activation , we have used Colocalization Single Molecule Spectroscopy ( CoSMoS ) ( Friedman et al . , 2006; Hoskins et al . , 2011 ) to study the binding dynamics of U4 relative to other spliceosome components . CoSMoS is particularly suited for the study of unsynchronized reactions such as spliceosomal activation since individual reaction trajectories can be followed in real time and subsequently grouped into common patterns of behavior during data analysis ( Larson et al . , 2014; Walter et al . , 2008; Chen et al . , 2012 ) . Here we followed the comings and goings of U4 , U5 and the NTC under conditions that permitted or prevented spliceosomal activation . By simultaneously observing multiple species , we were able to distinguish reaction pathways relevant to activation from off-pathway processes and define the order and reversibility of on-pathway binding and dissociation steps . Together the data reveal new features of spliceosomal activation , provide evidence for an ATP-dependent checkpoint that results in rapid tri-snRNP discard , and define the major pathway for NTC recruitment . In previous studies , we have shown that snRNPs and the NTC can dynamically engage pre-mRNAs under ATP concentrations that permit splicing ( 2 mM ) ( Hoskins et al . , 2011; Shcherbakova et al . , 2013 ) . In the absence of ATP , however , only U1 snRNP could stably interact with pre-mRNA ( Hoskins et al . , 2011 ) , consistent with previous spliceosome assembly models ( Legrain et al . , 1988; Wahl et al . , 2009 ) . In between these two extremes , lower amounts of ATP ( e . g . , 50 µM ) allow for spliceosome assembly up to tri-snRNP addition ( B complex formation ) but prevent loss of U4 snRNP and spliceosomal activation ( Figure 1A ) ( Chan et al . , 2003; Tarn et al . , 1993 ) . This has proven particularly useful for isolating B complex and investigating its composition ( Fabrizio et al . , 2009 ) . We wondered if a similar approach could be used to stall spliceosomes to observe B complex on single pre-mRNA molecules and gain insights into the mechanisms of spliceosomal activation in single molecule experiments . Based on the abundance of U1 snRNP in biochemically-purified , stalled B complex ( Fabrizio et al . , 2009 ) , we predicted that these stalled spliceosomes could be detected at the single molecule level by an increase in U1 snRNP dwell time at 50 µM ATP compared to dwell times measured under conditions that permit U1 release and splicing . We previously described procedures for monitoring U1 snRNP interactions with surface-tethered pre-mRNAs in yeast whole cell extract ( WCE ) using green-laser excitable ( Dy549 ) SNAP tag fluorophores on U1 components ( Snp1 , Prp40; Table 1 ) and RP51A pre-mRNA containing a 5' cap , 3' biotin handle , and a single , site-specifically incorporated Alexa647 ( red-laser excitable ) or Alexa488 ( blue-laser excitable ) fluorophore ( Figure 1B ) ( Hoskins et al . , 2011 ) . 10 . 7554/eLife . 14166 . 007Table 1 . Yeast strains and labeled proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 14166 . 007StrainSNAP Tag LabelsaDHFR Tag LabelsGenotypeReferencesyAAH0001 BJ2168NoneNoneMATa prc1–407 prb1–1122 pep4–3 leu2 trp1 ura3–52 gal2Parental StrainyAAH0006U1: Snp1 , Prp40NoneyAAH001 + snp1::snp1-SNAP-HYG + prp40::prp40-SNAP-NATHoskins et al . , 2011yAAH0020NTC: Cef1 , Ntc90NoneyAAH001 + cef1::cef1-SNAP-HYG + ntc90::ntc90-SNAP-NATHoskins et al . , 2011yAAH0067U4: Prp3 , Prp4NoneyAAH001 + prp3::prp3-SNAPf-Hyg + prp4::prp4-SNAPf-NATthis workyAAH0071U4: Prp3U5: Brr2 , Snu114yAAH001 + prp3::prp3-SNAPf-NAT + brr2::brr2-DHFR-HYG + snu114::snu114-DHFR-BLEthis workyAAH0077U4: Prp3NTC: Cef1 , Ntc90yAAH001 + prp3::prp3-SNAPf-NAT + cef1::cef1-DHFR-HYG + ntc90::ntc90-DHFR-BLEthis workaStrains yAAH0067 , 71 , and 77 were labeled with the fast SNAP ( SNAPf ) tag As expected , U1 snRNPs initially bound RP51A pre-mRNAs at both 2 mM and 50 µM ATP ( Figure 1C ) . However , the profile of U1 fluorescent spot accumulation under the two conditions was quite different: at 2 mM ATP , U1 snRNP spot numbers increased , then decreased as spliceosomes underwent activation; at 50 µM ATP , U1 spots reached a higher surface density that did not decrease with time ( Figure 1C ) . The latter result is consistent with U1 remaining bound to pre-activation spliceosomal B complexes . Inspection of individual U1 binding events on single pre-mRNAs confirmed this assumption . In 50 µM ATP , U1 spots often remained visible for tens of minutes ( e . g . , Figure 1D , light green and Figure 1—figure supplement 1 ) . This is much longer than previously observed either in the absence of ATP ( which prevents A complex formation ) or at an ATP concentration ( 2 mM ) that supports splicing ( Figure 1D , dark green and Figure 1—figure supplement 1; [Hoskins et al . , 2011] ) . The distribution of U1 dwell times fit a three component model . This model contained short and intermediate dwell time components ( ~0 . 15 and 1 . 3 min , respectively ) previously observed in 2 mM ATP ( Hoskins et al . , 2011 ) plus a much longer-lived component ( >11 min ) new to the 50 µM ATP condition ( Figure 1E and Figure 1—source data 1 ) . It is likely that these long-lived U1s are in stalled B complexes analogous to those previously purified from 50 µM ATP splicing reactions ( Fabrizio et al . , 2009; Tarn et al . , 1993; Chan et al . , 2003 ) . It has been previously found that the NTC does not co-purify with B complex but only with activated spliceosomes ( Tarn et al . , 1993; Chan et al . , 2003; de Almeida and O'Keefe 2015 ) . To investigate whether and for how long NTC associates with stalled B complexes , we used yeast extracts bearing green-excited SNAP tag fluorophores on NTC components ( Cef1 and Ntc90 ) to compare NTC association with pre-mRNA at 2 mM and 50 µM ATP ( Figure 1F , Table 1 ) . As previously observed ( Hoskins et al . , 2011 ) , spots of NTC fluorescence accumulated on a surface with tethered pre-mRNAs in the presence of 2 mM ATP . However , at 50 µM ATP we observed much less NTC accumulation ( Figure 1G ) . These results suggest that either the NTC cannot associate with B complex or it can associate only transiently , consistent with previous hypotheses ( Tarn et al . , 1993; Chan et al . , 2003 ) . Inspection of individual pre-mRNAs showed fewer bound NTC molecules in 50 µM ATP and that NTC binding events often lasted for only a few seconds ( Figure 1H , light green and Figure 1—figure supplement 2 ) . Analysis of the observed dwell times showed that at both 2 mM and 50 µM ATP , NTC binding events could be described by a distribution containing two exponential terms corresponding to short ( ~0 . 4 min ) and long ( ~5 min ) dwell time components ( Figure 1I and Figure 1—source data 1 ) . However , under 50 µM ATP the long dwell time component represented only a negligible fraction of the distribution ( amplitude of 0 . 06 vs . 0 . 46 for 50 µM and 2 mM ATP , respectively ) . Thus , the kinetically more stable interactions of NTC that are present at 2 mM ATP are largely absent at 50 µM ATP . In summary , the single-molecule U1 and NTC binding dynamics under low ATP conditions are consistent with the formation of stalled , B complex spliceosomes on surface tethered pre-mRNAs . To gain further information about the step at which spliceosome activation is blocked at 50 µM ATP , we incorporated fluorophores into U4/U6-snRNP proteins ( Prp3 and Prp4 , [Hardin et al . , 2015] ) that were previously shown to be lost during activation along with the U4 snRNA ( Fabrizio et al . , 2009 ) . We used homologous recombination to create a yeast strain harboring SNAPf tags ( Sun et al . , 2011 ) on both Prp3 and Prp4 ( Table 1 and Figure 2—source data 1 ) . This strain was viable , had a similar growth rate to the untagged parental strain ( Figure 2—figure supplement 1A ) , and yielded WCE in which both SNAPf tagged proteins could be specifically and rapidly labeled ( Figure 2—figure supplement 1B , C ) . Further , the labeled extract was active in bulk splicing reactions ( Figure 2—figure supplement 1D , E ) . We performed two-color CoSMoS experiments to detect U4 interactions with surface-tethered pre-mRNAs via the labels on Prp3 and Prp4 ( Figure 2A; Video 1 ) . At 50 µM ATP , U4 signals accumulated on the surface and reached a high density that did not decrease with time ( Figure 2B , light green ) , similar to the behavior we observed with U1 . At 2 mM ATP we saw little accumulation of U4 signals ( Figure 2B , dark green ) . As for U1 , these results suggest that at limiting ATP , U4 remains bound to pre-activation spliceosomal B complexes . However , we do not yet know if complexes formed at limiting ATP that contain U4 are the same complexes that also contain U1 . The properties of individual U4 binding events on single pre-mRNAs were consistent with B complex formation . In 50 µM ATP , some U4 particles interacted transiently while others remained associated for tens of minutes , whereas at 2 mM ATP nearly all U4 binding was transient ( Figure 2C; Figure 2—figure supplement 2 ) . Also consistent with these observations , at 2 mM ATP U4 exhibited two comparatively short lifetime components of ~0 . 2 and ~2 min ( Figure 2D; Figure 2—source data 2 ) . In contrast , at 50 µM ATP the measured dwell times fit to a tri-exponential distribution with two similar short components ( ~0 . 1 min and ~1 min ) , but with a substantial fraction in an additional much longer component ( >13 min ) . We hypothesize that this long component arises from the formation of stalled B complex spliceosomes at 50 µM ATP . Absence of this component at 2 mM ATP is consistent with the idea that binding of U4/U6 . U5 to form B complex is followed quickly by activation and concomitant U4 dissociation . 10 . 7554/eLife . 14166 . 008Figure 2 . Two-color CoSMoS observation of U4 dynamics at 2 mM ( dark green ) and 50 µM ( light green ) ATP . ( A ) Schematic of an experiment for observing U4 binding dynamics with a design similar to that in Figures 1B and 1F . ( B ) Time record of the number U4 fluorescence spots relative to the number of surface-tethered pre-mRNA molecules at 2 mM and 50 µM ATP . ( C ) Example fluorescence intensity records showing binding events of U4 to single pre-mRNA molecules at 2 mM and 50 µM ATP . ( D ) Probability density histogram of dwell times for the U4 at 2 mM ( N = 336 events on 229 pre-mRNAs ) and 50 µM ATP ( N = 151 events on 226 pre-mRNAs ) . Lines represent fits of the distributions of dwell times to multi-exponential equations; fit parameters for U4 are given in Figure 2—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 14166 . 008 10 . 7554/eLife . 14166 . 009Figure 2—source data 1 . Oligonucleotides used for generating yeast strains . DOI: http://dx . doi . org/10 . 7554/eLife . 14166 . 009 10 . 7554/eLife . 14166 . 010Figure 2—source data 2 . Fit parameters describing the distribution of dwell times observed for the U4 subcomplex . DOI: http://dx . doi . org/10 . 7554/eLife . 14166 . 010 10 . 7554/eLife . 14166 . 011Figure 2—figure supplement 1 . Characterization of U4-labeled yeast strains and splicing extracts . ( A ) Relative doubling times ( ± SD ) of strains containing SNAP tags on U4 proteins Prp3 and/or Prp4 compared to the parental , WT strain ( yAAH1 ) . ( B ) SDS-PAGE analysis of extracts made from U4-labeled yeast . The gel was first imaged by fluorescence ( bottom ) to identify fluorescently-labeled , SNAP-tagged U4 proteins . The gel was then subsequently stained with Coommassie blue ( top ) . SNAP-labeled U4 proteins migrated at their expected molecular weights . Lane 6 contains prestained molecular weight markers ( BioRad ) . Gel images show two separated regions from the same gel with intervening lanes removed for clarity . ( C ) Labeling kinetics of SNAP- or SNAPf-tagged U4 proteins Prp3 and Prp4 with benzylguanine dyes . ( D ) Results from in vitro splicing assays showing the accumulation of the fraction products that had completed the 1st chemical step in splicing ( lariat intron-3' exon and mRNA ) over time for U4-labeled extracts compared with the parental , WT strain . ( E ) Results from in vitro splicing assays showing the accumulation of the fraction of 2nd step product ( mRNA ) over time for U4-labeled extracts compared with the parental , WT strain . yAAH1 #1 and #2 in D and E represent two different preparations of WCE from the same parental , WT strain yAAH1 . DOI: http://dx . doi . org/10 . 7554/eLife . 14166 . 01110 . 7554/eLife . 14166 . 012Figure 2—figure supplement 2 . Additional examples ( to supplement those shown in Figure 2C ) of fluorescence intensity traces showing individual U4-SNAP subcomplexes co-localizing with surface-tethered pre-mRNA in the presence of 2 mM ( A–E , dark green ) or 50 µM ( F–J , light green ) ATP . DOI: http://dx . doi . org/10 . 7554/eLife . 14166 . 01210 . 7554/eLife . 14166 . 013Video 1 . Surface accumulation of the U4 snRNP on surface-tethered pre-mRNAs ( not shown ) under both low ( 50 µM ) and high ( 2 mM ) ATP conditions . Spliceosome activation is prevented at low ATP . White spots represent fluorescent U4 subcomplexes labeled with Dy549-SNAPf tags . This movie was obtained from recordings ( 60 min ) of 1 s duration frames recorded at 3 s intervals . To reduce the movie file size , only every third frame was included from 0 to 45 min in the movie file , and the movie is played back at 20 frames/s . Each of the two recordings shows a ~20 x 20 µm field of view , and the two individual movie files were combined using Apple Keynote software . DOI: http://dx . doi . org/10 . 7554/eLife . 14166 . 013 In WCE , U4 snRNP exists in three major complexes: U4 snRNP alone , the U4/U6 di-snRNP , and the U4/U6 . U5 tri-snRNP ( Raghunathan and Guthrie 1998a ) . Although it is presumed that only the tri-snRNP binds to the U1 . U2 pre-spliceosome , this has not been directly tested . Therefore , it is possible that the short-lived dwell time components reported above might represent transient interactions of U4 alone or U4/U6 di-snRNP with pre-spliceosomes . In order to specifically identify tri-snRNP binding events , we constructed a yeast strain bearing labels on both U5 and the U4/U6 component Prp3 ( Table 1 ) . This strain grew similarly to the parental strain and its WCE exhibited high levels of in vitro splicing activity ( Figure 2—figure supplement 1 ) . We then used the WCE in a three color CoSMoS experiment with red-excited DHFR tags on U5 ( Brr2 , Snu114 ) , a green-excited SNAPf tag reporting on U4 ( Prp3 ) , and blue-excited RP51A pre-mRNA ( Figure 3A , Video 2 ) . 10 . 7554/eLife . 14166 . 014Figure 3 . Three-color CoSMoS observation of U4 and U5 binding dynamics at 2 mM and 50 µM ATP . ( A ) Schematic of a three-color experiment in which U5 was labeled with two red-excited ( Cy5 ) fluorophores , U4 was labeled with a single green-excited ( Dy549 ) fluorophore , and the surface-tethered pre-mRNA was labeled with a single blue-excited ( Alexa488 ) fluorophore . ( B and C ) Representative time records at 2 mM ( B ) or 50 µM ATP ( C ) , each showing peaks in fluorescence intensity corresponding to colocalization of U4 ( green ) and U5 ( red ) with the same individual pre-mRNA molecule . Dashed rectangles mark examples of the simultaneous appearance of U4 and U5 spots; galleries show consecutive images ( ~0 . 7 × 0 . 7 µm ) taken from those parts of the recording showing that spot appearance is simultaneous . ( D and E ) Routes for loss of either the U4 or U5 fluorescent spots at 2 mM ( D , N = 128 ) or 50 µM ATP ( E , N = 207 ) in two typical experiments collected under otherwise identical conditions . Red and green shapes represent observation of fluorescence from the corresponding Dy549 ( green-excited ) or Cy5 ( red-excited ) fluorophores on U4 or U5 , respectively; grey shapes represent the absence of fluorescence . Percentages represent the fraction of U4/U5 complexes in which fluorescence disappeared by the indicated pathway; more prevalent pathways are emphasized with thicker arrows . DOI: http://dx . doi . org/10 . 7554/eLife . 14166 . 01410 . 7554/eLife . 14166 . 015Figure 3—figure supplement 1 . Additional examples ( to supplement those shown in Figure 3B ) of fluorescence intensity traces showing individual U5-DHFR subcomplexes ( red ) co-localizing with U4-SNAP subcomplexes ( dark green ) on surface-tethered pre-mRNA in the presence of 2 mM ATP . DOI: http://dx . doi . org/10 . 7554/eLife . 14166 . 01510 . 7554/eLife . 14166 . 016Figure 3—figure supplement 2 . Additional examples ( to supplement those shown in Figure 3C ) of fluorescence intensity traces showing individual U5-DHFR subcomplexes ( red ) co-localizing with U4-SNAP subcomplexes ( light green ) on surface-tethered pre-mRNA in the presence of 50 µM ATP . DOI: http://dx . doi . org/10 . 7554/eLife . 14166 . 01610 . 7554/eLife . 14166 . 017Figure 3—figure supplement 3 . Probability density histogram of dwell times ( points ) for U4 recorded at 150 ( blue ) and 450 µW ( red ) laser powers , and global fit to a two-exponential model that includes power-proportional photobleaching . The fit ( lines , superimposed with one another ) showed no evidence for photobleaching under these conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 14166 . 01710 . 7554/eLife . 14166 . 018Video 2 . Surface accumulation of the U4 and U5 snRNPs on surface-tethered pre-mRNAs ( not shown ) under both low ( 50 µM ) and high ( 2 mM ) ATP conditions . Spliceosome activation is prevented at low ATP . White spots represent fluorescent U4 or U5 subcomplexes labeled with Dy549-SNAPf tags or Cy5-TMP tags , respectively . This movie was obtained from recordings ( 60 min ) of 1 s duration frames recorded at 3 s intervals . To reduce the movie file size , only every third frame was included from 0 to 45 min in the movie file , and the movie is played back at 20 frames/s . Each of the four recordings shows a ~20 x 20 µm field of view , and the four individual movie files were combined using Apple Keynote software . DOI: http://dx . doi . org/10 . 7554/eLife . 14166 . 018 At 2 mM ATP , three-color CoSMoS experiments showed that the largest cohort of U5 spot arrivals ( 63%; 551 of 876 total ) on pre-mRNA molecules exhibited simultaneous appearance of a U4 spot at the same location ( e . g . , Figure 3B and Figure 3—figure supplement 1 ) . Similarly 46% of U4 spots ( 467 of 1016 total ) arrived simultaneously with a U5 spot appearance . Colocalization of simultaneously arriving U4 and U5 spots sometimes persisted for tens or hundreds of seconds; these events likely reflect tri-snRNP binding . Consistent with this interpretation , simultaneous appearances of U4 and U5 at control areas of interest ( AOIs ) with no detected pre-mRNA molecules were extremely rare ( ~3 × 10-4 events AOI-1 min-1 ) . In contrast , the U4-only and U5-only binding events were typically transient , often lasting only a single frame ( <3 s ) , and were not analyzed further ( data not shown ) . In contrast to the observations at 2 mM ATP , we observed at 50 µM ATP a higher proportion of coincident U5 and U4 binding [79% of U5 spots appeared simultaneously with a U4 spot ( 223/284 ) and 85% of U4 spots appeared simultaneously with a U5 spot ( 223/262 ) ; e . g . , Figure 3C and Figure 3—figure supplement 2] . Because the same preparation of labeled U4/U5 extract was used in both 2 mM and 50 µM ATP experiments , it is probable that differences in the amount of coincident U4 and U5 binding reflect differences in tri-snRNP relative abundance under the two conditions; differences that likely originate from tri-snRNP disruption that accompanies spliceosome activation and splicing occurring at 2 mM ATP but not at 50 µM ATP . In addition , the tri-snRNP itself is destabilized by the addition of ATP , leading to an decrease in its abundance relative to free U4 and U5 snRNPs and the U4/U6 di-snRNP ( Raghunathan and Guthrie 1998a; 1998b; Huang et al . , 2014 ) . Based on the idea that the U4 and U5 coincident arrivals in the three-color CoSMoS experiments represented tri-snRNP binding events leading to B complex formation , we selected just those events for further analysis . We observed four different outcomes from these coincident binding events ( Figure 3D , E ) : ( 1 ) the U4 snRNP spot disappeared while the U5 spot remained; ( 2 ) the U5 snRNP spot disappeared while the U4 spot remained; ( 3 ) both the U4 and U5 spots disappeared simultaneously; or ( 4 ) both the U4 and U5 signals persisted on the pre-mRNA until the experiment was terminated after one hour . In the two cases in which one spot disappeared before the other , disappearance could be caused by photobleaching of the dye label , or by dissociation of the individual snRNP from B complex , leaving the other behind . However , simultaneous disappearance of both spots most likely represents U4/U6 . U5 tri-snRNP dissociation because simultaneous photobleaching is improbable . In the presence of 2 mM ATP , the predominant outcome ( 62%; 79/128 events in a given experiment ) of coincident U4 . U5 binding was U4 spot loss followed by U5 spot loss ( Figure 3D ) . This outcome is consistent with the expected U4 loss accompanying B complex activation followed by U5 loss after splicing . Loss of the U5 signal prior to loss of U4 was rarely seen ( 2% ) , and in no case did the U4 . U5 signals persist until the end of the experiment . The remainder mostly ( 36% ) followed the route in which U4 and U5 spots disappeared simultaneously . These results are consistent with the majority of tri-snRNP interactions with pre-mRNA at 2 mM ATP resulting in activation and U4 dissociation . The smaller fraction with simultaneous spot disappearance may represent release of intact tri-snRNP without spliceosome activation . Flux through these pathways changed dramatically when the ATP concentration was reduced to 50 µM ( Figure 3E ) . The U4-first loss pathway was rarely observed ( 9%; 19/207 events in a given experiment ) , consistent with activation being blocked at 50 µM ATP . Instead , U4 and U5 most frequently ( 76% ) disappeared simultaneously , with a small fraction ( 9% ) not disappearing at all but persisting until the end of the experiment . These results suggest that when activation is suppressed by reducing ATP , tri-snRNP mostly releases intact but on rare occasions participates in formation of a kinetically stable complex on the pre-mRNA . The persistent complexes may correspond to the stalled B complexes previously isolated in ensemble studies ( Fabrizio et al . , 1989; Tarn et al . , 1993 ) . Further , the observation that U4-first loss is infrequently observed at 50 µM ATP confirms that most of the more frequent U4-first loss events observed at 2 mM ATP are due to U4 snRNP dissociation , not photobleaching . We next analyzed the kinetic behavior of the complexes formed by simultaneous arrival of U4 and U5 – i . e . , the tri-snRNP binding events – at 50 µM and 2 mM ATP . First , we measured the U4/U5 dwell times for those events in which U4 and U5 left the pre-mRNA simultaneously ( Figure 4A ) . These events presumably represent release of bound tri-snRNP without activation . At both 50 µM and 2 mM ATP the dwell time distributions were multi-exponential ( Figure 4B–D ) , suggesting the presence of multiple distinct types of tri-snRNP/pre-mRNA complexes . At 50 µM ATP , the dwell time distribution fit to the sum of two exponential terms ( Figure 4D ) , whereas the distribution at 2 mM ATP required three exponential terms for a satisfactory fit ( Figure 4C ) . The two time constants at 50 µM ATP ( 52 and 780 s ) appear to most closely correspond to the two longest time constants at 2 mM ATP ( 34 and 660 s; Figure 4B; Figure 4—source data 1 ) . However , at 2 mM ATP the majority of dissociation events are not in these longer components but rather in the shortest-lived component ( 4 s ) . Thus , our observations suggest that at 2 mM ATP most non-productively bound tri-snRNPs are quickly discarded by a pathway not operative or much slower at 50 µM ATP . 10 . 7554/eLife . 14166 . 019Figure 4 . Dwell time analysis at 2 mM and 50 µM ATP of complexes containing both U4 and U5 for which the two snRNPs both arrived at and departed from the pre-mRNA simultaneously . ( A ) Routes for loss of snRNP fluorescence ( as in Figure 3D , E ) ; shading indicates a subset of events , interpreted as tri-snRNP dissociation , that are analyzed in this figure . ( B ) Fit parameters ( ± S . E . ) describing the distribution of U4/U5 dwell times at 2 mM and 50 µM ATP ( see also Figure 4—source data 1 ) . The shortest dwell time component ( here described as τ1 ) appeared absent at 50 µM ATP ( ND , Not Determined ) . ( C , D ) Probability density histogram of U4/U5 dwell times observed at 2 mM ATP ( C; N = 128 events on 471 pre-mRNAs ) or 50 µM ATP ( D; N = 164 events on 210 pre-mRNAs ) . Lines represents fits of the dwell time distributions with equations containing three ( C ) or two ( D ) exponential terms that yielded the parameters reported in ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14166 . 019 10 . 7554/eLife . 14166 . 020Figure 4—source data 1 . Fit parameters describing the distribution of dwell times observed for U4/U5 complexes that arrived and departed simultaneouslyDOI: http://dx . doi . org/10 . 7554/eLife . 14166 . 020 10 . 7554/eLife . 14166 . 021Figure 4—figure supplement 1 . ATP dependence of discard of U4 . U5 complexes assembled at 50 µM ATP . ( A ) Flowchart depicting the experimental design . During each wash step the imaging chamber ( ~20–30 µL ) was washed with 100 µL of the same buffer used in single molecule assays but without yeast WCE and with the indicated amount of ATP . ( B ) Quantification of changes in U4 or U5 fluorescent spot density in fields of view after the second wash step with 50 µM or 2 mM ATP . ( C ) Analysis of the fraction of U4 fluorescent spots that colocalize with U5 fluorescence in fields of view in WCE and after each wash step . ( D ) Analysis of the fraction of U5 fluorescent spots that colocalize with U4 fluorescence in fields of view in WCE and after each wash step . Each bar plotted in B–D represent the average of data obtained from 5–10 different fields of view ±SD . DOI: http://dx . doi . org/10 . 7554/eLife . 14166 . 021 To probe the ATP-dependence of this potential discard pathway further , we assembled stalled complexes on surface-tethered pre-mRNAs at 50 µM ATP . We then exchanged the WCE containing 50 µM ATP with splicing buffer , also containing 50 µM ATP , to remove any unstable tri-snRNP-containing complexes ( Figure 4—figure supplement 1A ) . During this first buffer wash , the majority of U4 . U5 complexes remained bound to the slide ( 10–17% decrease in U4 . U5 spot density , N = 253 ± 18 U5 spots per field of view in WCE prior to the wash ) . We next carried out a second buffer exchange and flowed in buffer containing either 50 µM or 2 mM ATP . When a second 50 µM ATP wash was carried out , the majority of U4/U5 complexes remained bound ( ~19% decrease in spot density between the first and second wash steps , Figure 4—figure supplement 1B ) . In contrast , the majority of U4 and U5 spots were released when a 2 mM ATP wash was used: 74% and 68% decreases in U4 and U5 spot density , respectively , and only a small fraction of remaining U4 and U5 particles colocalized with one another ( Figure 4—figure supplement 1B–D ) . Release of U4 and/or U5 occurred very rapidly—within the deadtime of our experiment ( ~30 s ) . Loss of signals from both U4 and U5 is most consistent with ATP-dependent discard of these complexes since U5 should be retained in Bact spliceosomes ( Figure 1A ) . Furthermore , these results suggest that the ATPase responsible for the discard of U4 and U5 is stably associated with the pre-mRNA and remains bound when the WCE is removed and replaced by buffer . Our data indicate that the majority ( ~68% ) of stable U4 . U5-containing complexes assembled at 50 µM ATP in WCE will be discarded in the presence of 2 mM ATP in buffer and only a smaller fraction ( at most 32% ) can potentially form Bact spliceosomes . We next analyzed the pathway that is dominant at 2 mM ATP: events in which U4 and U5 fluorescence appeared simultaneously but the U4 fluorescence was lost prior to U5 loss ( Figure 5A ) . This cohort is expected to include the pre-mRNA molecules with productively bound tri-snRNPs that progress to spliceosomal activation and splicing . The U4 dwell time distribution for this activation cohort ( tU4release−tU4 . U5arrival; Figure 5B ) is distinct from that in the non-productive cohort that release tri-snRNP ( Figure 4 ) ; this demonstrates that the U4- and U5-containing complexes in the two cohorts must have different functional and structural properties even prior to dissociation . In the activation cohort , only a few U4 molecules left immediately after the tri-snRNP bound; indeed , the distribution of tU4release−tU4 . U5arrival values shows a dearth of short events ( Figure 5B and Figure 5—source data 1 ) . The distribution fit to a convolution of two exponentials , suggesting that U4 dissociation that accompanies activation is preceded by at least two sequential steps that are each partially rate limiting ( Colquhoun and Hawkes 1995 ) . This is not surprising , given that spliceosome activation encompasses multiple conformational transitions and points of regulation ( Brow 2002; Small et al . , 2006; Brenner and Guthrie 2006; Bellare et al . , 2008 ) . We also measured how long it took for each U5 spot to disappear after U4 disappearance ( tU5release−tU4release ) ; these U5-containing molecules include both on-pathway complexes that acquire NTC and subsequently splice , as well as those that are discarded prior to splicing completion . Consistent with the expected presence of multiple U5-containing species , the dwells fit to the sum of two exponentials ( Figure 5C and Figure 5—source data 1 ) . 10 . 7554/eLife . 14166 . 022Figure 5 . Dwell time analysis at 2 mM ATP of complexes containing both U4 and U5 for which the two snRNPs both arrived at the pre-mRNA simultaneously but in which U4 fluorescent spot disappearance preceded U5 spot disappearance . ( A ) Routes for loss of snRNP fluorescence ( as in Figure 3D , E ) ; shading indicates a subset of events , interpreted as spliceosome activation , that are analyzed in this figure . ( B ) Probability density histogram of dwell times for U4 molecules in the analyzed subset ( tU4release−tU4 . U5arrival; N = 244 events on 471 pre-mRNAs ) . The line represents a fit of the dwell time distribution to a convolution function . ( C ) Probability density histogram of dwell times for U5 molecules in the departure pathway subsequent to U4 departure ( tU5release−tU4release; N = 207 events on 471 pre-mRNAs ) . The line represents a fit of the dwell time distribution to an equation containing two exponential terms . ( D ) The dwell time of each U5 ( tU5release−tU4release ) plotted against the dwell time of U4 ( tU4release−tU4 . U5arrival ) in the same complex . Fit parameters for ( C ) and ( D ) are given in Figure 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 14166 . 022 10 . 7554/eLife . 14166 . 023Figure 5—source data 1 . Fit parameters describing the distribution of dwell times observed for U4 or U5 after simultaneous U4/U5 Arrival and U4 loss preceding U5 loss . DOI: http://dx . doi . org/10 . 7554/eLife . 14166 . 023 RNAs can adopt multiple secondary structures , and it is possible that after tri-snRNP binding there exist multiple spliceosome configurations that differ in their behavior after activation . We tested this idea by asking whether ( tU4release−tU4 . U5arrival; a reporter of activation ) was correlated with the following U5 dwell time ( tU5release−tU4release; which reflects the speed of the subsequent processes ) . Because there was no clear correlation between the rate of U4 loss and the subsequent U5 dwell time ( Figure 5D ) , there was no evidence for multiple configurations with differing intrinsic kinetics that persist through activation . Even under splicing conditions at 2 mM ATP , we often observed multiple sequential tri-snRNP binding and dissociation events on the same pre-mRNA molecule ( ~22% of pre-mRNAs showed multiple tri-snRNP binding events , Figure 6—figure supplement 1A ) . The fraction of pre-mRNAs that only showed one tri-snRNP binding event ( ~22% ) is much lower than the fraction of RP51A substrates that typically splice in our assays ( ~60% , Figure 2—figure supplement 1 ) , although splicing on slides may be lower . This suggests that at least some pre-mRNAs that do splice also bind the tri-snRNP more than once . When multiple tri-snRNP associations were observed they were nearly always sequential: a second tri-snRNP binding was very rarely seen unless both U4 and U5 fluorescence from the previous association had already disappeared ( simultaneous colocalization of two tri-snRNPs was potentially observed on only 6 out of 481 RNAs , <1% ) . Some tri-snRNP binding events on a pre-mRNA molecule resulted in simultaneous appearance of both the U4 and U5 signals followed by their simultaneous loss , usually a short time later ( Figure 6 , yellow highlights ) . Other tri-snRNP binding events on the same pre-mRNA molecule would show simultaneous appearance of both U4 and U5 followed by loss of U4 fluorescence prior to loss of U5 ( i . e . , spliceosome activation; Figure 6 , red highlights ) . Thus , a transcript is able to recruit the tri-snRNP multiple times during spliceosome assembly , but not every recruitment to a given pre-mRNA molecule leads to activation . 10 . 7554/eLife . 14166 . 024Figure 6 . Examples of pre-mRNA molecules in which different outcomes result from individual tri-snRNP binding events at 2 mM ATP . Events are color-coded according to the pathways shown in the Key . The same pre-mRNA molecule may interact with the tri-snRNP by releasing U4 and U5 simultaneously or proceeding through activation ( A-C ) . Some pre-mRNA molecules exhibited multiple rounds of activation ( B and C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14166 . 02410 . 7554/eLife . 14166 . 025Figure 6—figure supplement 1 . Analysis of multiple tri-snRNP binding events at 2mM ATP . ( A ) Frequency of tri-snRNP binding events observed on single pre-mRNAs ( N= 471 pre-mRNAs ) . ( B ) Probability density histogram of arrival times for the first U4 . U5 complex ( tU4 . U5arrival ) to bind each pre-mRNA . The line represents a fit of the arrival time distribution to a convolution function with fitted parameters of τ1 = 2 , 105±886 s and τ2 = 417±260 s ( N = 197 events ) . ( C ) Probability density histogram of arrival times for U4 . U5 complexes following events in which U4 . U5 bound simultaneously to the pre-mRNA and U4 release preceded U5 release ( tU4 . U5arrival−tU5 , activationrelease ) . The line represents a fit of the arrival time distribution to an equation with a single fitted parameter of τ = 617±66 s ( N = 87 events ) . ( D ) Probability density histogram of arrival times for U4 . U5 complexes following events in which U4 . U5 bound and released simultaneously from the pre-mRNA ( tU4 . U5arrival−tU4 . U5 , discardrelease ) . The line represents a fit of the arrival time distribution to an equation with a single fitted parameter of τ = 492±52 s ( N = 90 events ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14166 . 025 If loss of U4 snRNP from B complex commits the pre-mRNA to splicing , we would expect to see loss of U4 without simultaneous loss of U5 to occur only once on each pre-mRNA . Instead , we observed that approximately 10% ( 49 of 481 ) of pre-mRNAs under 2 mM ATP exhibited at least two repeats of U4/U5 simultaneous binding , followed by loss of the U4 , followed by loss of U5 ( Figure 6B , C ) . Since binding of the tri-snRNP is dependent on presence of an intron ( Shcherbakova et al . , 2013; Hoskins et al . , 2011 ) , we conclude that on these pre-mRNAs the first U4 release event did not result in productive splicing and that subsequent observations of U4 release result from spliceosome re-assembly on the same pre-mRNA . These data indicate that activation does not always commit the pre-mRNA to splicing , and are consistent with the presence of discard pathways that can prevent splicing subsequent to activation ( Mayas et al . , 2006; Koodathingal et al . , 2010 ) . Since tri-snRNP association is reversible ( Figure 4 and [Hoskins et al . , 2011] ) , it is not surprising that we observed tri-snRNP to re-associate multiple times with the same pre-mRNA . In contrast , re-association of U4 complexes containing the labeled protein Prp3 after prior U4 departure was exceedingly rare . Of 260 cases in which both U4 and U5 fluorescence appeared simultaneously followed by disappearance of U4 and then U5 , we observed only 15 instances ( 5 . 7% ) of reappearance of U4 while U5 remained visible . It is likely that some of these rare events arose from background binding of fluorescent molecules to the slide surface since control experiments revealed that the rate of background U4 binding is similar to the observed rate of U4 re-binding subsequent to U5 loss ( 2 . 3 × 10-3 vs . 9 . 8 × 10-3 events/min , respectively ) . In addition , only 5 cases were found in which the rebound U4 spot was stable enough to last more than a single frame . Taken together , these results are consistent with activation being largely irreversible , with only a slim likelihood ( <2% ) for rebinding of Prp3-containing U4 or U4/U6 snRNPs . We wondered if pre-mRNAs that bound the tri-snRNP more than once also recruited tri-snRNPs by different kinetic pathways . That is , do the kinetics of initial tri-snRNP binding to a pre-mRNA differ from the kinetics of tri-snRNP association after an activation attempt or discard ? For each pre-mRNA , we measured the arrival time of the first tri-snRNP binding event ( tU4 . U5arrival ) and plotted the distribution of arrival times in a histogram ( Figure 6—figure supplement 1B ) . The histogram shows few events at small values of tU4 . U5arrival and could be described by a function that is a convolution of two exponential terms . This is consistent with multiple partially rate-limiting steps being required prior to initial tri-snRNP binding to the pre-mRNA ( e . g . , U1 and U2 binding ( Hoskins 2011 ) . Interestingly , the distributions of tri-snRNP arrival times following U4 . U5 discard or U4 release events ( tU4 . U5arrival−tU4 . U5 , discardrelease or tU4 . U5arrival−tU5 , activationrelease ) both fit to distributions described by an equation with single exponential term and with similar fitted parameters ( Figure 6—figure supplement 1C , D ) . This suggests that rebinding of the tri-snRNP to pre-mRNA following discard of tri-snRNPs or activated spliceosomes involves fewer steps than initial tri-snRNP recruitment , possibly because they are rebinding to an already partially assembled complex . Thus , splicing efficiency may potentially be maintained by not only multiple tri-snRNP binding events occurring on the pre-mRNA but also by facilitating subsequent recruitment of the tri-snRNP to those pre-mRNAs . The NTC is required for spliceosomal catalysis , but the timing of NTC association relative to other steps in activation is unresolved ( Hogg et al . , 2010 ) . In order to address this issue , we carried out 3-color CoSMoS experiments in which the U4 snRNP was labeled with a green-excited fluorophore on Prp3 and red-excited fluorophores were attached to NTC components ( Cef1 and Ntc90 ) with the DHFR tag ( Figure 7A ) . This triply-tagged strain showed near-parental-strain growth kinetics and splicing activity ( Figure 2—figure supplement 1 ) . In 2 mM ATP , we observed transient U4 and longer-lasting NTC associations with pre-mRNA molecules , consistent with the behaviors seen in extracts in which U4 and NTC were individually labeled ( Figures 1 and 2 ) . We then measured for each pre-mRNA molecule the sequence of U4 and NTC appearance and disappearance ( Figure 7B , C and Figure 7—figure supplement 1 ) . In the majority of cases ( 79% ) , NTC signals appeared soon ( typically ~30 s ) after loss of U4 ( Figure 7D ) . A randomized control ( Figure 7D , red lines ) indicates that this distribution did not arise by chance . In addition , the data show evidence for a small subpopulation ( on order 10% of complexes ) where NTC bound first but U4 release followed almost immediately ( within a few seconds ) . Since ~53% of all molecules that recruit NTC subsequently are spliced ( Hoskins 2011 ) , the data suggest that the main pathway leading to functional spliceosomes is NTC association subsequent to U4 loss . 10 . 7554/eLife . 14166 . 026Figure 7 . Three-color CoSMoS observation of U4 and NTC binding dynamics at 2 mM ATP . ( A ) Schematic of the experiment , in which the NTC was labeled with two red-excited ( Cy5 ) fluorophores , U4 was labeled with a single green-excited ( Dy549 ) fluorophore , and the surface-tethered pre-mRNA contained a single blue-excited ( Alexa488 ) fluorophore . ( B and C ) Representative traces showing peaks in fluorescence intensity corresponding to colocalization of U4 ( dark green ) and NTC ( red ) with single pre-mRNA molecules . Raw images ( ~0 . 9×0 . 9 µm ) corresponding to portions of the trace segments enclosed by the dashed boxes are included above the traces . The times of U4 release ( tU4release ) and NTC arrival ( tNTCarrival ) are determined from these data as shown . Subtraction of tNTCarrival from tU4releasetime yields a positive number if the NTC arrived prior to U4 release or a negative number if the NTC arrived after U4 release . ( D ) Probability density histogram showing the delay between NTC arrival and U4 loss ( gray ) . The first and last bins ( limits of -3600 – -200 and 200 – 3600 s , respectively ) were truncated in the figure for clarity . Most often ( 79% of N = 293 total events on 402 pre-mRNAs ) , the NTC arrived soon after loss of the U4 signal ( tU4release−tNTCarrival<0 ) . A randomized control histogram ( see Materials and methods ) is also shown ( red ) . Comparison of gray and red curves suggests that U4 departed after NTC arrival in <19% of complexes . DOI: http://dx . doi . org/10 . 7554/eLife . 14166 . 02610 . 7554/eLife . 14166 . 027Figure 7—figure supplement 1 . ( A-E ) Additional examples ( to supplement those shown in Figure 7 ) of fluorescence intensity traces showing individual NTC-DHFR subcomplexes ( red ) co-localizing with U4-SNAP subcomplexes ( dark green ) on surface-tethered pre-mRNA in the presence of 2 mM ATP . DOI: http://dx . doi . org/10 . 7554/eLife . 14166 . 027 Inhibiting spliceosomal activation by limiting ATP has been used in multiple laboratories to accumulate and purify B complex spliceosomes for analysis of their function and composition ( Chan et al . , 2003; Tarn et al . , 1993; Fabrizio et al . , 2009 ) . It has often been assumed that the tri-snRNP binds to but cannot release from pre-mRNAs under these conditions . Our data show that this is not the case: the majority of tri-snRNP encounters with pre-mRNA molecules involved simultaneous U4 and U5 association and dissociation of both after just tens of seconds . While we did not examine U1 and U2 directly in these experiments , it is likely that these pre-mRNAs were assembled into pre-spliceosomes since single molecules of the tri-snRNP fail to accumulate on pre-mRNAs without ATP ( Hoskins et al . , 2011 ) and tri-snRNP association almost always requires prior U1 and U2 binding ( Shcherbakova et al . , 2013 ) . Only a small fraction of events resulted in tri-snRNP retention at the conclusion of our experiments . Based on our data , it is unclear what distinguishes complexes that released the tri-snRNP from those in which it was retained for tens of minutes . It is likely that the longest-lived complexes are the assemblies that have been purified and characterized as the spliceosomal B complex ( Fabrizio et al . , 2009; Tarn et al . , 1993 ) . Our experiments ( Figure 4—figure supplement 1 ) suggest that the majority of spliceosome complexes formed at 50 µM ATP and that survive removal of the WCE are not competent for Bact formation and are instead rapidly discarded by an ATP-dependent mechanism . However , neither our work nor previous studies have directly established what fraction of the remaining long-lived complexes are kinetically competent intermediates on the splicing pathway at physiological ATP . Under limiting ATP conditions ( Figure 8 ) , the tri-snRNP associates with the pre-spliceosome but activation and release of U4 rarely occurs . Unexpectedly , at physiological ATP conditions that permit activation and splicing ( Figure 8 ) , tri-snRNP release is much faster than at 50 µM ATP . The reason that splicing can nevertheless occur is because at high ATP U4 release is rapid , irreversible , and is thus able to compete with tri-snRNP dissociation even though that dissociation rate has also increased . Based on the observed fluxes through the U4 release and tri-snRNP dissociation pathways , a significant fraction ( one of every three ) tri-snRNP binding events ends in tri-snRNP dissociation without spliceosome activation . While many steps in splicing have been shown to be readily reversible including snRNP and NTC association ( Hoskins et al . , 2011 ) , pre-mRNA conformational changes ( Abelson et al . , 2010; Krishnan et al . , 2013 ) , and the chemical steps ( Tseng and Cheng 2008 ) , in our experiments the activation and loss of U4 appears to be almost irreversible . In >98% of events in which U4 was ejected after tri-snRNP binding , Prp3-containing U4 snRNPs did not reappear while U5 remained bound . This indicates effective irreversibility of U4 loss . Since U4 loss permits U6 to make both inter- and intramolecular basepairing interactions to form the spliceosome catalytic core and these are mutually exclusive with U4 basepairing , our results suggest that U4 is quickly recycled in the extract ( Raghunathan and Guthrie 1998a ) and/or that the conformational rearrangements within U6 occur rapidly , preventing re-base pairing with U4 directly or with the aid of the U4/U6 annealing protein Prp24 . Despite irreversible loss of U4 during the B to Bact transition , spliceosomal activation does not guarantee that the pre-mRNA will be spliced by that particular spliceosome . We observed multiple instances in which a pre-mRNA underwent several rounds of tri-snRNP binding , U4 release , then U5 release and tri-snRNP rebinding ( Figure 6 ) . Thus , release of U4 during activation does not always commit a particular spliceosome to catalysis and mRNA release . This result is consistent with the presence of proofreading and discard steps carried out by Prp16 and Prp22 occurring after spliceosomal U4 release , and with disassembly of discarded spliceosomes by Prp43 ( Burgess and Guthrie 1993; Koodathingal et al . , 2010; Mayas et al . , 2006 ) . Our data support a model in which the spliceosome cannot reverse its steps back to the B complex if discard occurs after U4 release . Instead , U5 ( and likely U6 ) are released and a new tri-snRNP must be recruited to the pre-mRNA for a new activation attempt . Taken together our data suggest that tri-snRNP rebinding compensates for discard at the B complex or subsequent stages , contributing to the overall efficiency of splicing by permitting multiple attempts at assembly and activation . The coordination of U4 release with other events during activation has not been well-studied . In particular , the relative timing of U4 release and NTC recruitment has remained ambiguous ( de Almeida and O'Keefe 2015; Hogg et al . , 2010 ) . Early immunoprecipitation experiments suggested that the NTC joins the spliceosome soon after or concomitant with U4 loss ( Chan et al . , 2003; Tarn et al . , 1993 ) . In contrast , later mass spectrometry analyses of purified spliceosomes stalled by limiting ATP detected small amounts of NTC in B complex , suggesting that stable U4 and NTC binding are not mutually exclusive ( Fabrizio et al . , 2009 ) . Our data explain this apparent contradiction by showing that at 2 mM ATP NTC recruitment soon after U4 release is the predominant pathway ( 79% of events ) . However , under limiting ATP ( where there is little U4 release ) transient interactions of NTC were still observed . More stable binding of NTC after U4 release is in agreement with both previous crosslinking results ( Rasche et al . , 2012 ) and more recent spliceosome structures ( Yan et al . , 2015 ) that show key binding sites for the NTC include the U6 internal stem loop ( ISL ) and the catalytic core . Because the ISL is not present when U6 is basepaired to U4 , active site formation may be the major means by which the spliceosome recruits and retains the NTC . In contrast with the predominant pathway , we less frequently ( 19% of events ) observed NTC arrival prior to U4 loss . Recently it has been shown that functional tri-snRNPs can form in the absence of stable U4/U6 basepairing , suggesting that protein/RNA interactions can play a major role in tethering U4 to U5 and U6 ( Burke et al . , 2015 ) . Thus , it is possible that the minor pathway we observe represents spliceosomes that have disrupted U4 . U6 basepairing to facilitate NTC recruitment but in which other interactions have remained intact that delay U4 release . Finally , at 50 µM ATP NTC failed to accumulate on the pre-mRNA molecules and instead displayed mostly transient interactions ( Figure 1 ) . This indicates that the NTC cannot stably bind spliceosomes stalled at B complex . The source of the small number of NTC peptides detected in mass spectrometry analyses of stalled B complex ( Fabrizio et al . , 2009 ) may reflect binding of the NTC to a small number of spliceosomes that may have been able to release U4 at 50 µM ATP ( Figure 3E ) . Our single molecule data are consistent with data from human spliceosomes showing that NTC recruitment is also inhibited in complexes that fail to form Bact when exchange of U1 for U6 pairing at the 5' SS is prevented ( Chiou et al . , 2013 ) . Here , direct observation of binding and release of spliceosomal subcomplexes has allowed us to define the predominant pathways leading to spliceosomal activation ( Figure 8 ) . The results reveal that ATP stimulates both U4 release and tri-snRNP discard . ATP-dependent U4 release is known to be mediated by the Brr2 helicase ( Raghunathan and Guthrie 1998b ) . The rapid tri-snRNP discard pathway is consistent with ATP-dependent proofreading by Prp28 during exchange of U1 for U6 basepairing at the 5' SS ( Yang et al . , 2013 ) . If exchange is successful the spliceosome can proceed through the next steps of activation . If unsuccessful , Prp28 may induce rapid tri-snRNP release . It is possible that the discard pathway observed here may be related to prior observations of yeast pre-spliceosomes containing weakly bound tri-snRNPs upon inhibition of U1/U6 exchange ( Staley and Guthrie 1999 ) and the labile tri-snRNPs present in human exon-defined splicing complexes ( Boesler et al . , 2015 ) . The existence of this ATP-dependent tri-snRNP release pathway may serve to promote efficient splicing by preventing cellular accumulation of non-functional assemblies of snRNPs on pre-mRNAs . By labeling protein components associated with U4 release , we have now demonstrated that all of the spliceosomal snRNPs and NTC are amenable to CoSMoS assays ( Rodgers et al . , 2015; Hoskins et al . , 2011 ) . Thus , there now exists a powerful toolset for studying compositional changes and dynamics of single spliceosomes in real time . Further mechanistic insights are likely to result from combining these tools with mutations in key spliceosomal components to better dissect the roles of individual proteins and snRNAs in the steps of spliceosomal assembly , activation , proofreading , chemistry , and recycling . Even in the absence of further studies , the data presented here define the sequence of major molecular events in spliceosome activation and explain how splicing can be efficient despite high fluxes through discard pathways after tri-snRNP binding and after activation . Capped RP51A pre-mRNA for in vitro splicing assays were transcribed in the presence of [α-32P]UTP and purified as previously described ( Crawford et al . , 2008 ) . Dye-labeled RP51A pre-mRNAs used in single molecule experiments were prepared by splinted ligation of a trace [32P]-labeled , capped RP51A transcript to a biotinylated 2'-O-methyl oligonoucleotide derivatized with a single Alexa Fluor 488 ( Alexa488 , ThermoFisher Scientific; Waltham , MA ) or Alexa Fluor 647 ( Alexa647 , ThermoFisher Scientific; Waltham , MA ) fluorophore as previously described ( Crawford et al . , 2008 ) . Yeast strains containing fast SNAP ( SNAPf ) ( Sun et al . , 2011 ) tags on U4 snRNP proteins ( Table 1 ) were prepared by homologous recombination as previously described ( Hoskins et al . , 2011; Shcherbakova et al . , 2013 ) . In brief , PCR products containing homology to the 3’ region of the PRP3 or PRP4 genes and downstream genomic DNA , a ( Gly-Ser-Gly ) 2 linker , the SNAPf tag , and a yeast selectable marker ( hygromycin/HygR or nourseothricin/NatR ) were generated using primers listed in Figure 2—source data 1 and the plasmids pAAH0034 ( containing the SNAPf gene and the NatR marker , [Shcherbakova et al . , 2013] ) or pAAH0013 ( containing the SNAPf gene and the HygR marker ) . The pAAH0013 plasmid was prepared by cloning the SNAPf gene into the HindIII and BamHI sites of pAG32 ( Euroscarf ) . The purified PCR products were then transformed into yeast using the lithium acetate method and colonies selected for growth in the presence of either hygromycin ( 300 µg/mL ) or nourseothricin ( 100 µg/mL ) . SNAPf tags were confirmed by PCR using primers listed in Figure 2—source data 1 as well as by labeling of the proteins with fluorophores ( see below ) . Yeast growth rates were determined using a protocol adapted from Biotek ( Held 2010 ) . Experiments were carried out in a Tecan multimode plate reader by dilution of overnight cell cultures into flat-bottomed , clear , 96-well plates ( 200 µL , typically 8 replicates ) and covering with optically clear TopSeal-A plate sealers . Shaking speed was set to slow with 1-mm amplitude , and plates were incubated at 30°C . Cell growth was monitored at 600 nm by recording measurements at ~ 2 min intervals . Growth rates were calculated as previously described ( Hoskins et al . , 2011 ) and compared to the parental control strain ( yAAH0001 ) . Yeast whole cell extract ( WCE ) was prepared as previously described ( Ansari and Schwer 1995 ) with the following modifications . The concentration of DTT was increased to 1 mM at all steps . Lysates were aliquoted ( 1 . 25 mL ) , frozen in liquid N2 , and stored at -80°C after the second high speed spin . The lysate buffer was exchanged into 50 mM HEPES/KOH pH 7 . 9 , 50 mM KCl , 10% ( v/v ) glycerol , and 1 mM DTT by gel filtration using a previously published protocol ( Anderson and Hoskins 2014 ) , aliquoted ( 42 µL ) , frozen in liquid N2 , and stored at -80°C . In cases where SNAP-tagged proteins were labeled , the lysate was incubated for 30 min at room temperature with the fluorophore ( e . g . , benzylguanine-Dy549/SNAP-Surface 549 , New England Biolabs; Ipswich , MA ) prior to gel filtration . A fluorophore concentration of 2 . 1 µM was used to label SNAP tags , and fast SNAP ( SNAPf ) tags were labeled using 1 . 1 µM fluorophore . SNAP-tagged proteins derivatized with fluorophores were visualized by denaturing polyacrylamide gel electrophoresis ( SDS-PAGE ) followed by imaging fluorescence on a Typhoon biomolecular imager ( GE Life Sciences; Pittsburgh , PA ) . Data were analyzed using ImageQuant software . The extents and rates of fluorophore incorporation into SNAP and SNAPf tags were determined using previously described protocols ( Hoskins et al . , 2011 ) . Splicing assays were carried out as previously described ( Crawford et al . , 2008 ) using 40% WCE and ~0 . 2 nM RP51A substrate . [32P]-labeled RNAs were visualized by denaturing PAGE followed by phosphorimaging . Data were analyzed using ImageQuant software ( GE Lifesciences; Pittsburgh , PA ) . Single molecule imaging chambers and buffers including oxygen scavengers and triplet quenchers were prepared as previously described ( Hoskins et al . , 2011; Crawford et al . , 2008; Anderson and Hoskins 2014 ) . Single-molecule data were collected on a homebuilt , micromirror TIRF microsope at room temperature as previously described ( Friedman et al . , 2006; Hoskins et al . , 2011 ) . To facilitate correction of stage drift , chambers also included fluorescent beads ( TransFluoSpheres , streptavidin-labeled , 40 nm , 488/645 ) to act as fiducial markers . Laser powers were typically set at either 150 or 450 µW for both the 532 and 633 nm lasers . The distributions of dwell times from both laser powers in experiments at 2 mM ATP were compared to assess the impact of photobleaching and showed little difference in the fitted parameters ( Figure 3—figure supplement 3 ) . In three-color CoSMoS experiments , the 488 nm laser was set to 800 or 1750 µW and RNAs containing 488-excitable fluorophores were only imaged at the beginning and end of the experiment to avoid photobleaching of other fluorophores . In three-color CoSMoS experiments , data were collected with a frame duration of 1 s , a spacing between frames of 3 s , and with simultaneous illumination with both the 532 and 633 nm lasers . The same frame duration and spacing was used to collect data for two-color CoSMoS experiments following U4 binding dynamics except that only the 532 nm laser was used . Two-color CoSMoS experiments of U1 and NTC binding dynamics under low ATP used a frame duration of 1 s , a spacing between frames of 5 s , and illumination with the 532 nm laser . In all experiments , autofocusing was carried out once per minute with a 785 nm laser ( Hoskins et al . , 2011 ) . A reference data set used for mapping the <635 nm and >635 nm fluorescence emission fields of view onto one another was acquired with each experiment ( Hoskins et al . , 2011 ) . Data was analyzed essentially as described ( Hoskins et al . , 2011 ) by mapping the fields of view onto one another , correcting for stage drift , determining locations of immobilized pre-mRNAs ( areas of interest , AOIs ) , and integrating pixel intensity at each AOI using custom Matlab software ( The Mathworks; Natick , MA; https://github . com/gelles-brandeis ) . Peaks in intensity were identified by changes occurring more than 3 . 2σ over the baseline noise , and signals ended when they fell below 1σ . Each peak was manually inspected to confirm the presence of a spot of fluorescence centered in the AOI . The distributions of observed dwell times for each subcomplex were displayed by constructing probability density plots in which the dwell times were binned and each bin divided by the product of the bin width and total number of events . Error bars for each bin were calculated as the error of a binomial distribution as previously described ( Hoskins et al . , 2011 ) . Distributions described by one or two exponential terms were fit by maximum likelihood methods to exponential probability density functions as previously described for single and double exponential distributions ( Equations 1 or 2 , respectively; Hoskins et al . , 2011 ) . Distributions described by three exponential terms or by a convolution function ( Lu et al . , 1998 ) were fit to Equations 3 or 4 , respectively . In all equations , tm represents the minimum detectable dwell time; tmax represents the duration of the experiment ( 60 min ) ; A1 and A2 the fitted amplitudes; and τ1 , τ2 , τ3 represent the fitted parameters . Errors in the fit parameter were determined by bootstrapping 1000 random samples of the data and determining the standard deviation of the resultant values . ( 1 ) [ ( A1⋅ ( e−tmτ1−e−tmaxτ1 ) ) ]−1⋅[ A1τ1e−tτ1 ] ( 2 ) [ ( A1⋅ ( e−tmτ1−e−tmaxτ1 ) ) + ( ( 1−A1 ) ⋅ ( e−tmτ2−e−tmaxτ2 ) ) ]−1⋅[A1τ1e−tτ1+1−A1τ2e−tτ2] ( 3 ) [ ( A1⋅ ( e−tmτ1−e−tmaxτ1 ) ) + ( A2⋅ ( e−tmτ2−e−tmaxτ2 ) ) + ( ( 1−A1−A2 ) ⋅ ( e−tmτ3−e−tmaxτ3 ) ) ]−1⋅[A1τ1e−tτ1+A2τ2e−tτ2+1−A1−A2τ3e−tτ3] ( 4 ) [ ( 1/τ1⋅1/τ2 ) ⋅ ( e−t/τ1−e−t/τ2 ) ]⋅[ ( 1/τ2⋅ ( e−tm/τ1−e−tmax/τ1 ) ) − ( 1/τ1⋅ ( e−tm/τ2−e−tmax/τ2 ) ) ]−1	The genes in an organism’s DNA may be expressed to form a protein via an intermediate molecule called RNA . In many organisms including humans , gene expression often begins by making a precursor molecule called a pre-mRNA . The pre-mRNA contains regions called exons that code for the protein product and regions called introns that do not . A machine in the cell called the spliceosome has the job of removing the introns in the pre-mRNA and stitching the exons together by a process known as splicing . The spliceosome is made up of dozens of components that assemble on the pre-mRNAs . Before a newly assembled spliceosome can carry out splicing , it must be activated . The activation process involves several steps that are powered by the cell's universal power source ( a molecule called ATP ) , including the release of many components from the spliceosome . Many of the details of the activation process are unclear . Spliceosomes in the yeast species Saccharomyces cerevisiae are similar to spliceosomes from humans , and so are often studied experimentally . Hoskins et al . have now used a technique called colocalization single molecule fluorescence spectroscopy to follow , in real time , a single yeast spliceosome molecule as it activates . This technique uses a specialized microscope and a number of colored lasers to detect different spliceosome proteins at the same time . Hoskins et al . found that one of the steps during activation is irreversible – once that step occurs , the spliceosome must either perform the next activation steps or start the processes of assembly and activation over again . Hoskins et al . also discovered that ATP causes some spliceosomes to be discarded during activation and not used for splicing . This indicates that before spliceosomes are allowed to activate , they may undergo 'quality control' , which may be important for making sure that gene expression occurs efficiently and correctly . Future studies will investigate how this quality control process works in further detail .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2016	Single molecule analysis reveals reversible and irreversible steps during spliceosome activation
Quantifying the relative impact of environmental conditions and host community structure on disease is one of the greatest challenges of the 21st century , as both climate and biodiversity are changing at unprecedented rates . Both increasing temperature and shifting host communities toward more fast-paced life-history strategies are predicted to increase disease , yet their independent and interactive effects on disease in natural communities remain unknown . Here , we address this challenge by surveying foliar disease symptoms in 220 , 0 . 5 m-diameter herbaceous plant communities along a 1100-m elevational gradient . We find that increasing temperature associated with lower elevation can increase disease by ( 1 ) relaxing constraints on parasite growth and reproduction , ( 2 ) determining which host species are present in a given location , and ( 3 ) strengthening the positive effect of host community pace-of-life on disease . These results provide the first field evidence , under natural conditions , that environmental gradients can alter how host community structure affects disease . Infectious disease is strongly influenced by host community structure and abiotic conditions ( Halliday et al . , 2020a; Halliday et al . , 2019 ) , both of which are undergoing unprecedented change as the climate is warming ( Pachauri et al . , 2014 ) and biodiversity is being reshuffled ( Díaz et al . , 2019; Hillebrand et al . , 2018 ) . Understanding how biotic and abiotic conditions interact to drive the emergence and spread of infectious disease is quickly emerging as one of the greatest research challenges of the 21st century and will be the key to limiting the impacts of infectious diseases on food production systems , wildlife , and humans . Disease ecology provides a framework for achieving this goal through careful examination of interactions among hosts , parasites , and the environment ( Johnson et al . , 2015a; McNew , 1960; Seabloom et al . , 2015; Figure 1a ) . Yet , we have a poor understanding of how this framework operates under natural conditions , in part because several mechanisms can operate simultaneously , making it difficult to tease apart their relative contributions to realized disease risk . Climate change involves increased environmental temperatures , which can profoundly alter disease risk ( Garrett et al . , 2006; Harvell et al . , 2002; Rohr et al . , 2011 ) . These effects can result from direct impacts of environmental factors on parasite growth , survival and reproduction that underpin disease risk . For example , in an experiment in the Rocky mountains , host plants that grew on heated research plots showed increased disease , largely by increasing the amount of time that environmental conditions were favorable for parasite growth and reproduction ( Roy et al . , 2004 ) . Importantly however , these same environmental factors can also indirectly influence disease risk by altering the composition of host or vector communities that are required for sustained parasite transmission ( Elad and Pertot , 2014; Garrett et al . , 2006; Harvell et al . , 2002; Mordecai et al . , 2019; Newton et al . , 2011; Yáñez-López , 2012; Rohr et al . , 2011 ) , or by altering host defenses ( Descombes et al . , 2017; Pellissier et al . , 2018; Wolinska and King , 2009 ) . Thus , shifts in parasite replication that are driven by changing host or vector distributions can also determine whether and how changing environmental conditions will alter disease risk . There is growing empirical evidence in support of both direct effects that alter parasite growth and replication , as well as indirect effects that are mediated by changing host or vector community structure . However , disentangling the relative impacts of these direct and indirect effects of environmental factors on disease risk has been historically challenging , because it often requires a priori knowledge of environmental constraints acting on host and parasite populations ( Garrett et al . , 2006; Harvell et al . , 2002; Mordecai et al . , 2019; Rohr et al . , 2011 ) . One way to disentangle direct and indirect effects of environmental conditions on disease is to consider these effects in the context of host functional traits . Host functional traits underlie ecologically important resource acquisition and allocation tradeoffs: hosts must balance allocating limited resources to maximize growth and reproduction , while also constructing tissue capable of withstanding stressful environmental conditions ( Díaz et al . , 2016; Reich , 2014; Reich et al . , 2003; Wright et al . , 2004 ) . Thus , linking environmental conditions with relevant functional traits has become a tractable way to predict the richness and composition of communities ( i . e . community structure ) ( Cornelissen et al . , 2003; Díaz and Cabido , 1997; Funk et al . , 2017; Kattge et al . , 2020; Lavorel and Garnier , 2002; McMahon et al . , 2011; Reich , 2014; Sundqvist et al . , 2013 ) . The functional traits expressed by those species that are able to colonize and persist in a given location can , in turn , affect disease risk ( Halliday et al . , 2019; Johnson et al . , 2013; Kirk et al . , 2019 ) . Specifically , an infected host’s ability to transmit disease to uninfected hosts , a trait often referred to as host competence , is often related to fast-growing , poorly defended tissues and short lifespans ( Becker and Han , 2021; Cronin et al . , 2014; Cronin et al . , 2010; Huang et al . , 2013; Johnson et al . , 2012; Martin et al . , 2019; Martin et al . , 2016; Parker and Gilbert , 2018; Stewart Merrill and Johnson , 2020; Welsh et al . , 2020 ) . Importantly , these functional trait values also underlie ecological tradeoffs related to host growth and defense , resource acquisition and allocation , and survival and reproduction ( i . e . life history ) ( Coley et al . , 1985; Herms and Mattson , 1992; Martin et al . , 2016; Reich , 2014; Reich et al . , 2003; Ricklefs and Wikelski , 2002; Stearns , 1992; Stearns , 1989; Wright et al . , 2004 ) . Thus , host community competence ( a community-level metric of host competence ) is expected to correspond to the same functional traits ( i . e . host pace-of-life ) that link host community structure to shifting environmental conditions . A trait-based framework of host community competence may explain why biodiversity loss is consistently associated with higher disease risk ( Halliday et al . , 2020b; Johnson et al . , 2013; LoGiudice et al . , 2003; Ostfeld and LoGiudice , 2003 ) , a relationship known as the ‘dilution effect’ of biodiversity ( Keesing et al . , 2010; Keesing et al . , 2006; Ostfeld and Keesing , 2000 ) . This is because host species that are most resistant to biodiversity loss or best able to colonize newly disturbed habitats often rely on the same life-history strategies that are associated with higher host competence ( Johnson et al . , 2013; LoGiudice et al . , 2003; Ostfeld and LoGiudice , 2003 ) . For example , species that are associated with habitat fragmentation , a key anthropogenic driver of biodiversity loss , are often characterized by life history strategies favoring a ‘fast pace-of-life’ ( i . e . fast growth rates , quick reproduction , and high dispersal ) ( Albrecht and Haider , 2013; Fay et al . , 2015; Gibbs and van Dyck , 2010; Hanski et al . , 2006; Keinath et al . , 2017; Merckx et al . , 2018; Ziv and Davidowitz , 2019 ) . But this fast pace-of-life often comes at the cost of reduced defense against parasites ( Cappelli et al . , 2020; Coley et al . , 1985; Cronin et al . , 2014; Cronin et al . , 2010; Heckman et al . , 2019; Herms and Mattson , 1992; Johnson et al . , 2012; Sears et al . , 2015 ) . Thus , habitat fragmentation can increase disease by increasing the density of fast pace-of-life , highly competent hosts , while slow pace-of-life , less-competent hosts are lost ( Johnson et al . , 2015b; Joseph et al . , 2013; Mihaljevic et al . , 2014 ) . This hypothesis has widespread empirical support in a variety of systems ( Johnson et al . , 2019; Johnson et al . , 2013; Liu et al . , 2018; Ostfeld and LoGiudice , 2003 ) . Shifting community structure during biodiversity loss may therefore predictably influence infectious disease risk ( Halliday et al . , 2020b ) . Although relationships between host community structure and disease risk are becoming increasingly appreciated , how these relationships change across environmental gradients remain unknown ( Halliday et al . , 2020b; Halliday and Rohr , 2019 ) . The relationship between host traits and host competence can be variable , and this relationship might also depend on the environmental context in which host-parasite interactions play out ( Figure 1b . , path d ) . For example , Welsh et al . , 2016 showed that when hosts were reared under novel resource conditions , trait-based models of host susceptibility became increasingly inaccurate , because novel resource conditions altered how traits covaried with one another and how raw trait values predicted infection . Thus , traits associated with host community competence in one environment might not predict host community competence across environmental gradients . We hypothesized that three non-mutually exclusive mechanisms would determine how environmental conditions influence disease risk in host communities: ( 1 ) directly , by altering parasite growth and reproduction ( i . e . through abiotic constraints; Figure 1b . , path a ) , ( 2 ) indirectly , by altering which host species occur in which locations ( i . e . mediated by shifting host community structure; Figure 1b . , paths b and c ) , and ( 3 ) indirectly , by altering how host traits influence parasite transmission ( i . e . moderated by altering the relationship between host traits and host competence , which we refer to as the trait-competence relationship; Figure 1b . , path d ) . Here , we test the relative contributions of these three mechanisms through which environmental conditions can drive infectious disease risk ( i . e . direct , mediated , and moderated ) by measuring foliar fungal disease in host plant communities along a roughly 1100 m elevational gradient in Southeastern Switzerland . Foliar fungal parasites are a widely used , tractable model of disease risk that respond to small-scale variation in host community structure and environmental conditions ( Cappelli et al . , 2020; Halliday et al . , 2019 , Halliday et al . , 2017; Liu et al . , 2018; Liu et al . , 2017; Liu et al . , 2016; Mitchell et al . , 2003; Mitchell et al . , 2002; Rottstock et al . , 2014 ) . Host community structure and environmental conditions , in turn , vary predictably with elevation ( Grinnell , 1914; Halbritter et al . , 2018; Malhi et al . , 2010; Sundqvist et al . , 2013; Whittaker , 1956 ) . Thus , an elevational gradient represents a natural laboratory for studying long-term , large scale changes in climate as well as interacting biotic and abiotic factors that are associated with climate change ( Alexander et al . , 2015; Fukami and Wardle , 2005; Sundqvist et al . , 2013 ) . Our study reveals strong evidence that increasing temperature associated with lower elevation can directly influence disease risk , which we attribute to well-established effects of abiotic conditions ( Avenot et al . , 2017; Garrett et al . , 2006; Harvell et al . , 2002; Tapsoba and Wilson , 1997; Waugh et al . , 2003 ) on parasite replication and growth , and can indirectly influence disease risk by shifting host community structure and by modifying the trait-competence relationship . Together , these results highlight the need to consider biotic and abiotic drivers jointly , in order to predict disease risk in the face of climate change . The elevational gradient captured by the CBO allows us to explore associations among abiotic factors and biodiversity while minimizing other confounding factors like day length , geology , and biogeographic history ( Halbritter et al . , 2018 ) . We assessed the association between elevation and environmental conditions by fitting linear models . Mean soil , soil surface , and air temperature strongly and consistently decreased with increasing elevation ( p < 0 . 001 , R2 = 0 . 88; p < 0 . 001 , R2 = 0 . 86; p < 0 . 001 , R2 = 0 . 89; respectively ) , while mean soil moisture was uncorrelated with elevation ( p = 0 . 72 , R2 = 0 . 006 ) . The mean soil surface temperature at sites located in the highest elevation meadow ( 1576 m–1749 m ) was , on average , 4 . 67 °C lower than sites located in the lowest elevation meadow ( 648 m–766 m ) . The altitudinal temperature lapse rate along the elevational gradient was –0 . 57 °C/100 m . In total , 188 host taxa were observed across the 220 small plots of the CBO . The communities consisted mostly of perennial herbs such as Salvia pratensis and Helianthemum nummularium , and were dominated by grasses that tolerate grazing such as Dactylis glomerata , Lolium perenne , and Phleum pratense . The most abundant species was Brachypodium pinnatum . An herbarium specimen of each taxon encountered is deposited at the University of Zürich . We assessed the relationship between abiotic conditions and species richness by fitting linear mixed models with large plots , sites , and meadows as nested random intercepts . Species richness in the small plots varied from 7 to 30 species ( median 20 ) , was uncorrelated with soil moisture ( p = 0 . 98 ) and increased as elevation increased and soil surface temperature declined ( p = 0 . 005; Marginal R2 = 0 . 10; Conditional R2 = 0 . 75 ) , with median species richness roughly 16% higher in plots located at the highest elevation meadow , characterized by the coolest environmental temperatures , compared to the lowest elevation meadow , which was characterized by the warmest environmental temperatures ( Supplementary file 1b ) . These effects were qualitatively similar when we included air temperature and elevation in place of soil-surface temperature , though the relationship became nonsignificant when we replaced soil-surface temperature with soil temperature in the model ( p = 0 . 15; Supplementary file 1b ) . We performed confirmatory factor analysis to assign six foliar functional traits associated with the worldwide leaf economics spectrum to a single axis representing host pace-of-life . One trait , photosynthetic rate , loaded particularly poorly on this axis ( factor loading 0 . 036 ) , and was therefore excluded from the latent factor . This resulted in a single factor , explaining 62% of the variance in specific leaf area , 51% of the variance in leaf chlorophyll content , 25% of the variance in leaf nitrogen , 10% of the variance in leaf phosphorus , and 2% of the variance in leaf lifespan ( χ² ( df = 5 ) = 4 . 24 , p = 0 . 52; CFI = 1 . 019; Figure 3—figure supplement 1 ) . Consistent with resource-acquisition and allocation tradeoffs ( Díaz et al . , 2016; Reich , 2014; Wright et al . , 2004 ) , higher values of host pace-of-life were associated with increases in specific leaf area , leaf chlorophyll content , leaf nitrogen , and leaf phosphorus , and with shorter leaf lifespans . We then used each species’ unique score on this pace-of-life factor to quantify the community-weighted mean host pace-of-life ( hereafter community pace-of-life ) for each small plot . We assessed the relationship between abiotic conditions and community pace-of-life by fitting linear mixed models with large plots , sites , and meadows as nested random intercepts . Although host community pace-of-life was unrelated to soil moisture ( p = 0 . 13 ) , host community pace-of-life declined with reduced soil-surface temperature associated with higher elevation ( p = 0 . 010; Marginal R2 = 0 . 11; Conditional R2 = 0 . 83; Supplementary file 1b; Figure 3—figure supplement 2 ) , consistent with expectations regarding shifting host community structure ( Descombes et al . , 2017; Hulshof et al . , 2013; but see Pellissier et al . , 2018 ) . These effects were qualitatively similar when we included soil temperature or air temperature in place of soil-surface temperature in the model , though the effect became marginally nonsignificant when we replaced temperature with elevation in the model ( p = 0 . 066; Supplementary file 1b; Figure 3—figure supplement 2 ) . We tested whether the relationship between host community structure ( i . e . host species richness and host community pace-of-life ) and disease would change as a function of environmental conditions by fitting a linear mixed model with square-root transformed community parasite load ( e . g . Halliday et al . , 2019 , Halliday et al . , 2017; Mitchell et al . , 2002 ) as the response . Soil-surface temperature , soil moisture , host community species richness , pace-of-life , and pairwise interactions between both measures of community structure and each abiotic variable were treated as fixed effects , with large plots , sites , and meadows as nested random intercepts . All variables that were treated as fixed effects in the model were centered so that the mean value of each variable was used as the reference value for interpreting the other variables' independent effects . This mixed model of disease revealed several independent and interactive effects of host community structure and environmental conditions on disease risk ( Marginal R2 = 0 . 227; Conditional R2 = 0 . 497; RMSE = 0 . 292; LOOCV RMSE = 0 . 311; Table 1 ) . Consistent with the hypothesis that host pace-of-life can determine host community competence , communities that were dominated by hosts with fast-paced life-history strategies exhibited the most disease , but this effect declined as elevation increased and temperature declined ( temperature × pace-of-life: p < 0 . 001 ) . This weakening effect of host community pace-of-life as soil-surface temperature declined is consistent with the hypothesis that abiotic conditions can alter which traits favor parasite transmission through the relationship between host competence and disease risk ( Figure 3 ) . These results therefore provide field evidence that an environmental gradient can alter the effect of host community structure on disease risk . The model also revealed significant independent effects of host community structure and abiotic conditions on disease risk . Specifically , the model revealed evidence supporting the dilution effect hypothesis: increasing species richness was associated with a reduction in disease ( p = 0 . 021 ) , and this effect was independent of soil-surface temperature ( temperature × richness: p = 0 . 10 ) . Community parasite load was also positively associated with increasing soil-surface temperature ( p = 0 . 006 ) , consistent with the hypothesis that environmental gradients can alter parasite growth and reproduction via abiotic constraints . These effects were qualitatively similar when we included soil temperature , air temperature , or elevation in place of soil-surface temperature in the model ( Supplementary file 1c ) . In contrast with results involving soil-surface temperature , there was no statistically significant linear relationship between soil moisture and disease ( p = 0 . 71 ) , nor was there a significant interaction between soil moisture and host richness ( p = 0 . 10 ) or community-weighted mean pace-of-life ( p = 0 . 45 ) on disease . Because soil moisture was unrelated to elevation , pace-of-life , species richness , and disease in our models , this factor was omitted from further analyses . Together , models of host community species richness and pace-of-life showed that declining temperature associated with increasing elevation could determine changes in host community structure , and the model of disease showed that host community structure and temperature could independently and interactively influence disease . To explore the relative influence of these direct and indirect effects on disease risk , we next constructed a structural equation model . Our data were well fit by this model ( Fisher’s C = 4 . 111; p-value = 0 . 662; 6 degrees of freedom , Supplementary file 1d; Figure 4 ) . The model leverages the strong , negative effect of elevation on soil-surface temperature ( standardized path coefficient = –0 . 91 , R2 = 0 . 85 ) to compare three separate pathways through which increasing temperature can increase disease: First , increasing temperature increased community parasite load directly ( standardized path coefficient = 0 . 24 ) . Second , increasing temperature increased community parasite load indirectly by reducing host species richness ( i . e . via mediation; product of standardized path coefficients = 0 . 045 ) . Third , increasing temperature increased community parasite load indirectly by simultaneously increasing host community pace-of-life ( i . e . via mediation; mean-centered standardized path coefficient = 0 . 39 ) and strengthening the relationship between host pace-of-life and disease risk ( i . e . via moderation; mean-centered standardized path coefficient = 0 . 18; Figure 4 ) . Together these results highlight the pressing need to consider host community context in predicting how shifting environmental gradients will alter disease risk . This study shows , to our knowledge , the first evidence under natural field conditions that , in addition to directly influencing disease risk , the abiotic environment can also indirectly influence disease both by altering host community structure ( i . e . mediation ) and by modifying how host community structure influences disease risk ( i . e . moderation ) . Furthermore , this linkage between the abiotic environment and host community structure suggests that any single factor would be inadequate for explaining disease risk along our environmental gradient . Together , these results reveal the role that host communities play in determining ecosystem health across environmental gradients , suggesting that predicting how shifting abiotic conditions will influence disease risk will require explicit consideration of how host and parasite communities jointly respond to the abiotic environment . Our results indicate that increasing temperatures associated with lower elevation , can independently influence disease . Specifically , increasing temperature increased disease , even after accounting for effects of host community structure on disease . We hypothesize that reduced temperature associated with increasing elevation may have reduced disease directly by constraining parasite growth , survival , and reproduction . Many foliar parasites grow and reproduce more successfully in warmer environmental temperatures ( Avenot et al . , 2017; Garrett et al . , 2006; Harvell et al . , 2002; Tapsoba and Wilson , 1997; Waugh et al . , 2003 ) . Warmer temperatures can also increase parasite overwintering success ( Burdon and Elmqvist , 1996; Pfender and Vollmer , 1999 ) or allow parasites to produce more generations during a longer growing season ( Garrett et al . , 2006 ) . These results corroborate past studies suggesting that environmental gradients can directly alter the strength of biotic interactions ( Descombes et al . , 2017; Hargreaves et al . , 2019; Pellissier et al . , 2014; Roslin et al . , 2017; Schemske et al . , 2009 ) , including host-parasite interactions ( Abbate and Antonovics , 2014; Allen et al . , 2020; LaManna et al . , 2017; Nunn et al . , 2005 ) . However , despite the strong and consistent effect of increasing temperature on disease , temperature was highly correlated with elevation , and we cannot rule out the possibility that these effects might be driven by some other factor associated with elevation that was not measured , such as changing humidity or soil nutrient availability . Thus , temperature effects should be interpreted with some caution . In addition to directly influencing disease , our results indicate that increasing temperature can also indirectly influence disease by altering host community structure . Specifically , increasing temperature reduced host species richness , which , in turn , reduced disease . The reduction in host species richness with increasing temperature might be attributable to the occurrence of both low-elevation and high-elevation adapted species occupying the coolest study sites , located at the highest elevation ( Colwell and Lees , 2000 ) . Communities in the highest elevation meadow , located just below the tree line , included plant species characteristic of low elevations ( e . g . Lathyrus pratensis , Lolium perenne , and Salvia pratensis ) and plant species that tend to occupy high elevation grasslands ( Soldanella alpina , Ranunculus montanus , and Carex sempervirens ) , indicating that these high-elevation sites represent an intermediate zone between subalpine and alpine vegetation communities . Host communities with higher species richness , in turn , experienced less disease ( i . e . a dilution effect; Keesing et al . , 2010; Keesing et al . , 2006 ) , even after accounting for the direct effects of temperature on disease and other measures of host community structure . Past studies indicate that increasing biodiversity is often associated with a decline in disease risk because host community structure shifts during biodiversity loss to favor more competent hosts ( Johnson et al . , 2019; Joseph et al . , 2013; Liu et al . , 2018; LoGiudice et al . , 2003; Ostfeld and LoGiudice , 2003; Rohr et al . , 2020 ) . However , in contrast with past studies focused on biodiversity loss , our study measured biodiversity change across a natural biodiversity gradient , which is not expected to consistently influence disease risk ( Halliday et al . , 2020b ) . We hypothesize that increasing species richness may have reduced disease risk in this system by reducing host density ( Keesing et al . , 2006; i . e . via encounter reduction; Mitchell et al . , 2002 ) . Encounter reduction might be particularly relevant in this system , because , in addition to altering host richness , reduced temperatures associated with increasing elevation also influence the length and timing of the growing season , which can affect peak prevalence and the duration of the epidemic season . In addition to direct and indirect effects via mediation , our results further indicate that increasing temperature can indirectly influence disease by modifying the effect of host community structure on disease ( i . e . via moderation ) . Specifically , disease was influenced by host community pace-of-life , but only at high temperature , low elevation sites . Because more competent hosts often exhibit fast-paced life history strategies ( Cronin et al . , 2010; Johnson et al . , 2012; Martin et al . , 2016; Parker and Gilbert , 2018; Welsh et al . , 2020 ) , we expected that host communities dominated by species with a fast pace-of-life would experience greater disease . However , a prior study suggested that the relationship between host traits and host competence might be sensitive to environmental conditions ( Welsh et al . , 2016 ) , which we hypothesized would cause the relationship between host community pace-of-life and disease risk to shift across environmental gradients . Our analysis was consistent with this hypothesis: increasing temperature not only modified host community pace-of-life , but the effect of host community pace-of-life on disease was also sensitive to increasing temperature . Host community pace-of-life most strongly predicted disease risk at the highest temperatures , associated with the lowest elevation , but this effect weakened and ultimately disappeared as elevation increased and temperature declined . These results indicate that warming temperatures can modify the effect of host community pace-of-life on disease risk , which we attribute to a change in the relationship between host traits and host competence across environmental conditions . However , we cannot rule out the possibility that the interaction between host pace-of-life and temperature could have also been driven by other mechanisms . For example , the values of functional traits expressed by a single species may have changed along the environmental gradient via a phenomenon known as intraspecific trait variation ( Albert et al . , 2011; Funk et al . , 2017; Messier et al . , 2010; Violle et al . , 2012 ) . Studies of functional traits ( including this study ) typically characterize each species with a single value for each trait , such as the species-level mean , under the assumption that ecologically important traits vary more among species than within species ( McGill et al . , 2006 ) . However , functional traits of individuals within a species can vary due to local adaptation and phenotypic plasticity driven by local context ( Albert et al . , 2011; Funk et al . , 2017; Messier et al . , 2010; Violle et al . , 2012 ) . Thus , intraspecific shifts in the expression of key functional traits across our elevational gradient could drive the apparent interaction between host community pace-of-life and temperature . Alternatively , a reduction in infection severity with cooling temperatures could weaken the importance of investment in disease resistance ( Benkman , 2013; Thompson , 1999 ) . Thus , host species may still form strong trade-offs in fast vs slow strategies for growth vs . survival , but this pace-of-life trait would have a weak link with disease severity . Future studies should explore these mechanisms by directly measuring host and parasite functional traits across environmental gradients like elevation . Together , the results of this study highlight the need to consider host community context in predicting how climate change will alter disease risk . Specifically , in this study , effects of the abiotic environment and changing environmental temperature on disease strongly depended on shifting host community pace-of-life . These results are consistent with a growing body of literature suggesting that the role of host communities in regulating ecosystem processes is at least partially explained by characteristics of species present in those ecosystems ( Allan et al . , 2015; Heilpern et al . , 2020; Le Bagousse-Pinguet et al . , 2019; Leitão et al . , 2016; Mouillot et al . , 2011; Start and Gilbert , 2019; Van de Peer et al . , 2018 ) , but that abiotic factors such as temperature can override the effects of biotic factors on ecosystem processes ( Cannone et al . , 2007; Laiolo et al . , 2018 ) . These results therefore suggest that predicting how climate change will influence disease may depend on complex relationships between environmental factors and the structure of host communities . The Calanda Biodiversity Observatory ( CBO ) consists of four publicly owned meadows located along a 1101 m elevational gradient ( 648 m to 1749 m ) below tree-line on the south-eastern slope of Mount Calanda ( 46°53′59 . 5″N 9°28′02 . 5″E ) in the canton of Graubünden ( Figure 2 ) . The mean annual temperature at 550 m altitude is 10°C and the mean annual precipitation is 849 mm ( MeteoSwiss , 2020 ) , with temperature declining and precipitation increasing as elevation increases ( e . g . in 2013 and 2014 , mean temp and precipitation at 1400 m were 7°C and 1169 mm , respectively; Alexander et al . , 2015 ) . The soil in the area is generally calcareous and has low water retention ( Alexander et al . , 2015; Eggenberg and Möhl , 2013 ) . The four CBO meadows are variable in size ( roughly 8–40 Ha ) , and separated by forests and at least 500 m elevation . Meadows are maintained through grazing and mowing , a typical form of land use in the Swiss Alps ( Bätzing , 2015 ) , and cover collinean ( < 800 m ) mountain ( 800 m–1500 m ) and subalpine ( 1500–2200 m ) vegetation zones ( Eggenberg and Möhl , 2013; Ozenda , 1985 ) . The CBO meadows are grazed by cattle twice per year as the cattle are moved between low and high altitudes . Increasing elevation is associated with changes in a variety of abiotic conditions , including a reduction in temperature . Temperature decreases approximately 0 . 4–0 . 7 °C for each 100 m increase in elevation because of lower air pressure in high elevations , a phenomenon known as the altitudinal temperature lapse rate ( Barry , 2008 ) . The altitudinal temperature lapse rate varies among years and even days , usually being lower in winters and during nights . Typical altitudinal temperature lapse rates in the Alps vary from –0 . 54°C/100 m to –0 . 58°C/100 m ( Rolland , 2003 ) . The CBO consists of a nested set of observational units ( Figure 2 ) . Each meadow contains 4–7 , . 25 ha sites ( n = 22 sites ) . Sites were selected to maximize coverage over each meadow , avoiding roads that would cross the sites and large trees , shrubs and rocks that could create a forest- or shrub-type habitat that differs from grassland , and were placed sufficiently far from forest edges so that they were not shaded by the forest canopy . Each site is 50 m x 50 m and contains a grid of nine evenly spaced , 4 m2 large-plots , with the exception of one site ( I3 ) , which is 100 m x 25 m and contains 10 large plots due to its shape . Altogether , there are 199 large plots . In each site , large plots are arranged in a grid with the center of each plot separated by at least 20 m distance from its nearest neighbor . The location of the grid was randomized within each site and always located at least 2 m from the site edge . Each large plot is subdivided into four , 1 m2 subplots ( n = 796 ) . At each site , five large plots were selected to contain an intensively surveyed module ( ISM ) , which consisted of two 50 cm-diameter , round small plots , placed in opposite subplots ( n = 110 ISMs consisting of 220 small plots ) . These intensively surveyed small plots are the smallest unit of observation used in this study ( Figure 2 ) . In July 2019 , we recorded the identity and visually quantified the percent cover of all plant taxa in each small plot ( n = 220 ) . Vegetation surveys entailed the same two researchers searching within the subplot area for all vascular plants present in the subplot , before jointly estimating the total percent cover of each species ( Halbritter et al . , 2020 ) . Plant individuals that were growing outside the small plot , but whose foliage extended into the small plot , were included in this survey . Plant taxa were identified with the help of plant identification literature ( Eggenberg et al . , 2018; Eggenberg and Möhl , 2013; Lauber et al . , 2018 ) . The survey started at the lowest elevation and continued higher in order to survey the meadows approximately at the same phase of the growing season in relation to one another . The survey was initiated at least 4 days after cows had grazed each meadow ( Supplementary file 1a ) . We evaluated changes in two components of host community structure to evaluate indirect effects of environmental conditions on disease: host species richness and community-weighted mean host pace-of-life . These two components of host community structure commonly respond to changing environmental conditions ( Descombes et al . , 2017; Hulshof et al . , 2013 ) , and represent important characteristics of host communities that influence disease risk ( Joseph et al . , 2013; Liu et al . , 2018; Liu et al . , 2017 ) . We quantified community-weighted mean host pace-of-life using the TRY database ( Kattge et al . , 2020 ) . We first extracted six traits for every host taxon in the database ( plant photosynthetic rate , leaf chlorophyll content , leaf lifespan , leaf nitrogen content , leaf phosphorus content , and specific leaf area ) , omitting tree seedlings , which are functionally dissimilar from the more dominant herbaceous taxa , and taxa that could not be identified to host genus , which together , never accounted for more than 7% cover in a plot ( mean = 0 . 04% ) . Unknown taxa that could be identified to the genus level were assigned genus-level estimates for each host trait , by taking the mean of the trait value for all members of that genus that had been observed on Mount Calanda during extensive vegetation surveys ( Supplementary file 1e ) . We then performed full-information maximum-likelihood factor analysis to produce a single axis representing covariation in the functional traits associated with host pace-of-life using the umxEFA function in r-package umx ( Bates et al . , 2019 ) . This approach allows each host taxon to be assigned a value for host pace-of-life , even if that taxon is missing some values for individual functional traits . Finally , we calculated a single value for each small plot ( n = 220 ) using the community-weighted mean of host pace-of-life ( hereafter community pace-of-life ) . The community weighted mean ( CWM ) was calculated as:CWM=∑i=1Nsppixi where Nsp is the number of taxa within a plot with a pace-of-life trait value in the dataset , pi is the relative abundanceof taxon , i , in the plot ( i . e . the absolute vegetative cover of taxon , i , divided by the total absolute cover of all taxa in the plot ) , and xi is the host pace-of-life value for taxon , i . A survey of foliar disease symptoms was carried out in August 2019 by estimating the percent of leaf area damaged by foliar fungal parasites on up to five leaves of twenty randomly selected host individuals per small plot ( n = 18 , 203 leaves on 4400 host individuals across 220 small plots ) . The disease survey was conducted by placing a grid of 20 equally spaced grill sticks into the ground , with each stick having a distance of 10 cm to its nearest neighbor ( Figure 2—figure supplement 1 ) . The 20 plant individuals that were most touching the sticks were then identified , and the five oldest non-senescing leaves on each plant were visually surveyed for foliar disease symptoms following the plant pathogen and invertebrate herbivory protocol in Halbritter et al . , 2020 . The survey was carried out on leaves , because symptoms are highly visible and easily grouped into parasite types on leaves . On each leaf , we estimated the leaf area ( % ) that was covered by disease symptoms . Some plant individuals had fewer than five leaves , so fewer than five leaves were surveyed on those plants . Unlike the vegetation survey , the disease survey was not conducted in elevational order due to logistical constrains related to site accessibility . Small plots were surveyed between 29 July and 19 August 2019 ( Supplementary file 1a ) , which we observed to be time of peak plant biomass in this system . Disease was assessed for each small plot using community parasite load , calculated as the mean leaf area damaged by all parasites on a host , multiplied by the relative abundance of that host species from the July vegetation survey , and then summed across all hosts in the plot ( Halliday et al . , 2019 , Halliday et al . , 2017; e . g . , Mitchell et al . , 2002 ) . Soil temperature ( 6 cm below the soil surface ) , soil surface temperature , air temperature ( 12 cm above the soil surface ) , and soil volumetric moisture content were recorded at 15 minute intervals for 22–37 days ( average 31 days ) in the central large plot of each site ( n = 22 ) using a TOMST-4 datalogger ( Wild et al . , 2019 ) . The total duration of measurement varied because some of the dataloggers had to be moved earlier or temporarily because of mowing or grazing activities ( Supplementary file 1a ) . All statistical analyses were performed in R version 3 . 5 . 2 ( R Development Core Team , 2015 ) . We assessed the association between elevation and environmental conditions by fitting linear models with the lm function . All other analyses consisted of fitting linear mixed models with an identity link and Gaussian likelihoods using the lme function in the nlme package ( Pinheiro et al . , 2016 ) . In order to meet assumptions of normality and homoscedasticity , we square-root transformed community parasite load and added an identity variance structure ( varIdent function ) for each site , which based on visual inspection of residuals of each model , exhibited considerable heteroscedasticity ( Pinheiro et al . , 2016; Zuur et al . , 2009 ) . Each model included large plots , sites , and meadows as nested random intercepts to account for non-independence among observations due to the sampling design of the CBO . Full equations and parameters for these models are available on Github ( https://github . com/fhalliday/Calanda19/tree/Calanda19_disease_submission; Halliday , 2021; copy archived at swh:1:rev:86ce01777c396840455fd67a3ff5cd8420e8df21 ) . We first explored the relationship between elevation and environmental conditions by constructing four models , each including one environmental factor ( either mean soil temperature , soil surface temperature , air temperature , or soil moisture ) as a response variable , and mean elevation of the site as the predictor . Next , we explored the relationship between each measure of host community structure ( i . e . host species richness and host community pace-of-life ) and environmental conditions by constructing two models , each consisting of one measure of host community structure as a response variable and one measure of soil-surface temperature and soil moisture as fixed effects . We only included a single measure of temperature in these models , and excluded elevation , to avoid problems associated with collinearity . We used soil-surface temperature , as this measurement represented the temperature that the majority of leaves ( and therefore foliar pathogens ) were exposed to ( Figure 2—figure supplement 1 ) . Results using soil temperature , air temperature , and elevation are reported in the Supplement . We then tested whether the relationship between host community structure and disease would change as a function of environmental conditions by constructing a mixed model with square-root transformed community parasite load as the response , and soil-surface temperature , soil moisture , host community species richness , and pace-of-life as fixed effects . To estimate whether the effect of host community structure depends on environmental conditions , we also included in the model the pairwise interactions between both measures of host community structure and either soil-surface temperature or soil moisture as additional fixed effects . As before , we only included a single measure of temperature in this model and excluded elevation to avoid problems associated with collinearity . Results using soil temperature , air temperature , and elevation are reported in the Supplement . To aid the interpretation of main effects in the model , we centered all variables so that the mean value of each variable was used as the reference value for interpreting the other variables' independent effects . To evaluate model fit , we calculated the root-mean-squared error ( RMSE ) of the model , the marginal and conditional pseudo-R2 of the model using the r . squaredGLMM function in the MuMIn package ( Bartoń , 2018 ) , and the RMSE using leave-one-out cross validation ( LOOCV RMSE ) . To test whether effects driven by host community pace-of-life were influenced by one or a few important functional traits , we repeated this analysis , including the community-weighted-mean of each leaf trait ( leaf chlorophyll content , leaf lifespan , leaf nitrogen content , leaf phosphorus content , and specific leaf area ) replacing host community pace-of-life . Individual leaf traits were measured using different units , and were therefore transformed to a common scale using a z-transformation . None of the models including individual leaf traits were improvements over the model including only host community pace-of-life ( Supplementary file 1f ) ; thus , individual leaf traits were excluded from further analyses . Finally , to compare direct and indirect effects of environmental conditions on disease risk , we performed confirmatory path analysis using the PiecewiseSEM package ( Lefcheck , 2016 ) . Specifically , we fit a structural equation model ( SEM ) that included the effect of elevation on soil-surface temperature , the effect of soil-surface temperature on square-root-transformed disease , the effect of soil-surface temperature on two endogenous mediators ( host community species richness and pace-of-life ) , which together measure changes in host community structure ( following Halliday et al . , 2020a; Halliday et al . , 2019 ) , and the effects of those two mediators on square-root-transformed community parasite load . We also tested the hypothesis that soil-surface temperature altered the relationship between host community structure and disease by fitting a second-stage moderated mediation ( Hayes , 2015 ) including the pairwise interaction between soil-surface temperature and community pace-of-life , omitting other potential interactions that were non-significant in the model testing whether effects of community structure on disease depend on environmental conditions . Soil moisture was excluded from the SEM because it was unrelated to all other variables in the model . To aid the interpretation of direct effects in the model , we mean-centered soil-surface temperature and host community pace-of-life , so that average soil-surface temperature and host community pace-of-life were used as the reference values for interpreting the other variable’s independent effects . We then explored the interaction between community pace-of-life and temperature by setting the reference temperature to one standard deviation above and below the mean temperature , and re-running the model .	Climate change is causing shifts in the ecology and biodiversity of different world regions at unprecedented rates . Global warming is also linked with changes in the risk for certain infectious diseases in humans , but also in animals and plants . There are several possible mechanisms for this . For one thing , changing weather patterns may affect how pathogens grow and reproduce . For another , the distribution ranges of animal and plant hosts of certain disease-causing pathogens are changing because of global warming . This means that the distributions of pathogens are also changing , and so is the severity of the diseases that they cause . Increasing temperatures may also influence the physiological traits that make host species suitable for pathogens . This is because the traits that allow species to survive or adapt to changes in their environment may also make them better at hosting and transmitting the pathogens that cause disease . For example , in plant communities , rising temperatures could favor species with faster growth rates , quicker reproduction and high dispersal , and these traits are often associated with more efficient spread of disease . Despite a lot of research into the effects of climate , it remains unclear how temperature , pathogen growth and reproduction , and host species’ traits and distributions combine and interact to alter infectious disease risk , especially in wild plant communities . To investigate this , Halliday , Jalo and Laine studied an area in southeast Switzerland where natural temperature and biodiversity change gradually through the region . The aim was to explore how relationships between plant biodiversity , pathogens and disease risk change with temperature , and to understand whether environmental or biological factors influence infectious disease risk more . Halliday , Jalo and Laine measured the levels of fungal diseases found in the leaves of plant communities spanning 1 , 100 meters of elevation , showing that higher temperatures increase disease risk both directly and indirectly . Directly , higher temperatures increased pathogen growth and reproduction , and indirectly , they influenced which plants were present and therefore able to act as disease hosts . The results also indicated that temperature can affect how the traits of plants drive the transmission rates of fungal pathogens . Important predictors of disease risk were traits relating to the growth rate of host plants , which tended to increase in areas with low elevation where the surface of the soil was warm . This study represents the first analysis , in wild plants , of how changing temperatures , the traits of shifting host species , and resident parasite populations interact to impact infectious disease risk . The insights Halliday , Jalo and Laine provided could aid in predicting how global climate change will influence infectious disease risk .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology" ]	2021	The effect of host community functional traits on plant disease risk varies along an elevational gradient
Homo naledi is a previously-unknown species of extinct hominin discovered within the Dinaledi Chamber of the Rising Star cave system , Cradle of Humankind , South Africa . This species is characterized by body mass and stature similar to small-bodied human populations but a small endocranial volume similar to australopiths . Cranial morphology of H . naledi is unique , but most similar to early Homo species including Homo erectus , Homo habilis or Homo rudolfensis . While primitive , the dentition is generally small and simple in occlusal morphology . H . naledi has humanlike manipulatory adaptations of the hand and wrist . It also exhibits a humanlike foot and lower limb . These humanlike aspects are contrasted in the postcrania with a more primitive or australopith-like trunk , shoulder , pelvis and proximal femur . Representing at least 15 individuals with most skeletal elements repeated multiple times , this is the largest assemblage of a single species of hominins yet discovered in Africa . The word naledi means ‘star’ in the Sotho language and refers to the Dinaledi Chamber's location within the Rising Star cave system . The Dinaledi chamber is located approximately 30 meters underground , within the Rising Star cave system at about 26°1′13′′ S; 27°42′43′′ E . The system lies within the Malmani dolomites , approximately 800 meters southwest of the well-known site of Swartkrans in the Cradle of Humankind World Heritage Site , Gauteng Province , South Africa . The present sample of skeletal material from the Dinaledi Chamber was recovered during two field expeditions , in November 2013 and March 2014 . Six specimens from an ex situ context can be identified as bird bones , and few fragmentary rodent remains have been recovered within the excavation area . Neither of these faunal constituents can presently be associated with the hominin fossil collection ( Dirks et al . , 2015 ) . Aside from these limited faunal materials , the Dinaledi collection is entirely composed of hominin skeletal and dental remains . The collection so far comprises 1550 fossil hominin specimens , this number includes 1413 bone specimens and 137 isolated dental specimens; an additional 53 teeth are present in mandibular or maxillary bone specimens . Aside from the fragmentary rodent teeth , all dental crowns ( n = 179 ) are hominin , recovered both from surface collection and excavation . Likewise , aside from the few bird elements , all morphologically informative bone specimens are clearly hominin . In all cases where elements are repeated in the sample , they are morphologically homogeneous , with variation consistent with body size and sex differences within a single population . These remains represent a minimum of 15 hominin individuals , as indicated by the repetition and presence of deciduous and adult dental elements . The geological age of the fossils is not yet known . Excavations have thus far recovered hominin material from Unit 2 and Unit 3 in the chamber ( Dirks et al . , 2015 ) . Surface-collected hominin material from the present top of Unit 3 , which includes material derived from both Unit 2 and Unit 3 , represents a minority of the assemblage , and is morphologically indistinguishable from material excavated from in situ within Unit 3 . In addition to general morphological homogeneity including cranial shape , distinctive morphological configurations of all the recovered first metacarpals , femora , molars , lower premolars and lower canines , are identical in both surface-collected and excavated specimens ( see Figure 14 later in the text ) . These include traits not found in any other hominin species yet described . These considerations strongly indicate that this material represents a single species , and not a commingled assemblage . The cranium of H . naledi does not have the well-developed crest patterns that characterize Australopithecus garhi ( Asfaw et al . , 1999 ) and species of the genus Paranthropus , nor the derived facial morphology seen in the latter genus . The mandible of H . naledi is notably more gracile than those of Paranthropus . Although maxillary and mandibular incisors and canines of H . naledi overlap in size with those of Paranthropus , the post-canine teeth are notably smaller than those of Paranthropus and Au . garhi , with mandibular molars that are buccolingually narrow . H . naledi differs from Australopithecus afarensis and Australopithecus africanus in having a pentagonal-shaped cranial vault in posterior view , sagittal keeling , widely spaced temporal lines , an angular torus , a deep and narrow digastric fossa , an external occipital protuberance , an anteriorly positioned root of the zygomatic process of the maxilla , a broad palate , and a small canine jugum lacking anterior pillars . The anterior and lateral vault of H . naledi differs from Au . afarensis and Au . africanus in exhibiting only slight post-orbital constriction , frontal bossing , a well-developed supraorbital torus with a well-defined supratoral sulcus , temporal lines that are positioned on the posterior rather than the superior aspect of the supraorbital torus , a root of the zygomatic process of the temporal that is angled downwards approximately 30° relative to the Frankfort Horizontal ( FH ) and which begins its lateral expansion above the mandibular fossa rather than the EAM , a mandibular fossa that is positioned medial to the wall of the temporal squame , a small postglenoid process that contacts the tympanic , a coronally oriented petrous , and a small and obliquely oriented EAM . The H . naledi mandible exhibits a more gracile symphysis and corpus , a more vertically inclined symphysis , a slight mandibular incurvation delineating a faint mental trigon , and a steeply inclined posterior face of the mandibular symphysis without a post incisive planum . The incisors of H . naledi overlap in size with some specimens of Au . africanus , though the canines and post-canine dentition are notably smaller , with relatively narrow buccolingual dimensions of the mandibular molars . The maxillary I1 lacks a median lingual ridge and exhibits a broad and uninflated lingual cervical prominence , the lingual mesial and distal marginal ridges do not merge onto the cervical prominence in the maxillary I2 , the mandibular canine exhibits only a weak lingual median ridge and a broad and uninflated lingual cervical prominence , and the buccal grooves on the maxillary premolars are only weakly developed . H . naledi exhibits a small and isolated Carabelli's feature in the maxillary molars , unlike the more prominent and extensive Carabelli's feature of Australopithecus . Moreover , the H . naledi mandibular molars possess small , mesiobuccally restricted protostylids that do not intersect the buccal groove , differing from the typically enlarged , centrally positioned protostylids that intersect the buccal groove in Australopithecus . The cranium of H . naledi differs from Australopithecus sediba ( Berger et al . , 2010 ) in exhibiting sagittal keeling , a more pronounced supraorbital torus and supratoral sulcus , a weakly arched supraorbital contour with rounded lateral corners , an angular torus , a well-defined supramastoid crest , a curved superior margin of the temporal squama , a root of the zygomatic process of the temporal that is angled downwards approximately 30° relative to FH , a flattened nasoalveolar clivus , weak canine juga , an anteriorly positioned root of the zygomatic process of the maxilla , and a relatively broad palate that is anteriorly shallow . The H . naledi mandible ( DH1 ) has a mental foramen positioned superiorly on the corpus that opens posteriorly , unlike the mid-corpus height , more laterally opening mental foramen of Au . sediba . The maxillary and mandibular teeth of H . naledi are smaller than those of Au . sediba , with mandibular molars that are buccolingually narrow . The lingual mesial and distal marginal ridges do not merge onto the cervical prominence in the maxillary I2 , the paracone of the maxillary P3 is equal in size to the protocone , the protoconid and metaconid of the mandibular molars are equally mesially positioned , and the lingual cusps of the molars are positioned at the occlusobuccal margin while the buccal cusps are positioned slightly lingual to the occlusobuccal margin . Also , Au . sediba shares with other australopiths a protostylid that is centrally located and which intersects the buccal groove of the lower molars , unlike the small and mesiobuccally restricted protostylid that does not intersect the buccal groove in H . naledi . The cranium of H . naledi differs from Homo habilis in exhibiting sagittal keeling , a weakly arched supraorbital contour , temporal lines that are positioned on the posterior rather than the superior aspect of the supraorbital torus , an angular torus , an occipital torus , only slight post-orbital constriction , a curved superior margin of the temporal squama , a suprameatal spine , a weak crista petrosa , a prominent Eustachian process , a small EAM , weak canine juga , and an anteriorly positioned root of the zygomatic process of the maxilla . Mandibles attributed to H . habilis show a weakly inclined , shelf-like post incisive planum with a variably developed superior transverse torus , unlike the steeply inclined posterior face of the mandibular symphysis of H . naledi , which lacks both a post incisive planum or superior transverse torus . The H . naledi mandible ( DH1 ) has a mental foramen positioned superiorly on the corpus that opens posteriorly , while the mental foramen of H . habilis is at mid-corpus height and opens more laterally . The maxillary and mandibular dentitions of DH1 are smaller than typical for H . habilis . The mandibular P3 of H . naledi is more molarized and lacks the occlusal simplification seen in H . habilis; it has a symmetrical occlusal outline , and multiple roots ( two roots: mesiobuccal and distal ) not seen in H . habilis . The molars of H . naledi lack crenulation , secondary fissures , and supernumerary cusps that are common to H . habilis . The protoconid and metaconid of the mandibular molars are equally mesially positioned . The cranium of H . naledi differs from Homo rudolfensis by its smaller cranial capacity , and by exhibiting frontal bossing , a post-bregmatic depression , sagittal keeling , a well-developed supraorbital torus delineated by a distinct supratoral sulcus , temporal lines that are positioned on the posterior rather than the superior aspect of the supraorbital torus , an occipital torus , an external occipital protuberance , only slight post-orbital constriction , a small postglenoid process , a weak crista petrosa , a laterally inflated mastoid process , a canine fossa , incisors that project anteriorly beyond the bi-canine line , and a shallow anterior palate . As in H . habilis , mandibles attributed to H . rudolfensis show a weakly inclined , shelf-like post incisive planum with a variably developed superior transverse torus , unlike the steeply inclined posterior face of the mandibular symphysis of DH1 , the latter of which lacks either a post incisive planum or superior transverse torus . The mandibular symphysis and corpus of H . naledi are more gracile than those attributed to H . rudolfensis , and the H . naledi mandible ( DH1 ) has a mental foramen positioned superiorly on the corpus that opens posteriorly , unlike the mid-corpus height , more laterally opening mental foramen of H . rudolfensis . The maxillary and mandibular dentition of H . naledi is smaller than that of most specimens of H . rudolfensis , with only KNM-ER 60000 and KNM-ER 62000 appearing similar in size for some teeth ( Leakey et al . , 2012 ) . The molars of H . naledi lack crenulation , secondary fissures , or supernumerary cusps common in H . rudolfensis . The buccal grooves of the maxillary premolars are weak in H . naledi , and the protoconid and metaconid of the mandibular molars are equally mesially positioned . H . naledi lacks the typically distinctive long and low cranial vault of Homo erectus , as well as the metopic keeling that is typically present in the latter species . H . naledi also differs from H . erectus in having a distinct external occipital protuberance in addition to the tuberculum linearum , a laterally inflated mastoid process , a flat and squared nasoalveolar clivus , and an anteriorly shallow palate . The parasagittal keeling that is present between bregma and lambda in H . naledi ( DH1 and DH3 ) is less marked than often occurs in H . erectus , including in small specimens such as KNM-ER 42700 and the Dmanisi cranial sample . Also unlike most specimens of H . erectus , H . naledi has a small vaginal process , a weak crista petrosa , a marked Eustachian process , and a small EAM . The mandible of H . erectus shows a moderately inclined , shelf-like post incisive planum terminating in a variably developed superior transverse torus , differing from the steeply inclined posterior face of the H . naledi mandibular symphysis , which lacks both a post incisive planum or a superior transverse torus . The mental foramen is positioned superiorly and opens posteriorly in DH1 , unlike the mid-corpus height , more laterally opening mental foramen of H . erectus . The maxillary and mandibular incisors and canines of H . naledi are smaller than typical of H . erectus . The mandibular P3 of H . naledi is more molarized and lacks the occlusal simplification seen in H . erectus , they reveal a symmetrical occlusal outline , and multiple roots ( 2R: MB+D ) not typically seen in H . erectus . Furthermore , the molars of H . naledi lack crenulation , secondary fissures , or supernumerary cusps common in H . erectus . H . naledi lacks the reduced cranial height of Homo floresiensis , and displays a marked angular torus and parasagittal keeling between bregma and lambda that is absent in the latter species . H . naledi further has a flat and squared nasoalveolar clivus , unlike the pronounced maxillary canine juga and prominent pillars of H . floresiensis . The mandible of H . floresiensis shows a posteriorly inclined post incisive planum with superior and inferior transverse tori , differing from the steeply inclined posterior face of the H . naledi mandibular symphysis , which lacks both a post incisive planum or a superior transverse torus . Dentally , H . naledi is distinguishable from H . floresiensis by the mesiodistal elongation and extensive talonid of the mandibular P4 , and the lack of Tomes' root on the mandibular premolars . The molar size gradient of H . naledi follows the M1 < M2 < M3 pattern , unlike the M3 < M2 < M1 pattern in H . floresiensis , and the mandibular molars are relatively mesiodistally long and buccolingually narrow compared to those of H . floresiensis . H . naledi differs from Middle Pleistocene ( MP ) and Late Pleistocene ( LP ) Homo ( here we include specimens sometimes attributed to the putative Early Pleistocene taxon Homo antecessor , and MP Homo heidelbergensis , Homo rhodesiensis , as well as archaic Homo sapiens and Neandertals ) in exhibiting a smaller cranial capacity . H . naledi has its maximum cranial width in the supramastoid region , rather than in the parietal region . It has a clearly defined canine fossa ( similar to H . antecessor ) , a shallow anterior palate , and a flat and a squared nasoalveolar clivus . H . naledi lacks the bilaterally arched and vertically thickened supraorbital tori found in MP and LP Homo . H . naledi also differs in exhibiting a root of the zygomatic process of the temporal that is angled downwards approximately 30° relative to FH , a projecting entoglenoid process , a weak vaginal process , a weak crista petrosa , a prominent Eustachian process , a laterally inflated mastoid process , and a small EAM . The H . naledi mandible tends to be more gracile than specimens of MP Homo . The mandibular canine retains a distinct accessory distal cuspulid not seen in MP and LP Homo . Molar cuspal proportions for H . naledi do not show the derived reduction of the entoconid and hypoconid that characterizes MP and LP Homo . The mandibular M3 is not reduced in DH1 , thus revealing an increasing molar size gradient that contrasts with reduction of the M3 in MP and LP Homo . H . naledi differs from H . sapiens in exhibiting small cranial capacity , a well-defined supraorbital torus and supratoral sulcus , a root of the zygomatic process of the temporal that is angled downwards approximately 30° relative to FH , a large and laterally inflated mastoid with well-developed supramastoid crest , an angular torus , a small vaginal process , a weak crista petrosa , a prominent Eustachian process , a small EAM , a flat and squared nasoalveolar clivus , and a more posteriorly positioned incisive foramen . The H . naledi mandible shows a weaker , less well-defined mentum osseum than H . sapiens , as well as a slight inferior transverse torus that is absent in humans . The mental foramen is positioned superiorly in H . naledi , unlike the mid-corpus height mental foramen of H . sapiens . The mandibular canine possesses a distinct accessory distal cuspulid not seen in H . sapiens . Molar cuspal proportions for H . naledi do not show the derived reduction of the entoconid and hypoconid that characterizes H . sapiens . The mandibular M3 is not reduced in H . naledi , thus revealing an increasing molar size gradient that contrasts with reduction of the M3 in H . sapiens . H . naledi possesses a combination of primitive and derived features not seen in the hand of any other hominin . H1 is differentiated from the estimated intrinsic hand proportions of Au . afarensis in having a relatively long thumb ( ( Mc1 + PP1 ) / ( Mc3 + PP3 + IP3 ) ) ( Rolian and Gordon , 2013; Almécija and Alba , 2014 ) . It is further distinguished from Au . afarensis , Au . africanus , and Au . sediba in having a well-developed crest for both the opponens pollicis and first dorsal interosseous muscles , a trapezium-scaphoid joint that extends onto the scaphoid tubercle , a relatively large and more palmarly-positioned capitate-trapezoid joint , and/or a saddle-shaped Mc5-hamate joint . H . naledi also differs from Au . sediba in that it lacks mediolaterally narrow Mc2-5 shafts ( Kivell et al . , 2011 ) . Manual morphology of Au . garhi is currently unknown . H1 is distinguished from H . habilis in having a deep proximal palmar fossa with a well-developed ridge distally for the insertion of the flexor pollicis longus muscle on the first distal phalanx , and a more proximodistally oriented trapezium-second metacarpal joint . It also differs from both H . habilis and H . floresiensis by having a relatively large trapezium-scaphoid joint that extends onto the scaphoid tubercle , and from H . floresiensis in having a boot-shaped trapezoid with an expanded palmar surface , and a relatively large and more palmarly-positioned capitate-trapezoid joint ( Tocheri et al . , 2005 , 2007; Orr et al . , 2013 ) . H1 is dissimilar to hand remains attributed to Paranthropus robustus/early Homo from Swartkrans ( Susman , 1988; Susman et al . , 2001 ) in having a relatively small Mc1 base and proximal articular facet , a saddle-shaped Mc5-hamate joint , and more curved proximal and intermediate phalanges of ray 2–5 . Manual morphology of H . rudolfensis is currently unknown , and that of H . erectus is largely unknown . Still , H1 differs from a third metacarpal attributed to H . erectus s . l . , as well as from Homo neanderthalensis and H . sapiens by lacking a styloid process ( Ward et al . , 2013 ) . H1 is further distinguished from H . neanderthalensis and H . sapiens by its relatively small facets for the Mc1 and scaphoid on the trapezium , its low angle between the Mc2 and Mc3 facets on the capitate , and by its long and curved proximal and intermediate phalanges on rays 2–5 . H1 is differentiated from all known hominins in having a Mc1 that combines a mediolaterally narrow proximal end and articular facet with a mediolaterally wide distal shaft and head , a dorsopalmarly flat and strongly asymmetric ( with an enlarged palmar-lateral protuberance ) Mc1 head , and the combination of an overall later Homo-like carpal morphology combined with proximal and intermediate phalanges that are more curved than most australopiths . H1 also differs from all other known hominins except H . neanderthalensis in having non-pollical distal phalanges with mediolaterally broad apical tufts ( relative to length ) . The femur of H . naledi differs from those of all other known hominins in its possession of two well-defined , mediolaterally-running pillars in the femoral neck . The pillars run along the superoanterior and inferoposterior margins of the neck and define a distinct sulcus along its superior aspect . The tibia of H . naledi differs from those of all other known hominins in its possession of a distinct tubercle for the pes anserinus tendon . The tibia differs from other hominins except H . habilis , H . floresiensis , and ( variably ) H . sapiens in its possession of a rounded anterior border . The foot of H . naledi differs from the pedal remains of Au . afarensis , Au . africanus , and Au . sediba in having a calcaneus with a weakly developed peroneal trochlea . F1 also differs from Au . afarensis in having a higher orientation of the calcaneal sustentaculum tali . F1 can be further distinguished from pedal remains attributed to Au . africanus in having a higher talar head and neck torsion , a narrower Mt1 base , a dorsally expanded Mt1 head , and greater proximolateral to distomedial orientation of the lateral metatarsals . The H . naledi foot can be further differentiated from the foot of Au . sediba in having a proximodistally flatter talar trochlea , a flat subtalar joint , a diagonally oriented retrotrochlear eminence and a plantar position of the lateral plantar process of the calcaneous , and dorsoplantarly flat articular surface for the cuboid on the Mt4 ( Zipfel et al . , 2011 ) . Pedal remains of Au . garhi are currently unknown , and those of P . robustus are too poorly known to allow for comparison . The H . naledi foot can be distinguished from the foot of H . habilis by the presence of a flatter , non-sloping trochlea with equally elevated medial and lateral margins , a narrower Mt1 base , greater proximolateral to distomedial orientation of the lateral metatarsals , and a metatarsal robusticity ratio of 1 > 5 > 4 > 3 > 2 . Pedal morphology in H . rudolfensis is currently unknown , and that of H . erectus is too poorly known to allow for comparison . The H . naledi foot can be distinguished from the foot of H . floresiensis by a longer hallux and shorter second through fifth metacarpals relative to hindfoot length , and higher torsion of the talar head and neck . The foot of H . naledi can be distinguished from the foot of H . sapiens only by its flatter lateral and medial malleolar facets on the talus , its low angle of plantar declination of the talar head , its lower orientation of the calcaneal sustentaculum tali , and its gracile calcaneal tuber . H . naledi exhibits anatomical features shared with Australopithecus , other features shared with Homo , with several features not otherwise known in any hominin species . This anatomical mosaic is reflected in different regions of the skeleton . The morphology of the cranium , mandible , and dentition is mostly consistent with the genus Homo , but the brain size of H . naledi is within the range of Australopithecus . The lower limb is largely Homo-like , and the foot and ankle are particularly human in their configuration , but the pelvis appears to be flared markedly like that of Au . afarensis . The wrists , fingertips , and proportions of the fingers are shared mainly with Homo , but the proximal and intermediate manual phalanges are markedly curved , even to a greater degree than in any Australopithecus . The shoulders are configured largely like those of australopiths . The vertebrae are most similar to Pleistocene members of the genus Homo , whereas the ribcage is wide distally like Au . afarensis . H . naledi has a range of body mass similar to small-bodied modern human populations , and is similar in estimated stature to both small-bodied humans and the largest known australopiths . We estimated body mass from eight femoral specimens for which subtrochanteric diameters can be measured ( ‘Materials and methods’ ) , with results ranging between 39 . 7 kg and 55 . 8 kg ( Table 3 ) . No femur specimen is sufficiently complete to measure femur length accurately , but the U . W . 101-484 tibia preserves nearly its complete length , allowing a tibia length estimate of 325 mm ( Figure 10 ) . Estimates for the stature of this individual based on African human population samples range between 144 . 5 and 147 . 8 mm . Again , this stature estimate is similar to small-bodied modern human populations . It is within the range estimated for Dmanisi postcranial elements ( Lordkipanidze et al . , 2007 ) , and slightly smaller than estimated for early Homo femoral specimens KNM-ER 1472 and KNM-ER 1481 . Some large australopiths also had long tibiae and presumably comparably tall statures , as evidenced by the KSD-VP 1/1 skeleton from Woranso-Mille ( Haile-Selassie et al . , 2010 ) . 10 . 7554/eLife . 09560 . 014Table 3 . Dinaledi body mass estimates from femur specimens preserving subtrochanteric diametersDOI: http://dx . doi . org/10 . 7554/eLife . 09560 . 014Specimen IDSideAP subtrochanteric breadthML subtrochanteric breadthMass ( a ) Mass ( b ) U . W . 101-002R18 . 523 . 640 . 044 . 7U . W . 101-003R21 . 631 . 454 . 555 . 8U . W . 101-018R18 . 123 . 839 . 744 . 4U . W . 101-226L19 . 124 . 041 . 345 . 7U . W . 101-1136R16 . 925 . 539 . 744 . 4U . W . 101-1391R18 . 823 . 940 . 845 . 3U . W . 101-1475L18 . 829 . 046 . 549 . 7U . W . 101-1482L20 . 728 . 949 . 752 . 1Regression equations described in ‘Materials and methods’ . Mass ( a ) based on forensic statures from European individuals . Mass ( b ) based on multiple population sample . The two estimates diverge somewhat for smaller femora . 10 . 7554/eLife . 09560 . 015Figure 10 . Maximum tibia length in H . naledi and other hominins . Maximum tibia length for U . W . 101-484 , compared to other nearly complete hominin tibia specimens . Australopithecus afarensis represented by A . L . 288-1 and KSD-VP-1/1 ( Haile-Selassie et al . , 2010 ) ; Homo erectus represented by D3901 from Dmanisi and KNM-WT 15000; Homo habilis by OH 35; Homo floresiensis by LB1 and LB8 ( Brown et al . , 2004; Morwood et al . , 2005 ) . Chimpanzee and contemporary European ancestry humans from Cleveland Museum of Natural History ( Lee , 2001 ) ; Andaman Islanders from Stock ( 2013 ) . Vertical lines represent sample ranges; bars represent 1 standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 09560 . 015 The endocranial volume of all H . naledi specimens is clearly small compared to most known examples of Homo . We combined information from the most complete cranial vault specimens to arrive at an estimate of endocranial volume for both larger ( presumably male ) and smaller ( presumably female ) individuals ( larger composite depicted in Figure 11 ) . The resulting estimates of approximately 560cc and 465cc , respectively , overlap entirely with the range of endocranial volumes known for australopiths . Within the genus Homo , only the smallest specimens of H . habilis , one single H . erectus specimen , and H . floresiensis overlap with these values . 10 . 7554/eLife . 09560 . 016Figure 11 . Virtual reconstruction of the endocranium of the larger composite cranium from DH1 and DH2 overlaid with the ectocranial surfaces . ( A ) Lateral view . ( B ) Superior view . The resulting estimate of endocranial volume is 560cc . Scale bar = 10 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 09560 . 016 Despite its small vault size , the cranium of H . naledi is structurally similar to those of early Homo . Frontal bossing is evident , as is a marked degree of parietal bossing . There is no indication of metopic keeling , though there is slight parasagittal keeling between bregma and lambda , and some prelambdoidal flattening . The cranial vault bones are generally thin , becoming somewhat thicker in the occipital region . The supraorbital torus is well developed , though weakly arched , and is bounded posteriorly by a well-developed supratoral sulcus . The lateral corners of the supraorbital torus are rounded and relatively thin . The temporal lines are widely spaced , and there is no indication of a sagittal crest or temporal/nuchal cresting . The temporal crest is positioned on the posterior aspect of the lateral supraorbital torus , rather than on the superior aspect as in australopiths . At the posteroinferior extent of the temporal lines , they curve anteroinferiorly presenting a well-developed angular torus . The crania have a pentagonal outline in posterior view , with the greatest vault breadth located in the supramastoid region . The nuchal region exhibits sexually dimorphic development of nuchal muscle markings and the external occipital protuberance , and there is a clear indication of a tuberculum linearum in addition to the external occipital protuberance . In superior view the vault tapers from posterior to anterior , though post-orbital constriction is slight . The squamosal suture is low and gently curved , and parietal striae are well defined . The lateral margins of the orbits face laterally . A small zygomaticofacial foramen is typically present near the center of the zygomatic bone . The root of the zygomatic process of the maxilla is anteriorly positioned , at the level of the P3 or the P4 . There is no indication of a zygomatic prominence , and the zygomatic arches do not flare laterally to any extent . The root of the zygomatic process of the temporal is angled downwards approximately 30° relative to FH . The root of the zygomatic process of the temporal begins to laterally expand above the level of the mandibular fossa , rather than above the level of the EAM as in australopiths . The mandibular fossa is somewhat large , and moderately deep . The articular eminence of the mandibular fossa is saddle-shaped , and oriented posteroinferiorly . Almost the entire mandibular fossa is positioned medial to the temporal squama . The entoglenoid process is elongated and faces primarily laterally . The postglenoid process is small and closely appressed to the tympanic , forming part of the posterior wall of the fossa . The petrotympanic is distinctly coronally oriented . The vaginal process is small but distinct . The crista petrosa is weakly developed and not notably sharpened . There is a strong Eustachian process . The external auditory meatus is small , oval-shaped , and obliquely oriented , and a distinct suprameatal spine is present . The mastoid region is slightly laterally inflated . The mastoid process is triangular in cross-section , with a rounded apex and a mastoid crest . The digastric groove is deep and narrow , alongside a marked juxtamastoid eminence . The canine juga are weakly developed and there is no indication that anterior pillars would have been present . A shallow , ill-defined canine fossa is indicated . The nasoalveolar clivus is flat and square-shaped . The parabolic-shaped palate is broad and anteriorly shallow , becoming deeper posteriorly . The mandibular dentition of H . naledi is arranged in a parabolic arch . The alveolar and basal margins of the corpus diverge slightly . A single , posteriorly opening mental foramen is positioned slightly above the mid-corpus level , between the position of the P3 and the P4 . The mandibular corpus is relatively gracile , with a well-developed lateral prominence whose maximum extent is typically at the M2 . A slight supreme lateral torus ( of Dart ) weakly delineates the extramolar sulcus from the lateral corpus . The superior lateral torus is moderately developed , running anteriorly to the mental foramen where it turns up to reach the P3 jugum . The marginal torus is moderately developed , and defines a moderate intertoral sulcus . The posterior and anterior marginal tubercles are indicated only as slight roughenings of bone . The gracile mandibular symphysis is vertically oriented . A well-developed mental protuberance and weak lateral tubercles are delineated by a slight mandibular incisure , thereby presenting a weak mentum osseum . The post-incisive planum is steeply inclined and not-shelf-like . There is no superior transverse torus , while a weak , basally oriented inferior transverse torus is present . The anterior and posterior subalveolar fossae are continuous and deep , overhung by a well-developed alveolar prominence . The extramolar sulcus is moderately wide . The root of the ramus of the mandible originates high on the corpus at the level of the M2 . Strong ectoangular tuberosities are indicated . A large mandibular foramen is present , with a diffusely defined mylohyoid groove . Like the skull , the dentition of H . naledi compares most favorably to early Homo samples . Yet compared to samples of H . habilis , H . rudolfensis , and H . erectus , the teeth of H . naledi are comparatively quite small , similar in dimensions to much later samples of Homo . With both small post-canine teeth and a small endocranial volume , H . naledi joins Au . sediba and H . floresiensis in an area distinct from the general hominin relation of smaller post-canine teeth in species with larger brains ( Figure 12 ) . 10 . 7554/eLife . 09560 . 017Figure 12 . Brain size and tooth size in hominins . The buccolingual breadth of the first maxillary molar is shown here in comparison to endocranial volume for many hominin species . H . naledi occupies a position with relatively small molar size ( comparable to later Homo ) and relatively small endocranial volume ( comparable to australopiths ) . The range of variation within the Dinaledi sample is also fairly small , in particular in comparison to the extensive range of variation within the H . erectus sensu lato . Vertical lines represent the range of endocranial volume estimates known for each taxon; each vertical line meets the horizontal line representing M1 BL diameter at the mean for each taxon . Ranges are illustrated here instead of data points because the ranges of endocranial volume in several species are established by specimens that do not preserve first maxillary molars . DOI: http://dx . doi . org/10 . 7554/eLife . 09560 . 017 In comparison to H . habilis , H . rudolfensis , and H . erectus , the teeth of H . naledi are not only small , but also markedly simple in crown morphology . Maxillary and mandibular molars lack extensive crenulation , secondary fissures and supernumerary cusps . The M1 has an equal-sized metacone and paracone , and has a slight expression of Carabelli's trait represented by a small cusp or shallow pit . I1 exhibits slight occlusal curvature with trace marginal ridges and variably small tuberculum dentale . I2 exhibits greater occlusal curvature and tuberculum dentale expression but neither upper incisor has double shovelling or interruption groove . The mandibular canines of H . naledi have a small occlusal area , and have a distal marginal cuspule as a topographically distinct expression of the cingular margin . The P3 is double-rooted , fully bicuspid with metaconid and protoconid of approximately equal height and occlusal area separated by a distinct longitudinal groove , has a distally extensive talonid , and an occlusal outline approximately symmetrical with respect to the mesiodistal axis . P4 likewise has a distally extensive talonid and approximately symmetrical occlusal outline ( Figure 5 ) . M1 and M2 lack cusp 6 and cusp 7 , except for very slight expression in a small fraction of specimens , and have a very faint subvertical depression rather than a distinct or extensive protostylid . Like australopiths and some early Homo specimens , H . naledi has an increasing molar size gradient in the mandibular dentition ( M1 < M2 < M3 ) . The lower limb of H . naledi is defined not only by a unique combination of primitive and derived traits , but also by the presence of unique features in the femur and tibia . Like all other bipedal hominins , H . naledi possesses a valgus knee and varus ankle . The femoral neck is long , anteverted , and anteroposteriorly compressed . Muscle insertions for the M . gluteus maximus are strong and the femur has a well-marked linea aspera with pilaster variably present . The patella is relatively anteroposteriorly thick . The tibia is mediolaterally compressed with a rounded anterior border , a large proximal attachment for the M . tibialis posterior , and a thin medial malleolus . The fibula is gracile with laterally oriented lateral malleolus , a relatively circular neck and a convex surface for the proximal attachment of the M . peroneus longus . Unique features in the lower limb of H . naledi include a depression in the superior aspect of the femoral neck that results in two mediolaterally oriented pillars inferoposteriorly and superoanteriorly , and a strong distal attachment of the pes anserinus on the tibia . The foot and ankle of H . naledi are largely humanlike ( Figure 9 ) . The tibia stands orthogonally upon the talus , which is moderately wedged , with a mediolaterally flat trochlea having medial and lateral margins at even height , a form distinct from the strong keeling seen in OH 8 ( H . habilis ) and several tali from Koobi Fora . The talar head and neck exhibit strong , humanlike torsion; the horizontal angle is higher than in most humans , similar to that found in australopiths . The calcaneus is only moderately robust , but possesses the plantar declination of the retrotrochlear eminence and plantarly positioned lateral plantar process found in both modern humans and Au . afarensis . The peroneal trochlea is small , unlike that found in australopiths and similar only to that in H . sapiens and Neanderthals . The talonavicular , subtalar joints and calcaneocuboid joints are humanlike in possessing minimal ranges of motion and evidence for a locking , rigid midfoot . The intermediate and lateral cuneiforms are proximodistally elongated . The hallucal tarsometatarsal joint is flat and proximodistally aligned indicating that H . naledi possessed an adducted , non-grasping hallux . The head of the first metatarsal is mediolaterally expanded dorsally , indicative of a humanlike windlass mechanism . The foot possesses humanlike metatarsal lengths , head proportions , and torsion . The phalanges are moderately curved , slightly more so than in H . sapiens . The only primitive anatomies found in the foot of H . naledi are the talar head and neck declination and sustentaculum tali angles , suggestive of a lower arched foot with a more plantarly positioned and horizontally inclined medial column than typically found in modern humans ( Harcourt-Smith et al . , 2015 ) . The axial skeleton presents a combination of derived ( mainly aspects of the vertebrae ) and seemingly primitive ( mainly the ribs ) traits . The preserved adult T10 and T11 vertebrae are proportioned similarly to Pleistocene Homo , with transverse process morphology most similar to Neandertals . The neural canals of these vertebrae are large in comparison to the diminutive overall size of the vertebrae , proportionally recalling Dmanisi H . erectus , Neandertals , and modern humans . The 11th rib is straight ( uncurved ) , similar to Au . afarensis , and the shape of the upper rib cage appears narrow , as assessed from first and second rib fragments , suggesting that the thorax was pyramidal in shape . The 12th rib presents a robust shaft cross-section most similar to Neandertals . This combination is not found in other hominins and might reflect allometric scaling at a small trunk size . The Dinaledi iliac blade is flared and shortened anteroposteriorly , resembling Au . afarensis or Au . africanus . The ischium is short with a narrow tuberoacetabular sulcus , and the ischiopubic and iliopubic rami are thick , resembling Au . sediba and H . erectus . This combination of iliac and ischiopubic features has not been found in other fossil hominins ( Figure 13 ) . 10 . 7554/eLife . 09560 . 018Figure 13 . Selected pelvic specimens of H . naledi . U . W . 101-1100 ilium in ( A ) lateral view showing a weak iliac pillar relatively near the anterior edge of the ilium , with no cristal tubercle development; ( B ) anterior view , angled to demonstrate the degree of flare , which is clear in comparison to the subarcuate surface . U . W . 101-723 immature sacrum in ( C ) anterior view; and ( D ) superior view . U . W . 101-1112 ischium in ( E ) lateral view; and ( F ) anterior view , demonstrating relatively short tuberacetabular diameter . Scale bar = 2 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 09560 . 018 The shoulder of H . naledi is configured with the scapula situated high and lateral on the thorax , short clavicles , and little or no torsion of the humerus . The humerus is notably slender for its length , with prominent greater and lesser tubercles bounding a deep bicipital groove , with a small , non-projecting humeral deltoid tuberosity and brachioradialis crest . Distally , the humerus has a wide lateral distodorsal pillar and narrow medial distodorsal pillar , and a medially-displaced olecranon fossa with septal aperture . The Dinaledi radius and ulna diaphyses exhibit little curvature . The radius has a globular radial tuberosity , prominent pronator quadratus crest , and reduced styloid process . The hand shares many derived features of modern humans and Neandertals in the thumb , wrist , and palm , but has relatively long and markedly curved fingers ( Kivell et al . , 2015 ) . The thumb is long relative to the length of the other digits , and includes a robust metacarpal with well-developed intrinsic ( M . opponens pollicis and M . first dorsal interosseous ) muscle attachments ( Figure 6 ) . The pollical distal phalanx is large and robust with a well-developed ridge along the distal border of a deep proximal palmar fossa for the attachment of flexor pollicis longus tendon . Ungual spines also project proximopalmarly from a radioulnarly expanded apical tuft with a distinct area for the ungual fossa . The wrist includes a boot-shaped trapezoid with an expanded non-articular palmar surface , an enlarged and palmarly-expanded trapezoid-capitate joint , and a trapezium-scaphoid joint that extends further onto the scaphoid tubercle . Overall , carpal shapes and articular configurations are very similar to those of modern humans and Neandertals , and unlike those of great apes and other extinct hominins . However , the H . naledi wrist lacks a third metacarpal styloid process , has a more radioulnarly oriented capitate-Mc2 joint , and has a relatively small trapezium-Mc1 joint compared to humans and Neandertals . Moreover , the phalanges are long ( relative to the palm ) and more curved than most australopiths . The overall morphology of H . naledi places it within the genus Homo rather than Australopithecus or other early hominin genera . The shared derived features that connect H . naledi with other members of Homo occupy most regions of the H . naledi skeleton and represent distinct functional systems , including locomotion , manipulation , and mastication . Locomotor traits shared with Homo include the absolutely long lower limb , with well-marked linea aspera , strong M . gluteus maximus insertions , gracile fibula and generally humanlike ankle and foot . These aspects of the lower limb suggest enhanced locomotor performance for a striding gait . The H . naledi hand shares aspects of Homo morphology in the wrist , thumb and palm , pointing to enhanced object manipulation ability relative to australopiths , including Au . sediba ( Kivell et al . , 2011; Kivell et al . , 2015 ) . H . naledi lacks the powerful mastication that typifies Australopithecus and Paranthropus , with generally small teeth across the dentition , gracile mandibular corpus and symphysis , laterally-positioned temporal lines , slight postorbital constriction and non-flaring zygomatic arches . The upper limb , shoulder and ribcage have a more primitive morphological pattern , but do not preclude affiliating H . naledi with Homo , particularly considering that postcranial remains of H . habilis appear to reflect an australopith-like body plan ( Johanson et al . , 1986 ) . Locomotor , manipulatory , and masticatory systems have both historical and current importance in defining Homo ( Wood and Collard , 1999; Holliday , 2012; Antón et al . , 2014 ) , and H . naledi fits within our genus in these respects . The structural configuration of the H . naledi cranium , beyond the functional aspects of mastication , is likewise shared with Homo . As in many specimens of H . erectus and H . habilis , the H . naledi vault includes a well-developed and moderately arched supraorbital torus , marked from the frontal squama by a continuous supratoral sulcus , frontal bossing . Further , as in many H . erectus crania , H . naledi exhibits a marked angular torus and occipital torus . The H . naledi face includes a flat and squared nasoalveolar clivus , comparable to H . rudolfensis ( Leakey et al . , 2012 ) , and weak canine fossae . While its anatomy places it unambiguously within Homo , the H . naledi cranium and dentition lack many derived features shared by MP and LP Homo and H . sapiens . The australopith-like features of the postcranium , including the ribcage , shoulder , proximal femur , and relatively long , curved fingers , also depart sharply from the morphology present in MP humans and H . sapiens . The similarities of H . naledi to earlier members of Homo , including H . habilis , H . rudolfensis , and H . erectus , suggest that this species may be rooted within the initial origin and diversification of our genus . The fossil record of early Homo and Homo-like australopiths has rapidly increased during the last 15 years , and this accumulating evidence has changed our perspective on the rise of our genus . Many skeletal and behavioral features observed to separate later Homo from earlier hominins were formerly argued to have arisen as a single adaptive package , including increased brain size , tool manipulation , increased body size , smaller dentition , and greater commitment to terrestrial long-distance walking or running ( Wood and Collard , 1999; Hawks et al . , 2000 ) . But we now recognize that such features appeared in different combinations in different fossil samples ( Antón et al . , 2014 ) . The Dmanisi postcranial sample ( Lordkipanidze et al . , 2007 ) and additional cranial remains of H . erectus from Dmanisi ( Gabunia et al . , 2000; Vekua et al . , 2002; Lordkipanidze et al . , 2013 ) and East Africa ( Spoor et al . , 2007; Leakey et al . , 2012 ) , demonstrate that larger brain size and body size did not arise synchronously with improved locomotor efficiency and adaptations to long-distance walking or running in H . erectus ( Holliday , 2012; Antón et al . , 2014 ) . Further , the discovery of Au . sediba showed that a mosaic of Homo-like hand , pelvis and aspects of craniodental morphology can occur within a species with primitive body size , limb proportions , lower limb and foot morphology , thorax shape , vertebral morphology , and brain size ( Berger et al . , 2010; Carlson et al . , 2011; Kivell et al . , 2011; Churchill et al . , 2013; DeSilva et al . , 2013; Schmid et al . , 2013 ) . H . naledi presents yet a different combination of traits . This species combines a humanlike body size and stature with an australopith-sized brain; features of the shoulder and hand apparently well-suited for climbing with humanlike hand and wrist adaptations for manipulation; australopith-like hip mechanics with humanlike terrestrial adaptations of the foot and lower limb; small dentition with primitive dental proportions . In light of this evidence from complete skeletal samples , we must abandon the expectation that any small fragment of the anatomy can provide singular insight about the evolutionary relationships of fossil hominins . A recent phylogenetic analysis of fossil hominins based on craniodental morphology placed Au . sediba at the base of the genus Homo ( Dembo et al . , 2015 ) , in agreement with earlier analyses of this species ( Berger et al . , 2010 ) . The cranial and dental affinities identified between Au . sediba and Homo include many features shared by H . naledi . But H . naledi and Au . sediba share different postcranial features with other species of Homo . Resolving the phylogenetic placement of H . naledi will require both postcranial and craniodental evidence to be integrated together . Such integration poses a challenge because of the poor representation of several key species both within and outside of Homo , most notably H . habilis , for which postcranial evidence is slight , and H . rudolfensis for which no associated postcranial remains are known . We propose the testable hypothesis that the common ancestor of H . naledi , H . erectus , and H . sapiens shared humanlike manipulatory capabilities and terrestrial bipedality , with hands and feet like H . naledi , an australopith-like pelvis and the H . erectus-like aspects of cranial morphology that are found in H . naledi . Enlarged brain size was evidently not a necessary prerequisite for the generally human-like aspects of manipulatory , locomotor , and masticatory morphology of H . naledi . Although it contains an unprecedented wealth of anatomical information , the Dinaledi deposit remains undated ( Dirks et al . , 2015 ) . Considering that H . naledi is a morphologically primitive species within our genus , an age may help elucidate the ecological circumstances within which Homo arose and diversified . If the fossils prove to be substantially older than 2 million years , H . naledi would be the earliest example of our genus that is more than a single isolated fragment . The sample would illustrate a model for the relation of adaptive features of the cranium , dentition and postcranium during a critical time interval that is underrepresented by fossil evidence of comparable completeness . A date younger than 1 million years ago would demonstrate the coexistence of multiple Homo morphs in Africa , including this small-brained form , into the later periods of human evolution . The persistence of such a species with clear adaptations for manipulation and grip , alongside MP humans or perhaps even alongside modern humans , would challenge many assumptions about the development of the archaeological record in Africa . The depth of evidence of H . naledi may provide a perspective on the variation to be expected within fossil hominin taxa ( Lordkipanidze et al . , 2013; Bermúdez de Castro et al . , 2014 ) . The entire Dinaledi collection is remarkably homogeneous . There is very little size variation among adult elements within the collection . Eight body mass estimates from the femur ( Table 2 ) have a standard deviation of only 4 . 3 kilograms , for a body mass coefficient of variation ( CV ) of only 9% . The CV of body mass within most human populations is substantially higher than this , with an average near 15% ( McKellar and Hendry , 2009 ) . Likewise , the size variation of cranial and dental elements is minimal . With 11 mandibular first molars , the CV of buccolingual breadth is only 3 . 2% and for 13 maxillary first molars the CV of buccolingual breadth is only 2 . 0% ( buccolingual breadth is used because it is not subject to variance from interproximal wear ) . Not only size , but also anatomical shape and form are homogeneous within the sample . Almost every aspect of the morphology of the dentition , including the distinctive form of the lower premolars , the distal accessory cuspule of the mandibular canines , and the expression of nonmetric features that normally vary in human populations , is uniform in every specimen from the collection . The distinctive aspects of cranial morphology are repeated in every specimen , and even the aspects that normally vary among individuals of different body size or between sexes exhibit only slight variation among the Dinaledi remains . One of the most unique aspects of H . naledi is the morphology of the first metacarpal; the derived aspects of this anatomy are present in every one of seven first metacarpal specimens in the collection ( Figure 14 ) . Unlike any other fossil hominin site in Africa , the Dinaledi Chamber seems to preserve a large number of individuals from a single population , one with variation equal to or less than that found within local populations of modern humans . 10 . 7554/eLife . 09560 . 004Figure 14 . First metacarpals of H . naledi . Seven first metacarpals have been recovered from the Dinaledi Chamber . U . W . 101-1321 is the right first metacarpal of the associated Hand 1 found in articulation . U . W . 101-1282 and U . W . 101-1641 are anatomically similar left and right first metacarpals , which we hypothesize as antimeres , both were recovered from excavation . U . W . 101-007 was collected from the surface of the chamber , and exhibits the same distinctive morphological characteristics as all the first metacarpals in the assemblage . All of these show a marked robusticity of the distal half of the bone , a very narrow , ‘waisted’ appearance to the proximal shaft and proximal articular surface , prominent crests for attachment of M . opponens pollicis and M . first dorsal interosseous , and a prominent ridge running down the palmar aspect of the bone . The heads of these metacarpals are dorsopalmarly flat and strongly asymmetric , with an enlarged palmar-radial protuberance . These distinctive features are present among all the first metacarpals in the Dinaledi collection , and are absent from any other hominin sample . Their derived nature is evident in comparison to apes and other early hominins , here illustrated with a chimpanzee first metacarpal and the MH2 first metacarpal of Australopithecus sediba . DOI: http://dx . doi . org/10 . 7554/eLife . 09560 . 004 The Dinaledi collection is the richest assemblage of associated fossil hominins ever discovered in Africa , and aside from the Sima de los Huesos collection and later Neanderthal and modern human samples , it has the most comprehensive representation of skeletal elements across the lifespan , and from multiple individuals , in the hominin fossil record . The abundance of evidence from this assemblage supports our emerging understanding that the genus Homo encompassed a variety of evolutionary experiments ( Antón et al . , 2014 ) , with diversity now evident for fossil Homo in each of the few intensively explored parts of Africa ( Leakey et al . , 2012 ) . But as much as it advances our knowledge , H . naledi also highlights our ignorance about ancient Homo across the vast geographic span of the African continent . The tree of Homo-like hominins is far from complete: we have missed key transitional forms and lineages that persisted for hundreds of thousands of years . With an increasing pace of discovery from the field and the laboratory , more light will be thrown on the origin of humans . In the differential diagnosis of H . naledi , we have compared the holotype DH1 , paratypes , and other referred material to fossil evidence from previously-identified hominin taxa . Our goal is to provide a diagnosis for H . naledi that is clear in reference to widely recognized hominin hypodigms . Different specialists continue to disagree about the composition and anatomical breadth represented by these hominin taxa and attribution of particular specimens to them ( see e . g . , Wood and Collard , 1999; Lordkipanidze et al . , 2013; Antón et al . , 2014 on early Homo taxa ) . We do not intend to take any position on such disagreements by our selection of comparative samples for H . naledi . We have been cautious in our attribution of postcranial specimens to hominin taxa , particularly in the African Plio-Pleistocene , where it has been demonstrated multiple hominin taxa coexisted in time , if not in geographical space . Because the purpose of this study is differential diagnosis in reference to known taxa , unattributed specimens are not germane , although in certain cases there are well-accepted attributions to genus for specimens ( e . g . , Homo sp . or Australopithecus sp . ) as cited below . We have included some specimens in comparisons because they are relatively complete , even if they cannot be attributed to a species , because few hominin taxa are represented by evidence across the entire skeleton . For some anatomical characters , parts are preserved only for MP or later hominin samples , so we have included such comparisons to make clear how H . naledi compares in these elements to the ( few ) known fossil examples . This study relies upon observations and measurements taken from original fossils by the authors , observations taken from casts , and observations taken from the literature . These observations are in large part standard anatomical practice; where features are specially described in previous studies we have referenced those here . For this study , a cast collection was assembled including the Phillip V . Tobias research collection at the University of the Witwatersrand and loans of cast materials from the University of Wisconsin–Madison , University of Michigan , American Museum of Natural History , New York University , University of Colorado–Denver , University of Delaware , Texas A&M University , and the personal collections of Peter Schmid , Milford Wolpoff and Rob Blumenschine . We extend our gratitude to the curators of fossil collections and the generosity of these institutions in facilitating this research , both in South Africa and throughout the world . This list of skeletal materials extends the list of craniodental comparative material used in diagnosing Au . sediba , with many of the hypodigms identical to that study ( Berger et al . , 2010 ) . Where we have had first-hand access to original specimens , we rely upon our own observations; we therefore do not refer readers to other sources for these data . The calvariae ( DH1-4 ) were scanned using a NextEngine laser surface scanner ( NextEngine , Malibu , CA ) at the following settings: Macro , 12 divisions with auto-rotation , HD 17k ppi . Depending on the complexity of the surface relief , either two or three complete scanning cycles were completed per specimen , resulting in multiple 360° scans . Each individual scan was trimmed , aligned , and fused ( volume merged ) in the accompanying ScanStudio HD Pro software . For each specimen , the individual 360° scans were then aligned and merged in GeoMagic Studio 14 . 0 ( Raindrop Geomagic , Research Triangle Park , NC ) , creating a final three-dimensional model of the specimen . Given the fragmented nature of the calvariae specimens , both the ectocranial and endocranial surfaces were captured in the scans . DH3 consisted primarily of portions of the right calvaria . However , a small section of the frontal and the parietal crossed the mid–sagittal plane . For this reason , it was possible to mirror image the surface scan to approximate the left calvaria and obtain a more complete visualization of the complete calvaria ( Figure 15 ) . The virtual specimen of DH3 was mirrored in GeoMagic Studio , and manually registered ( aligned ) using common points along the frontal crest and sagittal suture . The registration procedure in GeoMagic Studio is an iterative process that refines the alignment of specimens to minimize spatial differences between corresponding surfaces . In this manner , the program is able to match the position overlapping surfaces , in addition to their angulation and curvature . 10 . 7554/eLife . 09560 . 020Figure 15 . Posterior view of the virtual reconstruction of DH3 . The resultant mirror image is displayed in blue . The antimeres were aligned by the frontal crest and sagittal suture using the Manual Registration function in GeoMagic Studio 14 . 0 . DOI: http://dx . doi . org/10 . 7554/eLife . 09560 . 020 The same procedures were used to mirror image and create a virtual reconstruction of DH2 and the occipital portion of DH1 ( Figure 16 ) . The occipital and vault portions of DH1 were reconstructed based on the anatomical alignment of the sagittal suture , sagittal sulcus , parietal striae , and the continuation of the temporal lines across both the specimens . 10 . 7554/eLife . 09560 . 021Figure 16 . Virtual reconstruction of ( A ) DH2 and ( B ) occipital portion of DH1 . The actual specimen displays its original coloration and the mirror imaged portion is illustrated in blue . DOI: http://dx . doi . org/10 . 7554/eLife . 09560 . 021 In order to virtually estimate the cranial capacity , composite crania were constructed from the surface scans and mirror imaged scans of the calvariae . Two separate composite crania were created; the relatively smaller-sized calvariae ( DH3 and DH4 ) were combined into one composite , and the larger-sized calvariae ( DH1 and DH2 ) composed the larger composite cranium . The smaller composite cranium , DH3 was mirrored in GeoMagic Studio 14 . 0 , and merged with the original scan as outlined above . The surface scan of DH4 was uploaded and registered ( aligned ) to the DH3 model using overlapping temporal features ( e . g . , the external auditory meatus ) . No scaling was performed . DH4 was then mirror imaged to complete the occipital contour . The resultant model suggests a general concordance between the specimens in both size and shape with a close alignment of vault surfaces and anatomical features between specimens ( Figure 17 ) . 10 . 7554/eLife . 09560 . 022Figure 17 . Postero-lateral view of the virtual reconstruction of a composite cranium from DH3 and DH4 . ( A ) The surface scan of DH3 was mirror imaged and merged as described in Supplementary Note 8 . ( B ) The scan of DH4 was aligned to the DH3 model . ( C ) DH4 was then mirror imaged to complete the occipital contour ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09560 . 022 For the larger composite cranium , the surface model of DH2 and its mirror image was then uploaded , registered ( aligned ) , and merged with the mirror-imaged model of DH1 . No scaling was performed . The congruency between the specimens in the resultant model suggests that DH1 and DH2 are similar in both size and vault shape ( Figure 18 ) . 10 . 7554/eLife . 09560 . 023Figure 18 . Virtual reconstruction of a composite cranium from DH1 and DH2 . The surface model of DH2 ( blue ) , consisting of the original scan merged with the mirror image , was then uploaded and aligned with the mirror-imaged DH1 model ( pink ) . Note the similarity in size and shape between DH1 and DH2 observed in the posterior ( A ) anterior ( B ) lateral ( C ) and superior ( D ) views . DOI: http://dx . doi . org/10 . 7554/eLife . 09560 . 023 The composite model of DH3 and DH4 was used to estimate the cranial capacity for the smaller morphotype . In GeoMagic Studio 14 . 0 , the endocranial surface of the composite was carefully selected from the ectocranial surface and copied as a new object . In order to obtain a volume calculation the model has to be a closed surface , meaning that all of the holes in the surface model had to be filled . Small holes in the model were filled using the ‘Fill by Curvature’ function . Larger holes were filled in by sections . For example , the cranial base was filled in using a number of transverse sections , so that in the absence of the cranial base the contour of the various cranial fossae and the petrous portions of the temporal could be preserved as best as possible . When appropriate ( e . g . , around angular portions of the petrous bone ) , small sections were filled using a flat hole filling function . The new surfaces created by the hole-filling mechanism were carefully monitored and repeated until an acceptable model that appeared to best approximate the missing portions was obtained . The result is a closed model approximation of the endocranium , of which a volume can be calculated by GeoMagic Studio ( Figure 19 , Figure 20 ) . The volume of the smaller composite cranium ( DH3 and DH4 ) indicates a cranial capacity of approximately 465 cm3 . 10 . 7554/eLife . 09560 . 024Figure 19 . Virtual reconstruction of the endocranium of the composite cranium from DH3 and DH4 . ( A ) Lateral view . ( B ) Superior view . ( C ) Inferior view . In all views , anterior is to towards the left . DOI: http://dx . doi . org/10 . 7554/eLife . 09560 . 02410 . 7554/eLife . 09560 . 025Figure 20 . Virtual reconstruction of the endocranium of the composite cranium from DH3 and DH4 overlaid with the ectocranial surfaces . ( A ) Lateral view . ( B ) Superior view . DOI: http://dx . doi . org/10 . 7554/eLife . 09560 . 025 In order to determine whether significant errors were being introduced in the manner that the cranial base was filled in the above procedures , the endocranial volume of DH3/DH4 was also virtually calculated using the cranial base of Sts 19 as a model . A 3D model of Sts 19 was mirrored and aligned to the DH3/DH4 model using the external auditory meatus and common points on the internal surface of the petrous portion as a guide ( Figure 21 ) . The Sts 19 model was then scaled by 0 . 97 to obtain an optimal fit between the two models . 10 . 7554/eLife . 09560 . 026Figure 21 . Virtual reconstruction the DH3/DH4 cranial base using a model of Sts 19 . ( A ) Right lateral view . ( B ) Left lateral view . ( C ) Posterior view . ( D ) Inferior view . DOI: http://dx . doi . org/10 . 7554/eLife . 09560 . 026 After the Sts 19 model was merged with the DH3/DH4 model , the endocranial surface was extracted and reconstructed as described above ( Figure 22 ) . The resultant endocranial volume using the Sts 19 cranial base was 465 . 9 cm3 . This value is in agreement with the first estimate and suggests that using a model cranial base did not significantly alter the results . 10 . 7554/eLife . 09560 . 027Figure 22 . Virtual reconstruction the DH3/DH4 endocranial volume using a cranial base model of Sts 19 . Right lateral view . DOI: http://dx . doi . org/10 . 7554/eLife . 09560 . 027 The larger composite cranium , consisting of DH1 and DH2 , lacks most of the frontal region . In order to create a closed endocranial surface for a volume estimate , the frontal region from the smaller composite cranium was scaled by 5% , and then registered ( aligned ) and merged to the model of the larger composite cranium . As with the smaller composite cranium , the endocranial surface was then selected and converted to a new object , and the remaining holes filled based on the curvature of the surface . The volume of the closed endocranial model was calculated using GeoMagic Studio . The cranial capacity ( endocranial volume ) of the larger composite model is approximately 560cc . Eight femoral fragments from the Dinaledi collection allow a direct measurement of the subtrochanteric anteroposterior and mediolateral diameters ( Table 3 ) . We developed two regression equations to estimate body mass from these diameters based on the masses of modern human samples . MCE measured body masses of a sample of 253 modern European individuals , 128 males and 125 females , collected from the Institute for Forensic Medicine in Zurich , Switzerland . Body masses were taken at time of forensic evaluation . This sample yields the following regression equation relating body mass to subtrochanteric diameter , where FSTpr refers to the product of the femoral subtrochanteric mediolateral and anteroposterior breadths:Body Mass=0 . 060×FSTpr+13 . 856 , SEE=6 . 78 , r=0 . 50 , p=<0 . 001 . We further examined a broader sample of 276 modern humans taken from a number of populations around the world , with data measured by TWH . The body masses of individuals were estimated from femur head diameter , using the average of results obtained from Grine et al . ( 1995 ) and Ruff et al . ( 1997 ) . The sample includes 115 females , 155 males , and 6 individuals of indeterminate sex . Body Mass=0 . 046×FSTpr+24 . 614 , SEE=5 . 82 , r=0 . 82 , p<0 . 001 . We collected data from skeletal material representing two African population samples . We use only African populations in this comparison because the ratio of tibia length to femur length , and thereby the proportion of stature constituted by tibia length , varies between human populations both today and prehistorically . Although we do not know this proportion for H . naledi , we adopt the null hypothesis that they likely had tibia/femur proportions similar to other African population samples . 95 male and female Kulubnarti individuals from medieval Nubia are curated at the University of Colorado , Boulder . Data were collected by HMG , including estimates of living stature based on the Fully method ( Fully , 1956; Raxter et al . , 2006 ) , and these were used to develop a regression equation relating tibia length to stature . The resulting equation is:Stature=0 . 295×TML+48 . 589 , SEE=3 . 13 , r=0 . 90 , p<0 . 001 . We ( HMG and TWH ) collected measurements from 38 African males and 38 females curated within the Dart Collection of the University of the Witwatersrand . Specimens were randomly chosen with no preference for specific African ethnic groups . Cadaveric statures are documented for this collection , the regression equation relating tibia length to stature in this sample is:Stature=0 . 223×TML+75 . 350 , SEE=6 . 50 , r=0 . 63 , p<0 . 001 . The electronic edition of this article conforms to the requirements of the amended International Code of Zoological Nomenclature , and hence , the new name contained herein is available under that Code from the electronic edition of this article . This published work and the nomenclatural acts it contains have been registered in ZooBank , the online registration system for the ICZN . The ZooBank LSIDs ( Life Science Identifiers ) can be resolved and the associated information viewed through any standard web browser by appending the LSID to the prefix ‘http://zoobank . org/’ . The LSID for this publication is: urn:lsid:zoobank . org:pub:00D1E81A-6E08-4A01-BD98-79A2CEAE2411 . The electronic edition of this work was published in a journal with an ISSN ( 2050-084X ) and has been archived and is available from the following digital repositories: PubMed Central and LOCKSS . All Dinaledi fossil material is available for study by researchers upon application to the Evolutionary Studies Institute at the University of the Witwatersrand where the material is curated ( contact Bernhard Zipfel [Bernhard . Zipfel@wits . ac . za] ) . Three-dimensional surface renderings and other digital data are available from the MorphoSource digital repository ( http://morphosource . org ) .	Modern humans , or Homo sapiens , are now the only living species in their genus . But as recently as 100 , 000 years ago , there were several other species that belonged to the genus Homo . Together with modern humans , these extinct human species , our immediate ancestors and their close relatives , are collectively referred to as ‘hominins’ . Now Berger et al . report the recent discovery of an extinct species from the genus Homo that was unearthed from deep underground in what has been named the Dinaledi Chamber , in the Rising Star cave system in South Africa . The species was named Homo naledi; ‘naledi’ means ‘star’ in Sotho ( also called Sesotho ) , which is one of the languages spoken in South Africa . The unearthed fossils were from at least 15 individuals and include multiple examples of most of the bones in the skeleton . Based on this wide range of specimens from a single site , Berger et al . describe Homo naledi as being similar in size and weight to a small modern human , with human-like hands and feet . Furthermore , while the skull had several unique features , it had a small braincase that was most similar in size to other early hominin species that lived between four million and two million years ago . Homo naledi's ribcage , shoulders and pelvis also more closely resembled those of earlier hominin species than those of modern humans . The Homo naledi fossils are the largest collection of a single species of hominin that has been discovered in Africa so far and , in a related study , Dirks et al . describe the setting and context for these fossils . However , since the age of the fossils remains unclear , one of the next challenges will be to date the remains to provide more information about the early evolution of humans and their close relatives .	[ "Abstract", "Introduction", "Differential", "diagnosis", "Description", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology" ]	2015	Homo naledi, a new species of the genus Homo from the Dinaledi Chamber, South Africa
Insulin and IGF signaling are critical to numerous developmental and physiological processes , with perturbations being pathognomonic of various diseases , including diabetes . Although the functional roles of the respective signaling pathways have been extensively studied , the control of insulin production and release is only partially understood . Herein , we show that in Drosophila expression of insulin-like peptides is regulated by neprilysin activity . Concomitant phenotypes of altered neprilysin expression included impaired food intake , reduced body size , and characteristic changes in the metabolite composition . Ectopic expression of a catalytically inactive mutant did not elicit any of the phenotypes , which confirms abnormal peptide hydrolysis as a causative factor . A screen for corresponding substrates of the neprilysin identified distinct peptides that regulate insulin-like peptide expression , feeding behavior , or both . The high functional conservation of neprilysins and their substrates renders the characterized principles applicable to numerous species , including higher eukaryotes and humans . Neprilysins are highly conserved ectoenzymes that cleave and thereby inactivate many physiologically relevant peptides in the extracellular space , thus contributing considerably to the maintenance of peptide homeostasis in this compartment . Members of the neprilysin family generally consist of a short N-terminal cytoplasmic domain , a membrane spanning region , and a large extracellular domain with two highly conserved sequence motifs ( HExxH; ExxA/GD ) critical for zinc coordination , catalysis , and substrate or inhibitor binding ( Matthews , 1988; Oefner et al . , 2000 ) . Because of these characteristics , neprilysins are classified as M13 zinc metallopeptidases . For human Neprilysin ( NEP ) , the most well-characterized family member , identified substrates include endothelins , angiotensins I and II , enkephalins , bradykinin , atrial natriuretic peptide , substance P , and the amyloid-beta peptide ( Turner et al . , 2001 ) . Because of this high substrate variability , NEP activity has been implicated in the pathogenesis of hypertension ( Molinaro et al . , 2002 ) , analgesia ( Whitworth , 2003 ) , cancer ( Turner et al . , 2001 ) , and Alzheimer’s disease ( Iwata et al . , 2000; Belyaev et al . , 2009 ) . Recent clinical trials have demonstrated significant efficacy of Neprilysin inhibitors in the treatment of certain indications ( Jessup , 2014; McMurray et al . , 2014 ) . However , despite the clinical relevance of the neprilysins , the physiological function and in vivo substrates of most family members are unknown . In Drosophila melanogaster , at least five neprilysin genes are expressed ( Meyer et al . , 2011; Sitnik et al . , 2014 ) , two of the corresponding protein products , Nep2 and Nep4 , were reported to be enzymatically active ( Bland et al . , 2007; Meyer et al . , 2009; Thomas et al . , 2005 ) . With respect to Nep4 , a critical function of the enzyme's non-catalytic intracellular N-terminus has been demonstrated: when present in excess , the domain induces severe muscle degeneration concomitant with lethality during late larval development . Because the intracellular domain interacts with a carbohydrate kinase , impaired energy metabolism has been proposed as the underlying cause of the phenotype ( Panz et al . , 2012 ) . In addition , Nep2 has been implicated in the regulation of locomotion and geotactic behavior ( Bland et al . , 2009 ) , and neprilysin activity in general appears to be critical to the formation of middle- and long-term memory ( Turrel et al . , 2016 ) , as well as to the regulation of pigment dispersing factor ( PDF ) signaling within circadian neural circuits ( Isaac et al . , 2007 ) . However , despite these experiments and recent findings that suggest a critical role of neprilysins in reproduction ( Sitnik et al . , 2014 ) , the physiological functionality of these enzymes is still far from being understood . In this respect , the lack of identified substrates with in vivo relevance is a major hindrance . Herein , we describe the identification of numerous novel substrates of Drosophila Neprilysin 4 ( Nep4 ) and provide evidence that Nep4-mediated peptide hydrolysis regulates insulin-like peptide ( ILP ) expression and food intake . These results establish a correlation between neprilysin activity and ILP expression and thus clarify our understanding of the complex mechanisms that control the production and release of these essential peptides . In previous experiments , we showed that Nep4 is expressed in larval body wall muscles and that increased expression of the peptidase in this tissue interferes with muscle function and integrity and severely impairs movement of the larvae ( Panz et al . , 2012 ) . In the present study , we found that an increase in Nep4A in muscle cells ( mef2-Gal4 driver ) also induced biphasic lethality . An early phase occurred throughout embryonic and early larval development , and a late phase was evident by the end of larval development ( Figure 1A ) . Significantly , early lethality was observed only upon overexpression of the active enzyme; expression of catalytically inactive Nep4A , carrying a glutamine instead of an essential glutamate ( E873Q ) within the zinc-binding motif , did not affect viability at this point of development . By contrast , overexpression of catalytically active or inactive Nep4A constructs induced late larval lethality . These distinct effects demonstrate that lethality during early development is caused exclusively by a detrimental increase in catalytic activity , whereas late larval lethality appears to be a consequence of multiple physiological impairments . Comparable overexpression levels of the wild-type enzyme and the mutated construct were demonstrated previously ( Panz et al . , 2012 ) . Muscle-specific knockdown of nep4 slightly increased embryonic mortality , but the majority of the respective animals died during metamorphosis ( Figure 1A ) . To confirm RNAi specificity , we also analyzed flies expressing both the respective RNAi construct as well as the Nep4A overexpression construct . Simultaneous overexpression of Nep4A completely rescued the RNAi phenotypes ( embryonic/pupal lethality ) , thus confirming specificity of the knockdown ( Figure 1A ) . The result that respective animals exhibited a marginally , yet significantly increased lethality rate during third instar larval stage indicates that overexpression of Nep4A is somewhat more effective than knockdown , eventually resulting in slightly increased expression levels of the peptidase , which , as depicted above , result in elevated larval lethality . 10 . 7554/eLife . 19430 . 003Figure 1 . Modulating nep4 expression affects life span and body size . ( A ) Lethality assay . The percentages ( % ) of animals of a specific stage that did not develop into the next stage are shown . While muscle-specific overexpression of Nep4A ( mef2-Gal4 x UAS-Nep4A ) led to biphasic lethality with critical phases during embryonic and late larval development , overexpression of catalytically inactive Nep4A in the same tissue ( mef2-Gal4 x UAS-Nep4Ainact ) led to lethality only in the third instar larval stage . Muscle-specific nep4 knockdown ( mef2-Gal4 x UAS-nep4 RNAi ) slightly increased embryonic lethality , but the majority of the animals died as pupae . Glial cell-specific overexpression ( repo-Gal4 x UAS-Nep4A ) or knockdown of the peptidase ( repo-Gal4 x UAS-nep4 RNAi ) did not affect life span , which was also observed for neuronal overexpression or knockdown ( elav-Gal4 x UAS-Nep4A; elav-Gal4 x UAS-nep4 RNAi ) . mef2-Gal4 x w1118 , repo-Gal4 x w1118 , elav-Gal4 x w1118 , UAS-Nep4A x w1118 , UAS-nep4 RNAi x w1118 , and w1118 were used as controls . Asterisks indicate statistically significant deviations from the respective controls ( p<0 . 05 , p<0 . 01 , one-way ANOVA with pairwise comparisons ) . ( B ) Size and weight measurements . While muscle-specific overexpression of Nep4A ( mef2-Gal4 x UAS-Nep4A ) reduced the size and wet mass of third instar larvae , neither overexpression of catalytically inactive Nep4A in the same tissue ( mef2-Gal4 x UAS-Nep4Ainact ) nor muscle-specific nep4 knockdown ( mef2-Gal4 x UAS-nep4 RNAi ) significantly affected these parameters . Glial cell-specific overexpression of the peptidase ( repo-Gal4 x UAS-Nep4A ) did not alter size or weight , while downregulation of the peptidase in the same tissue ( repo-Gal4 x UAS-nep4 RNAi ) slightly , but significantly , reduced both parameters . Neuronal overexpression or knockdown of nep4 ( elav-Gal4 x UAS-Nep4A; elav-Gal4 x UAS-nep4 RNAi ) had no effect on size or weight . Control lines were the same as in A . Asterisks indicate statistically significant deviations from respective controls ( p<0 . 05 , *p<0 . 01 , one-way ANOVA with pairwise comparisons ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19430 . 00310 . 7554/eLife . 19430 . 004Figure 1—source data 1 . Lethality assay . Depicted are the percentages of animals of a specific stage that do not develop into the next stage . valuesmarked in bold indicate statistically significant deviations from the respective controls . DOI: http://dx . doi . org/10 . 7554/eLife . 19430 . 00410 . 7554/eLife . 19430 . 005Figure 1—source data 2 . Size and weight measurements . 3rd instar larvae where grouped into genotype-specific cohorts of 10 individuals . The weights of at least five cohorts per genotype were averaged to calculate the mean weight of one respective larva . P values marked in bold indicate statistically significant deviations from the respective controls . For size measurements , animals where photographed on scale paper and individual lengths were calculated with Adobe Photoshop . P values marked in bold indicate statistically significant deviations from the respective controls . DOI: http://dx . doi . org/10 . 7554/eLife . 19430 . 005 As shown previously , in addition to muscle tissue mef2 is expressed in distinct neurons , including clock neurons ( Blanchard , 2010 ) and Kenyon cells ( Schulz et al . , 1996 ) . To determine whether the effects described above ( using mef2-Gal4 as a driver ) are exclusively based on Nep4 activity in muscles , or if neuronal Nep4 is also involved , we used pan-neuronal elav-Gal4 as a driver to increase or reduce nep4 expression . In this line of experiments , neither overexpression nor knockdown of nep4 had any significant influence on viability ( Figure 1A ) . This result indicates that the effects observed with mef2-Gal4 are muscle-specific . In addition to muscle tissue , Nep4 is also expressed in glial cells of the central nervous system ( CNS ) ( Meyer et al . , 2009 ) . However , in contrast to the effects observed in muscle cells , neither increased nor reduced nep4 expression in glial cells , using glia-specific repo-Gal4 as a driver , significantly affected life span ( Figure 1A ) . Besides reduced viability , elevated Nep4A levels in muscle tissue affected body size . Interestingly , as with increased lethality during early development , the effects on body size depended on enzymatic activity . In third instar larvae , muscle-specific overexpression of the active enzyme decreased the size and weight of the animals relative to control animals , whereas overexpression of catalytically inactive Nep4A did not affect size or weight . Knockdown of the peptidase in the same tissue did also not significantly alter these parameters ( Figure 1B ) . In contrast to the muscle-specific effects , increased nep4A expression in glial cells or neurons did not affect the size or weight of the larvae . Glial cell specific nep4 knockdown slightly reduced both parameters , whereas neuronal knockdown had no effect ( Figure 1B ) . In line with the lethality assay , the effects of nep4 knockdown on size and weight were completely rescued by simultaneous overexpression of Nep4A , which again confirms specificity of the respective RNAi construct . The depicted results indicate essential functions of Nep4 in muscle tissue and glial cells . However , the effects of modifying nep4 expression were more severe in muscles , suggesting the active enzyme has a critical function particularly in this tissue . To understand the physiological basis of this function in more detail , we analyzed the metabolite composition in animals with increased or reduced nep4 levels and compared the respective compositions to those in control specimens . As shown in Figure 2A , increasing or decreasing the expression of nep4 in muscle tissue affected metabolite concentrations in transgenic third instar larvae . Of note , the depicted PCA scores plot is purely based on the amplitude of correlated between-sample variations , implicating that the strongest variations in metabolite composition are those separating the three genotypes . Further analysis of the respective data revealed that profound changes are related to the energy metabolism . Knockdown of nep4 increased the levels of fructose and a purine and decreased the levels of NAD , a purine nucleotide , and glutamine ( Figure 2B , C , Figure 2—source data 1 ) . Nep4A overexpression increased the signals of histidine , glutamine , and the same purine . In addition , a significant increase was observed in the spectral regions specific to glucose and fructose , indicating elevated levels of the two monosaccharides . Of note , only the glucose and fructose signals with contributions from both sugars ( depicted in Figure 2B ) were significantly affected , implying that there is a more stable response in the sum of the two than in either of them . However , evaluation of the corresponding individual spectra clearly suggested that both sugars are increased ( Figure 2—figure supplement 1 ) . On the other hand , lactate , NAD , trehalose , and tyrosine concentrations were reduced in Nep4A-overexpressing animals ( Figure 2B , C , Figure 2—source data 1 ) . Of note , increased formation of lactate and NAD is a hallmark of aerobic glycolysis , a specific metabolic program that starts approximately 12 hr before the end of embryogenesis . Aerobic glycolysis enables hatched 1st instar larvae to efficiently convert dietary carbohydrates into biomass , thereby supporting the considerable increase in body mass that occurs during larval development ( Tennessen et al . , 2014 ) . Inhibition of aerobic glycolysis in the course of this growth phase prevents the animals from metabolizing sufficient quantities of sugar , resulting in larval lethality ( Tennessen et al . , 2011 ) . The fact that animals overexpressing Nep4A die primarily during the embryonic-larval transition and during larval development ( Figure 1A ) and exhibit considerably reduced lactate and NAD levels indicates that an excess of Nep4A may interfere with this distinct metabolic program . OPLS-DA loading plots summarizing the respective NMR spectral changes are depicted in Figure 2C . 10 . 7554/eLife . 19430 . 006Figure 2 . Muscle-specific modulation of nep4 expression affects the metabolite composition in transgenic third instar larvae . ( A ) Score plot based on genotype-specific NMR spectra . PCA score plot showing the scores of six biological replicates for each genotype . Principal component analysis ( PCA ) was applied to identify metabolite changes in response to muscle-specific Nep4A overexpression ( mef2-Gal4 x UAS-Nep4A; red ) or knockdown ( mef2-Gal4 x UAS nep4-RNAi; blue ) , relative to control animals ( mef2-Gal4 x w1118; black ) . The score plot reveals genotype-specific clustering and thus distinct metabolite compositions in corresponding animals . One nep4 knockdown sample was distinctly different from the other five . The outlier is marked by a dotted border and was excluded from OPLS-DA identification of significantly affected metabolites . ( B ) Examples of NMR signals from significantly affected metabolites . Evaluation of the dataset revealed that Nep4A overexpression significantly reduced NAD and lactate concentrations , while glucose and fructose levels were elevated in the same animals . The effects of nep4 knockdown were less severe; NAD was reduced , and fructose was slightly elevated , compared to levels in control animals . The coloring is the same as in A . The knockdown outlier is marked by a dotted line . ( C ) OPLS-DA loading plots summarizing the NMR spectral changes induced by nep4 overexpression and knockdown . Depicted is an overview of the metabolomic changes induced by modifying the expression of nep4 . Positive and negative signals represent increases and decreases in metabolite concentrations , respectively . Significant alterations are color-coded from blue to red . Red represents the highest correlation between metabolite and genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 19430 . 00610 . 7554/eLife . 19430 . 007Figure 2—source data 1 . Chemical shifts and detected changes of significantly affected metabolites . Significant changes are based on correlations with predictive scores from cross validated OPLS-DA models ( Q2 = 0 . 95 and 0 . 74 , respectively ) . A cutoff value for R2 corresponding to p<0 . 05 with Bonferroni correction for an assumed number of 100 metabolites was used . DOI: http://dx . doi . org/10 . 7554/eLife . 19430 . 00710 . 7554/eLife . 19430 . 008Figure 2—figure supplement 1 . NMR-spectra of glucose and fructose . The depicted spectra are specific to either glucose or fructose . Signals with contributions from both sugars are excluded . The individual spectra indicate increased levels of both monosaccharides in response to muscle-specific Nep4A overexpression ( mef2-Gal4 x UAS-Nep4A; red ) , relative to control animals ( mef2-Gal4 x w1118; black ) . The effects of nep4 knockdown ( mef2-Gal4 x UAS nep4-RNAi ) are depicted in blue . One nep4 knockdown sample was distinctly different from the other five . The outlier is marked by a dotted line and was excluded from further analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 19430 . 008 Given that the described metabolic abnormalities are indicative of an impaired energy metabolism , we analyzed whether modulating nep4 expression affects feeding of corresponding animals . As depicted in Figure 3A , transgenes overexpressing the peptidase were characterized by considerably reduced food intake . After 10 min of feeding , the respective animals had ingested 47% less food than controls , after 20 min 59 . 5% less , and after 40 min 57% less , relative to controls . By contrast , nep4 knockdown did not affect food intake after 40 min; however , corresponding animals were characterized by significantly reduced food intake after 10 min ( 47% of control intake ) and 20 min ( 72% of control intake ) , indicating a delayed initiation of feeding . To investigate the possibility that the observed effects were caused by protein properties other than enzymatic activity , we also analyzed catalytically inactive Nep4A . Significantly , overexpression of this construct did not affect food intake , thus confirming abnormal catalytic activity as a causative factor ( Figure 3A ) . 10 . 7554/eLife . 19430 . 009Figure 3 . Muscle-specific modulation of nep4 expression affects food intake and dilp expression in transgenic third instar larvae . ( A ) The genotype-specific rates of food intake are depicted as percentages ( % ) relative to the intake in control specimens ( mef2-Gal4 x w1118 ) after 40 min of feeding , which was set to 100% . While nep4 knockdown animals ( mef2-Gal4 x UAS-nep4 RNAi ) exhibited reduced food intake after 10 and 20 min of feeding , Nep4A overexpression animals ( mef2-Gal4 x UAS-Nep4A ) were characterized by reduced intake throughout the whole measurement ( up to 40 min ) . Animals overexpressing catalytically inactive Nep4A ( mef2-Gal4 x UAS-Nep4Ainact ) did not exhibit any significant changes in food intake , when compared to controls . Values represent the mean ( ± s . d . ) of at least six independent biological replicates . Asterisks indicate statistically significant deviations from controls ( p<0 . 05 , one-way ANOVA with pairwise comparisons ) . The lower panel depicts representative images of the genotype-specific food intake at the indicated time points . ( B ) Changes in the expression of selected dilp genes are presented as percentages ( % ) relative to expression in control specimens ( mef2-Gal4 x w1118 ) , which was set to 100% . Muscle-specific overexpression of Nep4A ( mef2 x UAS-Nep4A ) reduced the expression of every dilp gene analyzed , while nep4 knockdown in the same tissue ( mef2 x nep4-RNAi ) resulted in upregulation of dilp2 . Animals overexpressing catalytically inactive Nep4A ( mef2 x UAS-Nep4Ainact ) did not exhibit any significant changes in dilp expression , when compared to controls . Values represent the mean ( + s . d . ) of at least three independent biological replicates , each consisting of at least three technical replicates . Asterisks indicate statistical significance ( p<0 . 1; p<0 . 05 , one-way ANOVA with pairwise comparisons ) ; n . s . indicates ‘not significant’ . DOI: http://dx . doi . org/10 . 7554/eLife . 19430 . 00910 . 7554/eLife . 19430 . 010Figure 3—source data 1 . Feeding assay . Depicted are dye intensities ( no . of detected pixels ) within the intestines of animals of the indicated genotype . At least six individuals per genotype and time point were analyzed . P values marked in bold indicate statistically significant deviations from the respective controls . DOI: http://dx . doi . org/10 . 7554/eLife . 19430 . 01010 . 7554/eLife . 19430 . 011Figure 3—figure supplement 1 . Glial cell-specific modulation of nep4 expression affects dilp expression in transgenic third instar larvae . Changes in the expression of dilp genes are depicted as percentages ( % ) relative to the expression in control specimens ( repo-Gal4 x w1118 ) , which was set to 100% . Glial cell-specific overexpression of Nep4A ( repo-Gal4 x UAS-Nep4A ) increased the expression of dilp5 by 104% , while nep4 knockdown in the same tissue ( repo x nep4-RNAi ) resulted in the upregulation of dilp2 by 54% and downregulation of dilp3 by 54% . Values represent the mean ( + s . d . ) of at least three independent biological replicates , each consisting of at least three technical replicates . Asterisks indicate statistically significant differences ( p<0 . 1; **p<0 . 05 , one-way ANOVA with pairwise comparisons ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19430 . 011 Since the increased glucose levels that are evident in Nep4A overexpression animals ( Figure 2B , C ) are symptomatic of impaired insulin signaling ( Broughton et al . , 2005; Rulifson et al . , 2002 ) , in a continuative set of experiments we analyzed whether altering nep4 levels also affected the expression of Drosophila insulin-like peptides ( dilps ) . We focused on dilps 1 , 2 , 3 , and 5 because they encode the major insulin-like peptides expressed by larval insulin-producing cells ( IPCs ) ( Rulifson et al . , 2002; Brogiolo et al . , 2001; Cao and Brown , 2001; Ikeya et al . , 2002; Lee et al . , 2008; Nässel et al . , 2013 ) . IPCs are located within the median neurosecretory cell cluster of the central brain and apparently function like pancreatic β-cells , since IPC ablation in Drosophila results in elevated levels of circulating glucose . In addition , animals with ablated IPCs are smaller than wild-type specimens , and they weigh less ( Broughton et al . , 2005; Rulifson et al . , 2002 ) . Significantly , these characteristic effects of IPC ablation were phenocopied by muscle-specific Nep4A overexpression ( Figures 1 and 2 ) . Furthermore , the respective transgenic animals exhibited considerably reduced expression of the selected dilps . In Nep4A-overexpressing animals , dilp1 expression decreased by 59% , dilp2 by 83% , dilp3 expression by 88% , and dilp5 expression by 84% , relative to expression in controls . On the other hand , muscle-specific nep4 knockdown had no effect on the expression of dilps 1 , 3 , and 5 , although expression of dilp2 increased by 82% , relative to expression in controls ( Figure 3B ) . The rather mild effect of nep4 knockdown on dilp expression , when compared to the effects of Nep4A overexpression , suggests that other , yet unknown peptidases can compensate for reduced Nep4 activity . In line with the results from the feeding assay ( Figure 3A ) , dilp expression is only affected by the wild type enzyme , while overexpression of catalytically inactive Nep4A did not significantly alter expression of the selected dilps ( Figure 3B ) . To determine if expression of insulin-like peptides is regulated exclusively by muscle-derived Nep4 or if intrinsic CNS signaling is also involved , we altered nep4 expression in a nervous-system-specific manner . As depicted in Figure 3—figure supplement 1 , glial-cell-specific overexpression of Nep4A increased the expression of dilp5 , while nep4 knockdown in the same cells resulted in an upregulation of dilp2 and downregulation of dilp3 . Although these effects were minor compared to the effects of modulating the expression of muscle-bound Nep4 ( Figure 3B ) , they demonstrate that proper regulation of dilp expression also requires adequate Nep4 levels within the CNS . In order to understand the physiological relation between dilp expression and Nep4 activity in more detail , we analyzed the expression pattern and the subcellular localization of the peptidase in larval body wall muscles and the larval CNS . As depicted in Figure 4 , in body wall muscles Nep4 exhibits a dual localization: in addition to localizing to membranes continuous with the nuclear membrane ( Figure 4A , arrowheads ) , which we previously identified as related to the sarco/endoplasmic reticulum ( Panz et al . , 2012 ) , the peptidase accumulates at the surface of the muscles ( Figure 4A , arrows ) . The latter localization is consistent with ectoenzymatic activity and indicative of a function in regulating the homeostasis of hemolymph circulating peptides . To confirm the specificity of the signal , we also stained the muscles of transgenic animals expressing nep4-specific RNAi ( mef2-Gal4 driver ) . In these transgenic animals , no signal above background was observed ( Figure 4B ) . In addition , staining wild-type muscles with secondary antibodies alone did not result in a distinct signal ( Figure 4C ) . Of note , the proteins expressed from the two overexpression constructs ( wild-type Nep4A and catalytically inactive Nep4A ) exhibited subcellular localizations identical to that of the endogenous protein ( Figure 4A , D , F ) , confirming that the observed overexpression phenotypes ( Figures 1–3 ) were not impaired by mislocalization of the respective constructs . In order to distinguish the ectopic proteins from the endogenous protein , the ectopic constructs were fused to a C-terminal HA-tag and labeled with corresponding antibodies . Antibody specificity was confirmed by the lack of staining in animals expressing only the Gal4 transgene but not the UAS-construct ( Figure 4E , G ) . 10 . 7554/eLife . 19430 . 012Figure 4 . Nep4 localizes to the surface of muscle cells . ( A ) Nep4 protein was labeled with a monospecific antibody ( red ) . In addition to membranes continuous with the nuclear membrane ( arrowheads ) , Nep4 accumulated at the surface of body wall muscles ( arrows ) . ( D , F ) Nep4 overexpression constructs ( mef2>Nep4A , mef2>Nep4Ainact ) exhibited subcellular localizations identical to that of the endogenous protein . The corresponding constructs were labeled with antibodies detecting the fused HA-tag . ( B , C , E , G ) Control stainings did not produce any signal above background . DOI: http://dx . doi . org/10 . 7554/eLife . 19430 . 012 To characterize expression in the CNS , we employed a reporter line that expresses nuclear GFP ( nGFP ) in a manner that recapitulates endogenous nep4 expression ( Meyer et al . , 2009 ) . As shown in Figure 5 , brain and ventral nerve cord tissue exhibited substantial reporter gene expression . With respect to the brain , expression was observed mainly in lamina ( Figure 5A , brackets ) and central brain cells ( Figure 5A , dashed line ) , while only a few medulla cells exhibited a distinct signal ( Figure 5A , bar ) . Within the ventral nerve cord , nep4 was detected in numerous cells along all segments . As confirmed by extensive colocalization with the glial cell marker Reversed-polarity ( Repo ) , nep4 was expressed primarily in this cell type; however , especially in the median region of the central brain , only partial colocalization was evident . Thus , in addition to glial cells , nep4 is expressed in certain neurons of the central brain ( Figure 5A-C ) . 10 . 7554/eLife . 19430 . 013Figure 5 . Nep4 is expressed in glial cells and neurons in the central nervous system . nep4 expression was visualized using a reporter construct that drives nuclear GFP ( nGFP ) expression in a nep4-specific manner ( nep4 > nGFP , green ) . Reversed polarity protein was labeled with a monospecific antibody ( α-Repo , red ) . ( A–C ) Optical projections of third instar larval whole brain-ventral nerve cord complexes . Scale bars: 100 µm; dorsal view , anterior up . Boxes indicate areas of higher magnification , as depicted in ( D–F ) and ( G–I ) . Within the brain , nep4 expression was strongest in the central brain ( A , dashed line ) and in lamina cells ( A , brackets ) , while only few nep4-positive medulla cells were observed ( A , bar ) . Within the ventral nerve cord , nep4 was expressed in numerous cells along all segments . ( D–I ) Optical projections of third instar larval brain hemisphere ( D–F ) and ventral nerve cord ( G–I ) . Scale bars: 20 µm; dorsal view , anterior up , midline to right . nep4 expression colocalized extensively with anti-Repo staining . DOI: http://dx . doi . org/10 . 7554/eLife . 19430 . 013 Interestingly , as confirmed by colocalization with dilp2-specific reporter gene expression , we found that these neurons included all IPCs , which reside within the median neurosecretory cell cluster of the brain hemispheres . The distinct localization of the respective signals is because both reporter constructs drive expression of a nuclear localized fluorophore ( Figure 6A–C ) . To assess the subcellular localization of Nep4 in IPCs , we performed double labeling experiments using a reporter line expressing eGFP in a dilp2-specific manner , thus labeling the IPC cytoplasm , together with Nep4-specific antibodies . As depicted in Figure 6D–F , the peptidase accumulated at the surface of numerous cells of the central brain , including IPCs . 10 . 7554/eLife . 19430 . 014Figure 6 . Nep4 localizes to the surface of insulin-producing cells . ( A–C ) nep4 expression was assessed using a reporter line that drives nuclear mCherry expression in a nep4-specific manner ( nep4 > mCherry , red ) . dilp2 expression was visualized using a reporter construct that drives nuclear GFP expression in a dilp2-specific manner ( dilp2 > nGFP , green ) . Depicted are optical sections ( 10 µm ) of a third instar larval central brain . Scale bars: 20 µm; dorsal view , anterior up . nep4 and dilp2 expression colocalized in IPCs . ( D–F ) Nep4 protein was labeled with a monospecific antibody ( red ) , and dilp2 expression was visualized using an eGFP reporter line ( dilp2 > eGFP , green ) . Depicted are optical sections ( 10 µm ) of a third instar larval central brain . Scale bars: 20 µm; dorsal view , anterior up . Nep4 accumulated at the surface of numerous cells , including IPCs ( D , F , arrowheads ) . The subcellular localization was assessed with fluorescence intensity measurements ( lower panel ) . The respective regions of evaluation are marked ( arrows in D–F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19430 . 014 The fact that major phenotypes described in this study strictly depend on the catalytic activity of Nep4 ( Figures 1 and 3 ) indicates that aberrant hydrolysis of peptides involved in regulating dilp expression and / or feeding behavior is primarily responsible for the phenotypes . This indication is corroborated by the localization of Nep4 to the surface of body wall muscles and IPCs ( Figures 4 and 6 ) , with the latter constituting the major site of Dilp synthesis in Drosophila ( Rulifson et al . , 2002; Brogiolo et al . , 2001; Ikeya et al . , 2002 ) . In order to identify the causative hydrolysis event ( s ) , we analyzed every peptide known to be involved in regulating dilp expression , feeding behavior , or both ( Nässel et al . , 2013; Pool and Scott , 2014 ) for susceptibility to Nep4-mediated cleavage . The only additional prerequisite for consideration as a potential substrate was a size of less than 5 kDa , which for steric reasons represents the maximum mass of a neprilysin substrate ( Oefner et al . , 2000 ) . As depicted in Table 1 , we found 23 peptides matching these criteria . Among these candidates , 16 were hydrolyzed at distinct positions by purified Nep4 , while the remaining seven peptides were not significantly cleaved . The identified substrates were adipokinetic hormone ( AKH ) , allatostatin A1-4 , corazonin , diuretic hormone 31 ( DH31 ) , drosulfakinins 1 and 2 , leucokinin , short neuropeptide F11–11 , short neuropeptide F14–11 ( also corresponding to sNPF212–19 ) , and tachykinins 1 , 2 , 4 , and 5 . No Nep4-specific cleavage was observed for hugin , neuropeptide F , proctolin , short neuropeptide F3 , short neuropeptide F4 , and tachykinins 3 and 6 . Analysis of the resulting hydrolysis products revealed that Nep4 preferentially cleaved next to hydrophobic residues , particularly with Phe or Leu at P1´ ( Table 1 ) . Identically treated control preparations lacking the peptidase did not exhibit any cleavage activity ( Figure 7 ) . Individual MS chromatograms are depicted in Figure 7 . Nep4B purity was confirmed with SDS-PAGE ( Figure 7—figure supplement 1 ) . 10 . 7554/eLife . 19430 . 015Figure 7 . Nep4 catalyzes the hydrolysis of peptides that regulate dilp expression or feeding behavior . Base peak all MS chromatograms of analyzed peptides . The respective sequences of unprocessed full-length peptides ( bold ) and of identified Nep4-specific cleavage products are indicated . Unlabeled peaks were not identified . Spectra corresponding to untreated peptides are indicated in black , spectra corresponding to peptides incubated with control preparations lacking Nep4B are indicated in green , and spectra corresponding to peptides incubated with Nep4B-containing preparations are indicated in red . X-axes depict retention time ( min ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19430 . 01510 . 7554/eLife . 19430 . 016Figure 7—figure supplement 1 . Heterologously expressed Nep4B can be purified to homogeneity . Coomassie-stained SDS-polyacrylamide gel of Ni-NTA elution fractions . Nep4B ( approx . 113 kDa ) was efficiently purified from whole cell lysates of nep4B-transfected SF21 cells ( arrow ) . In identically treated samples from untransfected control cells , only background binding is apparent . DOI: http://dx . doi . org/10 . 7554/eLife . 19430 . 01610 . 7554/eLife . 19430 . 017Table 1 . Nep4 hydrolyzes peptides that regulate dilp expression or food intake . Candidate peptides were analyzed for Nep4-specific cleavage . The individual molecular masses of full length peptides and cleavage products are depicted as the monoisotopic value . Cleavage positions and deviations from the respective theoretical masses ( ∆ ) are shown separately . Cleaved peptides are highlighted in blue , and non-cleaved peptides are depicted in red . Superscripts indicate the studies that biochemically characterized the respective peptides ( 1 ( Baggerman et al . , 2005 ) , 2 ( Wegener et al . , 2006 ) , 3 ( Wegener and Gorbashov , 2008 ) , 4 ( Predel et al . , 2004 ) , 5 ( Yew et al . , 2009 ) ) . n . d . indicates ‘not detected’ , thus the respective sequences represent genomic data based predictions . DOI: http://dx . doi . org/10 . 7554/eLife . 19430 . 017Name Sequence Mass ( Da ) Δ ( Da ) Sequence of cleavage products Mass ( Da ) Δ ( Da ) Cleavage position Allatostatin A1 VERYAFGLa4953 . 5−0 . 0676 VERYAFGVERYAF840 . 4783 . 4−0 . 0893 −0 . 0898 G/LF/GAllatostatin A2 LPVYNFGLa5920 . 5−0 . 0205 LPVYNFGLPVYNFLPVYN808 . 4751 . 4604 . 3−0 . 0492 −0 . 0148 −0 . 0223 G/LF/GN/FAllatostatin A3 SRPYSFGLa1 , 4924 . 5−0 . 0523 YSFGLa584 . 3−0 . 0241 P/YAllatostatin A4 TTRPQPFNFGLa1 , 4 , 51275 . 7−0 . 0629 TTRPQPFNFGTTRPQPFNFNFGLa1163 . 6959 . 5595 . 3−0 . 0850 −0 . 0790 −0 . 0301 G/LN/FP/FAKH QLTFSPDWa1 , 2 , 3 , 4992 . 50 . 0051TFSPDWaFSPDWa750 . 3649 . 3−0 . 0360 −0 . 0473 L/TT/FCorazonin QTFQYSRGWTNa1 , 2 , 3 , 4 , 51385 . 6−0 . 0582 FQYSRGWTNaQTFQYSRG1156 . 5985 . 5−0 . 0319 −0 . 0743 T/FG/WDH31TVDFGLARGYSGTQ-EAKHRMGLAAANFA-GGPan . d . 3149 . 5−0 . 0814 YSGTQEAKHRMGTVDFGLARG1363 . 6934 . 5−0 . 1761 −0 . 0198 G/Y; G/LG/YDrosulfakinin 1 FDDYGHMRFa1 , 4 , 51185 . 5−0 . 0572 FDDYGHMR1039 . 4−0 . 1147 R/FDrosulfakinin 2 GGDDQFDDYGHMRFa1 , 4 , 51657 . 7−0 . 0298 GGDDQFDDYGHMRFDDYGHMRFa1511 . 61185 . 5−0 . 1201 −0 . 0711 R/FQ/FLeucokinin NSVVLGKKQRFHSWGa1 , 3 , 4 , 51741 . 0−0 . 0905 NSVVLGKKQRFHSNSVVLGKKQRFHNSVVLGKKQRFHSWGa1498 . 31411 . 81127 . 7631 . 3−0 . 1474 −0 . 1094 −0 . 1121 −0 . 0100 S/WH/SR/FR/FsNPF11-11AQRSPSLRLRFa2 , 3 , 41328 . 8−0 . 0520 AQRSPSLRL1026 . 6−0 . 0962 L/RsNPF14-11/ sNPF212-19SPSLRLRFa1 , 2 , 3 , 4 , 5973 . 6−0 . 0859 SPSLRLRLRLRFa827 . 5702 . 5−0 . 1543 −0 . 1451 R/FS/LTachykinin 1 APTSSFIGMRa1 , 41064 . 5−0 . 0579 APTSSFIGFIGMRa778 . 4621 . 3−0 . 0434 −0 . 0706 G/MS/FTachykinin 2 APLAFVGLRa1 , 5941 . 6−0 . 0396 LAFVGLRaAPLAFVGFVGLRaAPLAF773 . 5673 . 4589 . 4517 . 3−0 . 0858 −0 . 0202 −0 . 0686 −0 . 0183 P/LG/LA/FF/VTachykinin 4 APVNSFVGMRa1 , 4 , 51075 . 6−0 . 0742 APVNSFVG789 . 4−0 . 0314 G/MTachykinin 5 APNGFLGMRa1 , 5960 . 50 . 0231FLGMRa621 . 3−0 . 0666 G/FHugin SVPFKPRLa1 , 2 , 3 , 4 , 5941 . 6−0 . 0776 NPF SNSRPPRKNDVNTMA-DAYKFLQDLDTYYGD-RARVRFan . d . 4278 . 20 . 50Proctolin RYLPTn . d . 648 . 4−0 . 0841 sNPF3 KPQRLRWa5981 . 6−0 . 05 sNPF4 KPMRLRWa5984 . 6−0 . 05 Tachykinin 3 APTGFTGMRa1935 . 5−0 . 0733 Tachykinin 6 AALSDSYDLRGKQQR-FADFNSKFVAVRan . d . 3087 . 6−0 . 1694 While the functional roles of insulin-like peptides ( ILPs ) and the corresponding insulin- and IGF-signaling have been intensively studied , the control of ILP production and release is not well understood . This study demonstrates that modulating the expression of a Drosophila neprilysin interferes with the expression of insulin-like peptides , thus establishing a correlation between neprilysin activity and the regulation of insulin signaling . A high physiological relevance is confirmed by the fact that altering nep4 expression phenocopies characteristic effects of IPC ablation , including reduced size and weight of corresponding animals , as well as increased levels of carbohydrates such as glucose and fructose ( Figures 1 and 2 ) . The result that the levels of these sugars are increased , although food intake rates are reduced ( Figure 3A ) presumably reflects the physiological impact of the diminished ilp expression that is also obvious in corresponding animals ( Figure 3B ) . In this respect , the impaired insulin signaling likely results in inefficient metabolization and thus accumulation of the sugars , which overcompensates the diametrical effects of reduced food intake . By identifying 16 novel peptide substrates of Nep4 , the majority of which are involved in regulating dilp expression or feeding behavior ( Table 1 , Figure 7 ) , and by localizing the peptidase to the surface of body wall muscles ( Figure 4 ) and IPCs within the larval CNS ( Figure 6 ) , we provide initial evidence that neprilysin-mediated hydrolysis of hemolymph circulating as well as CNS intrinsic peptides is the physiological basis of the described phenotypes . The finding that only the catalytically active enzyme affected dilp expression whereas the inactive construct did not ( Figure 3B ) , substantiates this evidence because it confirms aberrant enzymatic activity and thus abnormal peptide hydrolysis as a causative parameter . Interestingly , we observed the strongest effects on size and dilp expression with muscle-specific overexpression of Nep4; overexpression of the peptidase in the CNS was less detrimental ( Figure 1 , Figure 3B , Figure 3—figure supplement 1 ) . These results indicate that hemolymph circulating peptides accessible to muscle-bound Nep4 are mainly responsible for the observed effects , while CNS intrinsic peptide signaling is less relevant . The fact that all peptides cleaved by Nep4 ( Table 1 ) could be released into the hemolymph , either from enteroendocrine cells or from neurohormonal release sites ( Nässel and Winther , 2010 ) , substantiates this indication . Since the Drosophila midgut is the source of several neuropeptides ( Veenstra et al . , 2008; Reiher et al . , 2011 ) , it is conceivable that a main reason for the observed phenotypes is aberrant cleavage of certain gut-derived peptides that are required for proper midgut-IPC communication . Allatostatin A , neuropeptide F , diuretic hormone 31 , and some tachykinins are produced by endocrine cells of the gut ( Veenstra et al . , 2008; Reiher et al . , 2011; Lenz et al . , 2001 ) . Interestingly , all have been implicated in regulating dilp expression and/or feeding behavior ( Nässel et al . , 2013; Pool and Scott , 2014 ) , and most of them , namely allatostatin A1-4 , diuretic hormone 31 , and tachykinin 1 , 2 , 4 , and 5 , were cleaved by Nep4 ( Table 1 ) , indicating enzyme-substrate relationships . Thus , these results suggest that Nep4 activity at the surface of muscle cells is necessary to maintain homeostasis of distinct hemolymph circulating signaling peptides , probably gut-derived , thereby ensuring proper midgut-IPC communication . On the other hand , fat body-IPC feedback may be affected as well . However , the only factors known to mediate this process , Unpaired 2 ( Rajan and Perrimon , 2012 ) , DILP6 ( Bai et al . , 2012 ) , and Stunted ( Delanoue et al . , 2016 ) have molecular masses of more than 5 kDa , and thus exceed the maximum mass of a putative neprilysin substrate ( Oefner et al . , 2000 ) . Consequently , a direct regulatory influence of Nep4 on Unpaired 2 , DILP6 , or Stunted activity appears unlikely . In addition to body wall muscles , nep4 is expressed in numerous cells of the central nervous system , predominantly in glial cells ( Figure 5 ) . Interestingly , compared to the muscle-specific effects , modulating nep4 expression in this tissue has distinct and less severe effects on dilp expression ( Figure 3B , Figure 3—figure supplement 1 ) . This result suggests that CNS intrinsic Nep4 activity affects different neuropeptide regulatory systems than the corresponding muscle-bound activity . Considering the rather broad expression in glial cells , it is furthermore likely that the CNS regulation affects more than one system . However , localization at the IPC surface ( Figure 6 ) clearly supports a direct function in the regulation of dilp expression . In this context , spatial proximity of the peptidase may be necessary to ensure low ligand concentrations and thus tight regulation of specific neuropeptide receptors present at the surface of IPCs . Such receptors include an allatostatin A receptor ( Dar-2 ) ( Hentze et al . , 2015 ) , a tachykinin receptor ( DTKR ) ( Birse et al . , 2011 ) , and the short neuropeptide F receptor ( sNPFR ) ( Lee et al . , 2008 ) . All are essential to proper dilp expression ( Lee et al . , 2008; Hentze et al . , 2015; Birse et al . , 2011 ) . Interestingly , with respect to sNPFR , corresponding ligands ( sNPF11–11 , sNPF14–11 , and sNPF212–19 ) exhibit very high-binding affinities , with IC50 values in the low nanomolar range ( Garczynski et al . , 2006 ) , a finding that further emphasizes the need for effective ligand clearance mechanisms in order to prevent inadvertent receptor activation . Localization of Nep4 to the surface of IPCs ( Figure 6 ) and confirmation of Dar-2 , DTKR , and sNPFR ligands as substrates of the peptidase ( Table 1 , Figure 7 ) strongly indicate that Nep4 participates in such clearance mechanisms . Of note , sNPF species were detected in both , CNS and hemolymph preparations , with neuroendocrine functions of the respective peptides being suggested ( Veenstra et al . , 2008; Garczynski et al . , 2006; Baggerman et al . , 2005; Wegener and Gorbashov , 2008; Wegener et al . , 2006 ) . The dual localization is interesting because both compartments are accessible to Nep4 , either to the CNS resident or to the muscle-bound enzyme . Significantly , sNPF is a potent regulator of dilp expression . Increased sNPF levels result in upregulation of dilp expression , and decreased sNPF levels have the opposite effect ( Lee et al . , 2008 ) . The fact that these results inversely correlate with the effects of modulating nep4 expression ( Figure 3 ) suggests a functional relationship between sNPF and the neprilysin . Nep4-mediated cleavage of distinct sNPF species ( Table 1 , Figure 7 ) represents further evidence for this relationship . Besides sNPF , Nep4 also cleaves corazonin , drosulfakinins , and allatostatin A ( Table 1 , Figure 7 ) . Interestingly , corazonin promotes food intake ( Hergarden et al . , 2012 ) , while allatostatin A and drosulfakinins inhibit it ( Hergarden et al . , 2012; Söderberg et al . , 2012; Chen et al . , 2016 ) . This regulatory activity on peptides with opposing physiological functions indicates that Nep4 affects multiple aspects of feeding control , rather than promoting or inhibiting food intake in a mutually exclusive manner . Our finding that both , nep4 knockdown and overexpression larvae exhibit reduced food intake ( Figure 3A ) supports this indication since it suggests that regular Nep4 activity adjusts the general peptide homeostasis in a manner that promotes optimal food intake , with deviations in either direction being deteriorative . The result that nep4 knockdown animals exhibit reduced food intake for only up to 20 min of feeding ( Figure 3A ) may reflect this complex regulation since it indicates that at the onset of feeding reduced cleavage of peptides inhibiting food intake ( e . g . allatostatin A , drosulfakinins ) is a dominant factor . With ongoing feeding , accumulation of peptides promoting food intake ( e . g . corazonin ) may become decisive , thus restoring intake rates . In addition , Nep4 hydrolyzes numerous peptides that regulate dilp expression , including tachykinins , allatostatin A , and sNPF . However , AKH , a functional homolog of vertebrate glucagon that acts antagonistically to insulin , is also a substrate of Nep4 ( Table 1 , Figure 7 ) . This finding indicates that the Nep4-mediated regulation of dilp expression and sugar homeostasis can also not be attributed to a single substrate or cleavage event . Rather , it is a result of the concerted hydrolysis of several critical peptides , including both , hemolymph circulating and CNS intrinsic factors . Taking into account that overexpression and knockdown of nep4 have discrete effects on dilp expression ( Figure 3B ) , but comparable effects on feeding ( Figure 3A ) , it furthermore appears likely that dysregulation of the Nep4-mediated peptide homeostasis affects both processes somewhat independently of each other . The fact that among the novel Nep4 substrates we identified peptides that presumably affect either dilp signaling ( e . g . DH31 ) , or food intake ( e . g . leucokinin , drosulfakinins ) in a largely exclusive manner supports this indication . Because neprilysins and many of the novel substrates identified in this study are evolutionarily conserved factors , neprilysin-mediated regulation of insulin-like peptide expression and feeding behavior may be relevant not only to the energy metabolism in Drosophila , but also to corresponding processes in vertebrates , including humans . Interestingly , a critical function of murine Neprilysin in determining body mass has already been reported . The regulation depended primarily on the catalytic activity of peripheral NEP , while the CNS-bound enzyme was less important ( Becker et al . , 2010 ) . However , until now , the underlying physiology has been obscure , essentially because no causative hydrolysis event had been identified . Our finding that also in Drosophila mainly peripheral ( muscle-bound ) Nep4 activity affected body mass , while CNS-specific modulations had only minor effects on size or weight ( Figure 1 ) , indicates that the neprilysin-mediated regulation of food intake , body size and insulin expression involves similar physiological pathways in both species . Furthermore , the fact that altered catalytic activity and thus abnormal peptide hydrolysis is a critical factor in mice ( Becker et al . , 2010 ) and in Drosophila ( Figures 1 and 3 ) emphasizes the need to generate comprehensive , enzyme-specific lists of neprilysin in vivo substrates . In this context , the results of our screen for novel Nep4 substrates ( Table 1 , Figure 7 ) may be a valuable resource in order to identify corresponding substrates in vertebrates and humans . The following Drosophila lines were used in this work . Strain w1118 ( RRID:BDSC_5905 ) was considered wild type . The driver lines were mef2-Gal4 ( RRID:BDSC_27390 ) , repo-Gal4 ( RRID:BDSC_7415 ) , elav-Gal4 ( RRID:BDSC_8760 ) , and dilp2-Gal4 ( RRID:BDSC_37516 ) . UAS-lines were UAS-mCherry . NLS ( RRID:BDSC_38424 ) and UAS-2xEGFP ( RRID:BDSC_6874 ) . The nep4-nGFP reporter line was described previously ( Meyer et al . , 2009 ) . nep4 knockdown was achieved using line 100189 ( KK library , no off-targets , Vienna Drosophila Resource Center , VDRC ) . A high knockdown efficiency of the respective construct was shown previously ( Panz et al . , 2012 ) . To confirm specificity of the knockdown , a line being homozygous for both , the UAS-nep4 RNAi construct ( chromosome II ) and the UAS-Nep4A overexpression construct ( chromosome III ) was generated and crossed to either mef2-Gal4 or repo-Gal4 . Tissue-specific rescue of the respective RNAi phenotypes by simultaneous overexpression of Nep4A was used as readout for knockdown specificity . A second nep4 RNAi construct ( line 16669 , GD library , VDRC ) did not significantly reduce nep4 transcript levels ( Panz et al . , 2012 ) . It was therefore excluded from further analysis . Staged ( AEL 74–78 hr ) male third instar larvae where grouped into genotype-specific cohorts of 10 individuals . The weights of at least five cohorts per genotype were averaged to calculate the mean weight of one respective larva . For size measurements , larvae where exposed to 60°C water for 10 s , resulting in maximum relaxation of the body . Subsequently , animals where photographed on scale paper using a stereomicroscope ( Leica MZ16 FA ) , and individual lengths were calculated with the Adobe Photoshop CS5 measure tool using the scale paper as a reference . Animals of different genotypes were raised at 27 °C on apple agar plates supplemented with excess yeast paste . Stage-specific lethality rates were determined by calculating the percentage of animals of a specific stage that did not develop into the next stage . For each genotype and biological replicate , 550 embryos were analyzed . Three independent biological replicates were conducted . Staged ( AEL 74–78 hr ) male third instar larvae were starved for 1 hr . Subsequently , animals were fed with dyed yeast ( 0 . 3 mg Carmin , 4 mg dry yeast , dissolved in 10 ml H2O ) for 5 , 10 , 20 , or 40 min , respectively , washed , and photographed ( Stemi 2000-C , Zeiss , Jena , Germany ) . Dye intensities ( no . of detected pixels ) within the intestines were determined with Fiji software ( http://fiji . sc/ ) . At least six individuals per genotype and time point were analyzed . Staged ( AEL 74–78 hr ) male third instar larvae where grouped into genotype-specific cohorts , and six cohorts per genotype were independently analyzed to assess metabolite composition . Briefly , animals ( 50 mg/cohort ) were homogenized ( glass-Teflon homogenizer ) in 500 μl ice-cold ACN/H2O ( 50% ) and centrifuged ( 10 , 000 × g , 10 min ) to remove fly debris and precipitate . The resulting supernatant was lyophilized and frozen at −80°C for later use . Samples were rehydrated in 650 µl of 50 mM phosphate buffer in D2O ( pH 7 . 4 ) containing 50 mg/l 3-trimethylsilyl propionic acid D4 ( TSP ) as a chemical shift reference and 50 mg/l sodium azide to prevent bacterial growth . The NMR measurements were carried out at 25 °C on a Bruker Avance-III 600 spectrometer ( Bruker Biospin , Germany ) equipped with a double tuned 1H-13C 5 mm cryoprobe and operated at a 1H frequency of 600 . 13 MHz . The 1H NMR spectra were acquired using a single 90° pulse experiment with a Carr Purcell Meiboom Gill ( CPMG ) delay added , in order to attenuate broad signals from high molecular weight components . The total CPMG delay was 40 ms , and the spin echo delay was 200 µs . The water signal was suppressed by pre-saturation of the water peak during the relaxation delay of 4 s . A total of 96k data points spanning a spectral width of 20 ppm were collected in 128 transients . For assignment purposes , two-dimensional 1H-1H TOCSY and 1H-13C HSQC spectra were acquired . The spectra were processed using iNMR ( www . inmr . net ) . An exponential line broadening of 0 . 5 Hz was applied to the free induction decay , prior to Fourier transformation . All spectra were referenced to the TSP signal at −0 . 017 ppm , automatically phased and baseline corrected . The spectra were aligned using Icoshift ( Savorani et al . , 2010 ) , and the region around the residual water signal ( 4 . 88–4 . 67 ppm ) was removed . The integrals were normalized to total weight , and the data were scaled using pareto scaling ( Craig et al . , 2006 ) and centered . Initially , the whole dataset was subjected to principal component analysis ( PCA ) ( Stoyanova and Brown , 2001 ) . Afterwards , orthogonal projection to latent structures discriminant analysis ( OPLS-DA ) models were created to separate either larvae overexpressing nep4 from control larvae or nep4 knockdowns from control larvae . OPLS-DA models are multivariate models that predict group membership based on a multivariate input , in this case the NMR spectra . The model separates variations due to group membership from other ( orthogonal ) variations ( Bylesjö et al . , 2006 ) . The OPLS-DA models were validated by cross validation where models were made with randomly chosen groups of samples left out one at a time , and group membership was predicted for the left out samples . The predictability ( Q2 ) of the models , i . e . the correlation between predicted and actual classification , was 0 . 95 for the comparison between mef2-Gal4 x w1118 and mef2-Gal4 x UAS-Nep4A , and 0 . 74 for the comparison between mef2-Gal4 x w1118 and mef2-Gal4 x UAS nep4-RNAi , respectively , indicating high-quality models . The loadings and the correlation coefficient ( R ) between intensities at the individual frequencies and the predictive component were calculated . A cutoff value for R2 corresponding to p<0 . 05 with Bonferroni correction for an assumed number of 100 metabolites was calculated from the distribution of R2 values in 10 , 000 permutated data sets . Signal assignments were based on chemical shifts , using earlier assignments and spectral databases described elsewhere ( Cui et al . , 2008; Malmendal et al . , 2006; Pedersen et al . , 2008 ) . All multivariate analysis was performed using the Simca-P software ( Umetrics , Sweden ) . Heterologous expression was performed in SF21 cells ( RRID:CVCL_0518 ) using the Bac-to-Bac baculovirus expression system ( Life Technologies , Carlsbad , CA , USA ) . The nep4B coding sequence was fused to a C-terminal His-tag using appropriate primer design and cloned downstream of the polyhedrin promoter into an E . coli/S . cerevisiae/Baculovirus triple-shuttle derivative of the pFastBac Dual vector adapted for cloning by homologous recombination in vivo . The respective vector ( pJJH1460 ) was constructed similar to the vectors described in ( Paululat and Heinisch , 2012 ) . To track transfection efficiency , an egfp reporter gene was inserted into the same vector under the control of the p10 promoter . Transfected and non-transfected SF21 cells were cultured in 75-cm2 flasks for 72 hr and harvested by centrifugation ( 300 × g , 5 min ) . Subsequently , cells were resuspended in 5 ml binding buffer ( 50 mM NaH2PO4 , pH 7 . 9; 300 mM NaCl ) and lysed with a glass-Teflon homogenizer . The resulting homogenates were centrifuged ( 10 min , 10 , 000 × g ) , and the supernatants were subjected to gravity-flow-based His-tag purification according to the manufacturer’s instructions ( Protino Ni-NTA agarose , Macherey-Nagel , Düren , Germany ) . To measure enzymatic activity , 2 . 5 µl of Nep4B-containing ( 10 ng/µl , purified from nep4B transfected cells ) and non-containing ( from untransfected control cells ) preparations were supplemented with 3 . 5 µl ( 150 ng ) of individual peptides . After 5 hr of incubation ( 35°C ) , 1 µl of each respective preparation was analyzed with ESI mass spectrometry . Peptides were synthesized at JPT Peptide Technologies ( Berlin , Germany ) with more than 90% purity . Individual cleavage assays were repeated at least three times . Samples were loaded onto a trap column ( Acclaim PepMap C18 , 5 µm , 0 . 1 × 20 mm , Thermo Scientific , Sunnyvale , CA , USA ) and washed . The trap column was switched inline with a separation column ( Acclaim PepMap C18 2 µm , 0 . 075 × 150 mm , Thermo Scientific ) . Subsequently , bound substances were eluted by changing the mixture of buffer A ( 99% water , 1% acetonitrile , 0 . 1% formic acid ) and buffer B ( 80% acetonitrile , 20% water and 0 . 1% formic acid ) from 100:0 to 20:80 within 45 min . The flow rate was kept constant at 0 . 3 µl/min . Successively eluted compounds were analyzed with an ESI-ion trap ( Amazon ETD Speed with a captive spray ionization unit , Bruker Corporation , Billerica , MA , USA ) by measuring the masses of the intact molecules as well as the masses of the fragments , which were generated by collision-induced dissociation ( CID ) of the corresponding parent ion . All acquired data were used for determination of peptide-specific amino acid sequences with the Mascot search algorithm ( Matrix Science , Boston , MA , USA ) in combination with a custom-made database containing 37 different sequences of peptides . To avoid an increased false-positive identification rate the p-value was lowered to 0 . 005 ( resulting in an individual ion score > 18 ) . As enzyme , the option ‘none’ was chosen . Thus , every subsequence of every protein was used for identification . Brains prepared from staged male third instar larvae ( AEL 74–78 hr ) were fixed ( 3 . 7% formaldehyde , 1 hr ) and permeabilized ( 1% Triton X-100 , 1 hr ) . Subsequently , tissues were incubated in PBS containing 0 . 15% SDS ( 30 min ) , blocked with Roti-Block ( Carl Roth , Karlsruhe , Germany ) for 45 min , washed in PBT ( 4× , 10 min each ) , and incubated in Roti-Block ( 45 min ) and primary antibody ( overnight ) . Samples were washed in PBT ( 4× , 10 min each ) and blocked again as described above . Secondary antibodies were applied simultaneously for 90 min . Finally , samples were washed as described above and mounted in Fluoromount-G ( SouthernBiotech , Birmingham , USA ) . For staining of body wall muscles , male third instar larvae were dissected on Sylgard plates ( Sylgard 184 Elastomer Base and Curing Agent , Dow Corning , Michigan , USA ) , fixed in 3 . 7% formaldehyde in PBS for 1 hr , rinsed three times in PBS , and transferred into 1 . 5 ml reaction cups . Subsequently , tissues were permeabilized in 1% Triton X-100 for 1 hr , blocked in Roti-Block ( 45 min ) , and incubated with primary antibodies ( overnight ) . Samples were washed in PBT ( 3× , 10 min each ) and blocked again as described above . Secondary antibodies were applied for 90 min . Finally , samples were washed as described above and mounted in Fluoromount-G ( SouthernBiotech , Birmingham , USA ) . The primary antibodies used were: anti-Nep4 ( RRID:AB_2569115 , 1:200 , raised in rabbit , monospecificity was confirmed in [Meyer et al . , 2009] ) , anti-GFP ( RRID:AB_889471 , 1:500 , raised in mouse ) , anti-GFP ( RRID:AB_305564 , 1:2000 , raised in rabbit ) , anti-HA ( RRID:AB_262051 , 1:100 , raised in mouse ) , and anti-Repo ( RRID:AB_528448 , 1:5 , raised in mouse ) . The secondary antibodies were anti-mouse-Cy2 ( RRID:AB_2307343 , 1:100 , raised in goat ) , anti-mouse-Cy3 ( RRID:AB_2338680 , 1:200 , raised in goat ) , anti-rabbit-Cy2 ( RRID:AB_2338021 , 1:100 , raised in goat ) , and anti-rabbit-Cy3 ( RRID:AB_2338000 , 1:200 , raised in goat ) . Confocal images were captured with an LSM5 Pascal confocal microscope ( Zeiss , Jena , Germany ) . To exclude a possible bleed-through of the signals , sequential channel acquisition was performed starting with Cy3 channel by using single excitation at 543 nm and a long pass emission filter LP560 , followed by Cy2 channel acquisition with single excitation at 488 nm and a single bandpass filter BP 505–530 nm . There was no bleed-through of the Cy2 signal to the Cy3 channel because Cy2 is not excited by the 543 nm laser line . Using a narrow bandpass filter between 505 nm and 530 nm guaranteed that cross talk of Cy3 excitation by the 488 laser line is not detected during Cy2 channel acquisition . Z-stacks are displayed as maximum projections if not stated otherwise . Total-RNA ( RNeasy Mini Kit , Qiagen , Hilden , Germany ) from staged male third instar larvae ( AEL 74–78 hr ) was treated with DNase I ( Invitrogen , Carlsbad , CA , USA ) according to the manufacturer’s instructions and used as a template for cDNA synthesis ( AMV First Strand cDNA Synthesis Kit for RT-PCR , Roche ) . qRT-PCR was conducted according to standard protocols using DyNAmo ColorFlash SYBR Green qPCR Kit ( Biozym , Hessisch Oldendorf , Germany ) and an iCycler iQ Real-Time PCR System ( Bio-Rad , Munich , Germany ) . Data were evaluated as described in ( Simon , 2003 ) . All experiments were repeated at least three times ( individual biological replicates , each consisting of at least three technical replicates ) . The sequences of primers used were as follows: dilp1 , 5´-GGGGCAGGATACTCTTTTAG-3´ and 5´-TCGGTAGACAGTAGATGGCT-3´; dilp2 , 5´-GTATGGTGTGCGAGGAGTAT-3´ and 5´-TGAGTACACCCCCAAGATAG-3´; dilp3 , 5´-AAGCTCTGTGTGTATGGCTT-3´ and 5´-AGCACAATATCTCAGCACCT-3´; dilp5 , 5´-AGTTCTCCTGTTCCTGATCC-3´ and 5´-CAGTGAGTTCATGTGGTGAG-3´; rp49 , 5´-AGGGTATCGACAACAGAGTG-3´ and 5´-CACCAGGAACTTCTTGAATC-3´ . Statistical analysis ( one-way ANOVA with pairwise comparisons ) was performed using OriginPro 8 software ( OriginLab Corporation , Northampton , MA , USA ) .	The hormone insulin and similar molecules called insulin-like peptides act as signals to control many processes in the body , including growth , stress responses and aging . Disrupting these signaling pathways can cause many diseases , with diabetes being the most common of these . Although the roles of the signaling pathways have been well studied , it is less clear how the body controls the production of insulin and insulin-like peptides . Neprilysins are enzymes that can cut other proteins and peptides by a process known as hydrolysis . Their targets ( known as “substrates” ) include peptides that regulate a range of cell processes , and neprilysins have therefore been linked with many diseases . Fruit flies have at least five different neprilysin enzymes , but their substrates have not yet been identified . One of these , known as Nep4A , is produced in muscle tissue and appears to be important for muscles to work properly . Hallier , Schiemann et al . reveal that Nep4A regulates the production of insulin-like peptides . The experiments used fruit fly larvae that had been genetically engineered so that the level of Nep4A could be altered in muscle tissue . Larvae with very high or very low levels of Nep4A eat less food , have smaller bodies and produce different amounts of insulin-like peptides compared to normal larvae . Further experiments show that Nep4A can hydrolyze a number of peptides that regulate the production and the release of insulin-like peptides . This suggests that the enzymatic activity of neprilysins plays a direct role in controlling the production of insulin . The next challenge is to find out whether these findings apply to humans and other animals that also have neprilysins .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2016	Drosophila neprilysins control insulin signaling and food intake via cleavage of regulatory peptides
Animal groups have emergent properties that result from simple interactions among individuals . However , we know little about why animals adopt different interaction rules because of sparse sampling among species . Here , we identify an interaction rule that holds across single and mixed-species flocks of four migratory shorebird species spanning a seven-fold range of body masses . The rule , aligning with a one-wingspan lateral distance to nearest neighbors in the same horizontal plane , scales linearly with wingspan but is independent of nearest neighbor distance and neighbor species . This rule propagates outward to create a global flock structure that we term the compound-V formation . We propose that this formation represents an intermediary between the cluster flocks of starlings and the simple-V formations of geese and other large migratory birds . We explore multiple hypotheses regarding the benefit of this flock structure and how it differs from structures observed in other flocking species . The collective movements of animals—from schooling fish to swarming insects and flocking birds—have long excited intrigue among observers of nature . Collective motion arises as an emergent property of interactions between individuals ( reviewed by Herbert-Read , 2016 and by Vicsek and Zafeiris , 2012 ) . Thus , much attention has been placed on identifying local interaction rules ( Ballerini et al . , 2008a; Herbert-Read et al . , 2011; Katz et al . , 2011; Lukeman et al . , 2010 ) and how those rules affect group structure and movement ( Buhl et al . , 2006; Hemelrijk and Hildenbrandt , 2012 ) . However , comparative data across species are still limited , preventing us from testing hypotheses regarding the evolution and diversity of collective movement patterns . Hundreds of bird species fly in groups , but most quantitative research has focused on starlings ( Attanasi et al . , 2014; Ballerini et al . , 2008b; Cavagna et al . , 2010 ) , homing pigeons ( Nagy et al . , 2013; Nagy et al . , 2010; Pettit et al . , 2015; Pettit et al . , 2013; Usherwood et al . , 2011 ) and birds that fly in V-formations ( Badgerow and Hainsworth , 1981; Cutts and Speakman , 1994; Hummel , 1983; Lissaman and Shollenberger , 1970; Maeng et al . , 2013; Portugal et al . , 2014; Weimerskirch et al . , 2001 ) . These data indicate that smaller birds fly in relatively dense cluster flocks that facilitate group cohesion and information transfer ( Attanasi et al . , 2014; Ballerini et al . , 2008a ) , whereas larger migratory birds fly in highly structured V formations ( also known as line or echelon formations ) that provide aerodynamic and energetic benefits ( Lissaman and Shollenberger , 1970; Portugal et al . , 2014; Weimerskirch et al . , 2001 ) . However , descriptive accounts of flock structure over a greater range of species ( Heppner , 1974; Piersma et al . , 1990 ) cover a range of flock types , spanning the extremes of V-formation and large cluster flocks . The species whose flocking behavior have been studied quantitatively differ in many ways that could be important for flocking , including body size , ecology , the frequency of aggregation and its behavioral context . Therefore , on the basis of the available data , it is difficult to conclude what factors cause birds to adopt a specific group formation , or even what factors affect interaction rules , positioning and behavior within flocks . We aimed to address these questions by collecting three-dimensional ( 3D ) trajectories of the birds in flocks of four shorebird species that have similar ecologies ( all forage in large groups in coastal habitats and migrate long distances ) but that cover a seven-fold range of body mass and two-fold range of wingspan . Our study species include dunlin ( Calidris alpina Linnaeus 1758; 56 g , 0 . 34 m wingspan ) , short-billed dowitcher ( Limnodromus griseus Gmelin 1789; 110 g , 0 . 52 m wingspan ) , American avocet ( Recurvirostra americana Gmelin 1789; 312 g , 0 . 72 m wingspan ) , and marbled godwit ( Limosa fedoa , Linnaeus , 1758; 370 g , 0 . 78 m ) . Molecular dating indicates that these species diverged from their nearest common ancestor approximately 50 million years ago ( Mya ) ( Baker et al . , 2007 ) , providing time for evolutionary diversification of flocking behavior . By comparing the group structure of birds across a range of body sizes and by comparing our data with those in the literature , we aimed to determine the extent to which flock structure varies across species with different body sizes and ecologies . We employ three approaches: ( 1 ) identification of local interaction rules by quantifying the relative positions of birds and their nearest neighbors; ( 2 ) quantification of the degree of spatial structure within flocks; and ( 3 ) measurement of individual speeds and wingbeat frequencies to examine how local and global position within the flock affect flights behavior . On the basis of existing flock data , we hypothesized that flocks of larger shorebird species would be more structured than those of smaller species ( recapitulating the trend of larger birds flying in highly structured V formations ) and that larger species would also exhibit aerodynamic formations more frequently . Because a previous study showed that flying in a cluster flock is energetically costly in pigeons ( Usherwood et al . , 2011 ) , we hypothesized that birds flying in the middle and rear of flocks and birds flying closer to their nearest neighbor would have reduced flight performance ( lower speed relative to their wingbeat frequency ) . Surprisingly , we found that all four species studied here flew in a flock structure that we term the compound-V formation . We propose that this structure might be an adaptation for aerodynamic flocking in migratory species , and that ecology is an underappreciated driver of the evolution of avian flocking behavior . We examined flock structure by quantifying the position of each bird with respect to its nearest neighbor . We used modal values to characterize typical neighbor positions because position distributions were skewed as a result of values being cropped at zero . In all flocks , nearest neighbors flying within the same horizontal plane [an elevation slice of ±1 wingspan , a mean of 56% of nearest neighbors across all flocks ( range 35–76% ) ] exhibited a distinctly peaked distribution , where modal neighbor position was offset both in front-back and lateral distance ( Figure 2a , b ) . By contrast , nearest neighbor birds flying outside the horizontal elevation slice of ±1 wingspan were distributed randomly with a peak directly above or below the focal bird ( Figure 2c , d ) . This indicates that shorebirds adopt alignment rules for neighbors flying within their same elevation slice . On average , birds flew at approximately the same height as their nearest leading neighbor ( −0 . 01 ± 0 . 02 m , mean ± s . d . for the median trailing height across all 18 flocks ) . Both nearest-neighbor lateral distance and front-back distance differed among flocks and species ( Figure 3a ) . Species wingspan strongly predicted modal lateral neighbor position ( linear regression , slope = 0 . 85 , R2 = 0 . 93 , F = 228 . 29 , p<0 . 0001 ) . Wingspan also predicted front-back distance ( slope = 0 . 70 , R2 = 0 . 86 , F = 99 . 57 , p<0 . 0001 ) , although less strongly than lateral distance . After scaling alignment positions to wingspan ( i . e . , dividing neighbor distances by species wingspan ) , a distinctive pattern emerges ( Figure 3b ) . Specifically , the flocks adopted a modal lateral distance of approximately one wingspan ( mean 1 . 04 , range 0 . 88–1 . 24 wingspans ) . This non-dimensionalized lateral distance had a weak inverse relationship to species wingspan ( linear regression , slope = −0 . 37 , R2 = 0 . 37 , F = 9 . 38 , p=0 . 007 ) and was not related to flock density ( i . e . nearest neighbor distance , non-dimensionalized by wingspan; linear regression , R2 = 0 . 07 , F = 1 . 15; p=0 . 30 ) . Non-dimensionalized trailing distance was inversely proportional to species wingspan ( linear regression , slope = −0 . 40 , R2 = 0 . 33 , F = 7 . 93 , p=0 . 012 ) and increased with non-dimensional flock density ( linear regression , slope = 0 . 13 , R2 = 0 . 58 , F = 22 . 39; p=0 . 0002 ) . In summary , across all four species , shorebirds adhere to a non-dimensional spacing rule of aligning to neighbors with a lateral offset of approximately one wingspan while allowing trailing distance to vary with flock density . Data from mixed-species flocks of godwits and dowitchers further support the non-dimensional nature of the lateral spacing rule within individual flocks . Both dowitchers and godwits adjusted their lateral spacing depending on the species of their neighbor ( Figure 4 ) . Godwits following conspecifics had a modal lateral spacing of 0 . 76 m , or 0 . 97 godwit wingspans . When following the smaller dowitchers , godwits reduced the modal lateral distance to 0 . 60 m or 0 . 92 wingspans when calculated using the average wingspan of dowitchers and godwits ( Mann-Whitney U = 525 , 684; n1 = 1034; n2 = 81; p=0 . 0004 ) . Dowitchers following conspecifics flew with a modal lateral distance of 0 . 51 m , or 0 . 98 dowitcher wingspans . When following the larger godwits , dowitchers increased the modal lateral distance to 0 . 58 m or 0 . 89 average wingspans ( Mann-Whitney U test; U = 66 , 341; n1 = 743; n2 = 149; p<0 . 0003 ) . While recording the larger cluster flocks , we also recorded four godwit simple-V formations of between 16 and 44 individuals , which were recorded for between 42 and 211 frames ( Figure 5 ) . Here we compare the positioning of godwits in simple and compound-V formations . In both cases , nearest neighbors were most commonly in the same horizontal plane ( mean of 61% in godwit cluster flocks , 97 . 9% in godwit simple-V formations ) , defined as extending one wingspan above and below the focal bird , with the follower positioned over a narrow lateral range and wider range of trailing distances ( Figures 3b and 5b ) . The modal lateral position in the simple-V formations was slightly less ( mean of 0 . 8 wingspans ) than that in the compound-V formations , where the mean modal lateral position among godwit flocks was 0 . 96 wingspans ( Generalized Linear Model with terms for flock and simple versus compound-V formation; p<0 . 0001; Figure 5 , Figure 5—source data 1 ) . The modal trailing distance in simple-V formations was 0 . 50 wingspans; in compound-V formations of godwits , the mean modal trailing distance was 0 . 86 wingspans . We next examined how individual neighbor alignment rules relate to flock structure . We measured the angular distribution of neighbors at distances of two , four , six , and eight wingspans and at the maximum distance at which half of the flock remains in the flock’s core ( range 5 . 8–24 . 3 wingspans ) . This last measure was used as a proxy for whole flock structure while avoiding edge effects ( see 'Materials and methods' ) . For this analysis , we included all neighbors flying within a ±15 degree elevation slice relative to each focal bird . This was used instead of the ±1 wingspan slice used in other analyses ( e . g . , Figure 2 , Figure 3 ) because this metric corresponds to a decreasing proportion of the volume at further distances . At a distance of two wingspans , flocks were consistently asymmetrical , with trailing birds more frequently flying to the left of their leading neighbors in 12 of 18 flocks and to the right of their nearest leading neighbors in the remaining six flocks . This asymmetry persisted at all distances within the flock ( Figure 6a ) , including the overall flock shape ( Figure 6b ) . The direction of asymmetry was independent of relative camera viewing direction and flock turning direction but was positively correlated with relative wind direction ( see statistical results in Table 2 ) . We quantified several biomechanically relevant parameters from individual birds in flocks , including ground speed , estimated air speed , ascent or descent speed , wingbeat frequency and flapping phase . We created linear mixed effects ( LME ) statistical models to predict wingbeat frequency and airspeed from local and global flock positions and other flight parameters ( Table 3 ) . While speeds were measured for all individuals , flapping frequency and phase were only available from six cluster flocks in which birds were sufficiently close to the cameras to allow wingbeat measurements and for the simple-V formations . We examine only data for which wingbeat and estimated air speed data were available ( N = 3306 individuals ) . We were also unable to measure flapping parameters from Dunlin , the smallest species recorded here . We observed several individual and flock effects on flight speed and wingbeat frequency ( see Table 3 for full statistical results ) . As expected , different species flew with different characteristic flapping frequencies ( LME , p<0 . 00001 for all species ) and speeds ( LME , p<0 . 00001 for dowitcher and avocet , p=0 . 00022 for godwit ) , and climbing flight was associated with an increase in flapping frequency ( LME , p<0 . 00001 ) . Birds flying near the front of the flock along the direction of travel ( birds were given a continuous index with 0 being the frontmost and 1 the rearward-most position ) flew faster ( LME , p<0 . 00001 ) and with a lower flapping frequency than those near the rear ( LME , p<0 . 00006 ) . Birds flying near the edge of the flock also flew faster than those in the middle ( LME , p<0 . 00001 ) . Higher flapping frequencies were correlated with slower flight ( LME , p<0 . 00001 ) . Birds flying within a plausible range of locations for aerodynamic interaction ( 0 . 7–1 . 5 wingspans lateral distance and within two wingspans overall distance of their leading neighbor , coded as ‘Aerodynamic neighbor’ in Table 3 ) flew faster than expected after controling for the other effects described above ( LME , p<0 . 00001 , Figure 7 ) . However , positioning in this aerodynamic interaction region had no effect on flapping frequency , as it was not in our best model of wingbeat frequency based on Bayesian information criteria ( BIC; Table 3 ) . Adding the aerodynamic neighbor to the best model makes the term non-significant ( LME , p=0 . 30 ) and increases the BIC of the model by 6 . 8 . We examined the compound-V-flock data for evidence of flapping synchronization by examining the temporal and spatial phase offset between pairs of nearest neighbors for which synchronous wingbeat frequency data were available for at least 20 frames ( see 'Materials and methods' ) . We found no evidence for temporal ( Rayleigh test; N = 117; Z = 1 . 98; p=0 . 14 ) or spatial wingbeat synchronization ( Rayleigh test; N = 117; Z = 1 . 28; p=0 . 27 ) in the compound-V-formation shorebird flocks . We performed the same tests on the simple-V formation of godwits and again found no support for phasing relationships ( Rayleigh test; temporal phasing N = 39; Z = 0 . 09; p=0 . 90; spatial phasing; N = 39; Z = 0 . 46 , p=0 . 63 ) . Here , we report on the first cross-species quantitative analysis of bird flocking behavior . On the basis of previous studies , we predicted that larger species would adopt more structured flocks and would exhibit more frequent aerodynamic positioning . Neither of these hypotheses were supported by our data . Instead , we document a flock structure that we term the compound-V formation , in which birds in cluster flocks align to nearest neighbors within their same elevation slice ( ±1 wingspan ) with a one-wingspan lateral offset while allowing front-back distance to fluctuate with flock density ( Figures 2 , 3 and 4 ) . This flock type is similar to the shorebird ‘cluster’ and ‘bunch’ formations described by Piersma et al . ( 1990 ) and the ‘front cluster’ of Heppner ( 1974 ) . Here , this structure was observed in single- and mixed-species flocks of four shorebird species covering a seven-fold range in body mass . The simple alignment rule produces a flock structure that can be observed at all spatial scales within the flock , including overall flock shape ( Figure 6 ) . This is in contrast to flocks of other species , such as starlings , in which structure is only observed within each neighbor’s six nearest neighbors , equivalent to 1 . 2–2 . 7 wingspans ( Ballerini et al . , 2008a ) . Our data also show that shorebird global flock alignment is responsive to estimated local wind conditions ( Table 2 ) , and future work exploring this interaction may allow identification of the mechanism that governs the overall alignment . Wind conditions did not have discernible effects on local alignment , possibly because of the uncertainty in the measurement of the wind vector itself . Mixed-species assemblages typically represent around 10% of migratory shorebird flocks ( Piersma et al . , 1990 ) , possibly because species that have different preferred flight speeds would have to compromise their flight speed in order to remain together as a group . Our data include ~10% mixed-species flocks ( 2 of 18 ) and support the hypothesis that differences in preferred flight speed influence whether different species flock together . We documented two mixed-species flocks of godwit and dowitcher; these flocks had an airspeed of 9 . 02 ± 0 . 34 m s−1 ( mean ± s . d . , n = 2 ) . The single-species godwit and dowitcher flocks had airspeeds of 9 . 64 ± 1 . 53 ( n = 3 ) and 8 . 99 ± 1 . 92 m s−1 ( n = 3 ) , respectively . Dunlin , which were present at the same time as godwits and dowitchers but did not fly in mixed flocks with these species , had an airspeed of 7 . 58 ± 0 . 11 m s−1 ( n = 7 ) . Similarly , Avocets not observed to mix with other species at our field site , and they flew with airspeeds of 8 . 04 ± 0 . 13 m s−1 ( n = 3 ) . Thus , the similarity in preferred flight speeds among godwits and dowitchers might be important for these species to form mixed-species flocks . Dunlin and avocets also flew at similar airspeeds , but were not observed in mixed-species flocks , perhaps because of their large difference in body size , wingbeat frequency , and/or maneuverability . The flock data presented here also include other interesting results that lack clear explanations . Flight speed varied with position from front to rear and from center to margin ( Table 2 ) , implying that the flocks were not necessarily in equilibrium . This might cause larger flocks to separate into several smaller flocks over time , consistent with the observation that the arrival group size of migratory species is typically smaller than the departure group size ( Piersma et al . , 1990 ) . Because birds in simple and compound-V formations adopt similar neighbor alignment rules , functional hypotheses for simple-V formations might also apply to compound-V formations . These include collision avoidance and information transfer ( Dill et al . , 1997 ) . Collision avoidance is a plausible hypothesis to explain the formation of simple-V formations because they theoretically permit birds to keep all neighbors out of their direct path of travel . This is not the case for compound-V formations , where many birds are flying in front of and behind one another ( Figure 1b; Figure 6b ) . The problem of collision avoidance is exacerbated in compound-V formation because birds tend to fly in the same horizontal plane . A better strategy for collision avoidance is to fly in a three-dimensional shape , such as that adopted by flocks of chimney swifts ( Evangelista et al . , 2017 ) . In these flocks , the most common neighbor position is further lateral than in the shorebird flocks and with a shorter trailing distance , more completely moving those individuals out of the path of other flock members ( Figure 3—figure supplement 1 ) . Finally , even in the simple-V formation recorded here ( Figure 5 ) , birds flew with approximately 20% of wingspan overlap and so did not have an entirely clear forward path . Thus , collision avoidance appears to be an unlikely explanation for the structuring of both the compound-V and simple-V formations recorded here . Simple and compound-V formations might also be structured to maximize the observability of neighbors , facilitating information transfer by helping birds to detect and respond to changes in neighbor speed or direction , and improving flock cohesiveness by allowing information to propagate through the flock more quickly . Dill et al . ( 1997 ) proposed that birds in V formation should maximize the measurement of neighbor movements by aligning at a 35 . 3 degree angle ( relative to the direction of travel ) , or alternatively should maximize the measurement of neighbor speed by aligning at a 63 . 4 degree angle . The shorebird flocks examined here had modal neighbor-position alignment angles ranging from 33 . 7 to 51 . 8 with an average of 41 . 2 degrees . Neither this mean angle nor the nearly 20-degree range in alignment angle is consistent with Dill’s hypotheses or others calling for a single optimal alignment angle . Our finding that lateral spacing is uncorrelated with flock density , whereas trailing spacing increases with decreasing density , shows that the shorebird flocks are more organized in lateral distance than in trailing distance or alignment angle . Thus , hypotheses calling for organization based on alignment angle , whether to maximize information transfer or to keep lead birds in the visual fovea of trailing neighbors in a V formation ( Badgerow and Hainsworth , 1981 ) , are not well supported by our results . Theoretical ( Badgerow and Hainsworth , 1981; Hummel , 1983; Lissaman and Shollenberger , 1970; Maeng et al . , 2013 ) and empirical research ( Portugal et al . , 2014; Weimerskirch et al . , 2001 ) has provided support for the hypothesis that birds flying in simple-V formations gain aerodynamic and energetic benefits , and we propose that such benefits might also explain why birds adopt the compound-V formation . In both cases , birds fly with a lateral offset of approximately one wingspan while allowing trailing distance to vary ( Figures 3 and 5 ) , facilitating aerodynamic interaction . When compared to simple-V formations , compound-V formations allow greater flock densities , which should allow more rapid information transfer ( Attanasi et al . , 2014 ) , larger flock sizes , and improved predator defense ( Powell , 1974 ) . Analyses of the airspeeds and wingbeat frequencies of flocking shorebirds provide some support for the aerodynamic alignment hypothesis . Birds flying in positions where beneficial aerodynamic interactions are predicted to occur flew faster than expected after controling for other factors ( aerodynamic neighbor term in Table 3 , linear mixed effects model for airspeed , p<0 . 0001 ) . Over the entire dataset , 29 . 7% of nearest neighbor positions were in the ‘aerodynamic neighbor’ location ( Figure 3—figure supplement 2 ) , compared with only 3 . 4% of nearest neighbor in flocking chimney swifts ( Figure 3—figure supplement 1 ) . This faster flight should produce a reduced cost of transport , assuming there are no unmeasured compensating factors such as a simultaneous increase in stroke amplitude . Nevertheless , this speculative interpretation of the compound-V formation raises many new questions , such as how birds in the flock can maintain different speeds without separating and why an aerodynamic benefit would manifest as an increase in speed instead of , for example , a reduction in flapping frequency and airspeed as suggested by theoretical models ( Hummel , 1983 ) . Furthermore , despite the similarities in modal position among compound-V and simple-V flocks ( Figures 3 and 5 ) , it is not clear whether a single set of adjustment rules or responses to changes in neighbor position can produce both flock types . These questions , and a definitive explanation for why birds adopt a compound-V formation , cannot be answered with the current dataset . Progress in these areas will depend on new theoretical modeling and data collection from on-bird loggers measuring physiological , flock positioning and biomechanical data from a variety of species over a range of behavioral contexts . Further videographic flock surveys may also improve understanding of the variety of flock types , especially when collected with careful attention to behavioral context and with full measurement of local environmental conditions . We recorded multi-camera video of freely behaving , wild birds in Humboldt County , California between 17 April and 27 April 2017 and between 20 December 2017 and 1 January 2018 . Recordings were made at the Arcata Marsh Wildlife Sanctuary ( 40°51'25 . 35"N , 124° 5'39 . 37"W ) and above agricultural fields in the Arcata bottoms ( 40°53'51 . 98"N , 124° 6'55 . 85"W ) . No birds were captured or handled , and we made efforts to avoid influencing bird behavior . Video was captured at 29 . 97 frames per second and 1920 × 1080 pixel resolution using three Canon 6D cameras with 35 mm or 50 mm lenses . Cameras were set along a 10 m transect and staggered in elevation . We set cameras up to overlook locations where birds aggregated during high tide or when foraging in agricultural fields . Flocking events included birds moving with the tide , or flushing in response to predators ( e . g . , peregrine falcons ) or for unknown reasons . Cameras recorded continuously for up to 3 hr per day . For analysis , we selected flocks that included at least 100 individuals and that had an orientation and size that allowed visual discrimination of individuals within the flock . We used the MATLAB R2017a ( Natick , MA , USA ) computer vision toolbox to generate code for detecting birds in video recordings . A foreground detector first separated moving objects from the stationary background . A gaussian filter was then applied to the image with a diameter matched to bird size under the recording conditions . Two-dimensional peak detection found local peaks in the smoothed image that were taken as potential bird positions . Under some conditions , overlapping wings of adjacent birds prevented accurate detection of many individuals . To overcome this problem , we developed a frame-averaging algorithm that helped to obscure the wings and to emphasize the bodies . Here , optic flow determines the overall movement of the flock for each frame . Using the optic flow measurements and two-dimensional interpolation , the algorithm subtracts movement between frames . A rolling 5-frame window is then applied to the entire video . This procedure highlights pixels that are moving in the same direction as the flock , such as the birds’ bodies , while filtering pixels that are moving in other directions , such as the wings . Camera calibration followed established methodology ( Hedrick , 2008; Jackson et al . , 2016; Theriault et al . , 2014 ) , with the exception that the distance between cameras , instead of an object placed in view of the cameras , was used to scale the scene . This approach allowed us to record in locations where it was infeasible to place calibration objects in front of the cameras ( e . g . , over water ) . The in-camera horizontal alignment feature was used to align cameras to the horizon . The pitch of the camera was measured with a digital inclinometer with 0 . 1-degree precision . This allowed alignment of the scene to gravity in post processing , with the vertical ( Z axis ) origin placed at the level of the cameras . This permitted direct measurement of the elevation of the birds relative to one another . Background objects that were visible in the scene were used as calibration points . We developed a preliminary calibration using stationary objects such as trees , poles , and sitting birds . We then added flying birds , ensuring that points covered a wide range of distances and elevations relative to the cameras . Calibrations had low direct linear transformation ( DLT ) residuals ( <0 . 5–1 pixel ) , indicating high-quality calibrations . Cameras were synchronized by broadcasting audio tones over Walkie Talkies ( Motorola Talkabout MH230 ) to each camera . Audio tones were broadcast approximately once every five minutes during recording . A time offset was determined for each pair of cameras using cross-correlation of the audio tracks . This offset allowed camera synchronization within ±one half of a frame , or 16 . 6 ms . In recordings where birds were relatively close to the camera ( <50 m ) and moving at relatively high pixel speeds , we used sub-frame interpolation to achieve increased synchronization accuracy of one tenth of a frame , or ±1 . 7 ms . To determine the subframe offset , we interpolated tracks of moving birds used as background points in the calibration at 0 . 1 frame intervals from −1 to +1 frame ( −1 . 0 , –0 . 9 , etc ) . We then calculated the DLT residual for a calibration with each combination of subframe-interpolated points for the three cameras . The set of offsets generating the lowest DLT residuals was used for the final calibration and applied to birds tracked in the study . To reconstruct the three-dimensional positions of birds in a flock , 2D detections of individuals must be correctly assigned between cameras . We modified established software for this task ( Evangelista et al . , 2017; Wu et al . , 2009 ) . Briefly , the software first finds all combinations of 2D points having DLT residuals of <3 pixels . The software iteratively generates 3D points , starting with points that have the lowest DLT residuals and only allowing a 2D detection to be reused a single time . This helps with the problem of occlusion while limiting the number of ‘ghost’ birds ( bird positions created from incorrectly matching detections among cameras ) . This process is repeated twice . The first iteration allows the user to determine a bounding region in the 3D space in which the flock is contained . In the second iteration , three-dimensional positions outside this bounding region are filtered before they can be considered as potential 3D points . After 3D points have been generated , they are linked between frames to generate individual flight tracks . Here , a Kalman filter predicts the position of each bird in the subsequent frame for the 2D information from each camera and for the reconstructed 3D positions . In the first frame , the Kalman filter is seeded using optic flow measurements . For each frame step , a cost matrix is created from weighted sums of the 2D and 3D errors between predicted track positions and each reconstructed 3D point . The Hungarian algorithm is used to find a global optimum that minimizes the error in track assignment . A track that is not given an assignment is continued with a gap of up to four frames ( 0 . 13 s ) , after which it ends and any re-detection of the bird in question will start a new track . We measured wingbeat frequencies in a subset of recordings in which birds were both large enough and close enough to cameras to discern wingbeat oscillations . This excluded flocks of our smallest species , dunlin , and some flocks that were relatively distant from cameras . To measure wingbeat frequency , we used blob analysis to find a bounding box for each bird in each frame . We excluded blobs for which the bounding box included two or more birds as determined using the track-assignment algorithm described above . We averaged four components of the bounding box to measure wingbeat phase: height , inverse of the width , detrended X-coordinate of top-left corner , and inverse of detrended Y-coordinate of the top-left corner . This allowed quantification of wingbeat phase independent of bird orientation with respect to the cameras . Wingbeat phase was averaged across cameras and bandpass filtered before a 128-point Fast Fourier Transform ( FFT ) was applied to measure wingbeat frequency . The frame rate of the cameras ( 29 . 97 frames per second ) and the FFT window determined a wingbeat frequency bin size of 0 . 12 wingbeats s−1 . Our method is similar to that used in a recent study of two corvid species ( Ling et al . , 2018 ) . We recorded two mixed-species flocks of godwits and dowitchers . The size difference between species allowed species identification using the detected pixel area and distance of each bird ( Figure 8 ) . Here , blob analysis quantifies the pixel area for each bird in each tracked frame . Area data were excluded when two tracked birds were within a single blob bounding box . A low-pass filter was applied to the sequence of pixel area data across frames for each tracked bird to remove wingbeat effects . An object’s pixel area scales with the inverse of the square root of distance . Therefore , for each frame , the square root of the filtered pixel area was multiplied by bird’s distance to provide a distance-scaled area . This value was averaged across frames and across cameras for each bird track . In mixed-species flocks , a histogram of the scaled area had two distinct peaks with only a small amount of overlap ( Figure 8a ) . Fitting two normal distributions to these data revealed an expected error rate in species identification of 3 . 3% . The scaled area where the two normal distributions intersect was used as the threshold for species identification . We quantified the relative position of each bird and its nearest neighbor in the flock ( Figure 2 ) . This was done separately for neighbors within ±1 wingspan in flight elevation—the potential positions at which aerodynamic interactions and collisions are plausible—and for neighbors beyond ±1 wingspan . For each flock , we calculated the modal lateral distance and modal front-back distance by taking the peak of a probability density function generated with a kernel density estimator and a smoothing bandwidth of 0 . 25 wingspans . We used modal values because distance calculations are truncated at zero , producing skewed distributions . In a subsequent analysis ( Figure 6 ) , we quantified the angular distribution of neighbors at distances of two , four , six , and eight wingspans , and at a maximum radius depending on the size of the flock . Two-wingspan bins centered at the reference distance were used for selecting data points ( e . g . birds within 1–3 wingspans were included in the two-wingspan bin ) . Our aim was to examine the extent of internal structure within the flock . Because edge effects could create the appearance of internal structure , we excluded birds whose edge distance was less than the wingspan of the bin being analyzed . For example , for the two-wingspan analysis , all birds within three wingspans of the horizontal edge of the flock were excluded . The maximum radius was taken as the median horizontal edge distance of all birds in the flock ( Figure 9 ) . This ensured that our analysis always included at least half of the flock . We conducted an analysis to test for temporal and spatial wingbeat phase synchronization , following previously established methods ( Portugal et al . , 2014 ) . We selected pairs of nearest neighbors in flocks where simultaneous wingbeat frequency data were available for both individuals for at least 20 frames ( 0 . 66 s ) . Cross correlation was used to determine the temporal phase offset between the birds . This value was divided by 2πd , where d is wingbeat duration , to attain a value between 0 and 2π . The spatial phase offset equals the temporal phase offset minus 2πλ , where λ is wingbeat wavelength . We tested for temporal and spatial synchrony by applying Rayleigh’s test for homogeneity of circular data to the temporal and spatial phase delays . We estimated local wind speed and direction for each flock using observed variation of ground speeds from birds flying in different directions . Ground speeds and flight directions were calculated for each bird at one-wingbeat time intervals . Median ground speed was calculated for each 10-degree bin having at least 500 data points . A circle was then fit to these median values , with the center of the circle representing a vector of wind direction and magnitude . Ground speeds and wind direction and magnitude were then used to calculate airspeeds . This approach is similar in principle to that used to estimate local wind speed from the drift in the ground reference frame position of circling vultures ( Weinzierl et al . , 2016 ) , and shares the important assumption that airspeed is independent of wind direction . However , birds are theoretically expected and empirically known to vary airspeed with wind speed when flying in order to reach a destination efficiently ( Hedrick et al . , 2018; Shamoun-Baranes et al . , 2007 ) . Whether this is the case for shorebird flocks ( making shorter flocks around the stopover point ) is unknown , so we did not attempt to model this possible effect . We compared our wind estimates to data from nearby weather stations . Our estimated wind direction and speed was typically within ±45 degrees and ±2 m s−1 of weather station data ( weather station KCAARCAT25 ) . To avoid disturbing the birds , we did not attempt to release helium balloons to measure local wind conditions at altitude . Note that because our analysis here is based almost entirely on the positions and speeds of birds relative to their neighbors , our results are largely insensitive to the wind speed and direction . However , precise determination of bird airspeed and wind direction is required to model the expected position of the wake of the bird , and the absence of this information means that it is not possible to determine when or even if trailing birds interact with the wake of a leading neighbor , or to predict what flapping phase offset would be appropriate for aerodynamically beneficial interaction . Analyses were conducted using the statistical toolbox in MATLAB r2017b ( The Mathworks , Natick , MA , USA ) . We tested uniformity of circular distributions using Rao’s test ( Fisher , 1995 ) . Because multiple peaks were sometimes present , modal values were calculated using a circular kernel density estimator as an indicator of the predominant alignment direction . For the biomechanical analysis , we used linear mixed-effects models to predict individual wingbeat frequency and airspeed from seven fixed effects—nearest neighbor distance , nearest-neighbor lateral distance , edge distance , airspeed , vertical speed , nearest neighbor species , front-back flock position and hypothesized aerodynamic positioning . Bayesian information criterion ( BIC ) was used for model selection . All P-values were computed assuming two-tailed distributions .	Birds often fly in flocks ranging from very structured V-formations to unstructured clusters . Many studies have tried to prove what causes birds to flock and how it benefits them . Flocks , for example , may help birds to avoid predators and to navigate . Flying in a V-shaped formation likely also gives aerodynamic benefits that can make it easier to fly long distances . Few studies , however , have measured how the positions of birds in a flock relate to things like flying speed or the frequency of wing flaps . This is because it was difficult to take such measurements in large flocks of moving birds . Advances in cameras and computers are now making it easier to track individual birds flying in large flocks . The technology allows scientists to measure how birds position themselves in relation to other birds , or how flock-positioning varies by bird size , species , ecology , and behaviors . Such measurements may help scientists better understand why and how birds flock . Corcoran and Hedrick now show that four different types of shorebirds position themselves in the same way when flying in a flock . In the experiments , digital cameras recorded video of 18 cluster-like flocks of four different species of birds flying over a bird sanctuary or agricultural fields . The flocks ranged in size from a hundred to a thousand birds . Some flocks had two types of bird . The four types of birds – dunlin , short-billed dowitcher , American avocet , and marbled godwit – live in similar environments but greatly vary in size and fly at different speeds . Corcoran and Hedrick measured individual bird positions using three-dimensional computer reconstructions of the flocks . Each bird – regardless of size or species – most commonly flew about one wingspan to the side and between a half to one-and-a-half wingspans back from the bird in front of it . Birds flying in simple V-shaped formations follow similar rules . This suggests that birds flying in clusters may also gain aerodynamic benefits .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology" ]	2019	Compound-V formations in shorebird flocks
ISWI family chromatin remodeling motors use sophisticated autoinhibition mechanisms to control nucleosome sliding . Yet how the different autoinhibitory domains are regulated is not well understood . Here we show that an acidic patch formed by histones H2A and H2B of the nucleosome relieves the autoinhibition imposed by the AutoN and the NegC regions of the human ISWI remodeler SNF2h . Further , by single molecule FRET we show that the acidic patch helps control the distance travelled per translocation event . We propose a model in which the acidic patch activates SNF2h by providing a landing pad for the NegC and AutoN auto-inhibitory domains . Interestingly , the INO80 complex is also strongly dependent on the acidic patch for nucleosome sliding , indicating that this substrate feature can regulate remodeling enzymes with substantially different mechanisms . We therefore hypothesize that regulating access to the acidic patch of the nucleosome plays a key role in coordinating the activities of different remodelers in the cell . Eukaryotic genomes are packaged into chromatin , enabling large amounts of DNA to fit into the spatial constraints of the nucleus . This packaging has long been appreciated as a passive barrier to DNA access by nuclear factors . The discovery that chromatin regulators play critical roles in virtually all nuclear processes has informed a more nuanced view of chromatin as a dynamic regulatory platform that coordinates access to the genetic material . The smallest unit of chromatin is the nucleosome , a DNA-protein complex composed of ~150 bp of DNA wrapped around an octamer of histone proteins ( Luger et al . , 1997 ) . Nucleosomes can further interact with each other and with other factors to form higher-order structures ( Luger et al . , 2012 ) . Consequently , cells have evolved several sophisticated strategies to regulate chromatin structure at the nucleosome level . These include the covalent modification of histone proteins and DNA , as well as non-covalent changes to the position or composition of nucleosomes at specific genomic loci . Many of the non-covalent transformations , ranging from sliding nucleosomes to the complete disassembly of the histone octamer , are catalyzed by ATP-dependent chromatin remodeling enzymes ( Zhou et al . , 2016 ) . Underscoring their central role in chromatin regulation , remodeling enzymes play essential roles in many processes including transcription , DNA replication , and DNA repair ( Falbo and Shen , 2006; Hota and Bruneau , 2016; Price and D'Andrea , 2013 ) . How a relatively small number of remodeler types carry out such diverse regulatory functions remains an area of active research , not least because much remains unknown regarding remodeler mechanisms for substrate recognition and the coupling of that recognition to activity . Chromatin remodelers are members of the SF2 superfamily of nucleic acid motors , which catalyze noncovalent changes to nucleic acid substrates ( Zhou et al . , 2016 ) . Chromatin remodelers , however , are unique in that they specifically mobilize DNA in the context of the nucleosome , where DNA is tightly bound to histone proteins . Remodelers are further classified into families based on the domain architecture of their ATPase subunit . These families differ in their specific biochemical activities ( Zhou et al . , 2016 ) . Substantial progress in our general understanding of remodeling mechanisms has been made by asking what elements of the nucleosome are important for remodeling in different families . For example , maximal remodeling by ISWI family remodelers , which primarily slide nucleosomes , requires DNA flanking the nucleosome and the N-terminal tail of histone H4 ( Clapier et al . , 2001; Yang et al . , 2006 ) . Conversely , maximal remodeling by SWI/SNF family remodelers , which carry out the most diverse set of changes to nucleosome structure , does not require these nucleosomal epitopes ( Guyon et al . , 1999 ) . Less is known about how these substrate cues are recognized and mechanistically coupled to remodeling . Some important insights have come from biochemical analyses and structures of remodelers in the absence of nucleosomes , which suggest that in the ground state , chromatin remodelers are held in an inactive conformation by family-specific autoinhibitory motifs ( Clapier and Cairns , 2012; Hauk et al . , 2010; Xia et al . , 2016; Yan et al . , 2016 ) . Binding to specific nucleosomal epitopes is thought to relieve this autoinhibition via conformational changes in the remodeler , but the details of this process remain unclear . At the same time , structures of several nucleosome-protein complexes are revealing that many of these proteins interact with a conserved acidic patch formed by histones H2A and H2B on the top surface of the nucleosome ( McGinty and Tan , 2016 ) . These proteins interact with the acidic patch using an ‘arginine anchor’ which nestles into a pocket formed by the α1–2 helices of histone H2A ( McGinty and Tan , 2016 ) . It is therefore plausible that chromatin remodelers also recognize the acidic patch . Indeed recent work has indicated that mutating the acidic patch reduces the activity of ISWI , SWI/SNF and some CHD family remodeling enzymes ( Dann et al . , 2017 ) . To investigate the mechanistic role of the acidic patch in nucleosome remodeling by the ISWI family of enzymes we used a combination of ensemble and single molecule methods in the context of the human ISWI enzyme , SNF2h . We observe that interactions with the acidic patch activate SNF2h by relieving auto-inhibition mediated by two conserved domains of ISWI enzymes . Our results further suggest that contacts with the acidic patch helps control the distance the nucleosome is moved during translocation events . Finally we find that the acidic patch also stimulates the activity of the INO80 complex . Together , these results highlight the broad and essential role the acidic patch plays in chromatin regulation . The human ISWI ATPase , SNF2h , preferentially slides nucleosomes toward longer flanking DNA ( Yang et al . , 2006 ) . As a consequence of this activity , SNF2h slides mononucleosomes towards the center of a DNA strand , an activity that can be detected by a native gel mobility assay ( Yang et al . , 2006 ) . Recent work has shown that mutating residues in the nucleosome acidic patch reduces the ability of SNF2h to expose nucleosomal DNA to restriction enzymes ( Dann et al . , 2017 ) . To investigate the mechanism by which such mutations affect the nucleosome sliding activity of SNF2h and to determine if these mutations affected SNF2h binding or catalysis , we first used a mutant H2A in which four key residues in the H2A acidic patch are replaced with alanines ( E61A , E64A , D90A , and D92A ) ( Figure 1A ) . We call nucleosomes in which all four of these H2A residues have been mutated to alanines ‘APM’ ( acidic patch mutant ) nucleosomes . Using a native gel mobility assay , we measured remodeling rates of wild type ( WT ) and APM mononucleosomes containing 60 bp of DNA flanking one end of the nucleosome ( 0/60 ) ( Figure 1B ) . APM nucleosomes are centered substantially slower than WT nucleosomes under conditions where SNF2h is in excess of nucleosomes ( Figure 1B and C ) . Since the acidic patch has been shown to be critical for nucleosome binding by several chromatin proteins , we expected the apparent Km , Kmapp , to be increased with APM nucleosomes . Interestingly however , Kmapp is affected substantially less by mutation of the acidic patch than the maximal rate constant for remodeling , kmax ( Figure 1D , 61 nM to 280 nM , corresponding to ~4 . 5 fold increased Kmapp , versus 2 . 5 min−1 to 0 . 012 min−1 , corresponding to ~200 fold reduced kmax ) . These results indicate that the acidic patch plays a larger role in regulating maximal nucleosome sliding activity than nucleosome binding . We also measured ATP hydrolysis under conditions where nucleosomes are in excess of SNF2h . At saturating nucleosome concentrations , APM nucleosomes stimulate ATP hydrolysis 4-fold less than WT ( Figure 1G , Figure 1—figure supplement 2 ) . Together , these results imply that the acidic patch plays a critical role in coupling ATP hydrolysis to remodeling after the binding step . To further investigate the importance of the acidic patch , we asked whether binding by another factor to the acidic patch could compete for nucleosome sliding by SNF2h . We used a peptide derived from the latency associated nuclear antigen ( LANA ) from Kaposi’s sarcoma-associated herpesvirus that has previously been shown to interact directly with the acidic patch via an arginine anchor ( Barbera et al . , 2006 ) . We carried out these experiments using sub-saturating concentrations of SNF2h . Under these conditions most of the nucleosomes are unbound , allowing direct competition between the LANA peptide and SNF2h for the acidic patch . The presence of the LANA peptide dramatically slows remodeling by SNF2h in a dose-dependent manner ( ~200 fold reduction in rate constant at 50 µM LANA peptide ) , while a peptide with a mutant arginine anchor does not have a detectable inhibitory effect ( Figure 1E and F ) . We measured a KI of 1 . 2 µM for inhibition by the LANA peptide , within 5-fold of its published affinity for the nucleosome ( Figure 1F ) ( Fang et al . , 2016 ) . These results indicate that remodeling by SNF2h requires an interaction with the acidic patch that is mutually exclusive with the binding of acidic patch interacting factors such as the LANA peptide . The results are also consistent with recent work showing that the LANA peptide can inhibit the restriction enzyme accessibility activity of the SNF2h containing complex , ACF ( Dann et al . , 2017 ) . We next investigated if the acidic patch residues act independently or cooperatively by measuring the effects of individual mutations . Interestingly , except for residue E64 , all the single alanine mutants have comparable defects as the four point mutants combined ( Figure 1—figure supplement 3 ) . This observation suggests that three of the four acidic patch residues tested here act cooperatively to promote sliding by SNF2h . We next investigated which regions of SNF2h might functionally interact with the acidic patch . In principle , these would include ( i ) regions that directly contact the acidic patch and ( ii ) regions that do not directly contact the acidic patch , but whose function is energetically coupled to the presence of the acidic patch . To investigate regions that may directly contact the acidic patch we carried out cross-linking mass spectrometry using the zero-length , carbodiimide based reagent EDC . This method catalyzes the formation of new amide bonds between protein carboxylates , such as the side chains of aspartate and glutamate residues , and amino groups and is therefore well suited to probing electrostatic interactors of the acidic patch . Hundreds of high confidence cross-linked residue pairs were identified using this approach ( Supplementary file 1 ) . To focus on mechanistically meaningful domain-domain interactions , we employed a semi-quantitative mass spectrometry method to compare the extent of SNF2h-nucleosome cross-linking in the presence of different nucleotides ( Figure 2B ) . In previous work we have shown that ADP•BeFx mimics an activated state of the SNF2h-nucleosome complex ( Leonard and Narlikar , 2015; Racki et al . , 2014; Sinha et al . , 2017 ) . We therefore , focused on domain level interactions that were enriched at least two-fold in the presence of ADP•BeFx relative to ADP . Cross-links between the H4 tail and RecA lobe 2 of SNF2h are strongly enhanced in the ADP•BeFx state ( Figure 2B ) . This result is consistent with previous work showing that the H4 tail activates ISWI remodelers and promotes a restricted active site conformation in the presence of ADP•BeFx ( Clapier et al . , 2001; Hamiche et al . , 2001; Racki et al . , 2014 ) . We also found that the ADP•BeFx state promotes specific cross-links between the H2A/H2B acidic patch and lysines in the extended AutoN region , RecA lobe 1 , the NegC region , and the DNA binding HAND-SANT-SLIDE ( HSS ) region of SNF2h ( Figure 2—figure supplement 4 ) . While cross-links between the acidic patch and the AutoN , HSS and NegC regions are compatible with structural constraints from previous studies , the cross-links with the RecA lobe one are not easily explained . Multiple studies of ISWI remodelers suggest that their RecA lobes bind at a location two DNA helical turns from the nucleosome dyad ( SHL ±2 ) , quite far from the H2A/H2B acidic patch ( Dang and Bartholomew , 2007; Kagalwala et al . , 2004; Schwanbeck et al . , 2004; Zofall et al . , 2006 ) . We hypothesize that the RecA-acidic patch cross-links arise from a population of higher-order SNF2h-nucleosome aggregates in our mass-spec samples and therefore focus below on cross-links to the remaining three regions . To test the functional significance of cross-links to AutoN , we mutated the lysines that crosslink to the acidic patch in the ADP•BeFx state . However , none of the mutants significantly altered remodeling by SNF2h ( Figure 2—figure supplement 1 ) . These results suggest that the lysines in the AutoN region are in proximity to the acidic patch but do not make mechanistically significant interactions . We speculated , however , that other residues in AutoN might make functional interactions with the acidic patch . Previous work has indicated that many nucleosome binding proteins recognize the acidic patch via arginine residues . Near the AutoN lysines that crosslink to the acidic patch resides a key autoinhibitory motif specific to ISWI family remodelers that contains two arginine residues ( Figure 2A , Figure 2—figure supplement 4 , R142 and R144 ) . Autoinhibition by these arginine residues is relieved by a basic patch on the H4 tail , a nucleosomal epitope essential for maximal remodeling by ISWI-family remodelers ( Clapier and Cairns , 2012; Clapier et al . , 2002 ) . We tested whether R142 and R144 functionally cooperate with the acidic patch by generating a mutated version of SNF2h with the two critical arginines of AutoN mutated to alanine ( 2RA ) , and measured the remodeling activity of this mutant . Consistent with previous reports , 2RA SNF2h remodels WT nucleosomes ~ 2 fold faster than WT SNF2h ( Clapier and Cairns , 2012; Yan et al . , 2016 ) . However , 2RA SNF2h remodels APM nucleosomes ~ 50 fold faster than WT SNF2h ( Figure 2F ) . This corresponds to a ~ 25 fold reduced dependency on the acidic patch for remodeling with 2RA SNF2h ( Figure 2F ) . The same trend was also seen by an ensemble FRET remodeling assay ( Figure 2—figure supplement 2 ) . Together , these data suggest that the acidic patch contributes to relief of autoinhibition by R142 and R144 in AutoN . Since arginine residues are known to mediate this interaction in other systems , binding of either of the two arginines in AutoN to the acidic patch could provide a physical explanation for acidic patch recognition . However , given that neither carbodiimide chemistry nor any other commonly used cross-linking chemistry labels arginine residues , we cannot determine whether R142 and R144 make direct contacts with the acidic patch based on our mass-spectrometry data . We were unable to observe detectable binding between an AutoN peptide containing the 2R residues and the nucleosomal acidic patch through pull down assays ( data not shown ) , suggesting that either these residues do not physically interact with the acidic patch or the surrounding regions of the SNF2h protein are required for stable binding . To investigate the functional significance of cross-links to the NegC region of SNF2h , we determined the effect of replacing a stretch of 32 residues in NegC with a flexible serine-glycine linker ( mNegC ) . NegC is another autoinhibitory region of SNF2h that imposes flanking DNA length sensitivity on SNF2h by specifically slowing down remodeling of nucleosomes without flanking DNA ( Leonard and Narlikar , 2015 ) . Consistent with previous work , mNegC SNF2h slides WT 0/60 nucleosomes ~ 1 . 2 fold faster than WT SNF2h ( Figure 2E and F ) ( Leonard and Narlikar , 2015 ) . However , mNegC SNF2h slides APM nucleosomes ~ 100 fold faster than WT SNF2h ( Figure 2F ) . As a result , sliding of APM nucleosomes by mNegC SNF2h is only ~2 fold slower than WT nucleosomes . Thus the mNegC mutation almost completely rescues the defect of the acidic patch mutation . These results suggest that residues in NegC also link activation of nucleosome sliding to acidic patch recognition . The third category of cross-links entailed lysine residues in the HSS regions . Mutants in these lysines greatly reduced remodeling and , in contrast to the 2RA and mNegC SNF2h mutants , did not rescue the defects caused by the acidic patch mutations ( data not shown ) . It is possible that the defects caused by these lysine mutations reflect direct contacts between the HSS residues and the acidic patch . However , because the HSS also contacts flanking DNA , the reduction in sliding rate could also arise from defects in binding DNA . To better understand which crosslinks are dependent on H2A acidic patch binding , we also performed SNF2h-nucleosome cross-linking reactions with ADP•BeFx in the presence of the LANA peptide which competes for acidic patch binding ( Figure 1E ) . Addition of the LANA peptide substantially reduced crosslinks between the H2A acidic patch and both AutoN and NegC ( Figure 2—figure supplement 5 ) . In contrast , crosslinks between the HSS and the acidic patch were less strongly affected by LANA addition ( Figure 2—figure supplement 5 ) . This suggests that HSS positioning near the acidic patch in the ADP•BeFx state is not strictly dependent on direct binding to the region of the H2A acidic patch contacted by the LANA peptide . As a result , we cannot unambiguously interpret HSS-acidic patch crosslinks as reflecting mechanistically significant interactions . Together , these data suggest that acidic patch recognition is strongly linked to relief of autoinhibition by NegC and AutoN for nucleosome sliding by SNF2h . Furthermore , activation of SNF2h involves conformational changes that bring both of these autoinhibitory domains and the HSS in closer proximity to the acidic patch . To gain additional insight into which steps in the remodeling cycle involve interaction with the acidic patch , we turned to a single-molecule FRET assay ( smFRET; Figure 3A ) . This assay is analogous to the ensemble FRET remodeling assay described in Figure 2—figure supplement 2 , in that the activity of SNF2h in sliding nucleosomes away from DNA ends increases the distance between a donor and acceptor dye pair , and thus decreases the measured FRET efficiency . However , with smFRET we can follow the remodeling of individual , surface-immobilized nucleosomes , and thereby gain insights into the activity of SNF2h reaction steps that are obscured in asynchronous , population-averaged ensemble assays . smFRET has previously been used to study remodeling by several ISWI family members , as well as by remodeling complexes from the SWI/SNF and INO80 families ( Blosser et al . , 2009; Deindl et al . , 2013; Harada et al . , 2016; Hwang et al . , 2014; Zhou et al . , 2018 ) . A key insight revealed by these smFRET studies is that ISWI family remodelers do not slide nucleosomes in a continuous manner , such that FRET decreases continuously over time , but rather in a series of alternating phases: a ‘pause’ phase , in which FRET ( and therefore nucleosome position ) remain constant , and a ‘translocation’ phase , in which the nucleosome is moved relative to the DNA end . These repeating pause phases , which so far appear to be specific to ISWI , are essential to our understanding of the mechanism of action of ISWI remodelers , because the overall remodeling rates observed in ensemble assays are dominated by the durations of the pause phases , not the translocation rate itself . Moreover , substrate cues such as the H4 tail and flanking DNA have been shown to be sensed in the pause phase , not the translocation phase , for ISWI family remodelers ( Hwang et al . , 2014 ) . That is , shorter flanking DNA or mutation of the H4 tail increases the durations of the pauses , thereby decreasing the overall remodeling rate , while having no effect on the actual sliding rate of the nucleosome . These results suggest a separation of regulation and activity in ISWI remodelers: translocation events are regularly interrupted by pauses that allow for the periodic interrogation of substrate cues . This pausing behavior also explains the ability of ISWI remodelers such as ACF to keep nucleosomes centered: if the nucleosome is translocated off-center , the interruption of this translocation by a new pause can trigger translocation in the opposite direction , restoring the nucleosome to a centered position . As shown in Figure 3B , we find that SNF2h alone , like the ACF complex and the yeast ISWI enzymes , shares this alternating pause and translocation behavior at the single-molecule level . We first wished to ascertain whether the acidic patch , like other nucleosomal epitopes recognized by ISWI remodelers , affects the regulatory pause phase . Because the rate of remodeling of APM nucleosomes by SNF2h is significantly slower ( on the order of hours ) than the rate of dye photobleaching ( on the order of minutes; Figure 3—figure supplement 1B ) , we assembled nucleosomes with a single point mutation in the acidic patch ( E64R nucleosomes ) . We used the E64R mutation rather than the E64A mutation because the defect caused by this mutation was better rescued by the 2RA mutation in SNF2h than the defect in E64A ( Figure 1—figure supplement 3 ) . The single-point mutation ( E64R ) is less deleterious than mutating all 4 acidic residues ( APM ) , and SNF2h remodels E64R nucleosomes ~ 40 fold more slowly than WT nucleosomes as opposed to 200-fold more slowly with APM nucleosomes ( Figure 1—figure supplement 3 ) . However , this remodeling rate is still very slow relative to the timescales typically measured by smFRET , which posed two additional challenges: an increase in the number of noise events ( dye blinking , intensity fluctuations , etc ) per remodeling trajectory , and an increase in the amount of data to be analyzed . These challenges were addressed through the use of custom in-house smFRET analysis software ( ‘Traces’ , https://github . com/stephlj/Traces ) ( Zhou et al . , 2018; Johnson et al . , 2018; copy archived at https://github . com/elifesciences-publications/Traces ) to streamline the analysis of large data sets , and the adaptation of a computationally fast , versatile , open-source hidden Markov model ( HMM ) library called pyhsmm to quantify the durations of the pauses ( see Materials and methods ) . As shown in the example trajectories in Figure 3C , remodeling of E64R nucleosomes by SNF2h proceeds through the same alternating pause and translocation phases as does remodeling of WT nucleosomes . However , the pauses are noticeably longer with E64R nucleosomes , by at least a factor of 2 ( Figure 3D ) , indicating that the acidic patch epitope , like other substrate cues , is indeed sensed during the regulatory pause phase . We note that remodeling of E64R nucleosomes in ensemble assays is significantly slower than the photobleaching rate ( Figure 3—figure supplement 1B ) , so that by smFRET we only detect the fraction of remodeling events that are faster than photobleaching . As a result , the remodeling rate obtained for the E64R nucleosomes by smFRET represents an upper bound on the true remodeling rate ( i . e . the E64R nucleosomes appear to remodel faster by smFRET than by ensemble assays ( Figure 3—figure supplement 1B ) ) . Given the ability of the 2RA mutation of the AutoN motif to partially rescue remodeling defects in APM nucleosomes in ensemble assays ( Figure 2C ) , we next asked whether this rescue is due to a restoration of wild-type pause durations . In agreement with our ensemble results ( Figure 2C ) , 2RA SNF2h remodels WT nucleosomes slightly faster than SNF2h by reducing pause durations ( Figure 3D ) . The reduction is minor ( ~1 . 3 fold ) , consistent with a previous smFRET study of the effect of the 2RA mutation on pause durations for the ACF complex ( Hwang et al . , 2014 ) . Furthermore , 2RA rescues the deleterious effect of the E64R acidic patch mutation , by nearly restoring wild-type pause durations ( Figure 3D ) . The effects of the E64R nucleosomal mutation and the 2RA SNF2h mutation on pause durations therefore mirror the effects on overall remodeling rates of APM nucleosomes: 2RA remodels E64R nucleosomes nearly as fast as SNF2h remodels WT nucleosomes , but not as fast as 2RA remodels WT nucleosomes ( Figure 1—figure supplement 3 ) . These results are consistent with a model in which relief of autoinhibition of the AutoN motif of SNF2h through direct or indirect interactions with the acidic patch enable pause exit ( that is , promote the translocation phase ) . The example traces in Figure 3C suggest an additional defect in remodeling of E64R nucleosomes: a reduction in the distance the nucleosome is moved during each translocation event ( which we called the step size ) . Note that after two translocation events , WT nucleosomes with SNF2h or 2RA are moved from ~0 . 95 FRET to ~0 . 4 FRET ( Figure 3C top and second from bottom ) , whereas after two translocation events the nucleosome in the E64R/SNF2h example trace has moved from ~0 . 95 FRET to ~0 . 75 FRET . Step size , like pause duration , plays an important role in regulating ISWI remodeler activity: since the pause durations dominate the overall remodeling rate , a smaller step size , which means more pauses per unit distance that the nucleosome is translocated , will mean a significant reduction in the overall remodeling rate ( as observed in ensemble assays ) . Step sizes can be quantified by converting the change in FRET between subsequent pauses to a change in the number of base pairs of DNA between the Cy5-labeled DNA end and the edge of the nucleosome . We accomplish this conversion by means of a calibration curve , described previously ( Zhou et al . , 2018 ) . Like other ISWI family remodelers , SNF2h moves WT nucleosomes with an initial large step ( ~8 bp ) followed by a smaller ( ~5 bp ) step ( Figure 3E , Figure 3—figure supplement 3; [Blosser et al . , 2009; Deindl et al . , 2013] ) . However , SNF2h moves E64R nucleosomes a shorter distance in each translocation phase , as indicated by the leftward shift of the cumulative distributions in the red curves of Figure 3E , relative to the black curves ( on average , E64R nucleosomes move about 6 bp in the first translocation and about 4 bp in the second ( Figure 3—figure supplement 3 ) ) . Mutation of AutoN has little effect on the step size in the context of WT nucleosomes , but largely restores the step size in the context of E64R nucleosomes to that of SNF2h with WT nucleosomes ( Figure 3E , magenta and blue curves ) . Given that ISWI family remodelers have been shown to translocate a nucleosome in elementary steps of 1–2 bp ( Deindl et al . , 2013 ) , our results suggest that with E64R nucleosomes , SNF2h takes fewer of these elementary steps in succession during the translocation phase , before entering a new pause phase . The major contributions to the remodeling defects observed with SNF2h and APM nucleosomes can therefore be attributed to two effects: first , an increase in pause duration , and second , a decrease in the distance travelled per translocation event , meaning that there are more ( and longer ) pauses in APM remodeling reactions per unit distance that the nucleosome is moved . Both of these effects are rescued by the 2RA mutation of the AutoN motif of SNF2h . We propose that the acidic patch is important for relieving autoinhibition by AutoN and thereby promoting exit from the pause phase . Further , we propose that the acidic patch is also involved in keeping AutoN out of the active site until the nucleosome has been translocated to the full extent ( ~8 bp initially , ~5 bp subsequently ) and a new pause phase is entered ( Figure 5 ) . The ISWI ATPase forms complexes with several accessory proteins that regulate its localization and activity ( He et al . , 2006; Oppikofer et al . , 2017; Tsukiyama et al . , 1995; Varga-Weisz et al . , 1997 ) . ACF is one of the best studied of these complexes . ACF is a heterodimer of the ISWI ATPase subunit ( SNF2h in humans ) and the accessory subunit Acf1 , and is implicated in gene repression , DNA replication , and DNA repair ( Collins et al . , 2002; Fyodorov et al . , 2004; Lan et al . , 2010 ) . Biochemically , human ACF has the same core activity as SNF2h , but displays greater nucleosome affinity , enhanced sliding rates , and better kinetic discrimination of flanking DNA length ( He et al . , 2006; Yang et al . , 2006 ) . Despite these similarities , recent evidence suggests that ACF has some mechanistic differences from SNF2h on its own . For instance , in the context of ACF , AutoN regulates flanking DNA length sensing through interaction of an Acf1-specific domain called WAC ( Hwang et al . , 2014 ) . SNF2h alone has no comparable domain . Furthermore , recent work has suggested that mutating the nucleosomal acidic patch causes a smaller defect in remodeling by ACF compared to SNF2h ( Dann et al . , 2017 ) . Given this difference , we asked whether ACF requires the acidic patch for remodeling beyond binding . At saturating concentrations of ACF , where binding does not contribute to the overall remodeling rate , ACF slides APM nucleosomes 10-fold more slowly than WT ( Figure 4A ) , indicating that ACF also uses the acidic patch in a step after binding . However , consistent with previous work , ACF is less dependent on the acidic patch compared to SNF2h alone ( Dann et al . , 2017 ) . We next asked whether ISWI-family complexes uniquely use the acidic patch , or whether it is also used by other chromatin remodeling enzymes . Recently , it was shown that while some CHD family remodelers slide nucleosomes largely independent of the acidic patch , others are dependent on it ( Dann et al . , 2017; Levendosky et al . , 2016 ) . It has been further shown that SWI/SNF family enzymes also require the acidic patch for maximal activity ( Dann et al . , 2017 ) . These observations raise the possibility that the acidic patch is a feature used by most remodeling enzymes . To address this issue we determined if the acidic patch is required by yeast INO80 , which is in a distinct family from the CHD , ISWI and SWI/SNF families . INO80 , like ACF , slides nucleosomes preferentially in the direction of longer flanking DNA and can also create evenly spaced nucleosome arrays ( Udugama et al . , 2011 ) . This sliding activity is thought to be important for positioning the +1 nucleosome at transcription start sites ( Krietenstein et al . , 2016 ) . Interestingly , we find that INO80 slides APM nucleosomes ~ 200 fold more slowly than WT nucleosomes at saturating concentrations ( Figure 4B ) . This result indicates that INO80 also uses the acidic patch post-binding , but is more dependent on the acidic patch than the ACF complex . In this study , we investigate the role of the highly conserved H2A acidic patch in chromatin remodeling by ISWI enzymes . We find that the acidic patch is used post-binding in order to activate remodeling by both INO80 and ISWI family remodelers . Furthermore , using a combination of ensemble and single molecule methods , we show that the acidic patch is used by SNF2h to relieve autoinhibition by the conserved AutoN and NegC motifs . Below we explore the mechanistic and regulatory implications of these results . ATP-dependent chromatin remodeling enzymes carry out specialized reactions on a complex substrate . Understanding how recognition of this substrate is coupled to activity can provide a means to understanding common principles underlying ATP-dependent chromatin remodeling mechanisms . Owing to decades of study , ISWI enzymes provide a useful model system to address this question . On the basis of crosslinking and footprinting studies , the ATPase domain of ISWI enzymes is thought to bind and translocate DNA two helical turns from the nucleosomal dyad ( SHL ± 2 ) ( Dang and Bartholomew , 2007; Kagalwala et al . , 2004; Schwanbeck et al . , 2004; Zofall et al . , 2006 ) . Work from several groups has also shown that for ISWI remodelers , recognition of both the H4 tail and flanking DNA enhances remodeling activity post-binding ( Clapier et al . , 2001; Hamiche et al . , 2001; Yang et al . , 2006 ) . While the mechanisms by which these nucleosome cues activate remodeling are not well understood , the ISWI domains that recognize these cues are known . The C-terminal HAND-SANT-SLIDE ( HSS ) domain mediates flanking DNA recognition , while the H4 tail appears to be directly recognized by the second RecA lobe within the ATPase domain ( Dang and Bartholomew , 2007; Yan et al . , 2016 ) . The acidic patch resides on a surface near SHL ± 6 , far from where the ATPase domain engages the nucleosome . How then might the acidic patch be recognized and used by ISWI remodelers ? Our results suggest that two known autoinhibitory regions of ISWI enzymes , the AutoN region and the NegC region , functionally interact with the acidic patch , because mutating these regions dramatically reduces the dependence of SNF2h on the acidic patch for sliding . This suggests that a large role of the acidic patch is to relieve autoinhibition by both AutoN and NegC . While we do not have evidence for a direct interaction between the acidic patch and the two arginines in AutoN , our cross-linking mass spectrometry data suggests that activation of the enzyme places residues within both AutoN and NegC near this location . Our smFRET work here and previous smFRET work suggests that the AutoN region inhibits the transition from the pause phase to the translocation phase ( Hwang et al . , 2014 ) . It has been shown that flanking DNA and the H4 tail are both sensed in the pause phase ( Hwang et al . , 2014 ) . Further , previous ensemble work has suggested that the NegC region inhibits the transition between a flanking DNA sensing state of SNF2h and a translocation competent state of SNF2h ( Leonard and Narlikar , 2015 ) . We therefore propose that the acidic patch helps promote the translocation competent state of SNF2h by providing an alternative binding site for NegC and AutoN ( Figure 5 ) . In addition to an increase in pause durations with the acidic patch mutations , the amount of DNA translocated within a translocation phase is reduced compared to WT nucleosomes . We hypothesize that translocation is interrupted by premature reversion of the enzyme to the autoinhibited state in the absence of stabilizing interactions with the acidic patch ( Figure 5 ) . As a result , more pauses are encountered per distance translocated . Our results lead to a model in which the acidic patch provides a binding surface for NegC and AutoN that sequesters these regions from inhibiting SNF2h ( Figure 5 ) . Combined with previous work , our results underscore how the strong coupling of relief of autoinhibition to recognition of two conserved nucleosome cues ( the H4 tail and the acidic patch ) make this motor exquisitely specific for its complex substrate . The acidic patch increases Kmapp for SNF2h by ~5 fold , suggesting that interactions with the acidic patch also stabilize SNF2h binding . Thus the acidic patch appears to be used in at least two distinct steps of the SNF2h remodeling reaction . What residues in SNF2h might be interacting with the acidic patch in these two steps ? In contrast to the rescuing effects of AutoN mutations on maximal activity , AutoN mutations additively increase Kmapp beyond solely mutating the acidic patch ( Figure 2—figure supplement 3 ) . This suggests that AutoN and the acidic patch do not cooperate in the ground state . It is thus possible that different SNF2h domains contact the acidic patch in different steps of the remodeling reaction . The different set of cross-links observed between the acidic patch and SNF2h regions in the ADP vs . ADP•BeFx state are consistent with such a possibility . Based on previous work , we have hypothesized that SNF2h undergoes a large conformational change prior to adopting a translocation competent state that repositions the C-terminus from binding flanking DNA towards binding the nucleosome core ( Leonard and Narlikar , 2015 ) . Our crosslinking-MS data provides additional insights into the structural rearrangements that accompany enzyme activation . We find that crosslinks between the H2A/H2B acidic patch and the SANT domain increase in the ADP•BeFx-bound state compared to the ADP-bound state ( Figure 2B ) , consistent with the HSS binding to the nucleosome core . Interestingly , these crosslinks are not substantially reduced by LANA peptide binding , suggesting that the location of the HSS in the ADP•BeFx-bound state is not strongly dependent on direct contacts with the residues contacted by the LANA peptide ( Figure 2—figure supplement 5 ) . This result raises the possibility that the SANT domain and LANA peptide occupy adjacent regions on the nucleosome core . The SANT domain may then cross-link to the acidic patch when these sites are transiently exposed due to dynamics in LANA peptide binding . In contrast , NegC and AutoN crosslinks to the acidic patch are substantially reduced by the LANA peptide , suggesting LANA binding directly or indirectly displaces these regions from the nucleosome . Overall our results suggest that the key accessory regions of SNF2h namely , HSS , NegC and AutoN , are all positioned near the acidic patch in the activated state . In agreement with this observation , crosslinks between the C-terminus of SNF2h and NegC increase in the ADP•BeFx state ( Figure 2B ) . The close positioning of multiple SNF2h accessory domains near the acidic patch raises the possibility that contacts between accessory domains may play a role in promoting the translocation-competent state . While substantial future work would be needed to test this possibility , it is analogous to recent observations with the yeast CHD1 remodeling motor that contacts between its N-terminus and the C-terminal DNA binding domain regulate the sliding reaction ( Sundaramoorthy et al . , 2017 ) . Most chromatin remodeling ATPases also form large multi-subunit complexes , which regulate the activity and specificity of the remodeling reaction . Here we find that the human ISWI complex , ACF , requires the acidic patch for maximal activity but shows a ~ 20 fold smaller defect upon mutation of the acidic patch than observed with SNF2h alone . This result is qualitatively consistent with recent studies showing that the acidic patch has a smaller role in the activity of ACF vs . SNF2h ( Dann et al . , 2017 ) . Our results here provide a mechanistic framework for understanding these recent observations . In particular , our results suggest that the accessory protein Acf1 alters the mechanism of relief of autoinhibition by the acidic patch in a manner that makes the reaction less dependent on the acidic patch , perhaps by providing an alternative binding partner for the AutoN and NegC domains . Such a domain would be analogous to the WAC motif of Acf1 , which provides an alternative binding partner for the H4 tail ( Hwang et al . , 2014 ) . In contrast to ACF , we find that the yeast INO80 complex , a large multisubunit complex , is as dependent on the acidic patch for nucleosome sliding as the isolated SNF2h ATPase . Importantly , INO80 family remodelers are insensitive to the presence of the H4 tail and no AutoN-like or NegC like motif has been identified in the ATPase subunit of INO80 ( Udugama et al . , 2011 ) . The acidic patch must then activate INO80 through a mechanism distinct from ISWI complexes . Determining how the acidic patch is used by INO80 ATPases , and what roles the ATPase and accessory subunits play in this mechanism , are important areas of future study . Combined with previous results ( Dann et al . , 2017 ) , our results suggest that several families of chromatin remodelers require the acidic patch for remodeling . However , it is possible that this surface is not a universal requirement for chromatin remodeling , as yeast CHD1 can remodel nucleosomes largely independent of the acidic patch ( Levendosky et al . , 2016 ) . CHD1 instead uses an unidentified aspect of the histone H2A/H2B dimer to promote remodeling ( Levendosky et al . , 2016 ) . Yeast CHD1 and ISWI family remodelers have been thought to share a common remodeling mechanism , as these families share biochemical activities , like nucleosome sliding and spacing , and substrate cues required for maximal remodeling activity , such as the H4 tail and flanking DNA ( Ferreira et al . , 2007; Stockdale et al . , 2006 ) . However , the enormous difference in dependence on the acidic patch between yeast CHD1 and ISWI enzymes raise the possibility that these families remodel nucleosomes through distinct mechanisms ( Levendosky et al . , 2016 ) . Importantly , while yeast CHD1 shares domains with ISWI remodelers such as the SANT-SLIDE domains at the C-terminus and a version of the NegC region , called the C-terminal bridge ( Hauk et al . , 2010; Ryan et al . , 2011 ) , CHD family remodelers do not appear to possess an AutoN motif . Instead , remodelers like CHD1 have an N-terminal double chromodomain which has an analogous role as an autoinhibitiory domain that is relieved by H4 tail binding ( Hauk et al . , 2010 ) . At a primary level , the requirement of the acidic patch provides a powerful means for remodelers to sense and respond to chromatin structure and nucleosome content . Thus , nucleosomes lacking histone H2A-H2B dimers or containing a modified acidic patch through histone variants or covalent modifications may be recognized and remodeled differently than canonical nucleosomes by different remodelers ( Dann et al . , 2017 ) . Consistent with this possibility , nucleosomes containing histone H2Az , which have an extended acidic patch , are remodeled ~2 fold faster by ISWI complexes than canonical nucleosomes ( Goldman et al . , 2010 ) . Analogously , recent work has shown that INO80 preferentially slides H2AZ nucleosomes over H2A nucleosomes ( Brahma et al . , 2017 ) . Finally , given the growing list of factors that recognize the acidic patch , it is likely that remodelers and other chromatin binding proteins compete for access to the acidic patch . Indeed , binding by the LANA peptide to the acidic patch competes for remodeling by SNF2h . Sensitivity to the acidic patch could be a general mechanism to regulate the outcome of chromatin remodeling at loci where remodelers and other factors are jostling for access to their chromatin substrates . SNF2h was purified from E . coli as described previously with minor modifications ( Leonard and Narlikar , 2015 ) . DNA was precipitated following cell lysis by addition of 5% w/v polyethylenimine ( P3143 , Sigma-Aldrich , St . Louis , MO ) pH 7 . 9 dropwise to a final concentration of 0 . 1% and clarified by centrifugation . Following cobalt affinity purification , the 6xHis tag was cleaved overnight with TEV protease and dialysed into SEC Buffer . TEV-cleaved SNF2h was then purified by anion exchange chromatography using a HiTrap Q column and size exclusion chromatography ( GE Life Sciences , Pittsburgh , MA ) . SNF2h concentration was determined by SDS-PAGE with BSA protein standards and staining with SYPRO Red ( Thermo Fisher , Waltham , MA ) . Human ACF was expressed and purified recombinantly from Sf9 insect cells by FLAG immunoaffinity purification as described previously with minor modifications ( Aalfs et al . , 2001 ) . SNF2h-FLAG and Acf1 were expressed separately via infection with baculovirus . Nuclear extracts from each construct were generated and mixed together at a 10:1 Acf1:SNF2h-FLAG volume ratio to ensure full assembly of the complex . This mixture was then bound to FLAG M2-affinity resin ( Sigma-Aldrich , St . Louis , MO ) , washed with increasing KCl concentrations , and eluted with buffer with 100 mM KCl and 1 mg/mL FLAG Peptide . ACF concentration was determined by SDS-PAGE with BSA standards and based on the intensity of the Acf1 band . INO80 was purified by FLAG immunoprecipitation based on previously published methods ( Shen , 2004; Zhou et al . , 2018 ) . Briefly , S . cerevisae with endogenously flag-tagged INO80 was grown in YEPD at 30°C to saturation . Cells were pelleted by centrifugation for 10 min at 5000 rpm , resuspended with buffer H0 . 3 ( 25 mM HEPES , pH 7 . 5 , 1 mM EDTA , pH 8 . 0 , 10% glycerol , 0 . 02% NP-40 , 0 . 3 M KCl ) , and pelleted again . Pelleted cells were then extruded through a 60 mL syringe into liquid nitrogen to create ‘noodles’ . Cell ‘noodles’ were then lysed using a Tissue Lyser II ( Qiagen , Hilden Germany ) cooled in liquid nitrogen . Frozen lysate powder was resuspended in equal volume of H0 . 3 and spun in an SW28 rotor for 2 hr at 25 , 000 rpm at 4°C . Clarified lysate was mixed with equal volume buffer H0 . 3 and applied to FLAG M2-affinity resin ( 1 mL bead slurry per 40 mL of cleared lysate ) equilibrated with H0 . 3 and incubated for 3 hr at 4°C . Resin was washed with 3 × 50 mL buffer H0 . 5 ( H0 . 3 buffer except with 0 . 5 M KCl ) followed by 3 × 10 mL washes with buffer H0 . 1 ( 0 . 1M KCl ) and eluted with H0 . 1 supplemented with 1 mg/mL FLAG peptide . Eluate was concentrated , aliquoted , flash frozen in liquid nitrogen , and stored at −80°C . INO80 concentration was determined by SDS-PAGE with BSA standards , based on the intensity of the Ino80-flag band . Recombinant Xenopus laevis histones were expressed and purified from E . coli as previously described ( Luger et al . , 1999 ) . Histone H2A E61A , E64A , D90A , D92A expression plasmid was a generous gift from the Tan lab at Penn State . Purified histone H2A E64R was provided by the Wolberger lab . Histone octamer was reconstituted as previously described ( Luger et al . , 1999; Zhou and Narlikar , 2016 ) , except for smFRET nucleosomes where a 2:1 unlabeled:labeled H3 mixture was used during octamer assembly to generate nucleosomes with mostly one H3 or neither H3 labeled . Histone H3 with a cysteine introduced at position 33 was labeled with either Cy3 ( for smFRET ) or Cy5 ( for ensemble assays ) prior to histone octamer assembly via cysteine-maleimide chemistry . Cy3-labeled ( for ensemble assays ) and Cyanine 5 SE-labeled and biotinylated DNAs ( for smFRET ) were generated by PCR with HPLC-purified , labeled primers ( Cy5 primers: TriLink Biotechnologies , San Diego , CA; Cy3 and biotinylated primers: IDT , Coralville , IA ) and purified by PAGE . The strong , synthetic 601 nucleosome positioning sequence ( Lowary and Widom , 1998 ) was used to assemble all nucleosomes in this study , with an arbitrary sequence for DNA flanking the 601 positioning sequence ( Figure 1—figure supplement 1 ) . These DNAs were assembled with either wild-type or APM octamers by salt gradient dialysis and purified by glycerol gradient centrifugation ( Zhou and Narlikar , 2016 ) . All remodeling reactions were performed under single turnover conditions ( enzyme in excess of nucleosomes ) . Reactions with SNF2h were performed at 20°C with 20 nM nucleosomes , 12 . 5 mM HEPES pH 7 . 5 , 2 mM Tris pH 7 . 5 , 70 mM KCl , 5 mM ATP•MgCl2 , 3 mM MgCl2 , 0 . 02% NP40 , and ~3% ( v/v ) glycerol . Reactions with ACF and INO80 were performed as above at 30°C and with minor changes in buffer composition ( ACF: 10 nM nucleosomes , 12 . 5 mM HEPES pH 7 . 9 , 2 mM Tris pH 7 . 5 , 60 mM KCl , 2 mM ATP•MgCl2 , 3 mM MgCl2 , 0 . 02% NP40 , 0 . 3 mg/mL FLAG peptide , and ~9% glycerol; INO80: 10 nM nucleosomes , 40 mM Tris pH 7 . 5 , 60 mM KCl , 2 mM ATP•MgCl2 , 1 . 1 mM MgCl2 , 0 . 02% NP40 , 0 . 5 mg/mL FLAG peptide , and 1% glycerol ) . Reactions were started with addition of enzyme and time points were quenched with excess ADP and plasmid DNA . Time points were then resolved by native PAGE ( 6% acrylamide , 0 . 5XTBE ) and scanned on a Typhoon variable mode imager ( GE Life Sciences , Pittsburgh , PA ) by scanning for fluorescent labels . Gels were then quantified by densitometry using ImageJ . The fraction of nucleosomes end-positioned ( i . e . unremodeled ) at a given time point was determined by the ratio of fast-migrating nucleosomes to the total nucleosome intensity . This was fit to a single exponential decay using Prism 6 ( GraphPad , La Jolla , CA ) ( Equation 1 ) , ( 1 ) y= ( y0−p ) e−kobst+pwhere y0 is the initial fraction end-positioned , kobs is the observed rate constant , and p is the fraction end-positioned at plateau . Reactions in a given concentration series were fit constrained to a common y0 and p . Concentration series were fit to a cooperative binding model ( Equation 2 ) , ( 2 ) kobs=kmaxX ( Kmapp ) h+Xhwhere X is the concentration of SNF2h , h is the hill coefficient , Kmapp is the apparent Km , and kmax is the saturating rate constant . Competition assays were performed as described above with varying concentrations of LANA peptide and fit to a single exponential decay . This was then fit to a simple competition binding model ( Equation 3 ) , ( 3 ) kobs=k01+XK1where k0 is the rate constant without peptide , X is the concentration of peptide , and KI is the inhibition constant . ATPase reactions were performed under multiple turnover conditions ( nucleosomes in excess of enzymes ) . Reactions were performed with 12 . 5 nM SNF2h , 12 . 5 mM HEPES pH 7 . 5 , 2 mM Tris pH 7 . 5 , 70 mM KCl , 7 . 5 µM ATP•MgCl2 , 3 mM MgCl2 , 0 . 02% NP40 , ~3% ( v/v ) glycerol , and trace amounts of γ-32P-ATP . Reactions were started with addition of enzyme , and 2 . 5 µL time points were quenched with an equal volume of 50 mM Tris pH 7 . 5 , 3% SDS , and 100 mM EDTA . Inorganic phosphate was resolved from ATP on a PEI-cellulose TLC plate ( Select Scientific ) with 0 . 5 M LiCl/1M formic acid mobile phase . Plates were dried , exposed to a phosphorscreen overnight , and scanned on a Typhoon variable mode imager . Rate constants were determined by fitting a line through the first 10% of inorganic phosphate generated using Prism . Ensemble FRET remodeling assays were performed under the same conditions as gel remodeling assays . Reactions were initiated by addition of enzyme and then measured in a K2 fluorometer ( ISS ) equipped with a 550 nm short pass excitation filter and a 535 nm long pass emission filter . Reactions were excited at 515 nm and emission was measured at 665 nm . The resulting curves were fit to a two-phase exponential decay ( Equation 4 ) , ( 4 ) y= ( p+ ( y0−p ) ( ffaste−kfastt ) + ( 1−ffast ) e−kslowt ) where ffast is the fraction in the fast phase and kfast and kslow are the apparent rate constants of the fast and slow phase respectively . Crosslinking mass spectrometry samples were prepared by incubating 72 µg of mononucleosomes without flanking DNA at 9 µM final concentration with 24 µM SNF2h in buffer containing either ADP or ADP•BeFx ( 15 mM HEPES pH 7 . 5 , 140 mM KCl , 0 . 5 mM ADP•Mg , 0 . 5 mM MgCl2 , ±0 . 5 mM BeFx 1:5 BeCl2:NaF ) for 10 min at 30°C . The crosslinking reaction with the LANA peptide was performed the same as the ADP•BeFx condition with 30 µM peptide added . The samples were then reacted with 1 mM EDC and 20 µM N-hydroxysulfosuccinimide ( added as a 10x stock in water ) for 60 min at room temperature . Crosslinking reactions were then quenched by adding 10 mM Tris pH 7 . 5 and samples were acetone precipitated and washed once with cold acetone . The pellet was resuspended in 8M Urea , 5 mM TCEP , 100 mM ammonium bicarbonate and heated at 56°C for 25 min , followed by alkylation with 10 mM iodoacetamide for 40 min at room temperature . The sample was diluted 5-fold with 100 mM ammonium bicarbonate and digested with 1:25 trypsin for 4 hr at 37°C followed by addition of a second aliquot of trypsin and overnight digestion . Crosslinked peptides were desalted using 100 µl OMIX C18 tips ( Agilent ) , fractionated by size-exclusion chromatography ( SEC ) , and analyzed by LC-MS similarly to a previously described method ( Zhou et al . , 2017 ) . Briefly , trypsin digests were acidified to 0 . 2% TFA , desalted , and run over a Superdex Peptide PC 3 . 2/300 s column ( GE Healthcare ) . SEC fractions eluting between 0 . 9 ml and 1 . 4 ml were dried and resuspended in 0 . 1% formic acid for LC-MS . Each fraction was separated over a 15 cm x 75 μm ID PepMap C18 column ( Thermo ) using a NanoAcquity UPLC system ( Waters ) and analyzed by a Fusion Lumos mass spectrometer ( Thermo ) . Precursor ions were measured from 375 to 1500 m/z in the Orbitrap analyzer ( resolution: 120 , 000; AGC: 4 . 0e5 ) . Ions charged 3+ to 8+ were isolated in the quadrupole ( selection window: 1 . 6 m/z units; dynamic exclusion window: 30 s; MIPS Peptide filter enabled ) , fragmented by HCD ( Normalized Collision Energy: 28% ) and measured in the Orbitrap ( resolution: 30 , 000; AGC; 5 . 0e4 ) . The cycle time was set to 3 s . Peaklists were generated using PAVA ( UCSF ) and searched for crosslinked peptides with Protein Prospector 5 . 19 . 22 ( Trnka et al . , 2014 ) against a target database containing human SNF2h plus the four core histone sequences from X . laevis concatenated with a decoy database containing 10 randomized copies of each target sequence ( total database size: 55 sequences ) . Loss of the initiator methionine and carbamidomethylation of cysteine . Methionine oxidation , peptide N-terminal glutamine to pyroglutamate formation , acetylation at the protein N-terminus , and mis-annotation of the monoisotopic peak ( 1 Da neutral loss ) were treated as variable modifications . EDC was designated as a heterobifunctional crosslinking reagent with specificity of aspartate , glutamate , and the protein C-terminus on one side and lysine and the protein N-terminus on the other with a bridge mass corresponding to loss of H2O . A mass modification range of 400–5000 Da was specified on these residues and 85 product ion peaks from the peaklist were used in the search . Precursor and product ion tolerances were 8 and 25 ppm respectively . Crosslinked spectral matches ( CSMs ) were initially classified as in ( Zhou et al . , 2017 ) . The dataset was then aggregated into unique crosslinked residue-pair level data with a corresponding spectral count value . Due to the prevalence of multiple , closely spaced Asp and Glu residues in a typical tryptic peptide , site-localization of EDC crosslinks is more challenging than with homobifunctional lysine-directed reagents . To address this , when the site-localization was judged to be ambiguous , all possible residue-pairs were kept with an annotation noting the ambiguity . When calculating spectral counts , fractional spectral counts were assigned to these ambiguous site localizations so that a given spectrum was awarded exactly one spectral count . For instance , a product ion spectrum matching equally well to both K91 . H4-D65 . H2B or K91 . H4-E68 . H2B contributes 0 . 5 towards the spectral counts of each residue-pair . Decoy CSMs were retained throughout this aggregation and spectral counting process . A linear SVM model , built on five features of the Protein Prospector search output ( score difference , % of product ion signals matched , precursor charge , rank of peptide 1 , and rank of peptide 2 ) was constructed to sort crosslinked residue pairs into decoy and target classes . Crosslinked residue-pairs with an SVM score greater than 1 , score difference greater than 5 , and at least one spectral count are reported . The final residue-pair level data set is reported at specificity of 99 . 7% corresponding to 0 . 05% FDR . The number of unique crosslinks in the ADP condition was 974 , while 1470 crosslinks were unique to the ADP•BeFx condition and 707 crosslinks were common to both conditions . To determine which protein domains are involved in SNF2h mediated nucleosome sliding , residue level crosslink spectral counts were aggregated into domain level counts . Each domain pair was assigned a minimum spectral count of 1 to avoid dividing by zero , and the log2 ratio of spectral counts for each domain pair in the ADP•BeFx condition to the ADP condition were calculated . Domain pairs with a Log2 ratio of exactly 0 were treated as NA values . The remaining data were normalized such that the median value was set at 0 . Hence , most domain-domain interactions were assumed to not change substantially between conditions . Histone protein sequences were taken from Xenopus Laevis Uniprot Entries ( without the initiator methionine ) and domains were defined as follows: H2A N-term tail: 1–16 , H2A: 17–43 , H2A Acidic Patch: 44–100 , H2A C-term tail: 101–129; H2B tail: 1–34 , H2B: 35–99 , H2B Acidic Patch: 100–122; H3 tail: 1–44 , H3: 45–135; H4 tail: 25–102 . The sequence of SNF2h was identical to the Human entry in Uniprot ( O60264 ) with an additional two amino acids at the N-terminus to match the construct used . All residue numbers are therefore shifted from the Uniprot entry by two aa . SNF2h domains were defined as follows: Snf2h1: 1–83 , AutoN: 84–160 , Snf2h2: 161–183 , RecA1: 184–402 , RecA2: 403–641 , NegC: 642–703 , Snf2h4: 704–736 , HAND: 737–839 , SANT: 840–894 , SLIDE: 895–1013 , Snf2h5: 1014–1054 . Annotated Mass Spectra are available using MS-Viewer at: http://msviewer . ucsf . edu/prospector/cgi-bin/msform . cgi ? form=msviewer The ADP data set is accessed with search key: 2x0kr2kzq1 The ADP•BeFx data set is accessed with search key: fjamygr8pl The ADP•BeFx in the presence of LANA peptide is accessed with search key: c5o2mcxwum . Raw mass spectrometry data is available in the MassIVE repository at UCSD with accession key: MSV000082136 smFRET experiments were performed as previously described in ( Zhou et al . , 2018 ) with modifications to the reaction buffers as described below .	Every human cell contains nearly two meters of DNA , which is carefully packaged to form a dense structure known as chromatin . The building block of chromatin is the nucleosome , a unit composed of a short section of DNA tightly wound up around a spool-like core of proteins called histones . The tight structure of the nucleosome prevents the cell from accessing and ‘reading’ the genes in the packaged DNA , effectively switching off these genes . So the exact placement of nucleosomes helps manage which genes are turned on . Changing the position of the nucleosomes can ‘free’ the DNA and make genes available to the cell . Enzymes called chromatin remodelers move nucleosomes around – for example , they can make the histone core slide on the DNA strand . However , it is still unclear how these enzymes recognize nucleosomes . Previous research indicates that many proteins bind to nucleosomes by using a surface on the histone proteins called the acidic patch . Could chromatin remodelers also work by interacting with this acidic patch ? To address this further , Gamarra et al . investigate how a chromatin remodeler enzyme known as SNF2h interacts with a nucleosome . By default , SNF2h is inactive because two of its regions called AutoN and NegC act as brakes . The experiments show that the acidic patch helps to bypass this inactivation and switches on SNF2h . Gamarra et al . propose that , when SNF2h docks on to the nucleosome , the patch provides a landing pad for the AutoN and NegC modules; this interaction activates the enzyme , which can then start remodeling the nucleosome . However , another type of chromatin remodeler also uses the patch to interact with nucleosomes but it does not have the AutoN and NegC regions . This suggests that chromatin remodelers work with the acidic patch in different ways . Overall , the findings deepen our understanding of how DNA is packaged in cells , and how this process may go wrong and cause disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "structural", "biology", "and", "molecular", "biophysics" ]	2018	The nucleosomal acidic patch relieves auto-inhibition by the ISWI remodeler SNF2h
tRNAs are unique among various RNAs in that they shuttle between the nucleus and the cytoplasm , and their localization is regulated by nutrient conditions . Although nuclear export of tRNAs has been well documented , the import machinery is poorly understood . Here , we identified Ssa2p , a major cytoplasmic Hsp70 in Saccharomyces cerevisiae , as a tRNA-binding protein whose deletion compromises nuclear accumulation of tRNAs upon nutrient starvation . Ssa2p recognizes several structural features of tRNAs through its nucleotide-binding domain , but prefers loosely-folded tRNAs , suggesting that Ssa2p has a chaperone-like activity for RNAs . Ssa2p also binds Nup116 , one of the yeast nucleoporins . Sis1p and Ydj1p , cytoplasmic co-chaperones for Ssa proteins , were also found to contribute to the tRNA import . These results unveil a novel function of the Ssa2p system as a tRNA carrier for nuclear import by a novel mode of substrate recognition . Such Ssa2p-mediated tRNA import likely contributes to quality control of cytosolic tRNAs . Most cytoplasmic RNAs are exported unidirectionally from the nucleus across the nuclear envelope ( NE ) after their birth and appropriate processing in the nucleus ( Lei and Silver , 2002; Grünwald et al . , 2011 ) . tRNAs , however , are unique among major classes of cytoplasmic RNAs in that they shuttle between the nucleus and the cytoplasm ( Takano et al . , 2005; Shaheen and Hopper , 2005; for review; Yoshihisa , 2006; Phizicky and Hopper , 2010; Hopper , 2013 ) . The first indication of nuclear-cytoplasmic shuttling of tRNAs was the discovery of cytoplasmic splicing of pre-tRNAs in the yeast Saccharomyces cerevisiae ( Yoshihisa et al . , 2003 , 2007 ) in spite of the existence of small amounts of mature tRNAs in the nucleus ( Sarkar and Hopper , 1998; Grosshans et al . , 2000 ) . Subsequently , we and others demonstrated that mature tRNAs move from the cytoplasm back into the nucleus ( Shaheen and Hopper , 2005; Takano et al . , 2005 ) . Nuclear import of tRNAs has also been observed in mammals ( Zaitseva et al . , 2006; Shaheen et al . , 2007; Miyagawa et al . , 2012 ) , suggesting that eukaryotic cells are equipped with mechanisms that allow bidirectional movement of tRNAs between the nucleus and the cytoplasm . The export of tRNAs has been well studied . Importin β-family proteins have been identified as tRNA carriers across the nuclear pore complex ( NPC ) on the NE . Los1p/exportin-t is considered as a primary export carrier of mature tRNAs in yeast , plants , and vertebrates ( Arts et al . , 1998; Hellmuth et al . , 1998; Kutay et al . , 1998; Sarkar and Hopper , 1998; Li and Chen , 2003; Park et al . , 2005; Cook et al . , 2009 ) . Besides , yeast Los1p exports pre-tRNAs for cytoplasmic splicing ( Sarkar and Hopper , 1998; Yoshihisa et al . , 2003 ) . Msn5p/exportin-5 , another importin β homologue , provides an alternative export pathway in yeast and mammals ( Bohnsack et al . , 2002; Calado et al . , 2002; Takano et al . , 2005; Okada et al . , 2009 ) while Msn5p/exportin-5 is the main carrier of tRNAs in Drosophila ( Büssing et al . , 2010 ) . These importin β proteins appear to play different but overlapping roles in the export of mature tRNAs in the yeast; Los1p exports both newly synthesized and re-imported tRNAs while Msn5p exports only re-imported tRNAs ( Eswara et al . , 2009; Murthi et al . , 2010 ) . In both cases , the export of spliced or mature tRNAs is coupled to a tRNA quality control step governed by nuclear aminoacyl-tRNA synthetases ( ARSs ) to support efficient export of aminoacylated tRNAs ( Lund and Dahlberg , 1998; Azad et al . , 2001 ) . Efficient tRNA export relies on both nuclear and cytoplasmic factors , like Utp8p , Utp9p , Cex1p , so on . ( McGuire and Mangroo , 2007; Strub et al . , 2007; Eswara et al . , 2009; Nozawa et al . , 2013 ) . In contrast , mechanisms of tRNA import into the nucleus are only poorly characterized . Nuclear import of tRNAs can be blocked by depletion of intracellular ATP , indicating that the process is energy-dependent . However , tRNA import can also operate even in the absence of a gradient of GTP-bound Ran across the NE , which is the main energy source for many macromolecular transport mediated by importin β ( Takano et al . , 2005; Grünwald et al . , 2011 ) . It is known that various types of tRNAs are imported into the nucleus . Both authentic tRNAs and some types of damaged tRNAs , such as CCA-less tRNAs , are imported ( Takano et al . , 2005 ) . Wybutosine formation on tRNA-PheGAA is initiated by nuclear Trm5p after cytoplasmic splicing of pre-tRNA-PheGAA , indicating that nuclear import of a spliced intermediate precedes this modification ( Ohira and Suzuki , 2011 ) . In addition , spliced tRNA species but hypomodified and possessing 5′- and 3′ extensions , which are accidentally leaked from the nucleus in certain mutants , are retrograded into the nucleus to be repaired or degraded ( Kramer and Hopper , 2013 ) . These findings are in great contrast to the above-mentioned selective export of aminoacylated mature tRNAs . Collectively , tRNA import and export mechanisms with different specificities were predicted to maintain the quality of pre-existing mature tRNAs in the cytoplasm ( Yoshihisa , 2006; Kramer and Hopper , 2013 ) . However , identity of the carrier that mediates tRNA import remains totally unknown . One possible import carrier is an importin-β homologue Mtr10p , which was originally identified as an import carrier for Npl3p , as mRNA-binding protein ( Senger et al . , 1998 ) , and was shown to cause a defect in tRNA import when mutated ( Shaheen and Hopper , 2005 ) . Nevertheless , no evidence has been reported for tRNA binding by Mtr10p . Notably , balance between the import and export of tRNAs across the NE is determined by the physiological conditions . For instance , depletion of various nutrients , such as amino acids , phosphate , and glucose , results in nuclear accumulation of tRNAs in S . cerevisiae ( Hurto et al . , 2007; Whitney et al . , 2007 ) . Indeed , such alteration of tRNA localization affects translation efficiency of certain mRNAs that encode enzymes for amino acid biosynthesis , suggesting that retrieval of tRNAs from the cytosol has , at least , a regulatory role ( Chu and Hopper , 2013 ) . To alter the tRNA balance , signal transduction pathways mediated by PKA and TOR , but not Gcn2p , facilitate this transport regulation in the yeast ( Whitney et al . , 2007 ) . Although several kinases were shown to act on the regulation of tRNA transport , tRNA export carriers Los1p or Msn5p are not phosphorylated in vivo according to the nutrient status , and they are not substrates for PKA in vitro ( Pierce et al . , 2014 ) . Similar regulation could operate not only in fungi , but also in mammals ( Shaheen et al . , 2007; Miyagawa et al . , 2012 ) , while some reports argue that this is specific to some fungal species ( Chafe et al . , 2011 ) . A model was proposed that , while tRNA import is constitutive , export is fine-tuned , depending on the growth conditions ( Eswara et al . , 2009; Murthi et al . , 2010 ) . However , evidence is lacking for constitutive import of tRNAs at the same rate irrespective of nutrient conditions . It is thus essential to identify the factors mediating the tRNA import , which are possible targets of the regulation . Here , we describe the identification and characterization of Ssa2p , one of the major cytosolic Hsp70 in budding yeast ( Werner-Washburne et al . , 1987 ) , as a potential tRNA import carrier . While Hsp70 proteins are a class of molecular chaperones that bind and release proteins with exposed hydrophobic segments in an ATPase cycle-dependent manner to affect protein conformation ( Young et al . , 2004; Kampinga and Craig , 2010 ) , in vivo and in vitro data indicate that Ssa2p plays a pivotal role in nuclear import of tRNAs , and that this process is achieved through a novel mode of substrate recognition through its nuclear binding domain ( NBD ) . To search for carrier ( s ) of tRNA import into the nucleus , a biochemical approach was adopted to identify new tRNA-binding proteins . We previously showed that both full-length and CCA-less tRNAs are imported into the nucleus , and that this nuclear import is ATP-dependent ( Takano et al . , 2005 ) . Thus , we postulated that the putative import carrier ( s ) does not recognize the 3′ end of tRNAs and binds tRNAs in a nucleotide triphosphate-dependent or sensitive manner . To identify such proteins , tRNA-agarose was prepared , in which yeast tRNAs were immobilized via their 3′ ends ( Figure 1A ) . Next , a yeast cytosolic fraction depleted of endogenous tRNAs by anion exchange chromatography was applied to the tRNA-agarose in the presence or absence of 3 mM ATP . Bound proteins were subsequently eluted from the tRNA-agarose with 1 . 5 M NaCl . As shown in Figure 1B , the intensity of some bands varied depending on the absence ( Figure 1B closed arrowheads ) or presence of ATP ( open arrowheads ) . Bands in the eluates marked by arrows were then subjected to peptide fingerprinting . As summarized in Figure 1C , Tef1p ( eEF1A in the yeast ) and Eno2p ( enolase ) were identified , both of which are already known as tRNA binding proteins ( Nagata et al . , 1984; Entelis et al . , 2006 ) . In addition , RNA binding proteins , such as Pab1p , Gbp2p and Sro9p , were also detected . Among the proteins associated with the tRNA-resin in an ATP-sensitive manner , Ssa1p and/or Ssa2p , two major cytosolic Hsp70s with highly homologous sequences , were of interest because mammalian Hsp70 ( Hsc70 ) was shown to interact with an AU-rich element , which defines short-lived mRNAs ( Henics et al . , 1999; Laroia et al . , 1999; Lu et al . , 2006 ) . In addition , the yeast ssa1 mutant strain exhibits defective degradation of mRNAs possessing this AU-rich element ( Duttagupta et al . , 2003 ) . Therefore , the Ssa proteins were selected for further analyses . 10 . 7554/eLife . 04659 . 003Figure 1 . Purification of tRNA-interacting proteins with immobilized tRNA-resin . ( A ) A schematic diagram for preparation of tRNA-agarose by hydrazide coupling . ( B ) tRNA-binding proteins purified with the tRNA-agarose were analyzed by SDS-PAGE/CBB staining . Purification was performed in the absence ( − ) or presence ( + ) of 3 mM Mg-ATP . Bands appearing mainly in the ATP minus or plus lanes are marked by closed and open arrowheads , respectively . Bands that were identified by peptide mass fingerprinting after in-gel digestion are indicated by small numbered arrows . ( C ) A summary of the proteins identified in B . The numbers correspond to the arrows shown in B . *Sro9p , with a calculated molecular mass of 48 . 1 kDa , is known to migrate as a 60-kDa band on SDS-PAGE ( Sobel and Wolin , 1999 ) . ( D ) Yeast lysates prepared from strains expressing either Ssa1p-FLAG or Ssa2p-FLAG in addition to the wild type-strain were subjected to immunoprecipitation with anti-FLAG agarose in the absence ( − ) or presence ( + ) of 3 mM Mg-ATP . One-tenth of the eluates were analyzed by Western blotting with the anti-FLAG antibody ( WB ) , and the remainders of the eluates were subjected to RNA extraction and Northern blotting with a probe against mature tRNA-ProUGG ( NB ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04659 . 00310 . 7554/eLife . 04659 . 004Figure 1—figure supplement 1 . Identification of Ssa1p/Ssa2p as tRNA-binding protein by peptide mass fingerprinting . ( A ) The band marked as ‘1’ in Figure 1B was excised from the gel and in-gel-digested with trypsin . The resulting peptides were recovered by extraction with 30% vol/vol acetonitrile , 0 . 1% vol/vol TFA , and subjected to MALDI/TOF MS ( Voyager DE , Applied Biosystems , Foster City , CA ) . Peaks with red numbers indicate those derived from Ssa1p/Ssa2p . Protein identification was carried out with MS-Fit software ( http://jpsl . ludwig . edu . au/ucsfhtml3 . 4/msfit . html ) . ( B ) MS data were compared with the theoretical molecular masses of tryptic fragments from Ssa1p and Ssa2p . ( C ) A schematic drawing of the domain structures of Ssa1p and Ssa2p . The amino-acid positions of domain boundaries are indicated . The numbers between the bars indicate the identity of the two amino acid sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 04659 . 004 Because the peptide mass fingerprint of the tryptic fragments from the 70 kDa band could not discriminate between Ssa1p and Ssa2p ( Figure 1—figure supplement 1 ) , the SSA1 or SSA2 gene on the yeast chromosome was replaced with a FLAG-tagged version , and the capability of each FLAG-tagged Ssa protein to bind tRNAs in vivo was examined by RNA immunoprecipitation . When analyzed by Western blotting with the anti-FLAG antibody and by Northern blotting with an anti-mature tRNA-ProUGG probe ( Figure 1D ) , only immunoprecipitates of Ssa2p-FLAG in the absence of ATP contained tRNA-ProUGG at levels above background , indicating that the Ssa protein we identified as a tRNA-binding protein was Ssa2p . Next , the question whether Ssa2p is indeed involved in the nuclear import of tRNAs in vivo was addressed . tRNA import was assessed by measuring nuclear accumulation of tRNAs in a los∆1 msn5∆ double mutant ( Takano et al . , 2005 ) and by examining nuclear accumulation of tRNAs under nutrient starvation conditions ( Shaheen and Hopper , 2005 ) . First , an ssa1∆ or ssa2∆ mutation was introduced into the los1∆ msn5∆ strain , and the localization of tRNA-ProUGG , encoded by intron-containing genes and initiator tRNA-Met ( tRNA-iMet ) encoded by intronless genes was analyzed by FISH . Under normal growth conditions , the los1∆ msn5∆ cells accumulate large amounts of mature tRNAs in the nucleus . This tRNA gradient across the NE was abolished by treating the cells with 2-deoxyglucose and NaN3 and was re-established by removing these drugs in the presence of thiolutin , an inhibitor of all the three RNA polymerases ( Takano et al . , 2005; Figure 2—figure supplement 1 ) . If Ssa2p is involved in the nuclear import of tRNAs , the los1∆ msn5∆ ssa2∆ mutant would accumulate less mature tRNAs in the nucleus and fail to re-establish the tRNA gradient across the NE during energy recovery . As shown in Figure 2—figure supplement 1 , neither the ssa1∆ nor ssa2∆ mutants exhibits altered tRNA localization in a los1∆ msn5∆ background . Furthermore , ssa2∆ mutant cells re-established the tRNA gradient across the NE after energy depletion like the other strains ( Figure 2—figure supplement 1 , bottom row ) . These results indicate that neither Ssa2p nor Ssa1p plays a major role in tRNA import under normal growth conditions . Then , an effect of the ssa2 mutation on tRNA import was assessed under the conditions of nutrient starvation . Wild-type , ssa1∆ , and ssa2∆ cells cultured in the YPD medium were transferred to the amino acid-starvation medium SD+Ura , Ade ( SD ) , and the localization of tRNA-ProUGG and tRNA-iMet was analyzed by FISH . While the wild-type and ssa1∆ cells accumulated both tRNA-ProUGG and tRNA-iMet in the nucleus under these conditions within 2 hr , relatively lower amounts of tRNA-ProUGG and tRNA-iMet were observed in the ssa2∆ nuclei ( Figure 2A ) . We also examined three other tRNA species , tRNA-LysCUU , tRNA-LysUUU , and tRNA-TyrGUA , in FISH , and found that all the tRNAs were apparently affected by ssa2∆ mutation while the effects of ssa1∆ mutation were not obvious if any ( Figure 2—figure supplement 2 ) . The extent of nuclear accumulation under starvation conditions and the effect of ssa2∆ were variable among tRNA species ( Figure 2 and its supplement ) . To confirm difference of the two ssa mutations in tRNA accumulation upon starvation more in detail , quantitative analyses of the FISH images were carried out ( Figure 2A , table ) . While the average nuclear accumulation indices ( NAIs ) , rations of nuclear FISH signals against cytosolic signals , of tRNA-ProUGG in the wild-type , ssa1∆ , and ssa2∆ cells were nearly equal when the cells were grown in YPD , the NAIs of these strains became 1 . 52 ± 0 . 09 , 1 . 58 ± 0 . 10 , and 1 . 14 ± 0 . 25 , respectively , when the cells were starved , revealing an apparent difference between tRNA import in the wild-type cells and that in the ssa2∆ cells . We noticed that , under the starvation conditions , variation of NAIs of individual cells increased if compared with that in the rich medium ( Figure 2—figure supplement 3 ) while average NAIs of biological replicates fell into a narrower range . Deletion of either of the other two SSA genes , SSA3 and SSA4 , which are only expressed under stress conditions , had no impact on tRNA accumulation under starvation conditions ( Figure 2A , table ) . When the effect of simultaneous deletion of SSA1 and SSA2 genes was examined , the double deletion did not result in the obvious additive effect ( Figure 3 ) . Similar results were obtained when distribution of tRNA-iMet was quantified ( data not shown ) . 10 . 7554/eLife . 04659 . 005Figure 2 . Nuclear accumulation of tRNAs under starvation conditions is affected by SSA2 gene deletion . ( A ) Wild-type ( WT: W303-1A ) , ssa1∆ ( TYSC918 ) , and ssa2∆ ( TYSC920 ) cells were cultured in YPD until the log phase . The cells were then transferred to SD+Ura , Ade lacking all amino acids and cultured for additional 2 hr . The cells before ( YPD ) and after amino acid-starvation ( SD ) were subjected to FISH with a rhodamine-labeled probe against mature tRNA-ProUGG ( ProUGG ) and an FITC-labeled probe against tRNA-iMet ( iMet ) . The nucleus was visualized with DAPI . Fluorescence signals of FISH images of tRNA-ProUGG were quantified , and the ratio between the nuclear and cytosolic signals is expressed as the NAI ( see ‘Materials and methods’ ) . Three independent samples were analyzed , and their average NAIs ( NAI ) with standard deviations ( SDV ) are shown . Bar , 5 µm . ( B ) The wild-type , GAL7p-MTR10 ( mtr10↓: TYSC612 ) , ssa1∆ ( TYSC918 ) , ssa1∆ GAL7p-MTR10 double mutant ( ssa1∆ mtr10↓: ssa1∆ mtr10 ) , ssa2∆ ( TYSC920 ) , and ssa2∆ GAL7p-MTR10 double mutant ( ssa2∆ mtr10↓: ssa2∆ mtr10 ) strains pre-grown in YPGal were cultured in YPD for 18 hr , transferred to SD+Ura , Ade , and subjected to FISH with the anti-tRNA-ProUGG probe , as described in A . Average NAIs with SDVs from three independent experiments are shown in the table . ( C ) Localization of Ssa1p and Ssa2p was analyzed by immunofluorescence microscopy . Cells expressing SSA1-FLAG or SSA2-FLAG were grown in YPD ( YPD ) and were subsequently cultured in SD+Ura , Ade ( SD ) for 2 hr . The signal intensities of Ssa2p-FLAG on the lines shown in the pictures are shown in the far right graph . A solid line indicates a cell in SD , and a dashed gray line does a cell in YPD . Original microscopic images and individual data for quantitative FISH in this figure will be found in Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04659 . 00510 . 7554/eLife . 04659 . 006Figure 2—source data 1 . Zip file containing source data for Figure 2 . Yeast cells are processed as described in the Figure 2 legend and the ‘Materials and methods’ section . Images from three independent sets of FISH experiments are subjected to quantification . Each folder named as Fig2A_expX or Fig2B_expX contains gray-scale tif images with 16 bit depth ( acquired by MetaMorph ) of a set of the experiments . A file name consists of the strain name ( ‘wt ’or ‘ssa1’ for example ) and culture conditions ( ‘YPD’ or ‘SD’ ) with the last capital letter representing the recording channel ( ‘D’ for DAPI staining or ‘R’ for RNA FISH ) . If the number of cells suitable for quantification in one image was under 30 , those from two images were quantified . In such cases , two sets of images ( ‘ssa2_SD_a_R . tif’ and ‘ssa2_SD_b_R . tif’ for example ) are included . Raw quantification data and their processing to NAIs are summarized Excel files . Summary of the total experiments are shown in the ‘SUMMARY’ sheet in the file named ‘Figure 2A_data_summaryandexp1_DATA . xls’ or ‘Figure 2B_data_summaryandexp1_DATA . xls . ’ The same set of data for experiments with the wild-type strain are shown graphically in Figure 2—figure supplement 3 in the supplemental materials . All the tif images have 16-bit depth . DOI: http://dx . doi . org/10 . 7554/eLife . 04659 . 00610 . 7554/eLife . 04659 . 007Figure 2—figure supplement 1 . Nuclear import of tRNA under the los1∆ msn5∆ background was not affected by the deletion of SSA genes . Nuclear import of tRNA-ProUGG in ssa1∆ or ssa2∆ strains under non-starvation conditions was monitored by the nuclear import assay upon recovery from ATP depletion ( Takano et al . , 2005 ) . The strains shown at the top of the panels ( los1∆ msn5∆ , TYSC512; los1∆ msn5∆ ssa1∆ , TYSC1059; los1∆ msn5∆ ssa2∆ , TYSC1060 ) were grown in a rich medium ( YPD ) , exposed to 10 mM NaN3 and 10 mM 2-deoxyglucose for 1 hr ( YPAdG ) to deplete intracellular ATP , and then incubated in YPD with 5 µg/ml thiolutin , a transcription inhibitor , for 1 hr ( +thiolutin ) . At each time point , the cells were harvested and subjected to FISH with a probe against mature tRNA-ProUGG ( right columns ) . The nucleus was visualized by DAPI staining ( left columns ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04659 . 00710 . 7554/eLife . 04659 . 008Figure 2—figure supplement 2 . Defects in nuclear accumulation of several other tRNA species were observed in ssa2∆ cells . The wild-type ( W303-1A ) , ssa1∆ ( TYSC918 ) , and ssa2∆ ( TYSC920 ) strains treated as in Figure 2A were subjected to FISH analyses with anti-tRNA-LysCUU ( LysCUU ) , anti-mature tRNA-LysUUU ( LysUUU ) , or anti-mature tRNA-TyrGUA ( TyrGUA ) probes . Because tRNA-LysUUU and tRNA-TyrGUA are encoded by intron-containing genes while tRNA-LysCUU is by intronless genes , the probes for the former two tRNA species were designed to detect only spliced tRNAs . Bars , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 04659 . 00810 . 7554/eLife . 04659 . 009Figure 2—figure supplement 3 . Quantitative analyses of nuclear accumulation of tRNAs . A typical example of quantitative FISH data processing to acquire an average NAI and its standard deviation ( SDV ) is shown . In this case , wild-type cells ( W303-1A ) were cultured in the rich medium ( YPD ) and then were shifted to the nutrient starvation medium ( SD ) . Three independent sets of cell samples before and after incubation in the SD medium were subjected to FISH with the anti-mature tRNA-ProUGG probe . NAIs of 30 individual cells in each experiment are measured and shown as open circles ( YPD ) and closed squares ( SD ) . An average of NAIs of individual cells with an SDV in each experiment is shown in the bottom . The ‘average NAI’ appearing in the text is calculated from the NAIs of three independent experiments and shown in the right . DOI: http://dx . doi . org/10 . 7554/eLife . 04659 . 00910 . 7554/eLife . 04659 . 010Figure 3 . The ssa1∆ ssa2∆ double deletion does not cause a synergistic effect on the nuclear accumulation of tRNAs under starvation conditions . Wild-type ( W303-1A ) , ssa1∆ ( TYSC918 ) , ssa2∆ ( TYSC920 ) , and ssa1∆ ssa2∆ double mutant ( TYSC1013 ) strains were cultured in YPD ( YPD ) and transferred to SD+Ade , Ura ( SD ) for 2 hr . The cells were subsequently subjected to FISH analysis with anti-tRNA-ProUGG and tRNA-iMet probes . Bar , 5 µm . The fluorescence signals of tRNA-ProUGG images of three independent experiments were quantified , and the average NAIs with SDVs were calculated . Original microscopic images and individual data for quantitative FISH in this figure will be found in Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04659 . 01010 . 7554/eLife . 04659 . 011Figure 3—source data 1 . Zip file containing source data for Figure 3 . Yeast cells are processed as described in the Figure 3 legend and the ‘Materials and methods’ section . Images from three independent sets of FISH experiments are subjected to quantification . Each folder named as Fig3_expX contains gray-scale tif images with 16 bit depth ( acquired by MetaMorph ) of a set of the experiments . A file name consists of the strain name ( ‘wt , ’ ‘ssa1 , ’ for example ) and culture conditions ( ‘YPD’ or ‘SD’ ) with the last capital letter representing the recording channel ( ‘D’ for DAPI staining or ‘R’ for RNA FISH ) . If the number of cells suitable for quantification in one image was under 30 , those from two images were quantified . In such cases , two sets of images ( ‘ssa2_SD_a_R . tif’ and ‘ssa2_SD_b_R . tif’ for example ) are included . Raw quantification data and their processing to NAIs are summarized Excel files . Summary of the total experiments are shown in the ‘SUMMARY’ sheet in the file named ‘Figure 3_data_summaryandexp1_DATA . xls . ’ All the tif images have 16-bit depth . DOI: http://dx . doi . org/10 . 7554/eLife . 04659 . 011 An importin β , Mtr10p , was shown to participate in tRNA redistribution under starvation conditions ( Shaheen and Hopper , 2005 ) . Thus , we examined whether Ssa2p acts in the same pathway as Mtr10p , using an ssa2∆ GAL7p-MTR10 double mutant strain . As some controversy exists regarding the involvement of Mtr10p in the tRNA import ( Shaheen and Hopper , 2005; Chafe et al . , 2011 ) , the promoter shut-off strain was used to circumvent any possible adaptation effects due to the MTR10 deletion , which causes a strong growth defect . These strains were grown in YPGal medium and then grown in YPD for 18 hr to shut-off the expression of MTR10 before being subjected to nutrient starvation . As reported by Shaheen and Hopper ( 2005 ) , nuclear accumulation defects were evident in the MTR10 shut-off strain ( Figure 2B , mtr10↓ ) . The MTR10 shut-off had a stronger effect on tRNA distribution than the ssa2Δ single mutation . The double mutant showed an additive decrease in the signal intensity of nuclear tRNA-ProUGG ( Figure 2B , ssa2Δ mtr10↓ ) , and signal quantification revealed that the average NAI of the starved double mutant ( 0 . 87 ± 0 . 04 ) was clearly below those observed for the mtr10 and ssa2Δ single mutants ( 0 . 98 ± 0 . 04 and 1 . 11 ± 0 . 07 , respectively ) , and was far below that observed for the wild type strain ( 1 . 30 ± 0 . 10 ) . Student's t-test indicated that the probability that the NAIs of the mtr10 single and mtr10 ssa2∆ double mutants were the same was only 0 . 037 while the difference between those of the mtr10 single and mtr10 ssa1∆ double mutants was not significant ( p = 0 . 92 ) . Similar quantitative results were obtained for tRNA-iMet ( data not shown ) . These results support the idea that Ssa2p and Mtr10p act independently in parallel pathways for tRNA import under starvation conditions . If the nuclear import of tRNAs by Ssa2p is up-regulated under starvation conditions , localization of Ssa2p may be affected by starvation . Thus , localization of FLAG-tagged Ssa proteins was monitored by immunofluorescence . While clear exclusion of Ssa1p and Ssa2p from the nucleus was observed both in rich and poor media , a slight but distinct increase in the nuclear signals of both Ssa proteins was observed when the cells were incubated in SD+Ura , Ade ( Figure 2C ) . In summary , these results indicate that Ssa2p plays a pivotal role in the nuclear import of tRNAs under starvation conditions , while the other Ssa proteins , including Ssa1p , have a minor , if not at all , role in this process . Furthermore , Ssa2p may provide a novel nuclear transport pathway independent of Mtr10p . To investigate biochemical characteristics of the interaction between Ssa proteins and tRNAs , we examined the ability of recombinant Ssa proteins to bind tRNAs directly in vitro . Interactions between mammalian Hsc70 and short-lived RNAs with the AU-rich element were previously demonstrated by the label transfer assay ( Henics et al . , 1999 ) . A similar assay was first employed to test whether yeast Ssa proteins bind the AU-rich element in vitro . Both Ssa1p and Ssa2p received radioactivity from 32P-labeled ( AUUU ) 5 RNA but not from ( ACCC ) 5 RNA ( Figure 4—figure supplement 1 ) , and label transfer was dependent on UV-irradiation and sensitive to ATP . Besides , no label transfer was observed when BSA was used as a control protein . These results demonstrate that the yeast Hsp70s have the ability to bind certain RNA molecules . Next , tRNAs were used as label transfer substrates . Both Ssa1p and Ssa2p received radioactivity from 32P-labeled tRNA-ProUGG in an ATP-sensitive manner ( Figure 4A , B ) with only a marginal difference between the capacity of the two Ssa proteins to recognize the tRNA in vitro . This is in marked contrast to the effects of ssa1∆ and ssa2∆ mutations on tRNA import in vivo . Because a tRNA possesses an un-paired adenosine on its 3′ terminus , this adenosine might be recognized by Ssa proteins as an analogue of ADP or ATP , which is usually bound by the nucleotide-binding cleft of Hsp70s . To exclude this possibility , label transfer was examined using CCA-less tRNA-ProUGG , which starts with the 5′ guanosine and ends with the 3′ cytidine . As shown in Figure 4A , both Ssa1p and Ssa2p received radioactivity efficiently from the CCA-less tRNA . Label transfer with an intron-containing pre-tRNA was also performed and revealed that Ssa proteins bind tRNAs irrespective of their anticodon loop structure . When the specificity of tRNA recognition by Ssa proteins was tested by competition experiments with short RNAs , chemical amounts of in vitro-transcribed tRNA-ProUGG efficiently competed with the radioactive amount of tRNA-ProUGG for label transfer ( Figure 4C , lanes ‘tRNA’ ) while only limited competition was observed with single-stranded homo-oligo-ribonucleotides ( A30 , U30 , and G30 ) or a double-stranded homo-oligomer ( A-U ) 30 . These results indicate that Ssa proteins specifically recognize tRNAs in an ATP-sensitive manner via a mechanism that is not mediated through the 3′ end adenosine or the anti-codon loop of tRNAs , and that these properties resemble those observed in vivo . 10 . 7554/eLife . 04659 . 012Figure 4 . Ssa proteins directly and specifically interact with tRNAs in vitro . ( A ) In vitro-transcribed tRNA-ProUGG molecules shown in the lower panel ( full-length [full] , CCA-less [-CCA] , and precursor [pre] ) were subjected to label transfer assays with recombinant Ssa1p or Ssa2p in the absence ( − ) or presence ( + ) of 2 . 5 mM ATP . 32P-labeled Ssa proteins are indicated by an arrowhead . Bands marked with an asterisk are Ssa proteins cross-linked with partially digested tRNA species . In the lower part , nucleotides corresponding to the anticodon , the intron and the adenosine of the CCA end are shown as black circles , white circles , and a gray square , respectively . ( B ) Label transfer assays with 32P-labeled mature tRNA-ProUGG were performed in the presence of various concentrations of ATP . The label transfer efficiency without ATP was set to 100% . Data are presented as the averages of three independent experiments . Error bars represent SDVs . ( C ) The specificity of tRNA recognition by Ssa2p was examined by competition experiments with short RNAs . The indicated concentrations of in vitro-transcribed tRNA-ProUGG or homo-30mers shown on the top were added to the label transfer assays . Averages of relative band intensities in triplicate experiments are shown at the bottom with SDVs . ( D ) Label transfer assays were performed with equal radioactive amounts of three different tRNAs ( tRNA-PheGAA , tRNA-ProUGG , and tRNA-TrpCCA ) labeled with either 32P-UTP ( upper ) or 32P-CTP ( lower ) . ( E ) Label transfer from 32P-UTP-labeled tRNA-ProUGG to Ssa2p was monitored in the presence of increasing amounts of competitor tRNAs transcribed in vitro or purified from yeast . Label transfer efficiency without any competitor ( white bar ) was set to 100% . Original gel images and individual quantification data for the label transfer assays in this figure will be found in Figure 4—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04659 . 01210 . 7554/eLife . 04659 . 013Figure 4—source data 1 . Zip file containing source data for Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 04659 . 01310 . 7554/eLife . 04659 . 014Figure 4—figure supplement 1 . Label transfer from the AU-rich RNA to Ssa proteins . ( A ) Recombinant Ssa proteins were purified from E . coli . The indicated amounts of Ssa proteins were subjected to SDS-PAGE and stained with CBB . ( B ) Label transfer assays of Ssa proteins with RNA molecules with or without AU-rich elements . 32P-labeled ( AUUU ) 5- and ( ACCC ) 5-containing RNAs were transcribed in vitro with 32P-UTP . Each of these RNAs ( 1 . 0 × 106 cpm ) was incubated with 1 . 0 µg of either Ssa1p , Ssa2p or BSA in the absence of ATP . After UV-irradiation at 90 mJ/cm2 ( UV ‘+’ lanes ) or being held at room temperature without irradiation for the same duration ( UV ‘−’ lanes ) , the mixtures were digested with RNase Cocktail , and subjected to SDS-PAGE followed by radioimaging . The sequences of the RNA species used in the assay are shown at the bottom . DOI: http://dx . doi . org/10 . 7554/eLife . 04659 . 014 When different isoacceptor tRNAs were used as donors for radioactivity , the label transfer efficiencies were variable . In particular , tRNA-ProUGG was a potent substrate for label transfer while tRNA-PheGAA was a very poor substrate irrespective of 32P-labeled nucleotides used for in vitro transcription ( Figure 4D ) . These results suggest that Ssa proteins bind only a subset of tRNAs , or that appropriate positioning of radiolabeled nucleotides on a tRNA molecule is essential for efficient label transfer . To distinguish between these two possibilities , competition experiments were performed with radiolabeled tRNA-ProUGG and various non-labeled tRNAs . These experiments allowed us to examine the difference in affinity between unmodified and fully-modified tRNAs , as well . Chemical amounts of tRNAs were transcribed in vitro , and several isoacceptor tRNAs were purified from yeast by chaplet column chromatography ( Suzuki and Suzuki , 2007 ) . Competition experiments revealed that all tRNAs , including tRNA-PheGAA , competed with radiolabeled tRNA-ProUGG for label transfer even though their competition efficiencies were variable ( Figure 4E ) . It should be noted that unmodified tRNAs-ProUGG was a more potent competitor than the fully-modified tRNA . These results indicate that various tRNAs are recognized by Ssa proteins with different affinities and that certain nucleotide positions of a tRNA molecule may offer the sites for this recognition . The above results also show that Ssa proteins recognize differences in structural characteristics between unmodified and fully-modified tRNAs . We next examined if , in addition to the primary sequence , the higher-order structure of tRNAs is required for efficient label transfer to Ssa proteins . For this purpose , we first used hybridization of the tRNA substrate with an antisense oligonucleotide to disrupt the tRNA secondary structure . Indeed , an antisense oligonucleotide against the 5′ half of tRNA-ProUGG , but not an unrelated oligonucleotide , inhibited the label transfer only when hybridized with the substrate by denaturation and annealing ( Figure 5A ) . Heat denaturation and quick cooling of the tRNA substrate alone did not affect label transfer efficiency , probably due to rapid refolding of the tRNA molecule . The above possibility was further tested with tRNA-ProUGG mutants that exhibit altered tRNA structures ( Figure 5B ) . Introduction of G18A and U54C mutations , which disrupts the interaction between the D and TΨC loops , reduced label transfer efficiency . On the other hand , destabilization of the acceptor stem by introducing G69C or G68C substitutions resulted in higher label transfer efficiency , while a C67G mutation reduced the efficiency of label transfer . Both of these positive and negative effects induced by acceptor stem mutations were minimized by compensatory mutations ( Figure 5B , gray bars ) . Similar partial destabilization of the acceptor stem of other tRNA species is recognized as the degradation flag by CCA transferase to introduce an unusual CCACCA sequence instead of the normal CCA to the 3′-end of the tRNAs ( Wilusz et al . , 2011 ) . We tested whether Ssa proteins also prefer such mutant tRNAs . Indeed , replacement of the acceptor stem of human tRNA-LeuAAG with acceptor-like stems from unstable tRNA-like noncoding RNAs ( mouse MALAT1-associated cytoplasmic small RNA and MEM β RNA ) enhanced recognition by Ssa2p ( Figure 5C ) . This is also true even for single replacement of the G1-C71 Watson-Crick pair with the non-Watson-Crick pair ( G1•U71 ) in tRNA-ArgUGC . These results suggest that Ssa proteins recognize the overall structure of tRNAs , and prefer tRNAs with a destabilized acceptor stem but with a stable core structure . 10 . 7554/eLife . 04659 . 015Figure 5 . Ssa proteins recognize higher order structures of tRNAs and prefer a destabilized acceptor stem . ( A ) 32P-labeled tRNA-ProUGG was mixed without ( no , white bar ) or , with its anti-sense ( ProUGG , black bar ) or unrelated control ( anti-sense of TrpCCA , gray bar ) oligo-DNA . tRNA/oligo-DNA mixtures were heat-denatured and annealed to form DNA-RNA hybrids ( ‘heat and annealing’ ) as indicated on the left . A sample without oligo-DNA was also heated and quick-cooled ( the far right bar ) . The resulting samples were then subjected to the label transfer assay with Ssa2p . ( B ) Label transfer from tRNA-ProUGG mutants was assayed with Ssa2p . Mutation sites of tRNA-ProUGG used in the assay are indicated schematically in the left . Boldface characters indicate mutations introduced to cause destabilization , and italicized characters do mutations that compensate for mismatches caused by the former mutations . The label transfer efficiencies of mutant tRNAs to Ssa2p were summarized in the right graph . ( C ) Wild-type ( WT ) and mutant forms ( mut1 and mut2 ) of human tRNA-LeuAAG and tRNA-ArgUCG were subjected to the label transfer assay with Ssa2p . Sequences of the acceptor stems of the tRNAs are shown in the left . The tRNA-LeuAAG derivatives received replacements of the acceptor stem of tRNA-LeuAAG with those of tRNA-like ncRNAs such as MALAT1-associated small cytoplasmic RNA ( mascRNA , mut1 ) and MEM β RNA ( mut2 ) . For panels B and C , the label transfer efficiency of the wild-type tRNA was set to 100% . Similar results were obtained with Ssa1p ( not shown ) . Original gel images and individual quantification data for the label transfer assays in this figure will be found in Figure 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04659 . 01510 . 7554/eLife . 04659 . 016Figure 5—source data 1 . Zip file containing source data for Figure 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 04659 . 016 Hsp70 is composed of an N-terminal nucleotide-binding domain ( NBD ) , a substrate-binding domain ( SBD ) , and a C-terminal variable domain ( CVD ) ( Figure 1—figure supplement 1 ) . Hsp70 binds proteinaceous substrates through its SBD when its NBD binds ADP , and releases them when the NBD binds ATP ( Young et al . , 2004; Kampinga and Craig , 2010 ) . Thus , we examined whether Ssa proteins utilize their SBD for recognition of tRNAs . In the label transfer assay , increasing amounts of reduced and carboxymethylated lactalbumin ( RCMLA ) , a model unfolded protein often used in chaperone assays , were added up to 10-fold molar excess of Ssa proteins . As shown in Figure 6A , RCMLA addition did not inhibit , but rather moderately enhanced label transfer from tRNA-ProUGG . These results suggest that tRNAs are not recognized by Ssa proteins through their SBDs . To examine this notion further , purified glutathione-S transferase ( GST ) -fusions encompassing different Ssa domains were subjected to the label transfer assay . GST-NBD and GST-NBD-SBD fusions of Ssa1p and Ssa2p received radioactivity from tRNA-ProUGG while fusions without NBD did not , indicating that the NBD plays an essential role in tRNA recognition by Ssa proteins ( Figure 6B ) . It should be noted that the label transfer efficiencies of the GST-NBD and GST-NBD-SBD proteins were apparently lower than that of full-length Ssa proteins ( Figure 6B , the left most 6 lanes ) . Therefore , the CVD appears to contribute to the efficient recognition of tRNAs by the NBD . 10 . 7554/eLife . 04659 . 017Figure 6 . The NBD of Ssa proteins is essential for tRNA recognition . ( A ) Label transfer assays with 32P-labeled tRNA-ProUGG were performed in the presence of various concentrations of RCMLA . The amount of RCMLA is shown as the molar ratio against Ssa proteins . ( B ) Label transfer assays with full-length Ssa-His6 fusions or GST-fusions with partial Ssa proteins . Left , radioimaging; right , CBB staining . Arrowheads indicate the bands of GST-fusions that received radioactivity . ( C ) Label transfer assays were carried out with wild-type or mutant forms of GST-Ssa1p-NBD ( Ssa1p-NBD ) or GST-Ssa2p-NBD proteins ( Ssa2p-NBD ) . Quantitated data and raw gel images of a typical experiment are shown in the upper graph and the lower panels , respectively . The relative label transfer efficiency represents a ratio of label transfer of a mutant GST-Ssa-NBD to that of the corresponding wild type . The efficiency of the wild-type protein is set to 100% . All the experiments are done in triplicates , and error bars represent SDVs . Original gel images and individual quantification data for the label transfer assays in this figure will be found in Figure 6—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04659 . 01710 . 7554/eLife . 04659 . 018Figure 6—source data 1 . Zip file containing source data for Figure 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 04659 . 01810 . 7554/eLife . 04659 . 019Figure 6—figure supplement 1 . Mutations introduced into the NBD of Ssa proteins . ( A ) Sequence comparison among Ssa1p , Ssa2p , bovine Hsc70 ( BtHsc70 ) and E . coli Hsp70 ( DnaK ) . Asterisks , identical residues; colons , strongly conserved residues; periods , weakly conserved residues . Residues altered in mutants examined in Figure 6C by the label transfer assay are colored , and their positions in Ssa proteins are shown in boldface . ( B ) Mutation points are indicated in the structural model of the NBD of bovine Hsc70 ( 1–394 aa ) ( PDB ID , 2QWL; Jiang et al . , 2007 ) . The structure of bovine Hsc70 ( 1–394aa ) is shown in a ribbon model . Side chains of mutated residues are shown in wire drawing with the same color-code in A . An ADP molecule bound to the nucleotide binding cleft was shown in the green surface model . DOI: http://dx . doi . org/10 . 7554/eLife . 04659 . 019 Since many mutations of Hsp70s affecting their structures and/or functions have been identified in their NBD ( McClellan and Brodsky , 2000; Chang et al . , 2010 ) , we further examined whether such mutations alter the tRNA binding ability of the NBD of Ssa proteins . We introduced several mutations altering conserved amino acid residues in the nucleotide-binding cleft to GST-Ssa-NBD fusions ( Figure 6—figure supplement 1 ) . Mutations affecting efficiency of ATP hydrolysis and its activation , namely F66L and K69Q , enhanced tRNA binding while those compromising nucleotide binding/exchange ( G199D ) reduced the binding ( Figure 6C ) . Those mutations in the NBD that alter structural response to the nucleotide state ( E228Q , D229N ) caused virtually no changes in the tRNA binding . These results indicate that particular residues in the NBD are responsible for tRNA recognition by Ssa proteins , and that the structural features of the ATP/ADP binding state are important for the tRNA binding . Interestingly , the NBDs of Ssa1p and Ssa2p responded differently to these mutations; the Ssa1p-NBD responded more strongly to F66L and K69Q than the Ssa2p-NDB . More prominently , E173S and T201A , which affect the nucleotide binding/exchange , only caused negative effects on the Ssa2p-NBD , suggesting that these differences in in vitro-binding of tRNAs might be the reason why Ssa2p but not Ssa1p is responsible for nuclear import of tRNAs in vivo . If Ssa2p is directly involved in the transport of tRNAs through the NPC , Ssa2p needs to interact with nucleoporins ( Nups ) , components of the NPC . Thus , we tested this possibility by the pull-down assay . We mixed recombinant GST , GST-Nup100 ( 1–640 ) p , GST-Nup116 ( 165–715 ) p or GST-Nsp1 ( 1–601 ) p fusions with Ssa proteins , and recovered the GST fusions with Glutathione Sepharose beads . These portions of Nups are mostly composed of FG repeats , and are supposed to form a hydrogel phase in the NPC to allow selective transport of macromolecules ( Frey et al . , 2006; Frey and Görlich , 2007; Patel et al . , 2007 ) . In the control assays , both Ssa1p and Ssa2p were absent in the bound fraction of glutathione-beads pre-incubated with GST ( Figure 7A ) . Interestingly , both Ssa proteins specifically interacted with GST-Nup116 ( 165–715 ) p , but not with GST-Nup100 ( 1–640 ) p or with GST-Nsp1 ( 1–601 ) p . These results suggest that Ssa proteins can interact with the NPC through binding to certain Nups , such as Nup116p . 10 . 7554/eLife . 04659 . 020Figure 7 . Ssa proteins interact with a certain nucleoporin . ( A ) Interaction between Ssa proteins and nucleoporins was analyzed by the pull-down assay . 200 pmol of either GST , GST-Nup100 ( 1–640 ) p , GST-Nup116 ( 165–715 ) p or GST-Nsp1 ( 1–601 ) p were mixed with 40 pmol of either Ssa1p or Ssa2p , and were pull-downed with Glutathione Sepharose . The same portions of total ( T ) , unbound ( U ) , and bound ( B ) samples were subjected to SDS-PAGE . Ssa proteins in each fraction were detected by Western blotting with anti-Ssa protein antibodies ( upper , WB ) , and total proteins were visualized by CBB staining ( lower , CBB ) . Positions of GST fusions were indicated by arrowheads . ( B ) tRNA interaction with Nups were assayed by a variant of the low affinity binding assay . Alexa 488-labeled tRNA-ProUGG ( 0 . 50 µg ) was incubated with GST- or GST-Nup116 ( 165–715 ) p-coated Glutathione Sepharose in the absence ( none ) or presence of Ssa1p ( Ssa1p ) or Ssa2p ( Ssa2p ) . Binding of the fluorescent tRNA to the beads was monitored with a fluorescence microscope . DOI: http://dx . doi . org/10 . 7554/eLife . 04659 . 020 Then , we wanted to know whether Ssa proteins support NPC-interaction of tRNAs . To this end , the 3′-terminally Alexa 488-labeled tRNA-ProUGG and glutathione-beads coated with GST or GST-Nup116 ( 165–715 ) p were incubated in the absence or presence of the Ssa protein , and the mixtures were subjected to fluorescence microscopy ( Patel et al . , 2007 ) . Peripheral fluorescence on the glutathione-beads were observed only when the beads coated with GST-Nup116 ( 165–715 ) p were incubated with the fluorescent tRNA in the presence of Ssa proteins ( Figure 7B ) , while no such peripheral signal was seen in the absence of Ssa proteins and when GST-coated beads were used . We also performed similar experiments with Alexa 488-labled total yeast tRNAs , and obtained similar results ( data not shown ) . These results support the idea that Ssa2p drives interaction of tRNAs to the NPC . Hsp70s usually cooperate with co-chaperones to conduct their functions . Hsp40/DnaJ co-chaperones accelerate the ATPase activity of Hsp70s and pass specific substrates to corresponding Hsp70s to control their functionality in a spatiotemporal manner , while co-chaperones , GrpE , Bag-1 , and Hsp110 , act as nucleotide exchange factors ( NEFs ) ( Young et al . , 2004; Kampinga and Craig , 2010 ) . To examine whether certain yeast homologues of DnaJ are involved in nuclear accumulation of tRNAs under starvation conditions , three major cytoplasmic DnaJ proteins , Sis1p , and Ydj1p , which are specific for Ssa proteins , and Zuo1p , which is specific for another class of cytoplasmic Hsp70s , Ssb proteins , were chosen for further analyses . As shown in Figure 7 , both the sis1-151 and ydj1∆ mutants showed an apparent defect in the nuclear accumulation of tRNAs under starvation conditions while the zuo1∆ mutant cells were able to accumulate tRNAs in the nucleus . This is consistent with the finding that Ssb1p and Ssb2p are excluded from the nucleus and stay in the cytoplasm while Ssa1p and Ssa2p shuttles between the two compartments ( Shulga et al . , 1999 ) , and indicates that DnaJ proteins specific for Ssa1p and Ssa2p are required for tRNA import . If the ATPase activation of Ssa2p is required for tRNA import in the cytosol , a nuclear NEF for Ssa2p may contribute to nuclear import , as in the case of GEF for Ran GTPase ( Grünwald et al . , 2011 ) . In yeast , such an NEF is Snl1p , a Bag-1 homologue localized on the nucleoplasmic side of the NE ( Sondermann et al . , 2002 ) . In contrast to the case of sis1-151 and ydj1∆ , snl1∆ did not affect nuclear accumulation of tRNAs under starvation conditions , indicating that Snl1p is not involved in this process . Nevertheless , the effects of the sis1 and ydj1 mutations support the assumption that the ATPase cycle of Ssa2p plays an essential role in tRNA import . Hsp70 is a versatile Swiss Army knife for the protein world through its binding and releasing of protein substrates with destabilized structures ( Werner-Washburne et al . , 1987; Nelson et al . , 1992; James et al . , 1997; Daugaard et al . , 2007; Vos et al . , 2008 ) . Unexpectedly , in the biochemical search for factors driving the nuclear import of tRNAs , we identified a major cytosolic Hsp70 , Ssa2p , as a novel tRNA-binding protein that affects tRNA distribution under starvation conditions . Although Ssa2p has a highly homologous counterpart , Ssa1p , our findings , especially in vivo results , revealed difference between Ssa2p and Ssa1p in their involvement in nuclear import of tRNAs . And the following lines of evidence support the idea that Ssa2p has a novel function in intracellular dynamics of tRNAs as their nuclear import carrier . First , the ssa2∆ , but not ssa1∆ , mutant is defective in nuclear accumulation of the tRNAs analyzed so far under nutrient starvation ( Figure 2 ) . Besides , the ssa1∆ ssa2∆ double mutant does not show any additive defect in tRNA import ( Figure 3 ) . Because we have not analyzed all the isodecoder tRNAs in the yeast , there remains a possibility that nuclear import of certain tRNAs is driven by Ssa1p . Second , Ssa2p binds various tRNA species specifically and directly both in vivo and in vitro , while tRNA binding of Ssa1p was only observed in vitro ( Figures 1D , 4 ) . We noticed that mutations in the conserved residues in the NBD of the Ssa proteins affect the tRNA binding ability in vitro differently between Ssa1p and Ssa2p ( Figure 6C ) . A possible explanation for the difference between the in vivo and in vitro results is that , in vivo , an auxiliary factor ( s ) , such as co-chaperones regulating the ATPase cycle of Ssa2p , may confer specificity to Ssa2p as a tRNA import factor , or that the small difference in the tRNA binding of the two NBDs in vitro is enhanced to yield labor assignment between Ssa1p and Ssa2p in vivo when these multi-functional proteins are surrounded by variety of protein and RNA substrates . Third , Ssa proteins are known to shuttle between the cytoplasm and the nucleus , and their nuclear-cytoplasmic distribution is regulated according to environmental conditions ( Shulga et al . , 1999 ) . Indeed , we observed a marginal but reproducible increase in the nuclear pool of Ssa2p-FLAG upon amino acid starvation ( Figure 2C ) . Such alteration of transport carrier distribution was reported recently for Los1p , Msn5p , so on upon glucose starvation ( Huang and Hopper , 2014 ) . Fourth , Ssa proteins interact with the FG-repeat region of Nup116p , and support association of tRNAs to Nup116p , suggesting that Ssa2p can enter the NPC with tRNAs under certain conditions ( Figure 7 ) . Fifth , we demonstrated the involvement of Sis1p and Ydj1p , major cytoplasmic DnaJs for Ssa proteins , in tRNA import in vivo , while such effect is not observed in the zuo1∆ mutant , which is defective in Ssb-protein specific DnaJ ( Figure 8 ) . These results collectively suggest that tRNA loading onto Ssa2p may be assisted by these cytosolic co-chaperones specific for the Ssa proteins , and ATP hydrolysis by Ssa2p may drive upward transport of tRNAs across the NPC . Therefore , Ssa2p is the first factor that assists tRNA import by direct binding to tRNAs , shuttles between the cytosol and the nucleus , interacts with the NPC component , and couples the tRNA transport with energy release . 10 . 7554/eLife . 04659 . 021Figure 8 . Major cytosolic DnaJ homologues of Ssa proteins are involved in tRNA import under starvation conditions . tRNA localization under starvation conditions in mutants of major cytoplasmic DnaJ homologues ( SIS1 , YDJ1 and ZUO1 ) or a nuclear Bag-1 homologue ( SNL1 ) was analyzed by FISH . Pairs of YDJ1 ( PJ31-3A ) and ydj1∆ ( JJ160 ) strains , SIS1 ( TYSC950 ) and sis1-121 ( TYSC951 ) strains , ZUO1 ( BY4741 ) and zuo1∆ ( 5937 ) strains , and SNL1 ( W303-1B ) and snl1∆ ( SWY1353 ) strains were treated as described in Figure 2 except that all the strains but the SNL1 and snl1∆ strains were cultured at 23°C instead of 30°C . In each set of panels , upper two rows are the parental wild type and the lower two rows are the mutant . Bar , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 04659 . 021 There might be a possibility that Ssa2p acts as an import adaptor for tRNAs , like importin-α for proteins targeted to the nucleus , and Mtr10p , an importin β , functions as an import carrier of the tRNA-Ssa2p complex . However , the ssa2∆ and mtr10-shut off double mutant exhibited an additive defect in tRNA accumulation as compared with single mutants ( Figure 2B ) . While this additive effect was not so significant , the difference between the double mutant and the single mutants was statistically meaningful . Although we could not completely negate the above possibility so far , these data rather support the idea that Ssa2p and Mtr10p form parallel and independent pathways , as the case of Los1p and Msn5p in tRNA export . Another potential non-importin candidate for an Ssa2p import carrier is Opi10p , a yeast homologue of Hikeshi , which functions as a nuclear import carrier of mammalian Hsp70 under heat stress conditions ( Kose et al . , 2012 ) . However , our preliminary experiments suggest that the opi10∆ mutant does not affect localization of Ssa1p or Ssa2p under normal or starvation conditions ( data not shown ) . Although there are several other possibilities for involvement of the Ssa2p chaperone system in tRNA import , from the collective pieces of experimental results , we primarily propose that Ssa2p acts as an import carrier of tRNAs across the NPC . Since the ssa2∆ phenotype is only obvious under starvation conditions , Ssa2p may provide a secondary pathway to accommodate high demand for the tRNA import under these conditions . In other words , the Ssa2p pathway may confer a regulatory capability to the tRNA nuclear import , responding to growth conditions like the case of tRNA export ( Murthi et al , 2010; Huang and Hopper , 2014; Pierce et al . , 2014 ) . We speculate that Ssa2p also has some contribution to basal nuclear import of tRNAs under normal conditions by co-operating with the Mtr10p pathway , and that quantitative regulation of the Ssa2p pathway , but not complete on-and-off , enables enhancement of nuclear import under starvation conditions . As mentioned above , we observed a direct interaction between Ssa proteins and tRNAs , which is governed by a mode different from that for recognition of protein substrates by Hsp70 . tRNA binding is saturable and is competed specifically by other tRNA molecules but not by a protein substrate ( Figures 4 , 6A ) . We also found that tRNA binding takes place through the NBD , but not through the SBD , of Ssa proteins ( Figure 6B ) . On the other hand , Ssa proteins bind tRNAs that lack an adenosine at their 3' terminus , so that a tRNA molecule is not just a mimic of monomeric adenine nucleotides ( Figure 4A ) . Examination of the crystal structure of bovine Hsc70 ( Jiang et al . , 2005 ) indicates that the ATP-binding cleft is too narrow to accommodate the acceptor stem of the tRNA molecule while the several mutations introduced into this region affected tRNA binding of Ssa proteins , especially Ssa2p ( Figure 6C ) , suggesting that the ATP/ADP-binding state of Ssa2p alters an unidentified tRNA-binding interface on the NBD . While disruption of the overall tRNA structure decreases the tRNA affinity for Ssa proteins , some mutations on the acceptor stem enhance Ssa protein recognition ( Figure 5 ) . In addition , Ssa proteins prefer unmodified tRNAs to fully-modified tRNAs for their binding ( Figure 4E ) . Therefore , Ssa proteins appear to recognize some characteristics of the three-dimensional structure of tRNA molecules , though preferably loosely-folded structures of tRNAs . Although the tRNA import system ( s ) seems to accept a variety of tRNA species , including matured and stably folded tRNAs ( Takano et al . , 2005 ) , such preference for unmodified or loosely-folded tRNAs may well contribute to the quality control of cytosolic tRNAs . Recent studies of rapid tRNA decay ( RTD ) , which degrades hypomodified aberrant tRNAs in the cytosol , revealed that the RTD system also recognizes tRNAs with some defects in their acceptor stem , and a part of such aberrant tRNAs are degraded by the nuclear exonuclease Rat1p ( Alexandrov et al . , 2006; Whipple et al . , 2011; Wilusz et al . , 2011 ) . Because the above characteristics of the NBD of Ssa proteins are similar to those of the SBD with respect to protein binding ( Deshaies et al . , 1988; Tanaka et al . , 2002 ) , we postulate that the NBD possesses a chaperone-like activity for tRNAs in recognizing their structural features , and may contribute to substrate selection for the nuclear RTD through tRNA import , if such tRNAs escape from the cytoplasmic RTD . Although there may be other potential roles of Ssa2p in nuclear import of tRNAs , binding and releasing of tRNAs by Ssa2p , which is likely coupled with transport across the NPC , could constitute an intrinsic step for the tRNA import . An attractive idea is that tRNA import is driven by the ATPase cycle of the Hsp70 in a similar manner to the nuclear import driven by the Ran GTPase cycle . The Ssa2p-mediated pathway may function as a regulatory pathway for tRNA import to adapt nutrient stress by using the heat-shock protein in a novel mode of action . Future studies should reveal the detailed molecular mechanism , by which Hsp70 contributes to tRNA import into the nucleus and its regulation upon nutrient stress . Yeast genetic techniques are essentially described in Guthrie and Fink ( 1991 ) , and other molecular biological techniques are in Sambrook and Russell ( 2001 ) . S . cerevisiae strains used in this study are summarized in Supplementary file 2 . A yeast cytosolic fraction was prepared from logarithmically growing the S . cerevisiae strain W303-1A . First , the yeast cells were converted into spheroplasts and disrupted in 50 mM Tris–HCl , pH 7 . 4 , 350 mM NaCl , 5 mM MgCl2 supplemented with 0 . 5 mM PMSF , a 1/1000 volume of PIC ( Roche Diagnostics ) , and 1 . 0 mM β-mercaptoethanol by vigorous agitation with glass beads . The lysate was mixed with final 0 . 5% wt/vol of Triton X-100 , and was centrifuged at 100 , 000×g for 30 min . The recovered supernatant was passed through Q-Sepharose Fast Flow ( GE Healthcare ) to remove endogenous tRNAs . The flow-through fraction was applied to the tRNA-resin prepared as described previously in the presence or absence of 3 mM Mg-ATP , and bound proteins were eluted with 50 mM Tris–HCl , pH 7 . 4 , 1 . 5 M NaCl , 5 mM MgCl2 after extensive washing . ATP-dependent or ATP-sensitive tRNA-binding proteins were identified by peptide mass fingerprinting with a Voyager DE MALDI/TOF mass spectrometer ( Applied Biosystems , Foster City , CA ) . Full-length Ssa1p and Ssa2p recombinant proteins were expressed by a pCold2 vector ( Takara Bio ) with a C-terminal His6 tag in the Escherichia coli strain BL21 ( DE3 ) co-expressing trigger factor overnight at 15°C . The recombinant proteins were purified with Ni-NTA agarose ( QIAGEN , Hilden , Germany ) , and dialyzed against 20 mM Tris–HCl , pH 7 . 4 , 10 mM NaCl . For preparing GST fusions with partial Ssa proteins , PCR-amplified gene fragments encoding either Ssa1p-NBD ( 1–380 ) , Ssa1p-NBD-SBD ( 1–550 ) , Ssa1p-SBD ( 381–550 ) , Ssa1p-SBD-CVD ( 381–642 ) , Ssa1p-CVD ( 551–642 ) , Ssa2p-NBD ( 1–380 ) , Ssa2p-NBD-SBD ( 1–550 ) , Ssa2p-SBD ( 381–550 ) , Ssa2p-SBD-CVD ( 381–639 ) or Ssa2p-CVD ( 551–639 ) were cloned into pGEX-4T-2 ( GE Healthcare ) . Mutant forms of GST-Ssa-NBD genes were constructed by oligonucleotide-directed mutagenesis with overlap extension . Each GST fusion protein was expressed in BL21 ( DE3 ) cells and purified with Glutathione Sepharose ( GE Healthcare ) . The eluates were dialyzed against 20 mM Tris–HCl , pH 7 . 4 , 10 mM KCl , 2 . 0 mM DTT . To prepare GST-nucleoporin fusion proteins , DNA fragments encoding the 1–640 residues of Nup100p [Nup100 ( 1–640 ) p] , the 165–715 residues of Nup116p [Nup116 ( 165–715 ) p] , or the 1–601 residues of Nsp1p [Nsp1 ( 1–601 ) p] were amplified by PCR and inserted into pGEX-4T-2 . These fusion proteins in addition to GST were expressed in DH5α , and purified with Glutathione Sepharose . The resulting proteins were passed through NAP-10 desalting columns equilibrated with Buffer 88 ( 20 mM HEPES-KOH , pH 6 . 8 , 2 mM Mg ( OAc ) 2 , 150 mM KOAc [Baker et al . , 1988] ) supplemented with 0 . 10% wt/vol Tween-20 . RNAs were analyzed on 7 . 0 M urea/10% wt/vol polyacrylamide gels in the TBE buffer , and stained with Gel Red ( Biotium , Hayward , CA ) . The RNAs in the gels were then transferred to charged nylon membranes ( Hybond N+ , GE Healthcare ) , and specific RNAs were hybridized with an appropriate digoxigenin-labeled oligonucleotide probe produced with DIG Oligonucleotide Tailing Kit , Second Generation ( Roche Diagnostics ) . The signal was developed with ECF ( GE Healthcare ) and read with Storm 860 Image Analyzer ( GE Healthcare ) . Yeast cells were cultured in YPD until log phase and , if required , incubated in SD supplemented with only uracil and adenine ( SD+Ura , Ade ) for 2 hr . The cells were pre-fixed for 15 min with formaldehyde solution , fixed with a paraformaldehyde solution for 1 hr , and then subjected to FISH sample preparation with FITC or rhodamine-labeled oligonucleotide probes as described before ( Yoshihisa et al . , 2003 ) . In the case of DnaJ homologue mutants , some modifications were necessary to improve FISH images . Spheroplast formation was done in 0 . 90 M sorbitol-containing buffer with 18 µg/ml of Zymolyase 100T instead of standard 36 µg/ml Zymolyase . The resulting spheroplasts immobilized on poly-Lys-coated multiwell-slides were permeabilized with 0 . 10% wt/vol Triton-X100 , and were directly subjected to hybridization . Fluorescence images were recorded by a confocal system CSU-10 ( Yokogawa , Tokyo , Japan ) with a cooled CCD camera CoolSNAP HQ2 ( Photometrics , Tucson , AZ ) mounted on a BX-60 fluorescence microscope ( Olympus , Tokyo , Japan ) . The images were analyzed by Metamorph software ( Molecular Devices , Sunnyvale , CA ) . To calculate an NAI of a yeast cell , the average signal intensities of nuclear and cytosolic regions were measured in a cell , and a nuclear/cytosolic signal ratio was calculated . The NAI of each experiment is the average of individual NAIs measured in 30 or above cells ( see Figure 2—figure supplement 3 ) . The average NAI and its standard deviation ( SDV ) are calculated from NAIs obtained from three biological replicates of the same experimental conditions . The label transfer assay was performed essentially as described by Henics et al . ( 1999 ) . A 32P-labeled RNA transcribed with 32P-UTP or 32P-CTP was incubated with an appropriate protein at 30°C for 10 min in 12 mM HEPES-KOH , pH 7 . 9 , 15 mM KCl , 10% vol/vol glycerol , 0 . 20 mM DTT , 0 . 25 units/µl RNasin , and then UV was irradiated at 90 mJ/cm2 . After treatment with a 1/10 volume of RNase Cocktail ( Ambion ) , samples were subjected to SDS-PAGE and radioimaging with Imaging Plate ( Fujifilm , Tokyo , Japan ) and STORM 860 Image Analyzer ( GE Healthcare ) . For a typical pull-down assay , 200 pmol of either GST , GST-Nup100 ( 1–640 ) p , GST-Nup116 ( 165–715 ) p or GST-Nsp1 ( 1–601 ) p and 40 pmol of either Ssa1p or Ssa2p were incubated with Glutathione Sepharose in Buffer 88 with 0 . 10% wt/vol Tween-20 at 4°C for 2 hr . After washing the beads with the same buffer 3-times , bound proteins were eluted with the same buffer with 10 mM reduced glutathione . The glutathione beads binding assay coupled with microscopic observation is a variant of the low affinity binding assay developed by Patel et al . ( 2007 ) . The Alexa 488-labled-tRNA-ProUGG was mixed with Glutathione Sepharose beads that adsorbed GST or GST-Nup116 ( 165–715 ) p in advance in the buffer containing 10 mg/ml BSA and 0 . 50 % wt/vol 1 , 6-hexanediol . If indicated , final 5 . 0 µM Ssa1p or Ssa2p was added . The resulting mixture was observed under the fluorescence microscope .	Plants , animals , and fungi all store their DNA inside their cells within a structure called the nucleus , which is surrounded by a nuclear envelope that separates it from the rest of the cell . This DNA contains the instructions to build proteins , but proteins are actually built elsewhere in the cell , outside of the nucleus . This means that the instructions must first be copied and then carried out through pores in the nuclear envelope before they can be decoded to build the protein . Outside of the nucleus , molecules called transfer RNAs ( or tRNAs for short ) are involved in the decoding process by carrying the required building blocks to the cell's protein-making machinery . The tRNA molecules can also shuttle back and forth , in and out of the nucleus . In yeast and mammals , the localization of the tRNA molecules depends on the availability of nutrients; if nutrients are scarce , then more tRNAs are moved into the nucleus . While the mechanism by which tRNAs exit the nucleus is well characterized , little is known about the movement in the opposite direction . Takano et al . have now analyzed how tRNAs are transported into the nucleus in yeast cells by identifying the proteins that bind tightly to them . These experiments revealed that a protein called Ssa2p interacts with tRNAs . This protein belongs to a large family of proteins called Hsp70 chaperones that assist other proteins to fold into their correct shapes . Takano et al . observed that Ssa2p tends to bind to tRNAs that are poorly folded , suggesting that it may work like other well-known chaperones . Furthermore , cells that lack Ssa2p were shown to be unable to efficiently transport tRNAs into the nucleus when nutrients were limited . These results together demonstrate that the yeast protein Ssa2p acts as a carrier for transporting tRNAs into the nucleus; further work is now required to understand whether this mechanism is also found in other species .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2015	Cytosolic Hsp70 and co-chaperones constitute a novel system for tRNA import into the nucleus
Mutations in the cystic fibrosis transmembrane conductance regulator ( CFTR ) cause recurring bacterial infection in CF patients' lungs . However , the severity of CF lung disease correlates poorly with genotype . Antibiotic treatment helps dramatically prolong patients' life . The lung disease generally determines prognosis and causes most morbidity and mortality; early control of infections is thus critical . Staphylococcus aureus is a main cause of early infection in CF lungs . It secretes sphingomyelinase ( SMase ) C that can suppress CFTR activity . SMase C also inhibits voltage-gated K+ channels in lymphocytes; inhibition of these channels causes immunosuppression . SMase C's pathogenicity is further illustrated by the demonstration that once Bacillus anthracis is engineered to express high levels of SMase C , the resulting mutant can evade the host immunity elicited by a live vaccine because additional pathogenic mechanisms are created . By screening a chemical library , we find that the natural product tannic acid is an SMase C antidote . The cystic fibrosis transmembrane conductance regulator ( CFTR ) Cl− channel is activated when its regulatory ( R ) domain is phosphorylated by cyclic AMP-dependent protein kinase A ( PKA ) ( Tabcharani et al . , 1991 ) . Mutations of CFTR cause CF disease that involves multiple organs ( Knowles et al . , 1983; Kerem et al . , 1989; Riordan et al . , 1989; Rommens et al . , 1989 ) . In the lungs of CF patients , defective CFTR leads to production of thick mucus that obstructs airways and thus predisposes the patients to recurring bacterial infection . Although CF disease originates from genetically defective CFTR protein , the severity of CF lung disease is not well correlated with genotype . The CF lung pathology is not fundamentally different from those of many other types of chronic pulmonary infectious and inflammatory diseases . Aggressive antibiotic treatment and supportive measures have prolonged the median life span of patients from 5 to 37 years ( The Cystic Fibrosis Foundation , 2004 ) . These observations underscore the profound impact of infections on progression and severity of CF lung disease . Given that the lung disease generally determines CF patients' prognosis and causes ∼90% of their morbidity and mortality , it is extremely important to effectively control or mitigate lung infections in the early stages of CF disease . Staphylococcus aureus is a main cause of early infection in the lungs of CF patients ( The Cystic Fibrosis Foundation , 2008 ) . Most S . aureus strains , including methicillin-resistant S . aureus ( MRSA ) , secrete the pathogenic factor sphingomyelinase ( SMase ) C , which cleaves sphingomyelin into phosphocholine and ceramide ( Doery et al . , 1963 ) ( upper panel , Figure 1A ) . Our group has previously discovered that SMase C hydrolysis of sphingomyelin surrounding CFTR protein profoundly suppresses CFTR activity ( Ramu et al . , 2007 ) . Thus , bacteria can further diminish critical residual CFTR activity in CF patients , worsening the negative impact of mutations . By inhibiting CFTR , SMase C-producing bacteria can also create a temporary condition , analogous to CFTR deficiency , in non-CF patients with lung infection . Additionally , compounding this already very serious problem , SMase C strongly inhibits the activity of Kv1 . 3 voltage-gated K+ ( Kv ) channels of lymphocytes ( Xu et al . , 2008 ) . Inhibition of these channels is well known to cause immunosuppression ( Chandy et al . , 2004 ) . 10 . 7554/eLife . 03683 . 003Figure 1 . SMase C suppression of heretologously expressed and native CFTR currents . ( A ) Sphingomyelin hydrolysis reactions catalyzed by SMases C and D . ( B ) and ( C ) Currents of an oocyte injected with cRNA encoding CFTR before or after activation with 50 µM forskolin and 1 mM IBMX in the bath solution ( B ) and activated CFTR currents pre- or post-application of SMase C from S . aureus ( SaSMase C: 0 . 4 ng/μl ) ( C ) , elicited by stepping membrane voltage from the −30 mV holding potential to −80 mV and then to 50 mV . ( D ) Time course of normalized CFTR currents at 50 mV where the arrows indicate addition of forskolin plus IBMX ( arrow 1 ) or SaSMase C ( arrow 2 ) ( mean ± s . e . m , n = 6 ) . ( E ) Normalized current amplitude before or after activation by forskolin plus IBMX and post-SMase C treatment ( mean ± s . e . m , n = 6 ) . ( F ) and ( G ) Native currents of a Calu-3 cell before or after activation with 50 µM forskolin and 1 mM IBMX in the bath solution ( F ) and activated CFTR currents before and after addition of SaSMase C ( G ) , elicited by stepping membrane voltage from the 0 mV holding potential to −80 mV and then to 80 mV . ( H ) Time course of normalized CFTR currents at 80 mV where the arrows indicate addition of forskolin plus IBMX ( arrow 1 ) or SaSMase C ( 0 . 5 ng/μl; arrow 2 ) ( mean ± s . e . m , n = 7 ) . ( I ) Normalized current amplitude before or after activation by forskolin plus IBMX and post-SMase C treatment ( mean ± s . e . m , n = 7 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03683 . 003 SMase C's pathogenicity is further illustrated by the disturbing characteristics reported for a genetically engineered Bacillus anthracis mutant . Natural B . anthracis , against which the live STI-1 ( Sanitary Technical Institute , USSR ) vaccine provides effective protection , produces little SMase C due to a ‘defective’ regulatory gene . However , once B . anthracis is engineered to express high levels of SMase C , the resulting mutant not only remains lethal but also evades the host immunity elicited by the live vaccine because additional pathogenic mechanisms are generated ( Pomerantsev et al . , 1997 ) . Thus , it is important to find effective means to counteract the SMase C action . Experimentally , the CFTR channels are often activated by boosting the concentration of intracellular cAMP with a combination of the adenylate cyclase activator forskolin and the phosphodiesterase inhibitor isobutylmethylxanthine ( IBMX ) ( Csanady et al . , 2000 ) . Figure 1B shows current of CFTR heterologously expressed in a Xenopus oocyte bathed in a forskolin- and IBMX-containing solution . Following exposure to extracellular SMase C from S . aureus , the CFTR current was markedly diminished in a time-dependent manner ( Figure 1C–E ) . To investigate whether SMase C also suppresses native CFTR current , we tested the enzyme's effect on CFTR current in Calu-3 cells ( a lung epithelial cell line commonly used in CFTR research ) after activating their CFTR channels with forskolin and IBMX ( Figure 1F–H ) . As in oocytes , SMase C suppressed CFTR current in Calu-3 cells ( Figure 1G–I ) , a result confirming that this SMase C effect on CFTR activity is not unique to the oocyte heterologous expression system . Individual strains of a given bacterium species may be expected to secrete different isoforms of SMase C . We compared three isoforms from S . aureus to learn their behaviors ( Figure 2A; only isoform #1 was used elsewhere ) . All three recombinant isoforms suppressed CFTR current albeit with differing specific activity ( Figure 2B ) . Two isoforms lost activity over 96 hr , and one retained full activity; all three isoforms exhibited measurable activity for at least 48 hr . It follows that secreted SMase C can remain active for a considerable time following death of the bacteria . Thus , while antibiotics remain critical in combating the infection , an effective means to counteract SMase C action would be important in limiting the negative impact of the bacterial infection on the host . 10 . 7554/eLife . 03683 . 004Figure 2 . Effects of isoforms and mutations on SaSMase C suppression of CFTR current . ( A ) N-terminal sequences of three SaSMase C isoforms . ( B ) Percentage of SaSMase C suppression of CFTR current by individual isoforms at 1 , 48 and 96 hr after final purification of SMase C ( mean ± s . e . m , n = 5 ) , where the final concentration was 0 . 4 ng/µl for isoform 1 and 40 ng/µl for isoforms 2 and 3 . ( C ) Percentage of SaSMase C suppression of wild-type and mutant CFTR currents activated by expressing cRNA encoding the PKA catalytic subunit ( mean ± s . e . m , n = 7–10 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03683 . 004 We previously found that at levels beyond that required to achieve maximal CFTR activation , PKA's catalytic subunit significantly lessens suppression of CFTR current by SMase C ( Ramu et al . , 2007 ) . This observation suggests that SMase C modifying the head group chemistry of sphingomyelin molecules surrounding CFTR makes activation of CFTR more energetically difficult , and this SMase C action can be markedly overcome by ‘over-phosphorylating’ the channel's R-domain until all four sites are phosphorylated . This hypothesis implies that mutant CFTR channels with individual key phosphorylation sites ablated would suffer more pronounced suppression by SMase C at these ‘supersaturating’ levels of PKA's catalytic subunit . To test this prediction , we replaced each of the four phosphorylation sites , one at a time , with alanine and examined SMase C's effect on these single-mutant channels . As a control , we showed that in the presence of a supersaturating level of PKA's catalytic subunit , SMase C suppressed only 25% of the wild-type CFTR current ( Figure 2C ) . In contrast and as predicted , under the same experimental condition , SMase C suppressed 40–65% of the mutants' currents . A second corollary of our hypothesis is that a β adrenergic receptor agonist should alleviate the SMase C-caused suppression of CFTR current by boosting intracellular cAMP . In fact , β receptor agonists are routinely used as bronchodilators in patients suffering from pulmonary infectious and inflammatory diseases . Thus , application of β receptor agonists would not be contraindicated in cases of infection by SMase C-positive bacteria and likely be beneficial . However , given that a β receptor agonist unlikely causes overphosphorylation of the R domain , it may only modestly counteract the SMase C-dependent suppression of CFTR current . To develop a clinically suitable drug from scratch is difficult and costly . One shortcut is to find novel medical benefits in already approved drugs or edible natural products . As such , we screened a chemical library ( Spectrum 2000 ) consisting of approved drugs and natural products , each at 10 µM concentration . We employed a fluorescence-based assay ( Amplex Red ) involving a series of coupled enzymatic reactions . Figure 3A is a correlation plot of the Z scores obtained from two repeated screens . A positive result suggests an antagonizing effect of the tested compound on SMase C or on one or more of the coupled reactions . The 27 compounds with higher Z scores were then examined electrophysiologically for their antagonistic effects on SMase C ( Figure 3B ) . We found that , of those 27 compounds at 10 µM , only tannic acid exhibited a clear SMase C-antagonizing effect ( Figure 3B , C ) . Tannic acid was just as effective at concentrations down to 200 nM ( Figure 3D ) . Furthermore , we found that it also antagonized SMase C of B . anthracis ( Figure 4A–C ) . 10 . 7554/eLife . 03683 . 005Figure 3 . Tannic acid ( TA ) counteracts SaSMase C and SMase D suppression of CFTR current . ( A ) Correlation plot of Z scores ( raw scores minus the mean of the population , divided by the standard deviations of individual compounds ) of individual compounds tested in two repeated screens; 27 compounds with higher Z scores are colored orange . ( B ) Percentage SaSMase C suppression of CFTR currents in the presence of 27 tested compounds , each at 10 µM . ( C ) Time course of normalized CFTR currents at 50 mV without or with 10 µM tannic acid ( TA ) present in the bath; arrow indicates addition of SaSMase C ( 0 . 6 ng/µl ) . ( D ) and ( E ) Time courses of normalized CFTR currents at 50 mV without and with 100 or 200 nM ( D ) or 300 nM ( E ) tannic acid present in the bath; arrow indicates addition of SaSMase C ( D ) and SMase D ( E ) ( mean ± s . e . m , n = 3–5 ) . ( F ) Apparent activities of SaSMase C ( 1 and 2 ) or SMase D ( 0 . 6 ng/μl ) ( 3 ) plotted against the concentration of tannic acid , where the activities are expressed as half-time of the time course of sphingomyelin hydrolysis ( 1 ) or CFTR inhibition ( 2 and 3 ) , which were obtained by the Amplex red assay ( 1 ) or by electrophysiological recordings ( 2 and 3 ) . The half-time values were normalized to the corresponding ones obtained in the presence of relevant enzymes and the absence of tannic acid . DOI: http://dx . doi . org/10 . 7554/eLife . 03683 . 00510 . 7554/eLife . 03683 . 006Figure 4 . Tannic acid antagonizes suppression of CFTR and Kv1 . 3 currents by SMase C of B . anthracis ( BaSMase C ) . ( A ) and ( B ) Currents of an oocyte injected with cRNA encoding CFTR , which was activated as in Figure 1 , before ( pre ) and after ( post ) addition of BaSMase C ( 0 . 6 ng/μl ) without ( A ) and with ( B ) 200 nM TA present . ( C ) Time course of normalized CFTR currents at 50 mV without or with 200 nM TA present in the bath; arrow indicates addition of BaSMase C ( mean ± s . e . m , n = 10 ) . ( D ) and ( E ) Currents of an oocyte injected with cRNA encoding Kv1 . 3 before ( pre ) and after ( post ) addition of BaSMase C without ( D ) and with ( E ) 200 nM TA , elicited by stepping membrane voltage from the −80 mV holding potential to 20 mV and then back to −80 mV . ( F ) Time course of normalized Kv1 . 3 currents at 20 mV without or with 200 nM TA in the bath; arrow indicates addition of BaSMase C ( mean ± s . e . m , n = 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03683 . 006 Tannic acid is known to interact with many other proteins with apparent Kd or EC50 between 8 µM and 150 mM ( Gabbott , 2008 ) . Consequently , high concentrations of tannic acid ( 5–50 mM ) are often used to fix cells in EM studies ( Hayat , 1981; Afzelius , 1992; Bozzola and Russell , 1992 ) . To our knowledge , a EC50 range of 50–100 nM observed in our study is by far the lowest concentration range at which tannic acid is found to bind effectively to a protein and/or to produce a meaningful biological effect ( Figure 3F ) . Although tannic acid has been reported to bind to the choline group of phosphatidylcholine ( PC ) , it does not appear to antagonize SMase C activity by binding to PC and thereby preventing SMase C from accessing sphingomyelin because tannic acid binds to PC only at much higher concentrations than what is required to antagonize SMase C activity ( Kalina and Pease , 1977; Simon et al . , 1994 ) . Furthermore , the EC50 ( ∼50 nM ) of tannic acid for antagonizing the SMase C-catalyzed sphingomyelin hydrolysis in a biochemical reaction ( involving no PC ) is comparable to its EC50 ( ∼100 nM ) for antagonizing SMase C suppression of CFTR activity in a biological membrane ( Figure 3F ) . To assess the relative specificity of tannic acid's antagonizing effect on SMase C , we tested tannic acid against bacterial SMase D . Like SMase C , SMase D specifically hydrolyzes sphingomyelin ( bottom panel , Figure 1A ) . However , unlike SMase C , SMase D removes only the choline group rather than the entire phosphocholine head group from sphingomyelin , leaving ceramide-1-phosphate instead of ceramide behind in the membrane . Our group previously showed that SMase D also suppresses CFTR activity ( Ramu et al . , 2007 ) . Here , we show that tannic acid at 300 nM slightly slowed down SMase D suppression of CFTR activity ( Figure 3E ) . In contrast , as shown above , tannic acid at 200 nM essentially eliminated SMase C suppression of CFTR activity ( Figure 3D ) . Thus , tannic acid antagonizes SMase C and SMase D with differing potency ( Figure 3F ) . In lymphocytes , Kv1 . 3 channels help maintain a sufficiently negative resting membrane potential to sustain an adequate driving force for Ca2+ entry which acts as a key signal to activate lymphocytes ( DeCoursey et al . , 1984; Matteson and Deutsch , 1984; Cahalan and Chandy , 2009 ) . Consequently , inhibition of Kv1 . 3 channels causes immunosuppression , a finding that has stimulated the development of inhibitors of Kv1 . 3 as effective immunosuppressants to treat autoimmune diseases ( Chandy et al . , 2004 ) . On this backdrop , we previously showed that SMase C , by removing the phosphoryl head group of sphingomyelin molecules around Kv1 . 3 channels , profoundly diminishes these channels' activity ( Xu et al . , 2008 ) . The underlying mechanism is that removal of the negatively charged phosphoryl group makes it energetically more difficult for positively charged voltage sensors to transition to the activated state . Here , we find that tannic acid can also effectively counteract suppression of Kv1 . 3 activity by SMase C ( Figure 4D–F ) . As illustrated above , we have shown that tannic acid is an antidote to SMase C . It is a readily available and inexpensive natural product , widely and abundantly used in the food industry . The US Food and Drug Administration terms it Generally Recognized As Safe ( GRAS ) ( FDA , 2013 ) . The use of tannic acid to treat various diseases was reported as early as 1850 ( Alison , 1850; Allegrini and Costantini , 2012 ) , for example in the treatment of human diarrhea . Recently reported clinical doses are 0 . 5–1 g ( Bian et al . , 2009 ) , giving an estimated concentration in human body fluids of 8–16 µM , or 40–80 times higher than required to antagonize SMase C action . We suggest that tannic acid be considered in attempts to protect against B . anthracis mutants engineered to over-express SMase C , in the unfortunate event of an outbreak , or to lessen the harm caused by natural SMase C-positive bacteria including MRSA , in both CF and non-CF patients . The CFTR and PKA-C cDNAs were subcloned in the pGEMHE plasmid ( Liman et al . , 1992 ) . Mutant CFTR cDNAs were obtained through PCR-based mutagenesis and confirmed by DNA sequencing . The cRNAs were synthesized with T3 or T7 polymerase using the corresponding linearized cDNAs as templates . Channel currents were recorded from whole oocytes previously injected with the corresponding cRNAs and stored at 18°C , using a two-electrode voltage clamp amplifier ( OC-725C; Warner , Hamden , CT ) , filtered at 5 kHz and sampled at 50 kHz using a Digidata 1322 interfaced to a PC . The resistance of electrodes filled with 3 M KCl was 0 . 2–0 . 3 MΩ . Molecular Devices pClamp 8 software was used for amplifier control and data acquisition . Unless specified otherwise , the bath solution contained ( in mM ) : 95 NaCl , 5 KCl , 0 . 3 CaCl2 , 1 MgCl2 and 10 HEPES; pH was adjusted to 7 . 6 with NaOH . CFTR current from the Calu-3 cell line was recorded in the whole cell configuration with a patch-clamp amplifier ( 200B; Axopatch ) , filtered at 5 kHz and sampled at 50 kHz using a Digidata 1322 interfaced to a PC . Electrodes were fire polished ( 2–4 MΩ ) and coated with beeswax . Capacitance and series resistance were electronically compensated . Molecular Devices pClamp 8 software was used for amplifier control and data acquisition . The bath solution contained ( in mM ) 145 NaCl , 5 KCl , 0 . 3 CaCl2 , 1 MgCl2 , 10 HEPES ( pH 7 . 30 adjusted with NaOH ) and the electrode solution contained ( in mM ) 140 KCl , 10 EGTA , 1 CaCl2 , 1 MgCl2 , 10 HEPES ( pH 7 . 30 adjusted with KOH ) . SMases were manually added to the recording chamber to test their effects . The cDNAs of SMase C were produced with PCR , primed with a pair of oligonucleotides corresponding to the 5ʹ or 3ʹ translated regions against the genomic DNA isolated from B . anthracis and S . aureus ( isoform 1 , accession number AAB32218; isoform 2 , BAB43091 and isoform 3 , ZP_01242591 ) , respectively . To produce recombinant SMases C and D , Escherichia coli BL21 ( DE3 ) cells were transformed with the respective cDNAs cloned into pET30 vector ( Novagen , San Diego , CA ) , grown in LB medium to ∼0 . 6 OD at 600 nm , and induced with 1 mM IPTG for 2 hr . The bacteria were harvested , resuspended , and sonicated . The resulting samples were loaded onto a cobalt affinity column and eluted by stepping the imidazole concentration from 50 to 500 mM ( all SMase proteins contain N- and C-terminal His tags ) . The imidazole was later removed by dialysis . We screened the Spectrum 2000 compound library with Amplex red assay which is based on the following coupled reactions . First , sphingomyelinase hydrolyzes sphingomyelin , yielding ceramide and phosphocholine . Second , an alkaline phosphatase hydrolyses phosphocholine , yielding choline . Third , choline is oxidized by choline oxidase to betaine and H2O2 . Finally , H2O2 , in the presence of horseradish peroxidase , reacts with Amplex Red reagent in a 1:1 stoichiometry to generate the highly fluorescent product , resorufin . Resorufin has absorption and fluorescence emission maxima of approximately 571 nm and 585 nm , respectively . During the screen , individual wells of a PerkinElmer OptiPlateTM-384 HB black plate containing Spectrum 2320 compounds ( each at 10 μM ) dissolved in DMSO were added with 20 μl of a mixture containing sphingomyelin ( 100 μM ) , alkaline phosphatase ( 16 U/ml ) , choline oxidase ( 0 . 4 U/ml ) , horseradish peroxidase ( 4 U/ml ) , Amplex red ( 100 μM ) , and triton X ( 0 . 4% ) ( Amplex red sphingomyelinase assay kit A12220; Invitrogen , Grand Island , NY ) , using a Bio Tek microdispenser ( Janus Automated Workstation , PerkinElmer , Waltham , MA ) . To start the enzymatic reaction , 20 μl of a solution containing SaSMase ( 30 nM ) was dispensed into individual wells . For chemical library screening , SMase C activity was quantified from the intensity ratio of fluorescence excited at 570 nm and detected at 590 nm in the presence and absence ( DMSO only ) of tested compounds whereas it was quantified by monitoring absorbance at 570 nm in the tannic acid dose–response study .	Cystic fibrosis is an inherited disease that mainly affects the lungs and the intestine . It is caused by defective copies of a protein called CFTR , which normally allows chloride ions to flow in and out of cells through the cell membrane . The activity of the CFTR protein is important for transferring fluid in and out of cells , and the dysfunctional CTFR protein causes the cells lining the lungs and the intestine to produce thick mucus . This leads to problems with breathing and absorbing nutrients , and makes the lungs of people with cystic fibrosis more susceptible to infections . Recurring lung infections aggravate the symptoms of cystic fibrosis and worsen the predicted outcome for sufferers . A common skin bacterium , called Staphylococcus aureus , is one of the first to colonize the lungs of young cystic fibrosis patients . This pathogen releases an enzyme that can further reduce any residual activity of the defective CTFR protein . The enzyme—called sphingomyelinase C ( or SMase C for short ) —also interferes with the function of another protein that allows potassium ions to flow out of immune cells . This effect in turn reduces the body's ability to fight the infection . Now Ramu et al . have found , by using cells grown in a laboratory , that the enzyme released by the bacteria would remain active long after the bacteria had been killed . This finding suggests that a combination therapy of antibiotics that kill the bacteria and a drug that inhibits the enzyme function may greatly improve treatments for cystic fibrosis . Ramu et al . then tested a collection of over 2000 edible naturally-occurring chemicals and drugs , which are already approved for use in humans , to see if any could counteract the activity of the enzyme . A single natural chemical called tannic acid was shown to prevent both the negative effects on CTFR and those on the potassium channel protein . Other pathogenic bacteria also produce an SMase C enzyme and Ramu et al . showed that tannic acid can also interfere with the anthrax bacterium's enzyme . This suggests that treatment with tannic acid may improve the outcome in a number of bacterial infections . Further experimental work is now needed to establish whether tannic acid can alleviate the symptoms of infectious disease in animal models .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2014	Counteracting suppression of CFTR and voltage-gated K+ channels by a bacterial pathogenic factor with the natural product tannic acid
The motor system prepares for movements well in advance of their execution . In the gaze control system , the dynamics of preparatory neural activity have been well described by stochastic accumulation-to-threshold models . However , it is unclear whether this activity has features indicative of a hidden movement command . We explicitly tested whether preparatory neural activity in premotor neurons of the primate superior colliculus has ‘motor potential’ . We removed downstream inhibition on the saccadic system using the trigeminal blink reflex , triggering saccades at earlier-than-normal latencies . Accumulating low-frequency activity was predictive of eye movement dynamics tens of milliseconds in advance of the actual saccade , indicating the presence of a latent movement command . We also show that reaching a fixed threshold level is not a necessary condition for movement initiation . The results bring into question extant models of saccade generation and support the possibility of a concurrent representation for movement preparation and generation . The ability to interact with the world through movements is a hallmark of the animal kingdom . Movements are usually preceded by a period of planning , when the nervous system makes decisions about the optimal response to a stimulus and programs its execution . Such planning behavior is seen in a wide variety of species , including , insects ( Fotowat and Gabbiani , 2007; Card and Dickinson , 2008 ) , fish ( Preuss et al . , 2006 ) , frogs ( Nakagawa and Nishida , 2012 ) , and mammals ( Hanes and Schall , 1996; Churchland et al . , 2006a ) . A fundamental question in sensorimotor neuroscience is how planning activity is appropriately parsed in order to prepare and execute movements . In the primate gaze control system , premotor neurons that produce a volley of spikes to generate a movement are typically also active leading up to the movement . Build-up of low frequency activity prior to the saccade command , or its signatures , have been observed in a wide variety of brain regions involved in gaze control , including the frontal eye fields ( Hanes and Schall , 1996; Gold and Shadlen , 2000 ) , lateral intraparietal area ( Platt and Glimcher , 1999 ) , and superior colliculus ( SC ) ( Dorris et al . , 1997 ) . Since variability in the onset and rate of accumulation of low-frequency activity is correlated with eventual saccade reaction times ( Hanes and Schall , 1996; Ratcliff and Rouder , 1998; Usher and McClelland , 2001 ) , it is thought that this activity primarily dictates when the movement is supposed to be initiated . However , it is unclear how ( or if ) downstream motor networks distinguish activity related to movement preparation from the command to execute one . One possibility is that the activity of premotor neurons undergoes a transformation from representing preparation into movement-related commands at a discrete point in time ( Thompson et al . , 1996; Juan et al . , 2004; Schall et al . , 2011 ) , when the activity reaches a movement initiation criterion , thereby acquiring the potential to generate a movement . Indeed , this is the basis for stochastic accumulator models of saccade initiation , in which premotor activity must reach a threshold level in order to generate the movement ( Hanes and Schall , 1996; Ratcliff and Rouder , 1998; Zandbelt et al . , 2014 ) . Recent work in the skeletomotor system has also suggested that neuronal population activity undergoes a state space transformation just prior to the movement , thus permitting movement preparation without execution before the transformation ( Churchland et al . , 2012; Kaufman et al . , 2014; Elsayed et al . , 2016 ) . These related views dictate that movement planning and execution are implemented as serial processes in the motor system . Alternatively , it is possible that neurons in sensorimotor structures represent these signals concurrently , gearing up to execute a movement in proportion to the strength of the planning activity . In other words , preparatory build-up of neural activity can multiplex higher order signals while simultaneously relaying those signals to effectors; we call this latter property the ‘motor potential’ of preparatory activity . This idea is in fact the premise of the premotor theory of attention ( Rizzolatti et al . , 1987; Hoffman and Subramaniam , 1995 ) , and represents a latent behavioral manifestation of movement preparation . How might we test for the presence of a motor potential in low frequency preparatory activity ? The following thought experiment helps illustrate one approach . Consider the activity of a premotor neuron accumulating over time . Under normal circumstances , inhibitory gating on the saccadic system is released at an internally specified time , possibly when activity crosses a purported threshold level or when the population dynamics reach the optimal subspace , thus resulting in movement generation ( top row in Figure 1a ) . We can then infer that the high frequency burst during movement execution has motor potential if neural activity is correlated with dynamics of the ensuing movement , that is , saccade velocity is faster when burst activity is higher , and vice versa ( match gray traces between top and bottom rows in Figure 1a ) . Now , if inhibition was somehow removed at a prior time through an experimental manipulation instead of allowing the system its natural time course ( thick red line in Figure 1b ) , the occurrence of an early movement would indicate that ongoing low frequency preparatory activity also possesses motor potential . Importantly , this potential can be quantified by correlating neural activity with kinematics of the eye before the onset of the saccade proper ( pre-saccade velocity traces in Figure 1b ) , and comparing it to the previously estimated potential after saccade onset . Furthermore , the dynamics of activity following the manipulation would indicate whether the activity must cross a decision boundary ( i . e . , threshold or optimal subspace ) in order to produce the movement ( dashed traces in Figure 1b ) . This hypothetical manipulation would therefore simultaneously shed light on both concurrent processing of preparatory signals and the criterion for movement initiation . In this study , we used the trigeminal blink reflex to remove inhibition on the gaze control network during ongoing low-frequency activity . The omnipause neurons ( OPNs ) in the brainstem , which discharge at a tonic rate during fixation and are suppressed during saccades ( Figure 1c , Cohen and Henn , 1972; Keller , 1974 ) , also become quiescent during eye movements associated with blinks ( Schultz et al . , 2010 ) . Previous work in our lab has shown that removal of this potent source of inhibition on the saccade burst generator with reflex blinks triggers saccades at lower-than-normal latencies ( Gandhi and Bonadonna , 2005 ) , an observation that has been used to study latent sensorimotor processing in SC ( Jagadisan and Gandhi , 2016 ) , the motor potential of a target selection signal during visual search ( Katnani and Gandhi , 2013 ) , and the dynamics of movement cancellation during saccade countermanding ( Walton and Gandhi , 2006 ) . Here , we first established that the saccade-related burst in SC has motor potential under normal conditions , by correlating the activity during the burst to saccade kinematics on individual trials . Critically , when performing the same analysis in the perturbation condition , we found that the level of preparatory activity at the time of the blink was also strongly correlated with initial dynamics of the evoked movement , prior to the saccade proper , suggesting that ongoing sub-threshold activity in SC also possesses motor potential . Finally , we show that although these movements were preceded by an acceleration of ongoing activity following the perturbation , it is not necessary for preparatory activity in SC to reach a fixed threshold level before a saccade is produced – neural activity just prior to saccades triggered by reflex blinks was lower at both individual neuron and population levels . First , as one way to assay the behavioral manifestation of motor preparation , we verified that reflex blinks during the preparatory period produced low-latency saccades . Figure 3a shows saccade reaction time ( from GO cue ) as a function of the time of blink across all perturbation trials ( red circles ) . To visually compare reaction times on blink trials with those in control trials , it was necessary to include the distribution of control reaction times in this figure . To do this , we created a surrogate dataset by randomly assigning blink times to control trials , and plotted them on the same axes as blink trials in Figure 3a ( blue circles ) . Reaction times in perturbation trials were correlated with time of blink , and were significantly lower than control reaction times ( mean control reaction time = 278 ms , mean blink-triggered reaction time = 227 ms , p=2 . 2×10−197 , one-tailed t-test ) , consistent with previous observations ( Gandhi and Bonadonna , 2005 ) . We then verified whether saccades triggered by the blink were as accurate as normal saccades , in order to eliminate any potential confounds due to differences in accuracy . We calculated saccade accuracy as the Euclidean endpoint error normalized with respect to target location ( inset in Figure 3b ) . The distributions of relative errors for all control and blink trials are shown in Figure 3b . Blink-triggered saccade accuracy was not significantly different from control saccades ( mean control accuracy = 0 . 136 , mean blink-triggered accuracy = 0 . 133 , p=0 . 3 , two-tailed t-test ) , as reported previously ( Goossens and Van Opstal , 2000b; Gandhi and Bonadonna , 2005 ) . We also tested for and found no relationship between blink time and saccade accuracy ( Spearman’s correlation = 0 . 04 , p=0 . 22 ) . Note that the eyes are closed due to the blink , so visual feedback does not contribute to endpoint accuracy . Nonetheless , we have demonstrated previously that blanking the visual target during the blink-triggered movement does not compromise the accuracy on perturbation trials ( Gandhi and Bonadonna , 2005 ) . The blink perturbation thus provides an assay to study the question of motor potential without introducing confounding factors related to saccade metrics . Next , as a crucial prerequisite for our motor potential analysis , we examined how the motor burst in SC is correlated with saccade kinematics . Specifically , we computed the correlation between the trial-by-trial firing rates of a neuron and the corresponding velocities . This approach to estimating motor potential is illustrated in Figure 4a . Since our eventual goal was to study pre-saccade motor potential in blink-triggered saccades , we used the component of velocity in the direction of the saccade goal as our kinematic variable on a given trial ( inset in Figure 4a , see Methods – kinematic variables , for details ) . The choice to use projected kinematics is to maintain uniformity with analyses on blink-triggered movements ( see next section ) , but performing the analysis on the raw , unprojected kinematics for control saccades yielded very similar results ( Figure 4—figure supplement 1 ) . Further , to avoid assumptions about the efferent delay between SC activity and ocular kinematics , we computed the activity-velocity correlation at various potential delays between the two signals ( the blue bars in Figure 4a show an example delay of 12 ms – compare the similarly shaded bars in the velocity and activity panels ) . Figure 4b shows , for an example neuron , the trial-by-trial scatter of velocities at three time points ( 15 ms before , at , and 15 ms after saccade onset – light , medium , and dark blue circles , respectively ) plotted against neural activity preceding those respective velocities by 12 ms . Note that a strong across-trial correlation between activity and kinematics is seen only at the time point that is after saccade onset in this example . It is not surprising to see a lack of correlation with SC activity for pre-saccade velocities , since they are largely zero or constant , by definition , for normal saccades , because the inhibitory gating by OPNs has not been removed yet . In contrast , the strong correlation between kinematics and activity following movement onset indicates the presence of a motor potential in the saccade-related burst . We systematically explored the time course of this motor potential by computing the correlation at different time points before and during the saccade , for a range of delays between activity and kinematics . We did this for each neuron , and the population average correlation coefficients at each time point and delay are shown in the heat map in Figure 4c . To aid interpretation of this figure , the light blue asterisk in the heat map refers to the correlation computed at the time points with the corresponding asterisk in Figure 4a . Motor potential of SC activity emerged only after the onset of the saccade , and lasted throughout the movement ( streak of correlation below the unity line in Figure 4c ) . For each time point in the velocity signal , we also computed the activity time points at which correlation was highest , shown as the running mean ( black trace ) in the heat map . This provides a measure of the efferent delay at which neural activity is most effective in driving movement kinematics . Note that the black trace is roughly parallel to the unity line , suggesting that the efferent delay was consistent for the duration of the movement . To characterize this property better , we plotted optimal efferent delay , calculated as the difference between the black trace and the unity line , as a function time with respect to movement onset ( Figure 4d ) . The mean delay during the movement was −12 . 2 ms ( ±s . d . = 1 . 7 ms ) , meaning that instantaneous saccade kinematics were best predicted by SC activity roughly 12 ms before . This value is centered within the range of previous functional estimates of the conduction delay between SC and extra-ocular muscles ( 8–17 ms , Miyashita and Hikosaka , 1996 ) . The values for the delay before saccade onset are highly variable , likely due to noise in the pre-saccade velocities , and therefore should be ignored ( also note that these delays are positive and therefore non-causal ) . Finally , Figure 4e shows the correlation values at the −12 ms delay as a function of time . The gray region is the ±95% confidence intervals of the population average bootstrapped ( trial-shuffled ) distribution of correlations ( see Methods – surrogate data and statistics ) . Thus , for normal trials , motor potential , in the form of correlation between neural activity and eye velocity , only manifests after the onset of the saccade proper ( starting 3 ms after saccade onset , p<0 . 05 , bootstrap test ) . Having found a correlation between SC activity and ocular kinematics during saccades , we wanted to know whether the time course of such motor potential expands when the saccadic system is disinhibited at earlier time points . Specifically , we wanted to know whether ongoing preparatory activity contained any motor potential before its maturation into a motor command . We hypothesized that , if the low-frequency preparatory activity that precedes the high-frequency burst had motor potential , removing inhibitory gating on the system would result in a slow eye movement proportional to the level of activity before accelerating into a saccade . The rationale was that , because the downstream OPNs become quiescent during the blink , activity in SC neurons that possessed motor potential would be allowed to drive the burst generators , and subsequently , the eye muscles ( see circuit diagram in Figure 1c ) . To test this , we computed the correlation between SC activity and eye velocity before and during blink-triggered saccades in a manner like that for control saccades ( Figure 5a ) . It is important to note that we subtracted the BREM component from the blink-triggered saccade velocities before projecting these residuals in the direction of the saccade goal ( inset in Figure 5a ) . This was done to prevent independent variations in BREM kinematics ( unrelated to SC activity ) from masking any underlying motor potential-related correlation , which we found might be the case when we performed this analysis on the raw velocities for blink-triggered saccades ( Figure 5—figure supplement 1a–b ) . Figure 5b shows an example scatter plot of the trial-by-trial activities versus projected residual velocities for three time points with respect to saccade onset . In contrast with Figure 4b , activity was correlated with velocity at all three time points shown , including the ones at and 15 ms prior to saccade onset . We then computed the population average correlation at different time points before and after the onset of the high velocity saccade , at different efferent delays . Neural activity was highly correlated with movement kinematics after the onset of the high velocity saccade component ( after time 0 on the x-axis in the heat map in Figure 5c ) , in agreement with data from control saccades . Importantly , activity was also correlated with eye kinematics before saccade onset ( time points before 0 on the x-axis in Figure 5c ) , suggesting that upstream SC activity leaked through to the eye muscles as soon as the OPNs were turned off by the blink , causing activity-related deviations in the kinematics around the BREM . The black trace in Figure 5c also shows that the estimated efferent delay was similar to that observed in control trials and consistent before and after saccade onset . This is better observed in Figure 5d ( red trace ) – the mean delay after saccade onset was −11 . 9 ms ( ±s . d . = 2 . 3 ms ) , not significantly different from the mean delay before saccade onset ( mean ±s . d . = −11 . 8 ±1 . 9 ms , up to 30 ms before saccade onset; t-test , p=0 . 93 ) . The time course of efferent delay estimates for control trials from Figure 4d is also overlaid for comparison in Figure 5d ( blue trace ) . Figure 5e shows the correlation as a function of time during the blink-triggered movement at these mean pre- and post- saccade optimal delays ( red trace ) , with the time course for control trials from Figure 4e overlaid for comparison ( blue trace ) . The average correlation across the population was significant ( red trace above gray distribution , p<0 . 05 , bootstrap test ) starting 30 ms before onset of the saccadic component and lasting until the end of the movement , reiterating the key result of the study . Recall that , for control saccades , it made no difference whether the motor potential was computed based on the raw velocity traces or the component of velocity in the direction of the saccade ( compare Figure 4 and Figure 4—figure supplement 1 ) , an expected result since instantaneous eye velocity is predominantly in the direction of the saccade goal during control saccades . In contrast , for blink-triggered saccades , the motor potential was revealed only when considering kinematics in the direction of the saccade goal ( Figure 5 ) , and not with the unprojected kinematics ( Figure 5—figure supplement 1 ) . This observation is important for two reasons . First , it ensures that the pre-saccade motor potential seen during blink-triggered saccades is not a result of misestimating saccade onset within blink-triggered movements . If the observed pre-saccade correlation resulted solely from estimating saccade onset to be later than the ground truth , that is , it is actually a peri-saccade correlation in disguise , then it should persist even with the raw , unprojected blink-triggered velocity residuals , and at the same efferent delay as for the peri-saccade correlation . This is because we have already seen that motor potentials exist once the saccade has started . Second , it adds support to the notion of motor potential itself: if spikes from the preparatory activity of these neurons leaked through to the muscles , you would expect them to only drive kinematics in the neurons’ preferred direction ( as opposed to a non-selective impact on all movements ) . Additionally , we also verified whether the observed speeds of the pre-saccadic and saccadic components of blink-triggered movements were consistent with the neurons’ position along the rostro-caudal extent of the SC . The number of terminating boutons from SC neurons to the horizontal burst generator has been shown to increase monotonically along the rostro-caudal axis of SC ( Moschovakis et al . , 1998 ) . This increasing projection strength is reflected in the ‘main sequence’ of increasing saccade peak velocities with saccade amplitude ( and therefore its rostro-caudal position ) ( Gandhi and Katnani , 2011 ) . We reasoned that if the projection strength of a given neuron is fixed , then any kinematics , even pre-saccadic ones , caused by spikes emitted by that neuron must scale with the projection strength along the rostro-caudal axis . Figure 5—figure supplement 2a shows that the peak velocity of the saccade component in blink-triggered movements increased with amplitude of the eventual movement , re-confirming the well-known main sequence relationship for normal saccades . Consistent with this relationship and its implications for rostro-caudal projection strengths , the average pre-saccade velocities ( in a 20 ms window before detected saccade onset ) also showed a significant positive relationship with movement amplitude ( Figure 5—figure supplement 2b ) , although this relationship was weaker due to the relatively bigger spread of pre-saccade velocities for a given amplitude . We also verified whether the previously computed across-trial correlations between activity and velocity varied as a function of rostro-caudal position , and did not find a significant relationship ( not shown ) . This suggests that a strongly projecting neuron is as likely to have a correlation with kinematics as a weakly projecting one – but the actual kinematics are what are governed by the projection strengths . Note that the time course of motor potential , when it exists , and estimated efferent delay , are remarkably similar across all conditions and analyses ( Figure 5—figure supplement 1d–e ) , including the latent pre-saccade potential seen with the projected kinematics . Moreover , both subjects exhibited qualitatively similar properties in terms of pre-saccade motor potential and efferent delays , although the strength of the activity-velocity correlation was weaker in one of them ( not shown ) . Put together , these observations strongly suggest pre-saccade preparatory activity indeed possesses motor potential which is revealed when the appropriate correlations between neural activity and kinematic variables are computed . Our results so far show that like the saccade-related burst , preparatory activity in superior colliculus also has movement-generating potential , which is normally hidden due to downstream inhibitory gating . Next , we used the fact that blink-triggered saccades are evoked at lower latencies to study the related question of what factors determine movement initiation under normal circumstances . Specifically , we sought to test an influential model of saccade initiation – the threshold hypothesis ( Hanes and Schall , 1996 ) . We asked whether it is necessary for activity in SC intermediate layer neurons to reach a fixed activity criterion in order to generate a movement . Previous studies have estimated the threshold for individual neurons in SC and FEF by assuming a specific time at which the threshold could be reached before saccade onset or by computing the time , backwards from saccade onset , at which premotor activity starts becoming correlated with reaction time ( Hanes and Schall , 1996; Paré and Hanes , 2003 ) . Given the heterogeneity of the activity profiles of premotor neurons , we think this approach is too restrictive to obtain an unbiased estimate of the threshold , if any . Instead , we took a non-parametric approach and scanned through possible times at which a putative threshold might be reached prior to saccade onset ( Jantz et al . , 2013 ) . Figure 6a shows a snippet of the average population activity , normalized by the peak trial-averaged activity during control trials for each neuron , and aligned on saccade onset for control ( blue traces ) and blink ( red traces ) trials . For each neuron in this population ( n = 50 ) , we computed the average activity in 10 ms bins slid in 10 ms increments from 50 ms before to 20 ms after saccade onset ( colored windows at the bottom of Figure 6a ) . If activity on control trials reaches the purported threshold at any one of these times before saccade onset , a comparison with activity in blink trials at that time should reveal the existence , or lack thereof , of a fixed threshold . Figure 6b shows the activity in these bins for control trials plotted against blink trials , colored according to the bins in Figure 6a . Note that the majority of points for early time bins lie below the unity line . Activity on blink trials was significantly lower compared to control activity from 50 ms before to 10 ms after saccade onset ( square points , Wilcoxon signed-rank test , comparisons in each of these windows were significant at p=10−6 , at least ) . The systematic trend in the linear fits ( solid lines ) to these points suggests that the activity on blink trials gradually approaches that on control trials; however , the earliest time at which control activity was not different from activity in blink trials was 20 ms after saccade onset ( circles , Wilcoxon signed-rank test , p=0 . 8 ) – too late to be considered activity pertaining to a movement initiation threshold . Thus , activity at the population level need not reach a fixed threshold level in order to produce a movement . Nevertheless , we wanted to know if there exist individual neurons in the population that might obey the threshold hypothesis . For each neuron , we calculated whether activity on blink trials was higher , lower , or not significantly different from activity in control trials , at each time point from −50 before to 20 ms after saccade onset ( Wilcoxon rank-sum test , comparisons in each of these windows were significant at p=0 . 001 , at least ) . The three traces in Figure 6c represent the proportion of neurons that showed each of those three characteristics as a function of time . As late as 10 ms before saccade onset , more than 60% of the neurons had lower activity on blink trials compared to control trials ( blue trace ) , inconsistent with the idea of a fixed threshold . Roughly 30% of the neurons did not exhibit significant differences in activity on blink and control trials at that time point ( black trace ) ; however , this observation is insufficient to conclude that the activities in the two conditions were identical , or that it must reach a threshold . Of course , it is possible that some of these neurons belong to a class for which fixed thresholds have been observed in previous studies . Together , these results suggest that it is not necessary for premotor activity in SC intermediate layers to reach a fixed threshold at the individual neuron or population level in order to produce a movement . Since SC neurons do not necessarily cross a fixed threshold to produce movements , as we saw above , it is possible that blink-triggered saccades are initiated directly off the ongoing level of preparatory activity . Alternatively , low frequency SC activity may be altered by the blink , even if the saccade is triggered at a lower level compared to control trials . Therefore , we studied whether the dynamics of SC activity are modulated by the blink prior to saccade initiation . Since we wanted to test for a change in dynamics before the actual saccade started , we restricted our analysis to the subset of trials in which saccade onset occurred at least 20 ms after blink onset . This restriction reduced our population to 38 neurons . For each neuron , we estimated the rate of accumulation of activity in 20 ms windows before and after blink onset with piecewise linear fits ( Figure 7a , dashed red trace and solid lines ) . It is important to note that while the evolution of premotor activity is commonly modelled as a linear process , the actual dynamics of accumulation may be non-linear , causing spurious changes in linear estimates of accumulation rate over time . To account for this , we created a surrogate dataset of control trials for each neuron , with blink times randomly assigned from the actual distribution of blink times for that session . We then performed linear accumulation fits for the control dataset as well ( dashed blue trace and solid lines in Figure 7a ) . Changes in accumulation rate on control trials following the pseudo-blink should reflect the natural evolution of activity at typical blink times and provide a baseline for comparing any changes observed in blink trials . Figure 7b shows a scatter plot of pre- and post- blink accumulation rates on control and blink trials . Pre-blink rates were not different between the two conditions ( light circles , Wilcoxon signed-rank test , p=0 . 4 ) , but post-blink rates were significantly higher on blink trials ( dark circles , Wilcoxon signed-rank test , p=2 . 5×10−6 ) . Next , we tested for a change in accumulation rates following the blink by calculating a rate change index , defined as the difference of post- versus pre- blink rates divided by their sum , for each condition ( Figure 7c ) . This index was positive for most neurons , even for control trials , highlighting the natural non-linear dynamics mentioned above . The change in accumulation rate was significantly higher following the actual blink on blink trials ( Wilcoxon signed-rank test , p=4 . 4×10−6 ) compared to after the pseudo-blink on control trials . Thus , removal of inhibition seems to cause an acceleration in the dynamics of ongoing activity in the lead up to a saccade . Studies on the neural correlates of movement generation have largely focused on the divide between preparatory and executory activity , mainly due to the substantial natural latencies between the cue to perform a movement and its actual execution . Since variability in the properties of post-cue neural activity is correlated with eventual movement reaction times ( Hanes and Schall , 1996; Churchland et al . , 2006b ) , it is thought that this activity is purely preparatory in nature , influencing only when the movement is supposed to be initiated ( Kaufman et al . , 2016 ) , and is devoid of the potential to generate a movement , until just before its execution . Indeed , there is some evidence that movement preparation and execution have distinct neural signatures with a well-defined boundary separating them . Studies of movement preparation in the skeletomotor system have shown that cortical activity reaches an optimal subspace before undergoing dynamics that produce a limb movement ( Afshar et al . , 2011; Churchland et al . , 2012 ) . A related idea suggests that activity pertaining to movement preparation evolves in a region of population space that is orthogonal to the optimal subspace , and this dissociation confers neurons the ability to prepare the movement by incorporating perceptual and cognitive information without risking a premature movement ( Kaufman et al . , 2014; Elsayed et al . , 2016 ) . However , in the absence of perturbations to the natural time course of movement generation , or explanatory models linking them to downstream processing , it is unclear whether such neural correlates are the sole causal links to the process of initiating movements . How do we reconcile such previous observations with the finding in this study of a latent motor potential in preparatory activity ? One possibility is that the oculomotor system operates differently from the skeletomotor system . More generally , it is possible that it is necessary for population activity to be in an optimal , ‘movement-generating’ subspace in order to release inhibition and/or effectively engage downstream pathways leading up to activation of the muscles , but once the motor system has been disinhibited by another means ( e . g . , the blink perturbation in this study ) , preparatory activity is read out as if it were a movement command . Based on the results in our study , we hypothesize that the activity is likely in the movement preparation subspace when the slow eye movement is produced after the blink perturbation , and its subsequent transition into the movement execution subspace results in a high velocity movement . Note that this hypothesis is consistent with the observation that preparatory activity in premotor and motor cortices , even while not resulting in execution , can be tuned to movement parameters , e . g . , direction ( Churchland et al . , 2010 ) . However , this correlation between preparatory activity and spatial metrics of the upcoming movement - which is also present in SC since the locus of activity determines where the movement is going to be directed - is not directly indicative of a motor potential . In our framework , movement-generating potential requires that the activity be correlated with concurrent movement dynamics , ensuring that the activity actually makes its way down to the muscles . More studies that causally delink evolving population activity from physiological gating are needed to clarify these mechanisms . The threshold hypothesis is the leading model of movement preparation and initiation in the gaze control system ( Hanes and Schall , 1996 ) . Inspired by stochastic accumulator models of decision-making ( Carpenter and Williams , 1995; Ratcliff and Rouder , 1998 ) , this theory posits that saccade initiation is controlled by accumulation to threshold of a motor preparation signal in premotor neurons . However , it is unclear whether the preparatory activity must rise to a fixed biophysical threshold at the level of individual neurons or a population of neurons in order to initiate a saccade ( Hanes and Schall , 1996; Zandbelt et al . , 2014 ) , or whether there exists a dynamic equivalent to such a threshold ( Lo and Wang , 2006; Cisek et al . , 2009 ) . Another recent study has shown that the fixed threshold hypothesis does not hold true for most neurons in SC and FEF ( Jantz et al . , 2013 ) , finding that the effective threshold varies based on the task being performed by the subject . However , comparison of thresholds across tasks is subject to the confound that the network may be in a different overall state , thereby modulating the threshold . For instance , the presence or absence of the fixation spot at the time of movement initiation , or the presence of other visual stimuli or distractors , may affect how downstream neurons receiving premotor activity from the whole network decode the level of activity , thus influencing the effective threshold . In our study , thresholds are compared between interleaved perturbation and control trials in the same behavioral paradigm , eliminating the effect of preset network-level confounds in the premotor circuitry within SC and upstream thereof . Our results show that the low-frequency preparatory activity of individual or population of SC neurons does not have to transition into a high-frequency mode to trigger a movement ( Figure 6a , b ) . If gates downstream of the SC are removed , then reduced SC activity is sufficient to move the eyes . The effective reduction in threshold ( or equivalently , non-existence of a fixed threshold ) that we observe is likely due to reduced activity of the OPNs , which are a potent source of inhibition on the pathway downstream of the SC . The OPNs are inhibited by the reflex blink , and thus premotor activity needs to overcome lesser inhibition and is able to trigger movements off a lower level . It is also important to note that while there is some evidence that premotor activity in SC is attenuated when saccades are perturbed by a reflex blink ( Goossens and Van Opstal , 2000a ) , we did not observe suppression during movements that were triggered by the blink , as seen in the firing rate profile in Figure 7a . Nevertheless , the presence of any attenuation would only strengthen the result , since the occurrence of saccades despite attenuated SC activity goes against the notion of a rigid threshold . It has been hypothesized that the role of a threshold may perhaps be to arbitrate between speed and accuracy during decision making and movement planning ( Heitz and Schall , 2012 ) . Studies have provided contrasting evidence for this idea , with some suggesting a collapsing bound reflecting the urgency of a movement ( Cisek et al . , 2009; Thura et al . , 2012 ) , and others placing the onus of balancing speed with accuracy on non-threshold features of neural activity , such as baseline and gain ( Hanks et al . , 2014; Salinas et al . , 2014 ) . In our study , the accuracy of reduced latency blink-triggered movements was unaltered relative to normal saccades ( Figure 3b ) , despite the effective reduction in threshold we observed . However , the accumulation rate increased following disinhibition by the blink , suggesting that threshold and rate may be independently modulated to achieve the requisite balance between speed and accuracy . In addition , the observation that reaction times can be reduced without affecting accuracy suggests that saccades may not be subject to a strict speed-accuracy tradeoff , at least in the context of a simple single target task such as the one used in this study . Strikingly , for such tasks , even higher accuracies can be observed in conjunction with faster responses if an appropriate behaviorally relevant parameter is manipulated ( e . g . , reward in Takikawa et al . , 2002 ) . Thus , while speed-accuracy tradeoffs have been implicated in optimal control of saccades when speed refers to saccade velocity ( Harris and Wolpert , 2006 ) , more complex tasks which invoke cognitive processes such as decision-making may be necessary to observe a tradeoff of accuracy with reaction speed . Studies have shown that the latencies of saccades and hand movements are reduced under experimental conditions that introduce a startling stimulus ( Valls-Solé et al . , 1995 ) , present targets with predictable spatial and temporal features ( Paré and Munoz , 1996 ) , or provide instructions to time the movement ( Haith et al . , 2016 ) . We speculate that some combination of early onset of preparatory activity , higher baseline , or fast accumulation to a trial-specific threshold must be achieved before the movement is triggered . In terms of the population dynamics framework discussed a couple of sections back , the reduction in latency can be viewed as speeding up the transition of population signals from movement preparation to movement generation subspaces . It is important to note that the dynamics of movements evoked under the conditions used in these experiments are remarkably similar to control movements . In our view , reduced latency blink-triggered movements are not produced by the same neural mechanisms . We have reported previously that delivering an air-puff to the ear or neck does not reduce the latency of saccades , thus discounting the startle stimulus explanation ( Gandhi and Bonadonna , 2005; Jagadisan and Gandhi , 2016 ) . Here , we randomly interleaved blink and control trials , so preparatory signals cannot start to accumulate earlier and/or faster on perturbation trials . Finally , the dynamics of blink-triggered movements and control saccades are not similar . Our method thus reveals a motor potential component during the preparatory period that is not disclosed in the other studies . Although it is well-known that the locus of activity on the SC map determines amplitude and direction of the saccade vector ( see Gandhi and Katnani , 2011 ) , for a review ) , it has thus far been unclear whether the instantaneous firing rate of SC neurons directly drives saccade kinematics , i . e . , instantaneous velocity . In our view , the mini-vector model of saccade execution comes closest to specifying a direct relationship between SC spiking and saccade metrics ( Van Gisbergen et al . , 1987; Arai et al . , 1994; Goossens and Van Opstal , 2006 ) . This model proposes that each spike in SC contributes to a fixed length desired displacement of the eye during the window when gating is open , and has been tested with blink perturbations ( Goossens and Van Opstal , 2006 ) . But it is unclear how this translates to the motor potential we observe , either before or during the saccade , since we estimate this potential from correlated variability between SC activity and eye kinematics across trials , not within a trial . Moreover , in Figure 5 , this correlation is computed between activity and residual velocity projected onto the direction of the saccade target after subtraction of the BREM template . This can cause the kinematic variable to be instantaneously negative ( i . e . , going away from the saccade target ) on some trials , but as long as it is less negative on trials when the activity is higher , we can say that the activity has motor potential . In contrast , the mini-vector model will predict that the eye moves in fixed vector increments towards the saccade target ( which happens roughly to be the optimal vector of the recorded neuron ) . Moreover , in the previous study , not all neurons show the fixed spike count property , and at the population level , spike counts on perturbation trials are slightly higher than on control trials . This is entirely consistent with our observation that spikes can leak through when the gating is open ( leading sometimes to excess spikes as in Goossens and Van Opstal , 2006; also see Buonocore et al . , 2017 ) . Previous work has shown that the discharge profiles of burst generator neurons in the reticular formation are a scaled version of the observed eye velocity waveforms ( Cullen and Guitton , 1997 ) . Moreover , in the presence of a perturbation - for example , when a natural blink accompanies a saccade ( Gandhi and Katnani , 2011b ) or when a brief torque is applied to the head during eye-head gaze shifts ( Sylvestre and Cullen , 2006 ) - burst neuron activity is also modified to account for the observed changes in eye velocity . Given these results , we predict that the lower brainstem burst generator neurons will exhibit low frequency activity to produce the slow eye movement leaked by premature inhibition of OPNs , followed by a high frequency burst that generates the saccade . We would like to emphasize that the observed results - preparatory motor potential , reduced threshold , accelerated activity dynamics - are most likely indirect effects of the trigeminal blink reflex , via inhibition of the OPNs , and not directly due to the reflex itself . Prior work has shown that the activity of SC neurons is not affected by the BREM produced during fixation ( Goossens and Van Opstal , 2000a; Jagadisan and Gandhi , 2016 ) . For blinks produced after the saccade target is presented , some SC neurons in fact exhibit attenuation ( Goossens and Van Opstal , 2000a ) , although we did not see it in our dataset ( and even if that did happen , it is counter-intuitive to and does not explain the motor potential and acceleration of activity ) . These observations collectively suggest that the acceleration of activity that leads to a reduced latency saccade is not directly due to the trigeminal blink reflex but indirectly due to OPN inhibition . In our study , we found the correlation between SC activity and eye kinematics during the saccade to be maximal at a time shift of 12 ms between the two signals , providing us with an estimate of the optimal efferent delay between SC and extraocular muscles . This is in line with previously estimated ranges for the efferent delay ( 8–17 ms , Miyashita and Hikosaka , 1996 ) . In that study , single pulse microstimulation was delivered to the SC during an ongoing saccade , and the latency to deviation from normal saccade kinematics provided an estimate of the time it takes for a spike in SC to reach extraocular muscles while the gating in the pathway downstream of SC is open . The fact that we observe the same efferent delay ( i . e . , 12 ms ) prior to saccade onset , when the OPNs are already quiescent due to the reflex blink , fits neatly within this picture and adds credibility to the notion of a latent motor potential in preparatory spikes . An influential idea in systems neuroscience is the premotor theory of attention , which posits that spatial attention is manifested by the same neural circuitry that produces movements ( Rizzolatti et al . , 1987 ) . Consistent with this hypothesis , studies with cleverly designed behavioral tasks have attributed low-frequency build-up of activity in premotor neurons to a number of cognitive processes , including target selection ( Schall and Hanes , 1993; Horwitz and Newsome , 1999; McPeek and Keller , 2002; Carello and Krauzlis , 2004 ) , attention ( Goldberg and Wurtz , 1972; Ignashchenkova et al . , 2004; Thompson et al . , 2005 ) , decision-making ( Newsome et al . , 1989; Gold and Shadlen , 2000; Ramakrishnan and Murthy , 2013 ) , working memory ( Sommer and Wurtz , 2001; Balan and Ferrera , 2003 ) , and reward prediction ( Platt and Glimcher , 1999; Hikosaka et al . , 2006 ) . However , such multiplexing of cognitive signals along with movement preparation and execution-related activity by premotor neurons only provides circumstantial evidence in support of the premotor theory , leaving open the question of whether the low-frequency activity exclusively represents cognitive and preparatory processes , devoid of the concurrent ability to generate a movement ( a . k . a motor potential ) . Efforts to delineate spatial attention and movement intention by means of causal manipulations have produced a mixed bag of results , with some studies supporting disjoint processing ( Juan et al . , 2004 ) and others supporting concurrent processing ( Moore and Armstrong , 2003; Katnani and Gandhi , 2013 ) . The strongest piece of evidence yet for concurrent processing is the observation that many of these premotor neurons also discharge following the onset of a visual stimulus ( Wurtz et al . , 2001 ) , which can make its way down to effectors resulting in an electromyographic response , e . g . , in the neck ( Corneil et al . , 2004 ) . Such ‘leakage’ of decision-related activity down to the muscles has also been observed in other effectors , including reaches ( Corneil and Munoz , 2014; Servant et al . , 2015 ) , and even across effectors ( Joo et al . , 2016 ) . The discovery of a latent motor potential in the preparatory activity of SC neurons significantly advances this debate by suggesting that while the low-frequency build-up may not trigger movements under normal conditions , movement intention and motor programming signals are also encoded by these neurons in parallel . Moreover , unlike manipulations such as microstimulation or pharmacological inactivation that introduce extrinsic signals that may corrupt the natural processing of this activity ( Katnani and Gandhi , 2013 ) , reflex blinks are non-invasive and are likely to provide a more veridical readout of ongoing processes . It is worthwhile to end on a note of caution . The results in this study are based on experiments performed in one node , SC , in a distributed network of brain regions involved in gaze control . Traditional knowledge imposes a hierarchy on the sensorimotor transformations that need to occur before a gaze shift is generated ( Wurtz et al . , 2001 ) . It is possible that sensorimotor neurons in SC , and to some extent , FEF , which project directly to the brainstem burst generator ( Segraves , 1992; Rodgers et al . , 2006 ) , are more likely to exhibit signatures of a motor potential in preparatory activity compared to regions higher in the cascade . Neurons in other regions may still need to signal the initiation of a movement by reaching a threshold , optimal subspace , or other similar decision bound . Furthermore , it is known that movement initiation thresholds in SC and FEF can vary based on task context ( e . g . , Jantz et al . , 2013 ) . The results presented here are based on a relatively simple task – the delayed saccade task . The mechanisms of movement initiation , and the presence of motor potential in preparatory activity , could in principle be different in more complex tasks , e . g . those that involve competitive spatial selection of movements or sequential movements . Future studies that take causal approaches to perturbing intrinsic population dynamics in various premotor areas across different tasks and effector modalities are essential in order to gauge whether the findings in this study point to a fundamental and generalizable property of sensorimotor systems . All experimental and surgical procedures were approved by the Institutional Animal Care and Use Committee at the University of Pittsburgh and were in compliance with the US Public Health Service policy on the humane care and use of laboratory animals . We used two adult rhesus monkeys ( Macaca mulatta , 1 male and 1 female , ages 8 and 10 , respectively ) for our experiments . Under isoflurane anesthesia , a craniotomy that allowed access to the SC was performed and a recording chamber was secured to the skull over the craniotomy . In addition , posts for head restraint and scleral search coils to track gaze shifts were implanted . Post-recovery , the animal was trained to perform standard eye movement tasks for a liquid reward . Visual stimuli were displayed by back-projection onto a hemispherical dome . Stimuli were white squares on a dark grey background , 4 × 4 pixels in size and subtended approximately 0 . 5° of visual angle . Eye position was recorded using the scleral search coil technique , sampled at 1 kHz . Stimulus presentation and the animal’s behavior were under real-time control with a LabVIEW-based controller interface ( Bryant and Gandhi , 2005 ) . After initial training and acclimatization , the monkeys were trained to perform a delayed saccade task . The subject was required to initiate the trial by looking at a central fixation target . Next , a target appeared in the periphery but the fixation point remained illuminated for a variable 500–1200 ms , and the animal was required to delay saccade onset until the fixation point was extinguished ( GO cue ) . Trials in which fixation was broken before peripheral target onset were removed from further analyses . The animals performed the task correctly on >95% of the trials . During each session , we presented the targets in one of two locations – either inside the response field of the recorded neuron , or in the diametrically opposite location ( see below ) . On 15–20% of trials , fixation was perturbed by delivering an air puff to the animal’s eye to invoke the trigeminal blink reflex . Compressed air was fed through a pressure valve and air flow was monitored with a flow meter . To record blinks , we taped a small Teflon-coated stainless steel coil ( similar to the ones used for eye tracking , but smaller in coil diameter ) to the top of the eyelid . The air pressure was titrated during each session to evoke a single blink . Trials in which the animal blinked excessively or did not blink were excluded from further analyses . The air-puff delivery was randomly timed to evoke a blink either during fixation ( 400–100 ms before target onset ) or 100–250 ms after the GO cue , during the early phase of the typical saccade reaction time . The blink evoked during fixation produced a slow and loopy blink-related eye movement ( BREM , gray traces in Figure 2a ) . The eyes returned to near the original position and fixation was re-established for 400–100 ms before a target was presented in the periphery and the remainder of the delayed saccade task continued . The window constraints for gaze were relaxed for a period of 200–500 ms following delivery of the air puff to ensure that the excursion of the BREM did not lead to an aborted trial . The blink evoked after the GO cue typically produced a blink-triggered movement that can be described as a combination of a BREM and a saccade to the desired target location ( colored traces in Figure 2a ) . We used deviations from the BREM profile to determine true saccade onset , as described in more detail in the next section . Data were analyzed using a combination of in-house software and Matlab . Eye position signals were smoothed with a phase-neutral filter and differentiated to obtain velocity traces . Normal saccades , BREMs , and blink-triggered gaze shifts were detected using standard onset and offset velocity criteria ( 50 deg/s and 30 deg/s , respectively ) . Onsets and offsets were detected separately for horizontal and vertical components of the movements and the minimum ( maximum ) of the two values was taken to be the actual onset ( offset ) . Saccade onset times within blink-triggered movements were detected using a non-parametric approach ( Katnani and Gandhi , 2013 , also see Figure 2a ) . We first created an estimate of the expected BREM distribution during each session by computing the instantaneous mean and standard deviation of the horizontal and vertical BREM velocity profiles . Then , for each blink-triggered movement in that session , we determined the time point at which the velocity exceeded the ±2 . 5 s . d . bounds of the BREM profile distribution , and remained outside those bounds for at least 15 successive time points . We did this separately for the horizontal and vertical velocity profiles , and took the earlier time point between the two components as the onset of the saccade . We further manually verified that the detected saccadic deviations on individual trials were reasonable , esp . , in the spatial domain . Figure 2a illustrates this approach for three example trials . During each recording session , a tungsten microelectrode was lowered into the SC chamber using a hydraulic microdrive . Neural activity was amplified and band-pass filtered between 200 Hz and 5 kHz and fed to a digital oscilloscope for visualization and spike discrimination . A window discriminator was used to threshold and trigger spikes online , and the corresponding spike times were recorded . The location of the electrode in the intermediate layers of SC was confirmed by the presence of visual and movement-related activity as well as the ability to evoke fixed vector saccadic eye movements at low stimulation currents ( 20–40 µA , 400 Hz , 100 ms ) . Before beginning data collection for a given neuron , its response field was roughly estimated . During data collection , the saccade target was placed either in the neuron’s response field or at the diametrically opposite location ( reflected across both axes ) in a randomly interleaved manner . Response field centers , and therefore , target locations ( also consequently , saccade amplitudes and directions ) varied between 9–25 degrees in eccentricity and spanned all directions . Spike density waveforms were computed for each neuron and each trial by convolving the raw spike trains with a Gaussian kernel . We used a 3 ms wide kernel for the motor potential and threshold analysis ( involving across-trial correlations or trial-averaged neural activity ) and a 10 ms kernel for the accumulation rate analysis ( for better rate estimation on individual trials ) . For a given neuron and target location , spike densities were averaged across trials after aligning to target and saccade onsets . Neurons were classified as visual , visuomovement or movement-related , based on the presence of a significant target and/or saccade-related response . We only analyzed visuomovement and movement neurons for this study , the majority of which were visuomovement ( 47/50 ) . Where necessary , we normalized the trial-averaged spike density of each neuron to enable meaningful averaging across the population . The activity of each neuron was normalized by its peak trial-averaged firing rate during normal saccades . Overall , we recorded from 64 neurons for 12339 control trials and 2364 blink trials over 50 sessions . For all analyses , we only considered neurons for which we had at least 7 blink perturbation trials with the target in the response field . Since we only introduced the blink perturbation on a small percentage of trials in a given session in order to prevent habituation , this restricted our population to 50 neurons ( 7891 control trials and 1615 blink trials over 43 sessions ) . We used all 50 neurons for the threshold analysis ( Figure 6 ) . For the motor potential analysis ( Figure 5 ) , since our aim was to correlate neural activity with eye kinematics before saccade onset , we used only the subset of trials where the onset of the saccade was delayed with respect to overall movement onset by at least 20 ms ( see Figure 2b ) . To ensure that the correlation values were reliable , we used only neurons which had at least 7 trials meeting the above criterion . This restriction reduced the number of neurons available for the motor potential analyses to 38 ( 6771 control trials and 869 blink trials over 32 sessions ) , and we used the same neurons for control trials to enable meaningful comparison ( Figure 4 ) . For the same reason , we also used this subset of neurons for the accumulation rate analysis ( Figure 7 ) , where we compared the dynamics of neural activity in 20 ms windows before and after blink onset , and we wanted to ensure that the post-blink window did not include activity co-occurring with the saccade . For the motor potential analyses in Figures 4 and 5 , we computed the across-trial correlation between instantaneous movement kinematics and neural activity for each neuron . We computed the kinematic variable of relevance for each analysis as follows . In all cases , we used the raw or modified ( see below ) horizontal and vertical velocity signals to compute a single vectorial velocity signal using the Pythagorean theorem: vt=vh2 ( t ) +vv2 ( t ) . For the analyses in Figure 4—figure supplement 1 , for control trials , we used the raw , unmodified velocity signals to compute vectorial velocity as a function of time , which we then used as the instantaneous kinematic variable to correlate with neural activity . For perturbation trials in Figure 5 and Figure 5—figure supplement 1 , we first noted that the blink-related eye movement ( BREM ) contributes a substantial velocity component to the overall movement , since the initial phases of velocity and spatial profiles of blink-triggered movements look largely like those of a BREM ( see Figure 2a ) . Thus , in order to extract only the saccadic component of a blink-triggered movement , we subtracted from it the mean BREM template on a given session , and used only the horizontal and vertical residuals to compute the vectorial residual velocity: v~t=v~h2 ( t ) +v~v2 ( t ) , which was used for the correlation analysis in Figure 5—figure supplement 1 . A potential pitfall when using residual velocities by just subtracting out the mean BREM , given the variability in BREM profiles across repetitions , is that intrinsic variability of the BREM itself may mask any correlated variability that might be present between ocular kinematics and neural activity . In other words , if the BREM is driven by an independent pathway compared to the saccade/SC activity , it represents an orthogonal source of variability in the kinematics relative to the activity-driven variability that is being examined . Therefore , for the perturbation trial analysis in Figure 5 , we used the component of residual velocity in the direction of the saccade goal , to isolate variability in the direction of the saccade . The kinematic variable for this analysis is thus defined as: vθ~t=v~h2 ( t ) +v~v2 ( t ) cos⁡θ , where θ is the angle between the instantaneous residual velocity vector and the direction of the saccade goal ( e . g . , between the green and black vectors in the inset in Figure 5a ) . For the sake of consistency , we used a similar variable: vθt=vh2 ( t ) +vv2 ( t ) cos⁡θ , for the equivalent control analysis in Figure 4 , even though the instantaneous direction of velocity is largely towards saccade goal in this condition . To estimate motor potential , we computed the correlation between instantaneous neural activity and eye kinematics ( according to the variables defined above ) across trials , across a span of efferent delays ( time shifts ) between activity and velocity . For each neuron , we computed the Pearson’s correlation coefficient ct+Δ , t=corr ( at+Δ , vt ) between the activity vector a ( t ) =[a1 ( t ) , … , an ( t ) ] and the kinematics vector v ( t ) =v1 ( t ) , … , vn ( t ) at time separations or lags Δ∈[-50 , 50] , where n is the number of trials for that condition for that neuron . Each point in panel c in Figures 4–5 and associated figure supplements 1 represents the average correlation coefficient across the population of neurons between firing rate and kinematics at the corresponding time points t , t+Δ . To estimate the optimal efferent delay between activity and velocity , we computed the time at which this population average correlation peaks along the activity axis ( vertical axis in the panel c heatmaps ) for each time point during the movement . A moving average ( 5 ms window ) of this efferent delay trace is shown as the black trace in the heatmaps . The vertical distance of this trace from the unity line is equal to the the actual value of the optimal efferent delay at each movement time point , shown in panel d in Figures 4–5 and associated figure supplements 1 . We then calculated the mean efferent delay during the movement after saccade onset ( and before saccade onset in the perturbation condition ) , and plotted the population average correlation at this delay ( panel e in the aforementioned figures ) . For the accumulation rate analysis in Figure 7 , we created a surrogate dataset of control trials with blink times randomly sampled from the distribution of blink occurrences in perturbation trials for that session and assigned to individual control trials . For each neuron , we created 1000 such pseudo-trials by resampling from and reassigning to control trials . We then fit the accumulation rates 20 ms before and after the blink with piecewise-linear functions . The slopes of the linear fits were taken to be the pre- and post-blink rates of accumulation for each neuron . We then compared the change in accumulation rates before and after the pseudo-blink in control trials and the blink in perturbation trials by computing the rate modulation index for each condition as ratepost-ratepreratepost+ratepre . Note that we also created a similar surrogate dataset for control trials for the purpose of visualization alone in Figure 3a . To estimate the significance of the correlation profile in the motor potential analyses , we performed a bootstrap analysis on a trial-shuffled dataset using the motor potential estimation procedures laid out in the previous section . For each neuron , we shuffled trial identities between the across-trial activity and velocity vectors , and computed population average correlation as before . We performed this for 100 different shuffles , and the resultant across-shuffle mean correlation trace at the optimal efferent delay ( as estimated from the true data ) and the 95% confidence interval bounds are shown in panel e in Figures 4–5 and associated figure supplements 1 . At each time point , we calculated the difference between the actual correlation profile and each iteration of the bootstrap , and computed the 95% confidence interval of this difference distribution . The correlation at a particular time was considered significant if this interval did not contain 0 . For comparisons of threshold and accumulation rate between control and blink-triggered conditions , we used appropriate non-parametric statistical tests – the Wilcoxon rank sum test when comparing distributions across trials , and the Wilcoxon signed-rank test for trial-averaged comparison across a population .	Most of us are familiar with the experience of revving up the engine in anticipation of a red traffic light turning green . The revving , which enables us to move forward as soon as the brakes are released , reflects our ability to plan actions in advance . The brain shows broadly analogous behavior when preparing to move parts of the body . A few hundred milliseconds before we move our eyes , for example , brain regions responsible for eye movements start to gradually ramp up their activity . Returning to the car analogy , revving up the engine will not cause the vehicle to move until we also release the brakes . In the brain , inhibitory mechanisms analogous to a brake prevent the increasing neural activity from triggering movement . The brain is thought to apply the brakes until this preparatory activity reaches a threshold , at which point the activity becomes a command to move . This raises the question: would the preparatory activity be able to trigger movement before reaching threshold if the brakes were no longer being applied ? To find out , Jagadisan and Gandhi devised experiments in which they could essentially “release the brakes” at any time . Monkeys were trained to stare at a central spot on a screen and , only when this spot disappeared , to then move their eyes to a target that appeared elsewhere on the screen . To release the brakes , Jagadisan and Gandhi used a puff of air to make the monkeys blink reflexively . Reflex blinks turn off the inhibition , which in the case of eye movements originates in a structure called the brainstem . This in turn enabled the monkeys to move their eyes while the neural activity was still ramping up . Further analysis showed that preparatory activity in another region of the brain that sends signals to the brainstem – the superior colliculus – predicted the speed of the resulting eye movement . Together these results show that the neural activity involved in planning movements also has the potential to generate movement when released from inhibition . Understanding how the brain starts to produce a movement will allow scientists to probe why this process sometimes goes awry , for example during impulsive movements in ADHD and schizophrenia . It should also help decode the patterns of activity that the brain uses to represent movements before those movements occur . This could lead to improvements in technologies that enable patients to use brain activity to control artificial limbs .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Removal of inhibition uncovers latent movement potential during preparation
The sensation of breathlessness is the most threatening symptom of respiratory disease . The different subdivisions of the midbrain periaqueductal gray ( PAG ) are intricately ( and differentially ) involved in integrating behavioural responses to threat in animals , while the PAG has previously only been considered as a single entity in human research . Here we investigate how these individual PAG columns are differently involved with respiratory threat . Eighteen healthy subjects were conditioned to associate shapes with certain or uncertain impending respiratory load , and scanned the following day during anticipation and application of inspiratory loading using 7 T functional MRI . We showed activity in the ventrolateral PAG ( vlPAG ) during anticipation of resistive loading , with activity in the lateral PAG ( lPAG ) during resistive loading , revealing spatially and temporally distinct functions within this structure . We propose that lPAG is involved with sensorimotor responses to breathlessness , while the vlPAG operates within the threat perception network for impending breathlessness . Continued respiratory function is crucial for sustaining life , and perceived threat to respiration can induce an integrated stress reaction and crippling anxiety . A potentially pivotal nucleus within the breathlessness perception pathway is the midbrain periaqueductal gray ( PAG ) . The PAG has been implicated in many basic survival behaviours including cardiovascular , motor and pain responses ( De Oca et al . , 1998; Mobbs et al . , 2007; Pereira et al . , 2010; Tracey et al . , 2002; Benarroch , 2012; Paterson , 2014 ) , and is well situated to play a role in the integrative response to breathlessness , with major cortical inputs from areas involved with emotional regulation such as medial prefrontal , insula , anterior cingulate cortices and amygdala ( Beitz , 1982; Gabbott et al . , 2005; Rizvi et al . , 1991 ) , and descending connections to the respiratory nuclei of the medulla ( Huang et al . , 2000; Sessle et al . , 1981; Hayward , 2004 ) . These medullary nuclei projections include those to the ventrolateral medulla for the switch between inspiration and expiration ( Subramanian et al . , 2008; Subramanian , 2013 ) , and the nucleus retroambiguus for pharyngeal , laryngeal , thoracic and abdominal pressure control ( Holstege , 1989 ) . The PAG is subdivided into four columns either side of the aqueduct ( ventrolateral ( vlPAG ) , lateral ( lPAG ) , dorsolateral ( dlPAG ) and dorsomedial ( dmPAG ) ) , each with distinct functions for the control of respiration ( Subramanian et al . , 2008; Subramanian , 2013 ) . These columns are proposed to act within active coping strategies for escapable threat ( lPAG and dlPAG ) associated with fight and flight responses , or passive coping strategies to inescapable threat ( vlPAG ) often associated with freezing behaviours ( Keay and Bandler , 2001; Bandler and Shipley , 1994; Bandler et al . , 2000 ) . However , the roles of the different columns in human respiratory threat perception are yet to be investigated . Clinical populations such as those with chronic obstructive lung disease ( COPD ) , asthma , heart failure , cancer and panic disorder suffer from debilitating breathlessness , that contributes to a downward spiral of reduced physical activity , physical deconditioning and worsening breathlessness ( Hayen et al . , 2013a ) . Therefore , a better understanding of its neural basis has the potential to lead to new treatments with wide ranging impact ( Herigstad et al . , 2011 ) . Importantly , conditioned anticipation of environmental cues associated with breathlessness is integral to its threat detection and designated response . Anticipation of threatening sensations relies on cues from the environment . Conditioning is a process of learning an association between two unrelated stimuli , such that a previously neutral stimulus ( conditioned stimulus , CS ) may evoke anxiety due to learned associations with an aversive stimulus ( unconditioned stimulus , US ) ( Pavlov et al . , 2003 ) . Descending modulatory systems during anticipation of a stimulus have even been shown to modulate the response to the stimulus itself , such as those demonstrated with pain ( Porro et al . , 2002; Price et al . , 1999; Wager et al . , 2004 ) . Therefore , each individual PAG column may play differential and important roles in both the anticipation and response to a threatening respiratory stimulus , and thus are potentially pivotal in our understanding and treatment of the neural basis of breathlessness . Interestingly , PAG activity has been identified in a recent paper investigating brain responses to breathlessness-related word cues in patients with COPD , although without sufficient resolution to differentiate activity within specific columns ( Herigstad et al . , 2015 ) . Our understanding of anticipation of conditioned threat has been substantially enhanced by modern neuroimaging techniques , and despite differences in conditioning paradigms , a consistent network of brain areas has been identified , including the amygdala , insula , and anterior cingulate cortex ( Sehlmeyer et al . , 2009 ) . However , despite the proposed integral role of the PAG in threat perception , the ability to scrutinise contributions of smaller nuclei is often limited in neuroimaging by resolution and statistical power , and thus key structures such as the PAG have not yet been investigated in humans . The aim of this study was to investigate the roles of the individual PAG columns during both the perception of breathlessness and its anticipation . We used an aversive delay-conditioning paradigm to associate neutral shapes with upcoming resistive loaded breathing . To investigate if uncertainty of an aversive breathing stimulus altered the threat response ( Rhudy and Meagher , 2000; Ploghaus et al . , 2003 ) , we used three separate anticipation cues with a CS-US contingency pairing of 100% , 50% and 0% . In accordance with animal models , we hypothesised that the vlPAG would be active during anticipation of resistance , as threat is detected and passive coping strategies are employed to manage an upcoming inescapable stressor . Conversely , we hypothesised activity in the lPAG during inspiratory resistance with slowed , deep breathing , corresponding with results in animals ( Holstege , 1989; Keay and Bandler , 2001 ) and our previous work in breath holds ( Faull et al . , 2015 ) . Mean anxiety and intensity scores for conditioned responses to the respiratory tasks are given in Table 1 . Anxiety scores were significantly higher for the certain anticipation cue compared to the uncertain cue , and subsequent resistance was rated at a greater intensity following the certain cue . 10 . 7554/eLife . 12047 . 003Table 1 . Mean ( ± SD ) anxiety and intensity ratings to the conditioned respiratory tasks . DOI: http://dx . doi . org/10 . 7554/eLife . 12047 . 003No impending resistanceUncertain impending resistanceCertain impending resistanceAnxiety ( % ) 4 . 3 ( 5 . 1 ) 36 . 7 ( 22 . 3 ) 48 ( 26 . 7 ) Intensity ( % ) 4 . 7 ( 3 . 1 ) 55 . 5 ( 20 . 9 ) 62 . 9 ( 21 . 5 ) *Significantly ( p<0 . 05 ) different from ‘no impending resistance’ condition;Significantly ( p<0 . 05 ) different from ‘no impending resistance’ and ‘uncertain impending resistance’ . Group average heart rate ( ± SD ) during the brainstem BOLD scanning was 68 ( ± 9 ) beats per minute . Ventilatory variables during each of the respiratory conditions are given in Table 2 . Certain anticipation of resistance was associated with a greater decrease in PETCO2 and increase in PETO2 and respiratory volume per unit of time ( RVT ) than uncertain , indicating preparatory increases in respiration with more effective conditioning . 10 . 7554/eLife . 12047 . 004Table 2 . Mean ( ± SD ) physiological variables across conditioned respiratory tasks . DOI: http://dx . doi . org/10 . 7554/eLife . 12047 . 004AnticipationResistanceNo impending resistanceUncertain impending resistanceCertain impending resistanceAveragePeakPressure ( cmH2O ) -0 . 14 ( 0 . 11 ) -0 . 17 ( 0 . 12 ) -0 . 18 ( 0 . 24 ) -5 . 80 ( 3 . 64 ) -14 . 67 ( 8 . 28 ) PETCO2 ( % ) 4 . 41 ( 0 . 71 ) 4 . 41 ( 0 . 67 ) 4 . 32 ( 0 . 68 ) 4 . 46 ( 0 . 67 ) 4 . 62 ( 0 . 66 ) PETO2 ( % ) 18 . 1 ( 1 . 0 ) 18 . 1 ( 1 . 0 ) 18 . 3 ( 1 . 1 ) 18 . 5 ( 1 . 0 ) 18 . 9 ( 1 . 0 ) Respiratory rate ( min-1 ) 12 . 8 ( 3 . 7 ) 12 . 5 ( 3 . 8 ) 12 . 4 ( 3 . 6 ) 11 . 2 ( 4 . 6 ) 13 . 8 ( 5 . 9 ) RVT increase ( % ) -4 . 4 ( 7 . 4 ) 7 . 8 ( 19 . 6 ) 11 . 0 ( 23 . 0 ) -16 . 1 ( 21 . 6 ) 16 . 6 ( 28 . 5 ) **Significantly ( p<0 . 05 ) different from ‘no impending resistance’ condition . Abbreviations: Pressure , average mouth pressure across all ventilatory cycles; PETCO2 , pressure of end-tidal carbon dioxide; PETO2 , pressure of end-tidal oxygen; RVT , respiratory volume per unit time . The results of the targeted PAG subdivision analyses ( in which certain anticipation of resistive loading was contrasted with anticipation of no loading ) revealed significant increased BOLD activity in the vlPAG , and decreased BOLD in the lPAG during inspiratory resistance ( Figure 1 ) . A further analysis of the whole PAG showed that these activations were isolated to the vlPAG and lPAG in these conditions , although certain anticipation of resistance was now analysed against baseline for adequate statistical power ( Figure 2 ) . Furthermore , activity in the lPAG during certain anticipation of resistance was found to scale with intensity ratings across subjects ( Figure 3 ) . No areas of the PAG or cortex significantly scaled with intensity or anxiety ratings during inspiratory loading , possibly due to insufficient statistical power necessary to observe these scaled activations across subjects during the noisy stimulus of inspiratory loading . When comparing uncertain and certain anticipation of breathlessness , no significant difference was found in the PAG between the two conditions , possibly due to insufficient statistical power to detect a difference . However , during uncertain anticipation of resistance , subthreshold PAG activity ( p=0 . 11 ) was identified in the same area of the right vlPAG as the significant cluster found with certain anticipation of resistance ( Figure 4 ) . Activity in neither the vlPAG , nor the lPAG scaled with anxiety across subjects . 10 . 7554/eLife . 12047 . 005Figure 1 . Targeted PAG columnar analysis . Left: Schematic representation of the columns of the midbrain periaqueductal gray ( PAG ) , which almost surrounds the aqueduct . Middle: Ventrolateral PAG ( vlPAG ) activation during anticipation of resistance contrasted with anticipation of no resistance . Right: Lateral PAG ( lPAG ) deactivation during inspiratory resistance . Statistics are small-volume-corrected for multiple comparisons using highlighted PAG column masks , adapted from Ezra et al . ( 2015 ) , and the images consist of a colour-rendered statistical map superimposed on a standard ( MNI 1 mm3 ) brain . Line drawing originally published in Ezra et al . , 2015 . DOI: http://dx . doi . org/10 . 7554/eLife . 12047 . 00510 . 7554/eLife . 12047 . 006Figure 2 . Periaqueductal gray ( PAG ) analysis . Left: 3D representation of the PAG activations on the right . Top row sagittal view , bottom row axial view of activation in the vlPAG during anticipation of certain resistance ( against baseline: p=0 . 021 ) and deactivation during inspiratory resistance in bilateral lPAG ( p=0 . 007 ) . The key on the right shows location of PAG mask and orientation of displayed slices . Statistics are small-volume-corrected for multiple comparisons using highlighted PAG mask , and the images consist of a colour-rendered statistical map superimposed on a standard ( MNI 1 mm3 ) brain . DOI: http://dx . doi . org/10 . 7554/eLife . 12047 . 00610 . 7554/eLife . 12047 . 007Figure 3 . Scaled BOLD activity during 100% certain anticipation with intensity and anxiety . Right: Positive correlation in the lPAG with intensity ratings ( green uncorrected Z score , red/yellow TFCE-corrected for lPAG activity , PAG displayed in light grey ) but not anxiety . Top: Cortical correlations with average intensity score . Bottom: Cortical correlations with anxiety score for certain anticipation . Images consist of a colour-rendered statistical map superimposed on a standard ( MNI 1 mm3 ) brain . DOI: http://dx . doi . org/10 . 7554/eLife . 12047 . 00710 . 7554/eLife . 12047 . 008Figure 4 . vlPAG activation with 100% certainty of resistance . vlPAG activations during uncertain ( A ) and certain ( B ) anticipation of impending breathlessness . Uncertain anticipation produces subthreshold vlPAG activation in a consistent area to the certain condition . PAG mask displayed by light grey region . Images consist of a colour-rendered statistical map superimposed on a standard ( MNI 1 mm3 ) brain . Orientations marked on the image . DOI: http://dx . doi . org/10 . 7554/eLife . 12047 . 008 All respiratory tasks: We observed significant BOLD signal increases bilaterally in the motor cortex , supplementary motor cortex , primary sensory cortex , middle and posterior cingulate cortices , operculum , medulla and middle insular cortex , and decreased BOLD signal in the bilateral hippocampus and IX cerebellar lobe , for both certain and uncertain anticipation against baseline , and during inspiratory resistance ( Figure 5 ) . 10 . 7554/eLife . 12047 . 009Figure 5 . Cortical activity with functional tasks . Mean cortical activations and deactivations identified during inspiratory resistance , 100% certain anticipation , 50% uncertain anticipation and finger opposition . The images consist of a colour-rendered statistical map superimposed on a standard ( MNI 1 mm3 ) brain . The bright grey region represents the coverage of the coronal-oblique functional scan . Significant regions are displayed with a threshold Z>2 . 3 , with a cluster probability threshold of p<0 . 05 ( corrected for multiple comparisons ) . Abbreviations: VPL , ventral posterior lateral nucleus ( thalamus ) ; M1 , primary motor cortex; S1 , primary sensory cortex; CN , caudate nucleus; Put , putamen; Hipp , hippocampus; STN , subthalamic nucleus; PCC , posterior cingulate cortex; MCC , middle cingulate cortex; p-In , posterior insular; m-In , middle insular; OP , operculum; SMC , supplementary motor cortex; PCG , paracingulate gyrus; PN , posterior nuclei of the thalamus; PAG , periaqueductal gray; M , solitary nucleus of the medulla; Cu , cuneate nucleus ( medulla ) ; I-IV , I-IV cerebellar lobe; IX , IX cerebellar lobe . Source files providing peak voxel locations are provided ( Figure 5—source data 1–3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12047 . 00910 . 7554/eLife . 12047 . 010Figure 5—source data 1 . Co-ordinates of local maxima of significant increases ( activations ) and decreases ( deactivations ) in the BOLD response to inspiratory loading . Values derived from cluster-based analysis . The most significant maximum is listed for each anatomical location . Co-ordinates are in mm in standard space of MNI ( 1 mm3 ) . x , distance right ( + ) or left ( - ) of the mid saggital line; y , distance anterior ( + ) or posterior ( - ) from a vertical plane through the anterior commissure; z , distance above ( + ) or below ( - ) the intercommisurial plane . Abbreviations: VPL , ventroposterolateral nucleus of the thalamus . DOI: http://dx . doi . org/10 . 7554/eLife . 12047 . 01010 . 7554/eLife . 12047 . 011Figure 5—source data 2 . Co-ordinates of local maxima of significant increases ( activations ) and decreases ( deactivations ) in the BOLD response during certain and uncertain anticipation of inspiratory loading . Values derived from cluster-based analysis . The most significant maximum is listed for each anatomical location . Co-ordinates are in mm in standard space of MNI ( 1 mm3 ) . x , distance right ( + ) or left ( - ) of the mid saggital line; y , distance anterior ( + ) or posterior ( - ) from a vertical plane through the anterior commissure; z , distance above ( + ) or below ( - ) the intercommisurial plane . DOI: http://dx . doi . org/10 . 7554/eLife . 12047 . 01110 . 7554/eLife . 12047 . 012Figure 5—source data 3 . Co-ordinates of local maxima of significant increases ( activations ) and decreases ( deactivations ) in the BOLD response to a finger opposition task . Values derived from cluster-based analysis . The most significant maximum is listed for each anatomical location . Co-ordinates are in mm in standard space of MNI ( 1 mm3 ) . x , distance right ( + ) or left ( - ) of the mid saggital line; y , distance anterior ( + ) or posterior ( - ) from a vertical plane through the anterior commissure; z , distance above ( + ) or below ( - ) the intercommisurial plane . Abbreviations: VPL , ventroposterolateral nucleus of the thalamus . DOI: http://dx . doi . org/10 . 7554/eLife . 12047 . 012 Finger opposition resulted in consistant significant signal increases in both the brainstem and motor cortex with previous research ( Faull et al . , 2015; Lee et al . , 1999; Pattinson et al . , 2009 ) including bilateral activation in the motor cortex ( more extensive activation in the contralateral left motor cortex ) , supplementary motor cortex , middle cingulate and paracingulate cortices , primary sensory cortex , anterior insula cortex , operculum , caudate nucleus and putamen ( Figure 5 ) . Bilateral signal increases were seen in the thalamic VPL nuclei , as well as the left thalamic VPM nucleus . In addition , activations were observed in the left subthalamic and red nuclei , right ( ipsilateral ) cuneate nucleus of the medulla ( Figure 7 ) , and bilateral cerebellum ( VI and VIIIa lobules ) . 10 . 7554/eLife . 12047 . 014Figure 7 . Finger opposition functional localiser . Demonstration of the use of finger opposition as a functional localiser in brainstem FMRI in the current study compared to previous results , displaying hypothesised activation in the ipsilateral cuneate nucleus of the medulla ( z -54 ) . The 7 T 1 mm3 voxel data is derived from previously-published results ( Faull et al . , 2015 ) ( 14 repeats of 15 sec finger opposition , 1 mm3 voxels and TR=5 s ) , while the 1 . 5 mm3 voxel data is from the current study ( 10 repeats of 15 sec finger opposition , 1 . 5 mm3 voxels and TR=3 . 11 s ) . This technique provides confidence in the analysis model and registration accuracy of the current 7 T study . The images consist of a colour-rendered statistical map superimposed on a standard ( MNI 1 mm3 ) brain . Significant regions are displayed with a threshold Z>2 . 3 , with a cluster probability threshold of p<0 . 05 ( corrected for multiple comparisons ) . The sagittal image on the right displays the position of slices , for clarity only displayed from the 7 T 1 mm3 acquisition . Abbreviations: R , raphe nuclei; ret , nuclei reticularis; VII , facial nucleus; Amb , nucleus ambiguous; IX , glossopharyngeal nucleus; NTS , nucleus tractus solitaries; GC , gracile ( medial ) and cuneate ( lateral ) nuclei ( in blue ) . R ( right ) and L ( left ) indicate image orientation . Original line drawings adapted from Duvernoy , 1995 . DOI: http://dx . doi . org/10 . 7554/eLife . 12047 . 014 In this study we identified differential activity in the lateral and ventrolateral columns of the PAG relating to different aspects of the aversive stimulus of resistive inspiratory loading . We observed bilateral decreased BOLD activity in the lPAG during resistive inspiratory loading , and during cued anticipation activity in this area correlated with behavioural ratings of breathlessness intensity . Conversely , positive BOLD activity in the right vlPAG was identified during the cued anticipation of certain impending resistance , while uncertain anticipation activity remained subthreshold . Anxiety ratings , intensity scores and the ventilatory response were lower in the uncertain vs . certain condition , indicating a reduction in the conditioned threat response to a 50% ( uncertain ) predictive cue , compared to the 100% ( certain ) predictive cue . Significantly , recent work using diffusion tractography has revealed consistent columnar structure to animal models within the human PAG ( Ezra et al . , 2015 ) . During response to threat , functional organisation of these animal PAG columns has been hypothesised to consist of active and passive coping strategies ( Bandler et al . , 2000; Hayen et al . , 2013a; Herigstad et al . , 2011 ) . The lPAG and dlPAG are thought to employ active coping strategies for escapable stressors , consistent with the tachypnea observed in animals on stimulation of these columns ( Holstege , 1989 ) , while the vlPAG employs passive coping strategies for inescapable stressors ( vlPAG ) such as that seen with a range of physical stimuli ( Bandler et al . , 2000 ) . In the current investigation of the threat response to breathlessness , the aversive resisted breathing stimulus was an upcoming inescapable stressor , activating the vlPAG , while during the active stimulus response we observed lPAG activity . These results are the first in humans to adhere to the current models of distinctive threat perception of the animal PAG columns , although the cytoarchitecture and autonomic functions produced within the rostro-caudal axes of these columns is in humans not yet known . One recent study by Satpute and colleagues used 7 Ttesla functional MRI to identify highly localised activity in areas of the PAG along the rostro-caudal axis during exposure to aversive images in humans [Satpute et al . , 2013] however this study was based upon pre-defined divisions within the PAG that neither adhered to its known columnar structure , nor considered the characteristic functions of these columns within threat perception . Therefore , while there is much work to be done to accurately map the cytoarchitecture and functional localisation of autonomic functions within the human PAG columns , we will now discuss each of the activated columns in the current study , as a starting point towards understanding their potential role in the specific threat response to an aversive breathing stimulus as a model of breathlessness . The decrease in BOLD signal in the lPAG during an inspiratory resistance found in this study is consistent with previous PAG findings . Prior work by our group identified decreased BOLD signal in the human lPAG during the respiratory challenge of breath holds ( Faull et al . , 2015 ) , and animal studies have proposed the lPAG may play a role in respiratory behaviours such as prolonged inspirations and expirations ( Holstege , 1989 ) . Thus , it is possible that the lPAG is an integral nucleus within the somatomotor pathways of respiratory control in the active response to threat , and anatomical evidence exists to support this hypothesis ( Ezra et al . , 2015 ) . The lPAG has been reported to receive somatotopically organised spinal sensory afferents ( Bandler et al . , 2000; Craven , 2011 ) , which could provide sensory information from the chest , and it propagates direct efferent connections to the midline medulla ( Cowie and Holstege , 1992 ) for possible descending respiratory motor commands . Diffusion tractography in humans demonstrates preferential connectivity between somatomotor regions , such as between primary sensory and motor cortices and the lPAG , compared to the vlPAG ( Ezra et al . , 2015 ) . Our findings of activity in the lPAG whilst producing elevated inspiratory pressure supports the idea that this column of the PAG is involved in altered respiratory work , although whether this is in a motor or sensory capacity ( or both ) is currently unknown . Interestingly , activity in the lPAG during anticipation was found to scale with perceived stimulus intensity across subjects . Anticipation of a stimulus allows system preparation and response selection , and activity in the lPAG that scales with the perceived intensity of the forthcoming stimulus indicates a possible top-down control during preparation for the threat of inspiratory resistance . The cortical structures that scaled alongside the lPAG with perceived intensity included the premotor cortex and hippocampus , which may indicate increased motor preparatory activity ( Grafton et al . , 1998; Rizzolatti et al . , 2002 ) and greater working memory of the stimulus between the hippocampus and prefrontal cortex ( Laroche et al . , 2000 ) . Conversely , lPAG activity during anticipation did not correlate with anxiety scores . This suggests that lPAG activity is less likely to be involved in the emotional component of resistance anticipation ( Critchley et al . , 2004; Gray et al . , 2007 ) . Future work towards understanding whether the role of the lPAG is causative within this anticipatory breathlessness intensity network may be integral to pinpointing perceptual disruptions in chronic sufferers of breathlessness . The increase in BOLD signal identified in the vlPAG implies a role of the vlPAG in the conditioned response to anticipation of breathlessness . Anticipation of resistance also activated a cortical network of motor , sensory and interoceptive areas , indicating the potential position of the vlPAG within a threat detection and passive preparatory network stimulated by a conditioned breathlessness cue . Additionally , although prefrontal cortical areas were not imaged , diffusion tractography has demonstrated that the vlPAG receives the predominant proportion of the input from the prefrontal cortex ( Ezra et al . , 2015 ) , and animal models report direct connections between the posterior orbital frontal and anterior insula cortices to the vlPAG ( An et al . , 1998 ) . Therefore , it is possible that communication between the vlPAG and areas of executive function , interoception and motor preparation are vital to the threat detection and response selection that occur during the cued anticipation of breathlessness , which will be investigated in future work . While this study has made inroads into functionally differentiating the columns of the PAG at high resolution , further research into the intricacies of these communications is needed to fully understand the role of the vlPAG within this network . Interestingly , there did not appear to be any significant differences ( both within the vlPAG and superior cortical network ) between uncertain and certain anticipation of resistance , but rather subthreshold vlPAG activity with uncertain anticipation . Furthermore , the reduction in vlPAG activity was paralleled by reduced anxiety and intensity scores in uncertain anticipation , indicating a smaller conditioned response to this cue . This supports the idea that the vlPAG is involved within the threat perception network for breathlessness , and the magnitude of this activity reflects greater conditioning and increased anticipatory preparation . Interestingly , it does not appear that the uncertainty induced in this study drives hypersensitivity and resultant increased anxiety or rating scores ( Table 1 ) , differing from previous research in pain ( Rhudy and Meagher , 2000; Ploghaus et al . , 2003 ) . However , previous pain research has often used no anticipation cue in conjunction with an unpredictably intense stimulus , while in this study we used an uncertain prediction of a known stimulus . The current methodology allowed us to manipulate the conditioned response to a cue , permitting investigation into the role of the PAG during reduced perception of threat without changing the intensity of the stimulus . This study also revealed a cortical and subcortical network of structures that co-activated with PAG columnar activity in these conditions . Anticipation and conscious changes in respiration involve both sensorimotor and affective processing , as adequate ventilation is integral to sustaining life and thus closely monitored by homeostatic mechanisms ( Brannan et al . , 2001; Dempsey et al . , 1985 ) . Within the limited field of view of this study , the cortical network associated with breathing against an inspiratory load covered a network of primary motor and sensory structures , and the subcortical basal ganglia and insula ( Figure 5 ) as well as the lPAG , consistent with previous research using breath holds ( Faull et al . , 2015; McKay et al . , 2008; Pattinson et al . , 2009 ) and hypercapnia-stimulated hyperventilation using PET ( Brannan et al . , 2001 ) . Conversely , during anticipation , vlPAG activity was paired with less extensive activation of cortical primary motor and sensory structures compared with inspiratory loading , while activity was maintained in preparatory motor structures such as the supplementary motor cortex and basal ganglia ( Groenewegen , 2003; Mink , 1996; Alexander et al . , 1986 ) . While further research is required to investigate the role of prefrontal brain activity that will be simultaneously occurring within this respiratory threat network during anticipation and inspiratory loading , what will be of great interest is how these distinctly different PAG columnar activations are functionally interacting within this extensive cortical network to influence the perception of respiratory threat during these two conditions . It is common practice within the learning literature to contrast the conditioned cue that is paired with the stimulus with a cue that is unpaired with the stimulus ( Büchel et al . , 1998; Gottfried et al . , 2002; Gottfried and Dolan , 2004; LaBar et al . , 1998 ) , which in this case would be the contrast of certain anticipation of resistance with the anticipation of no resistance , respectively . However , this contrast is not feasible beyond targeted PAG column analysis in the current study , as the length of the inspiratory resistance stimulus required to amass statistical power limits the number of possible repeats of each condition . Therefore , beyond the targeted analysis of the vlPAG during certain anticipation of resistance greater than anticipation of no resistance , the anticipation conditions have been analysed against baseline . However , the inclusion of three anticipatory cue conditions does allow greater decorrelation of the general ‘cue response’ to each anticipation condition in the model . Further studies in this area may look to include more subjects , or fewer anticipation conditions to allow more repeats , enabling contrasts of anticipation of loading against anticipation of no loading in the whole PAG and wider cortex . The finger opposition task was used as both a control motor task and a methodological validation . Consistent with previous research ( Faull et al . , 2015; Pattinson et al . , 2009 ) , we hypothesised to see a localised increase in BOLD signal in the ipsilateral cuneate nucleus of the medulla , which is a sensory nucleus in the fine touch and proprioception pathway prior to decussation ( Craven , 2011 ) . This activation demonstrates the accuracy of registration required to align activations within small brainstem nuclei for group analysis . Brainstem fMRI is particularly susceptible to low signal to noise when compared to cortical areas . Physiological noise can present a significant problem , due to bulk susceptibility changes with the respiratory cycle , pulsatile movement with the cardiac cycle , and proximity to fluid-filled spaces ( Brookes , et al . , 2013; Harvey et al . , 2008; Hayen et al . , 2013b ) . Special care was taken in this study to address these issues , with ICA denoising used for movement and scanner artefact , and physiological noise modelling and RETROICOR used for slice-wise removal of cardiac and respiratory noise . The results of this study suggest that the individual columns of the PAG may be differentially involved in the perception of breathlessness . This study corroborates with recent findings that the lPAG may be involved with the sensorimotor aspect of breathing control during the active response to breathlessness , and top-down anticipatory activity may influence intensity perception of breathlessness . Conversely , the vlPAG appears to be only activated during anticipation of breathlessness , consistent with freezing behaviours reported in animals , with decreased anticipatory cue conditioning resulting in reduced vlPAG activity . We propose that the vlPAG is involved with the learned anticipatory threat detection of a breathlessness stimulus , corroborating with the proposed model of the vlPAG in the passive threat response to an inescapable stressor . In this study we have discriminated differential functional activity within the columns of the PAG in response to threat for the first time in humans , demonstrating the key differential roles of individual columns within the perception of breathlessness . The Oxfordshire Clinical Research Ethics Committee approved the study and volunteers gave written , informed consent . Eighteen healthy , right-handed volunteers ( 12 males , 6 females; mean age ± SD , 28 ± 4 years ) undertook one training session , followed by one MRI scanning session within 24 hr . Prior to scanning , all subjects were screened for any contraindications to magnetic resonance imaging at 7 Tesla . A breathing system was constructed to remotely administer periods of inspiratory resistance during scanning ( Figure 8 ) . During rest periods , compressed medical air was delivered to the breathing system and gas flow was maintained at a rate that was adequate to allow free breathing , sufficient that the reservoir bag never collapsed on inspiration . During inspiratory resistance , delivery of compressed air was stopped , and once the reservoir bag collapsed , inspiration was through the resistance arm of the circuit inhaling atmospheric air ( see Figure 8 for details ) . 10 . 7554/eLife . 12047 . 015Figure 8 . Breathing system . Schematic diagram of breathing system that allows remote administrations of inspiratory resistance . Medical air is supplied to the subject , with a reservoir of 2 L . Excess flow and expiration escapes through the one-way expiratory valve , close to the mouth to minimise rebreathing ( inspiratory and expiratory valves: Hans Rudolf , Kansas City , MO , USA ) . Resistive loading is induced by discontinuing the delivery of medical air , forcing the subject to draw air through the resistor ( porous glass disc ) . A diving mouthpiece ( Scubapro UK Ltd , Mitcham , UK ) connects to a bacterial and viral filter ( GVS , Lancashire , UK ) , sampling lines ( Vygon SA , Ecouen , France ) , connect to a pressure transducer ( MP 45 , ± 50 cmH2O , Validyne Corp . , Northridge , CA , USA ) and amplifier ( Pressure transducer indicator , PK Morgan Ltd , Kent , UK ) for inspiratory pressure readings , and to a gas analyser ( Gas Analyser; ADInstruments Ltd , Oxford , United Kingdom ) for respiratory gases . A mildly hyperoxic state was achieved through a constant administration of oxygen at a rate of 0 . 5 L/min . Periodically throughout scanning carbon dioxide challenges were administered to raise PETCO2 to match the PETCO2 rise during inspiratory loading periods . DOI: http://dx . doi . org/10 . 7554/eLife . 12047 . 015 To minimise the effect of changing arterial oxygen ( O2 ) and carbon dioxide ( CO2 ) levels upon the BOLD signal , the following steps were employed: additional medical oxygen was delivered , and the flow rate was manually adjusted to minimise fluctuations in pressure of end-tidal oxygen ( PETO2 ) , aiming to keep PETO2 at 18 kPa ( very slightly above normal ) . At designated time points during rest periods of the functional scan , CO2 challenges were administered by switching the flow of compressed air for a 10% CO2 mixture ( 10% CO2; 21% O2; balance nitrogen ) at 20 L/min for periods of 5–10 s , aiming to raise PETCO2 an equivalent amount as observed during the inspiratory loading periods . The subject’s nose was blocked using foam earplugs and they were asked to breathe through their mouth for the duration of the experiment . The experimental protocol was completed on two occasions; during the conditioning session and repeated in the scanner the following day . The purpose of the conditioning session was for subjects to learn to associate a different symbol ( star , triangle or square; randomised order ) to three breathing conditions , and the conditioned response to these symbols was then investigated by repeating the protocol with fMRI . The breathing conditions were as follows: The certain or uncertain resistance symbol was presented on the screen for 30 sec , which included a 5–15 s anticipation period before the resistance was applied ( where applicable ) . The no resistance symbol was presented for 20 s , and each condition was repeated 10 times in a semi-randomised order ( Figure 9 ) . A finger opposition task was also included in the protocol , as a brainstem functional localiser for confidence in image registration and analysis techniques ( Faull et al . , 2015; Pattinson et al . , 2009 ) , where an opposition movement was conducted with the right hand , with the cue ‘TAP’ presented for 15 s ( 10 repeats ) . 10 . 7554/eLife . 12047 . 016Figure 9 . Experimental protocol . Study overview ( top ) and example four minutes of the experimental protocol ( bottom ) , repeated throughout the conditioning and fMRI scanning sessions . Anticipation periods were 5–15 s duration , and resistance periods 15–25 s , and there were 10 repeats of each stimulus . DOI: http://dx . doi . org/10 . 7554/eLife . 12047 . 016 Rating scores of breathing difficulty were recorded after every symbol and at the beginning and end of the task , using a visual-analogue scale ( VAS ) with a sliding bar that the subjects moved between ‘Not at all difficult’ ( 0% ) and ‘Extremely difficult’ ( 100% ) . Subjects were also asked to rate how anxious each of the symbols made them feel using a VAS between ‘Not at all anxious’ ( 0% ) and ‘Extremely anxious’ ( 100% ) immediately following the experimental protocol . Using MRI to investigate respiratory control presents methodological challenges that require consideration , particularly in the brainstem ( Brookes , et al . , 2013 ) . We used previously-established methods to decorrelate the effects of hypercapnia from the localised BOLD responses associated with breathing against an inspiratory resistance , using additional , repeated CO2 challenges interspersed during rest periods in the fMRI protocols ( Faull et al . , 2015; Pattinson et al . , 2009 ) . Additionally , chest movements were measured using respiratory bellows surrounding the chest at the approximate level of the 10th rib , and heart rate was measured using a pulse oximeter ( 9500 Multigas Monitor , MR Equipment Corp . , NY , USA ) . PETCO2 and PETO2 were sampled via a port beside the mouth piece of the breathing system . Expired gases were determined using a rapidly-responding gas analyser ( Gas Analyzer; ADInstruments Ltd , Oxford , United Kingdom ) , and pressure at the mouth was measured using a pressure transducer ( MP 45 , ± 50 cmH2O , Validyne Corp . , Northridge , CA , USA ) connected to an amplifier ( Pressure transducer indicator , PK Morgan Ltd , Kent , UK ) . All physiological measurement devices were connected to a data acquisition device ( Powerlab; ADInstruments Ltd , Oxford , United Kingdom ) coupled to a desktop computer with recording software ( Labchart 7; ADInstruments Ltd , Oxford , United Kingdom ) . MRI was performed with a 7T Siemens Magnetom scanner , with 70 mT/m gradient strength and a 32 channel Rx , single channel birdcage Tx head coil ( Nova Medical ) . The fMRI experimental design is illustrated in Figure 9 .	Many people find feeling breathless one of the most upsetting symptoms of respiratory diseases , and breathlessness often causes anxiety that makes the condition seem more threatening than it is . Studies in animals suggest that a small cluster of neurons called the periaqueductal gray is important for responding to threats . This cluster , located at the top of the brainstem , is divided into parallel columns running from top to bottom . In animals , these columns are known to have distinct roles , but human research has tended to consider the periaqueductal gray as a single , uniform entity . Faull et al . wanted to find out whether different columns of the human periaqueductal gray have distinct roles in the perception of respiratory threat . During the study , participants breathed through a tube while watching shapes appear on a screen . This tube could be altered to make breathing more or less difficult – much like breathing through a narrow drinking straw . A conditioning session was first conducted so that participants learned that certain shapes on the screen signalled that their breathing was about to become difficult , while other shapes signalled normal breathing . A second session was then conducted in a brain scanner , using a technique called functional magnetic resonance imaging . This allowed Faull et al . to compare brain activity during the anticipation of difficult breathing with the brain activity during the breathing challenge itself . The results show that the column at the front of the periaqueductal gray ( the ventrolateral column ) was more active when participants saw the shape that signaled upcoming breathing difficulty . In contrast , difficult breathing was associated with activity in the lateral column ( at the side of the periaqueductal gray ) . Thus , the different columns of the human periaqueductal gray have different roles in the response to respiratory threat . Future studies could investigate how these columns interact with each other and with other brain regions . Such understanding is important for a range of conditions that may be influenced by the activity of the periaqueductal gray , including disruptions in bladder control , hypertension , chronic pain , and asthma .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Conditioned respiratory threat in the subdivisions of the human periaqueductal gray
Site-specific histone ubiquitylation plays a central role in orchestrating the response to DNA double-strand breaks ( DSBs ) . DSBs elicit a cascade of events controlled by the ubiquitin ligase RNF168 , which promotes the accumulation of repair factors such as 53BP1 and BRCA1 on the chromatin flanking the break site . RNF168 also promotes its own accumulation , and that of its paralog RNF169 , but how they recognize ubiquitylated chromatin is unknown . Using methyl-TROSY solution NMR spectroscopy and molecular dynamics simulations , we present an atomic resolution model of human RNF169 binding to a ubiquitylated nucleosome , and validate it by electron cryomicroscopy . We establish that RNF169 binds to ubiquitylated H2A-Lys13/Lys15 in a manner that involves its canonical ubiquitin-binding helix and a pair of arginine-rich motifs that interact with the nucleosome acidic patch . This three-pronged interaction mechanism is distinct from that by which 53BP1 binds to ubiquitylated H2A-Lys15 highlighting the diversity in site-specific recognition of ubiquitylated nucleosomes . Protein ubiquitylation is a versatile signal that controls a wide range of cellular functions including protein degradation , transcription , immunity , inflammation , endocytosis and DNA repair ( Komander and Rape , 2012; Jackson and Durocher , 2013 ) . The covalent attachment of ubiquitin ( ub ) , a 76 residue polypeptide , to a target protein is achieved by the sequential activity of ubiquitin-activating ( E1 ) , ubiquitin-conjugating ( E2 ) and ubiquitin-ligating ( E3 ) enzymes resulting in the formation of an isopeptide bond between the terminal carboxyl group of ubiquitin and an acceptor protein . Conjugated ubiquitin can be targeted for additional ubiquitylation cycles , producing polyubiquitin chains of various lengths and topologies ( Komander and Rape , 2012 ) . Translating the resulting ubiquitin signal into a specific cellular response typically involves interactions between ubiquitin and proteins containing ubiquitin-binding domains ( UBDs ) ( Dikic et al . , 2009 ) . UBDs , which are small ( 20–150 amino acids ) structurally diverse protein modules that interact with ubiquitin in a non-covalent manner , have been identified in more than 150 cellular proteins ( Hicke et al . , 2005 ) . The specificity of UBD-ubiquitin interactions is achieved by a variety of mechanisms that can involve distinct affinities for certain linkages and lengths of ubiquitin chains , avidity through combination of UBDs , and contributions by UBD-independent sequences within a ubiquitin binding protein ( Dikic et al . , 2009 ) . This latter mechanism is especially prevalent during the response to DSBs where some proteins harbor one such ‘specificity’ sequence adjacent to the UBD , termed an LR-motif or LRM ( Panier et al . , 2012 ) . The DSB response is a useful model to study ubiquitin-based signaling . In this process , DSBs elicit a complex chromatin modification cascade , illustrated in Figure 1A , which ultimately controls DNA repair ( Panier and Durocher , 2013 ) . The cascade begins with the phosphorylation of the C-terminal tail of histone H2AX , by the phosphatidylinositol 3-kinase-related kinase ATM and related kinases . Phosphorylated H2AX ( known as γ-H2AX ) is ‘read’ by the BRCT domains of the MDC1 , which is itself a target of ATM . The ubiquitin ligase RNF8 then binds ATM-phosphorylated MDC1 where , together with the E2-conjugating enzyme UBC13 , it catalyzes K63-linked polyubiquitin chains on linker histone H1 . These chains on H1 are recognized by the ubiquitin-dependent recruitment module 1 ( UDM1 ) of RNF168 ( Figure 1B ) , a second E3 ligase that plays a key function in DSB repair ( Thorslund et al . , 2015 ) . The RNF168-UDM1 module is composed of two ubiquitin binding domains , MIU1 ( motif interacting with ubiquitin ) and UMI ( MIU- and UIM-related ubiquitin binding motif ) along with the LRM1 specificity module . RNF168 catalyzes mono-ubiquitylation of H2A/H2AX on Lys13 or Lys15 yielding H2AK13ub and H2AK15ub , respectively ( Gatti et al . , 2012; Mattiroli et al . , 2012 ) . The action of RNF168 on chromatin results in the subsequent recruitment of additional DSB repair factors including 53BP1 , RAP80 , RAD18 and the RNF168 paralog RNF169 . RNF168 also promotes its own recruitment to DSB sites via a second UDM module , UDM2 , consisting of the MIU2 UBD and the LRM2 specificity motif ( Figure 1B ) . In contrast , RNF169 only contains a functional UDM2 module ( Figure 1B ) , explaining the strict dependence on RNF168 for its recruitment to DSB sites ( Panier et al . , 2012; Chen et al . , 2012; Poulsen et al . , 2012 ) . 10 . 7554/eLife . 23872 . 003Figure 1 . RNF168 and RNF169 bind RNF168-ubiquitylated NCPs . ( A ) Schematic of RNF8-mediated DNA DSB repair pathway . ATM: Ataxia telangiectasia mutated , MDC1: mediator of DNA damage checkpoint 1 , BRCT: breast cancer 1 C-terminal , H1: linker histone H1 , RNF: ring finger proteins , 53BP1: p53 binding protein 1 , Ub: ubiquitin , P: phosphate group , Me: methyl group . ( B ) Domain architecture of RNF168 ( 1-571 ) and RNF169 ( 1-708 ) . Domains and motifs are indicated . R: RING domain , MIU: motif interacting with ubiquitin , UIM: ubiquitin-interacting motif , UMI: UIM- , MIU-related ubiquitin binding motif , LRM: LR motif . ( C ) MBP pull-down assays of RNF168-ubiquitylated nucleosome core particles ( H2AK13/K15ub-NCP ) with the indicated MBP fusion proteins ( RNF168-UDM1 ( 110–201 ) , RNF168-UDM2 ( 374–571 ) , RNF169-UDM2 ( 662–708 ) , RAP80 ( 60-124 ) and RAD18 ( 201-240 ) ) . Input: 5% of the amount of ubiquitylated NCPs used in the pull-down . The migration of molecular mass markers ( kDa ) is indicated on the left . ( D ) Pull-down assays of NCPs ubiquitylated with the indicated E3s by either MBP–RNF169 ( UDM2 ) ( left ) or MBP–RNF168 ( UDM2 ) ( right ) . A reaction without E3 ( - ) acts as a negative control . B/R: BMI1/RING1b . ( E ) Structure of the nucleosome ( PDB: 2PYO [Clapier et al . , 2008b] ) . One copy of H2A and H2B is labeled in yellow and in red , respectively . Lysines that are ubiquitylated by RNF168 ( H2A K13/K15 ) and BMI1/RING1B ( H2A K118/K119 ) are indicated in space filling representation . DOI: http://dx . doi . org/10 . 7554/eLife . 23872 . 00310 . 7554/eLife . 23872 . 004Figure 1—figure supplement 1 . RNF169 ( UDM2 ) binds with higher affinity than RNF168 and does not discriminate between H2AK13 and K15 ubiquitylation . ( A ) Pull-down assays of NCP ubiquitylated by RNF168 using either MBP-RNF169 ( UDM2 ) or MBP-RNF168 ( UDM2 ) in the presence of varying concentrations of the acidic patch interacting KSHV LANA peptide . In addition , the 8LRS10 mutant ( mut: L8A R9A and S10A ) that does not bind to the NCP is added in one case ( concentration of the LANA peptide in μM ) . ( B ) Pull-down assay of NCPs conjugated to ubiquitin with MBP-RNF169 ( UDM2 ) . H2A mutants K15R and K13R were used to catalytically produce monoubiquitylated H2AK13-ub and H2AK15-ub , respectively . Catalytic ubiquitylation of K13R and K15R species by RNF168 produces a small amount of diubiquitylated H2A with the second ubiquitin conjugated to position K36 , as indicated on the gel ( Wilson et al . , 2016 ) . Input: 5% of the amount of ubiquitylated NCPs used in pull-down; the migration of molecular mass markers ( kDa ) is indicated on the left . DOI: http://dx . doi . org/10 . 7554/eLife . 23872 . 004 While 53BP1 , RNF168 , RNF169 , RAP80 and RAD18 accumulate on the chromatin surrounding DSB sites in an RNF168- and ubiquitin-dependent manner , how these factors recognize the products of RNF168 action are only beginning to emerge . 53BP1 , while lacking a recognizable UBD , specifically interacts with H2AK15ub , but not with H2AK13ub , in the context of nucleosomes also containing H4 Lys20 dimethylation ( Fradet-Turcotte et al . , 2013 ) . The structural basis for the site-specific recognition of H2AK15ub/H4K20me2 marks by 53BP1 was recently elucidated by electron cryomicroscopy ( cryo-EM ) ( Wilson et al . , 2016 ) . The resulting structure showed that a short 53BP1 peptide segment , the ubiquitin-dependent recruitment ( UDR ) motif , interacts with the nucleosome core particle ( NCP ) surface , sandwiched by a ubiquitin molecule ( Wilson et al . , 2016 ) . Like 53BP1 , both RNF168 and RNF169 interact with ubiquitylated H2A ( Panier et al . , 2012 ) , but it is not known if this interaction is selective for H2AK13/K15ub . Here , we first establish that RNF168 and RNF169 are site-specific readers of H2AK13/K15 ubiquitylation . Then , we make use of an integrative approach to elucidate the molecular basis for the specific recruitment of RNF168/169 to DSBs by generating a structural model of the RNF169 ( UDM2 ) -ubiquitylated nucleosome complex . Central to these efforts have been methyl-TROSY NMR , which enables the study of protein complexes with aggregate molecular masses as large as 1 MDa ( Sprangers and Kay , 2007; Rosenzweig and Kay , 2014; Gelis et al . , 2007 ) , and replica averaged molecular dynamics simulations ( Cavalli et al . , 2013a ) using restraints derived from NMR and site-directed mutagenesis experiments . Our model establishes that the previously identified bipartite interaction module of RNF169 ( Panier et al . , 2012 ) is composed of a classic MIU helix , that docks onto the canonical hydrophobic surface of ubiquitin , while the LRM2 region responsible for specificity is highly disordered . The LRM2 motif binds to the acidic patch region on the nucleosome surface using two basic regions with key arginine residues . We then use cryo-EM to validate this model . These results demonstrate the synergistic behavior of MIU-ubiquitin and LRM-NCP acidic patch interactions in facilitating recruitment of histone ubiquitylation readers RNF168/RNF169 to regions of DSBs and in providing the specificity of the interaction . To identify additional proteins that specifically recognize H2AK13/K15ub in the context of nucleosomes , we prepared MBP fusion proteins with the RNF168 UDM1 and UDM2 modules , the RNF169 UDM2 module and regions of RAD18 and RAP80 that are necessary and sufficient for their ubiquitin-dependent recruitment to DSB sites ( Panier et al . , 2012 ) . We observed that the RNF168 and RNF169 UDM2 modules were the only proteins , among those tested , that bound to NCPs catalytically ubiquitylated with the RNF168 ( 1-113 ) -UBCH5a complex in MBP pull-down assays ( Figure 1C ) . These data suggest that the RNF168 and RNF169 UDM2 modules might represent selective readers of H2A Lys-13/Lys-15 ubiquitylation . H2A is the most abundant ubiquitylated protein in the nucleus with 5–15% of H2A conjugated to ubiquitin at the K119 residue in vertebrate cells ( Cao and Yan , 2012 ) . As the BMI1-RING1b complex can catalyze this mark ( Wang et al . , 2004 ) , we tested whether the RNF168 or RNF169 UDM2 modules interact with H2AK119ub-containing NCPs prepared using the BMI1-RING1b enzyme complex . While both RNF169 ( UDM2 ) and RNF168 ( UDM2 ) were able to pull-down H2A K13/K15ub-NCPs , they were unable to interact with BMI1/RING1b-ubiquitylated NCPs ( Figure 1D ) . Although both NCP substrates provide accessible ubiquitin molecules , since they are attached to unstructured histone tails , the ubiquitin moieties reside on opposite ends of the NCP surface ( Figure 1E ) . Therefore , these results imply that the UDM2 module makes specific contacts with the N-terminal region of H2A or with the surrounding NCP surface , rather than non-specific interactions with DNA , to establish a strong and selective association with ubiquitylated NCPs . Having identified an interaction module required for binding of H2AK13/K15ub-NCPs , involving both MIU and LRM domains , we then used solution NMR spectroscopy to establish the secondary structures of these elements in the unbound state . The primary sequences of RNF169 ( UDM2 ) and RNF168 ( UDM2 ) , encompassing the MIU2 and LRM2 domains , are shown in Figure 2A . Residues 665–682 of RNF169 and 442–459 of RNF168 comprise the conserved primary sequence specific to MIUs , while the 13 residue long LRM2 regions contain the critical upstream arginine ( R689 in RNF169 and R466 in RNF168 ) and the LR pair ( L699/R700 in RNF169 and L476/R477 in RNF168 ) required for the stable interaction with NCPs ubiquitylated by RNF168 ( Panier et al . , 2012 ) . Despite the fact that RNF168 and RNF169 have high sequence similarity within their analogous MIU2-LRM2 modules , we observed a more robust interaction between RNF169 ( UDM2 ) and H2AK13/K15ub-NCPs in MBP pull-down experiments ( Figure 1—figure supplement 1A ) . Consequently , we focused our NMR experiments and subsequent structural models on the RNF169 ( UDM2 ) -H2AK13ub-NCP interaction since we ascertained that the UDM2 module , unlike 53BP1 , can bind to either H2AK13ub or H2AK15ub ( Figure 1—figure supplement 1B ) ( Fradet-Turcotte et al . , 2013 ) . Assignment of RNF169 ( UDM2 ) backbone resonances was carried out using standard triple-resonance solution NMR experiments ( Sattler et al . , 1999 ) at 35°C , with the assigned 1H-15N HSQC spectrum shown in Figure 2B . The secondary structural propensity program SSP was used to determine the type and probability of secondary structural elements based on measured 1Hα , 13Cα , 13Cβ , and 13CO chemical shifts ( Marsh et al . , 2006 ) . The results , shown in Figure 2C , indicate a relatively high propensity for helical secondary structure over much of the MIU2 region ( 665-682 ) and some propensity for helix into the region linking the MIU2 and LRM2 domains ( 683-685 ) . The remaining linker region , LRM2 , and residues C-terminal to the LRM2 appear to lack any secondary structure , aside from a modest increase in helical propensity peaking at residues Y697 and L698 , with an average 28% helical probability . These two residues are dynamic with Chemical Shift Index ( CSI ) based order parameters ( Berjanskii and Wishart , 2005 ) of 0 . 56 and 0 . 46 , respectively , where the order parameter is a metric used to quantify the amplitude of backbone motion , ranging from 0 ( isotropic motion ) to 1 ( rigid ) ( Lipari and Szabo , 1982 ) . In contrast , an average value for CSI-based order parameters of 0 . 80 ± 0 . 04 is obtained for the MIU2 sequence , indicating that it is significantly more rigid . On the basis of MIU2 sequence conservation and the formation of helical structure in the MIU2 region , it is likely that the MIU2 of RNF169 ( UDM2 ) interacts with ubiquitin via a classic MIU-ub interaction , whereby the helical arrangement aligns critical hydrophobic residues on one face of the helix so as to form contacts with the canonical hydrophobic patch of ubiquitin ( Penengo et al . , 2006 ) . 10 . 7554/eLife . 23872 . 005Figure 2 . Solution NMR analysis of RNF169 ( UDM2 ) reveals a flexible LRM2 . ( A ) Primary sequence of MIU2 ( orange ) and LRM2 ( blue ) modules of both RNF169 and RNF168 . Conserved residues are indicated with vertical bars . ( B ) Assigned 1H-15N HSQC spectrum of 15N , 13C-labeled RNF169 ( UDM2 ) . Data collected at 11 . 7 T , 35°C . ( C ) Output from SSP program ( Marsh et al . , 2006 ) using RNF169 ( UDM2 ) backbone chemical shifts as input . Corresponding primary sequence of RNF169 ( 662-708 ) displayed along horizontal axis , with MIU2 ( orange ) and LRM2 ( blue ) regions highlighted . Positive and negative SSP values indicate α-helical and β-strand secondary structure propensities , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 23872 . 00510 . 7554/eLife . 23872 . 006Figure 2—figure supplement 1 . Modified RNF169 ( UDM2 ) construct exhibits higher thermo stability . ( A ) Primary sequence of the modified RNF169 ( UDM2 ) construct used for NMR experiments at 45°C; addition of four N-terminal residues shown in bold . ( B ) Temperature melt of wild-type RNF169 ( UDM2 ) and the N-terminal extended variant using circular dichroism spectroscopy . Helical percentage calculated as described in Materials and Methods . DOI: http://dx . doi . org/10 . 7554/eLife . 23872 . 006 In order to improve the thermal stability of the MIU2 helix , we engineered an RNF169 protein containing the UDM2 sequence as before but where an aspartic acid , acting as a N-terminal helix cap , was introduced , followed by three alanine residues prior to K662 , Figure 2A ( Figure 2—figure supplement 1A ) . This peptide was used in all NMR studies with NCPs , since it tolerates temperatures of 45°C , where spectral quality is improved . It is noteworthy that the helical content of the thermostable variant at 45°C is near that of the wild-type peptide at 35°C , as shown by circular dichroism spectroscopy ( Figure 2—figure supplement 1B ) . To experimentally establish the molecular basis for the interaction between the MIU2 of RNF169 and ubiquitin , we performed NMR-based titrations of isotope-labeled ubiquitin , both free and chemically-attached to reconstituted nucleosomes . Chemical ubiquitylation of H2A at position K13 was achieved by converting G76 of ubiquitin and K13 in H2A to cysteine residues followed by conjugation via a sidechain-sidechain disulfide linkage ( Figure 3—figure supplement 1A ) . The ubiquitylated histone was then assembled into octamers and wrapped with DNA in the absence of any reducing agent to produce H2AK13Cub-NCPs , as previously described ( Dyer et al . , 2004 ) . The significant mass difference between free ubiquitin and ubiquitin conjugated to NCPs enabled the use of NMR-based 13C-edited diffusion experiments ( Choy et al . , 2002 ) , focusing on ubiquitin labeled with 13CH3 methyl groups ( see below ) , to confirm the conjugation product and monitor sample stability ( Figure 3—figure supplement 1B ) . The diffusion constant so obtained , D = 3 . 4 ± 0 . 2 × 10−7 cm2/s at 25°C , is consistent with a particle of molecular mass of approximately 250 kDa and is slightly smaller than for the 1/4 proteasome comprising a heptameric ring of α-subunits ( 180 kDa ) ( Religa et al . , 2010 ) where D = 4 . 2 ± 0 . 1 x 10−7 cm2/s has been measured under identical conditions . That the C76ub-C13H2A sidechain disulfide linkage supports RNF169 binding was demonstrated by showing that MBP-RNF169 ( UDM2 ) pulled down H2AK13Cub-NCPs as efficiently as catalytically prepared H2AK13ub-NCPs containing the isopeptide linkage ( Figure 3—figure supplement 1C ) . Having established both the suitability and the stability of our H2AK13Cub-NCPs , we carried out NMR titration experiments with RNF169 ( UDM2 ) ( unlabeled RNF169 in the case of free ubiquitin and perdeuterated for NCPs ) using 13C-labeled free ubiquitin or highly deuterated , Ile-δ1-13CH3 , Leu , Val-13CH3 , 12CD3-ubiquitin ( referred to as ILV-methyl labeled in what follows ) attached to H2AK13C-NCPs ( perdeuterated histone NCPs ) . Figure 3A shows a series of overlaid 1H-13C HMQC spectra of ILV-methyl labeled ubiquitin in H2AK13Cub-NCPs as a function of increasing amounts of RNF169 ( UDM2 ) . Chemical shift perturbations ( CSPs ) in both free ( Figure 3—figure supplement 2A ) and NCP-bound ubiquitin were of the same direction and magnitude , indicating very similar interactions for RNF169 ( UDM2 ) with both free and NCP-bound ubiquitin . Peaks exhibiting the largest CSPs are derived from residues that contribute to the canonical ubiquitin hydrophobic patch , including I44 , L8 and V70 ( Dikic et al . , 2009 ) . Indeed , NCPs conjugated to ubiquitin harboring an I44A point mutation failed to interact with the UDM2 module of both RNF168 and RNF169 in MBP pull-downs ( Figure 3—figure supplement 2B ) . CSP-derived binding curves were fit to extract equilibrium dissociation constants for RNF169 ( UDM2 ) with free ubiquitin and H2AK13Cub-NCPs of 956 ± 340 μM , 45°C , and 24 ± 7 μM , 45°C , respectively ( Figure 3B ) . It is noteworthy that only residues with relatively small CSPs , corresponding to fast exchange on the NMR chemical shift timescale , were included in the analysis since in this limit chemical shifts are directly related to dissociation constants ( KD ) and are not influenced by kinetics , allowing reliable estimates of affinity from peak positions . The low affinity interaction between RNF169 ( UDM2 ) and free ubiquitin is characteristic of ubiquitin binding domains , having dissociation constants generally larger than 100 μM ( Hurley et al . , 2006 ) . The presence of the NCP increases the affinity by approximately 40-fold , presumably through specific contacts between the LRM2 and the NCP surface near K13 of H2A . The use of two cooperative , relatively low affinity interactions to impart specificity is a common theme of ubiquitin signaling ( Chen et al . , 2009 ) . 10 . 7554/eLife . 23872 . 007Figure 3 . Thermodynamics and kinetics of the RNF169 ( UDM2 ) -H2AK13C ub-NCP interaction . ( A ) Selected regions of 1H-13C HMQC spectra of ILV-methyl labeled ubiquitin in H2AK13Cub-NCPs with increasing amounts of unlabeled RNF169 ( UDM2 ) . Arrows indicate direction of peak movement . Data collected at 14 . 1 T , 45°C . ( B ) Chemical shift derived binding curves for selected residues in ILV-methyl labeled ubiquitin in H2AK13Cub-NCPs ( left panel ) and free ILV-methyl labeled ubiquitin ( right panel ) upon addition of unlabeled RNF169 ( UDM2 ) ( circles ) , with best fits ( solid lines ) shown . Ratio of RNF169 ( UDM2 ) to ubiquitin indicated on horizontal axis . ( C ) Fitted line shapes for I44δ1 ( extracted from 1H-13C correlation spectra by taking traces along the 13C-dimension ) . Experimental data in black , with simulated line shapes in red . Ratio of ubiquitin to RNF169 ( UDM2 ) indicated in each panel and vertical grey dashed lines denote the resonance positions of I44δ1 in the absence ( free ) and presence ( bound ) of saturating amounts of RNF169 ( UDM2 ) . The extracted kinetic parameters for the RNF169 , ub-NCP binding reaction are shown above the traces . DOI: http://dx . doi . org/10 . 7554/eLife . 23872 . 00710 . 7554/eLife . 23872 . 008Figure 3—figure supplement 1 . Preparation and validation of disulfide linked H2AK13Cub-NCPs . ( A ) Schematic outlining the chemical activation of H2AK13C and conjugation with ubG76C . ( B ) NMR derived diffusion constants for free ubiquitin ( black ) , H2AK13Cub-NCP ( red ) and ¼ proteasome ( green ) . Plot of 1/Tln ( I/Io ) as a function of squared gradient strength ( circles ) , where T is the diffusion time ( 100 ms for ubiquitin , 250 ms for NCP and ¼ proteasome ) and I , Io are the intensities in the presence and absence of encoding gradients , with best fits ( lines ) and calculated diffusion constants indicated . ( C ) Pull-down assays of NCPs ubiquitylated by RNF168 ( ubCAT ) or prepared through the disulfide directed approach ( K13C-Ubss ) . MBP-RNF169 ( UDM2 ) was used in the pull-down . One pull-down in the absence of RNF168 ( - E3-ligase ) is used as a negative control . The gel was run in the presence of a reducing agent ( dithiothreitol ) so the H2AK13C-Ubss linkage is reduced to H2AK13C . Antibody against the acidic patch of H2A was used to detect H2A from Drosophila . Catalytic ubiquitylation of K15C species by RNF168 produces a small amount of diubiquitylated H2A with the second ubiquitin conjugated to position K36 , as indicated on the gel ( Wilson et al . , 2016 ) . The * indicates an artifact from the gel . Input: 5% of the amount of ubiquitylated NCPs used in pull-down . The migration of molecular mass markers ( kDa ) is indicated on the right . DOI: http://dx . doi . org/10 . 7554/eLife . 23872 . 00810 . 7554/eLife . 23872 . 009Figure 3—figure supplement 2 . RNF169 ( UDM2 ) binds ubiquitin through its canonical hydorphobic face in both the free and NCP-bound context . ( A ) Overlay of 1H-13C HMQC spectra of ILV-methyl labeled ubiquitin with increasing amounts of RNF169 ( UDM2 ) , as indicated . Data recorded at 11 . 7 T , 45°C . Final RNF169 ( UDM2 ) to ubiquitin ratio of 42:1 . ( B ) Pull-down assay of NCPs conjugated to wild-type or I44A ubiquitin with MBP-RNF169 ( UDM2 ) ( top panel ) or MBP-RNF168 ( UDM2 ) ( bottom panel ) . Ub-NCP binding-deficient RNF169 ( L699A/R700A ) and RNF168 ( L476A/R477A ) mutants , denoted by LR , were used as negative controls . Input: 5% of the amount of ubiquitylated NCPs used in pull-down; the migration of molecular mass markers ( kDa ) is indicated on the left . DOI: http://dx . doi . org/10 . 7554/eLife . 23872 . 00910 . 7554/eLife . 23872 . 010Figure 3—figure supplement 3 . Selected regions of CT-1H-13C HSQC spectra of 15N , 13C wild-type RNF169 ( UDM2 ) with increasing amounts of unlabeled ubiquitin . Arrows indicate direction of peak movement . Data collected at 11 . 7 T , 35°C . DOI: http://dx . doi . org/10 . 7554/eLife . 23872 . 010 Notably , for I44 , that has the largest CSP upon binding ( ~126 Hz ) , exchange between RNF169 ( UDM2 ) free and bound NCP states is in the intermediate regime on the chemical shift timescale , facilitating the extraction of kinetic parameters via lineshape analysis ( Tugarinov and Kay , 2003 ) . Figure 3C shows a comparison of experimental lineshapes ( black ) extracted from the 13C dimension of 1H-13C HMQC spectra for I44 ( δ1 methyl ) as a function of increasing amounts of added RNF169 ( UDM2 ) , with fitted simulated lineshapes that include the effects of chemical exchange in red . Using the value of KD = 24 ± 7 μM estimated from the titration data , on and off-rates are fit to be 2 . 4 ± 0 . 7×107 M−1 s−1 and 5 . 4 ± 0 . 4×102 s−1 respectively . Notably , the RNF169 ( UDM2 ) on-rate is an order of magnitude faster than the diffusion limit ( Alsallaq and Zhou , 2007; Schlosshauer and Baker , 2004 ) , emphasizing the importance of electrostatic interactions in contributing to the association . To investigate the regions of the NCP involved in the interaction with the LRM2 , we monitored CSPs in 1H-13C HMQC spectra upon addition of RNF169 ( UDM2 ) . As discussed previously methyl-HMQC spectra are significantly enhanced in sensitivity and resolution by a methyl-TROSY effect that preserves signal ( Tugarinov et al . , 2003 ) . This aspect is of particular importance in applications to high molecular weight proteins and protein complexes ( Sprangers and Kay , 2007; Rosenzweig and Kay , 2014; Gelis et al . , 2007 ) , such as the NCP ( Kato et al . , 2011a ) . Figure 4A shows 1H-13C HMQCs of ILV-methyl labeled H2A and H2B in the context of H2AK13Cub-NCPs , unbound or with saturating amounts of RNF169 ( UDM2 ) . Corresponding weighted CSPs are tabulated in Figure 4B ( see Materials and methods ) . Upon binding RNF169 ( UDM2 ) L64δ2 of H2A and V41γ1 , V45γ1 , V63b , L99a and L103δ1/δ2 of H2B showed significant CSPs . Note that in the absence of stereospecific assignments isopropyl methyl groups are referred to as ‘a’ or ‘b’ to denote upfield and downfield resonating moieties . Interestingly , the methyl-bearing residues mentioned above surround a highly negative surface of the H2A/H2B dimer , known as the acidic patch . The acidic patch is comprised of eight residues in total; six in H2A ( E55 ( E56 ) , E60 ( E61 ) , E63 ( E64 ) , D89 ( D90 ) , E90 ( E91 ) and E91 ( E92 ) in Drosophila melanogaster ( Homo sapiens ) ) and two in H2B ( E102 and E110 ) , creating a contoured and highly charged groove ( Kalashnikova et al . , 2013 ) . The acidic patch is a well-characterized interaction surface for a multitude of NCP binding proteins with a broad range of functions ( Kato et al . , 2011a; Barbera et al . , 2006; McGinty et al . , 2014; Makde et al . , 2010; Armache et al . , 2011; Morgan et al . , 2016 ) . Interestingly , no significant CSPs were observed in H3 ( Figure 4—figure supplement 1 ) , consistent with localized interactions involving the H2A/H2B dimer face of the NCP ( Figure 4C ) . 10 . 7554/eLife . 23872 . 011Figure 4 . NMR and mutagenesis identify the nucleosome acidic patch as the binding interface for RNF169 ( UDM2 ) . ( A ) Superimposed 1H-13C HMQC spectra of ILV-methyl labeled H2A ( left panel ) and ILV-methyl labeled H2B ( right panel ) in the context of the H2AK13Cub-NCP without ( black ) and with ( yellow , left; red , right ) wild-type RNF169 ( UDM2 ) , respectively . RNF169 ( UDM2 ) was added at 2 . 5-fold excess relative to ubiquitin . Arrows indicate peak movement . Data collected at 14 . 1 T , 45°C . ( B ) Chemical shift perturbations ( CSPs ) in ILV-methyl labeled H2A ( yellow , top panel ) and ILV methyl-labeled H2B ( red , bottom panel ) H2AK13Cub-NCPs . Residues with CSP values 1σ above the average are indicated ( black line ) . CSPs were calculated as described in Materials and Methods . ( C ) Location of residues with significant CSPs in H2A ( yellow ) and H2B ( red ) shown in space filling representation and indicated with arrows on nucleosome crystal structure ( 2PYO ) ( Clapier et al . , 2008a ) . Acidic patch residues are shown in stick representation and coloured black . H2A: light yellow , H2B: salmon , H3: light blue and H4: light green . ( D ) Selected isoleucine regions of 1H-13C HMQC spectra of free ( black ) and 2 . 5-fold excess RNF169 ( UDM2 ) bound ( red ) ILV-methyl labeled ub H2AK13Cub-NCP . Spectra of acidic patch mutant NCPs , H2AK13C ( D89A/E91A ) ub-NCP ( purple , left panel ) and H2AK13C ( E60A/E63A ) ub-NCP ( teal , right panel ) , with 5-fold excess RNF169 ( UDM2 ) to ubiquitin are overlaid and highlight the resulting binding deficiency . All data are recorded at 14 . 1 T , 45°C . DOI: http://dx . doi . org/10 . 7554/eLife . 23872 . 01110 . 7554/eLife . 23872 . 012Figure 4—figure supplement 1 . Overlay of 1H-13C HMQC spectra of ILV-methyl labeled H3 H2AK13Cub-NCPs with ( red ) and without ( black ) RNF169 ( UDM2 ) . No significant CSPs are observed . Data recorded at 14 . 1 T , 45°C . DOI: http://dx . doi . org/10 . 7554/eLife . 23872 . 012 While the methyl-TROSY NMR results implicate the acidic patch surface as the main contact point between the NCP and LRM2 , we made use of a combined site-directed mutagenesis – NMR approach to directly monitor the effect of acidic patch residues . Specifically , we prepared H2AK13Cub-NCPs with pairwise mutations in H2A including D89A/E91A and E60A/E63A , effectively neutralizing two regions of the acidic patch surface . The first mutant , H2A D89A/E91A , alters a region of the acidic patch that is responsible for making several contacts with the canonical residue common to all known acidic patch-binding proteins ( McGinty and Tan , 2015 ) . The second mutant , H2A E60A/E63A , was designed to neutralize the region of the acidic patch closer to the DNA , where additional contacts have been identified in NCP-binding proteins ( Kalashnikova et al . , 2013 ) . The large chemical shift changes of isoleucine residues in ubiquitin upon binding RNF169 ( UDM2 ) provide spectral signatures of free and RNF169 ( UDM2 ) -bound H2AK13Cub-NCPs , enabling a straightforward comparison of the binding capacity of the H2A mutants described above with wild-type H2A containing NCPs . The isoleucine region of ubiquitin in H2AK13C ( D89A/E91A ) ub- and H2AK13C ( E60A/E63A ) ub-NCPs , after addition of equivalent amounts of RNF169 ( UDM2 ) , are shown in Figure 4D , overlaid with free ( black ) and RNF169 ( UDM2 ) -bound wild-type H2AK13Cub-NCP ( red ) spectra . For both H2AK13C ( D89A/E91A ) ub- and H2AK13C ( E60A/E63A ) ub-NCPs the binding capacity for RNF169 ( UDM2 ) is reduced considerably , with KD values of approximately 280 ± 100 μM and 230 ± 100 μM relative to 25 μM for binding to wild-type NCPs . The compromised binding displayed by the H2A mutant-NCPs provides strong validation of the methyl-TROSY CSP data and indicates that at least one member of each pairwise H2A mutant plays an important role in the interaction . Moreover , the mutagenesis data extend the range of NCP residues that can be investigated , since the NMR analysis is limited to only methyl-containing amino acids . Previous work has identified several highly conserved residues within the LRM2 of both RNF168 and RNF169 that are vital to their accumulation at DSBs ( Panier et al . , 2012 ) . Individual substitutions of R689 , Y697 , and pairwise substitution of L699/R700 to alanine were found to abrogate recruitment of GST-RNF169 ( UDM2 ) to IR-induced foci ( Panier et al . , 2012 ) . In agreement with these findings , R689A , Y697A , L699A and R700A were all found to reduce the capacity of RNF169 to inhibit 53BP1 focus formation after irradiation with a 2 Gy dose of X-rays ( Figure 5—figure supplement 1 ) . While methyl-TROSY NMR was instrumental in identifying the acidic patch of the NCP surface as an important component in the interaction , a similar CSP-based analysis of the LRM2 was hampered by the presence of only three methyl-bearing residues in this motif ( V694 , L698 and L699 ) and by the complete resonance overlap of L698 and L699 ( Figure 5—figure supplement 2 ) . To confirm the results of the biological assay involving 53BP1 that established key LRM2 residues , we used a mutagenesis and methyl-TROSY NMR approach , similar to that described above . As before , we made use of the free and RNF169 ( UDM2 ) -bound signature spectra of isoleucine residues in ubiquitin to qualitatively evaluate the ability of single residue substitutions in RNF169 ( UDM2 ) to form a complex with wild-type H2AK13Cub-NCPs ( Figure 5 ) . As expected , R689 , L699 and R700 substitutions reduce the ability of RNF169 to form a complex with H2AK13Cub-NCPs , while Y697A was also observed to impede binding to a significant degree , and S701A exhibited the smallest decrease in binding capacity . 10 . 7554/eLife . 23872 . 013Figure 5 . R689 and L699/R700 are critical to the formation of the complex . Isoleucine region of 1H-13C HMQC spectra of ILV-methyl labeled ub H2AK13C-ubNCP without ( black ) and with ( red ) wild-type RNF169 ( UDM2 ) and the indicted RNF169 ( UDM2 ) LRM2 mutants . The ratio of wild-type or mutant RNF169 ( UDM2 ) to ubiquitin was 2 . 5:1 in all cases . DOI: http://dx . doi . org/10 . 7554/eLife . 23872 . 01310 . 7554/eLife . 23872 . 014Figure 5—figure supplement 1 . Alanine scanning of LRM2 reveals criticial residues in RNF169 ( UDM2 ) -ubNCP interaction . ( A ) Quantification of 53BP1 foci in U2OS cells transfected with GFP-RNF169 ( UDM2 ) wild-type or mutant , as indicated along the horizontal axis . CTRL: untransfected cells , LARA: L699A/R700A . Error bars represent the mean ±1 standard deviation for n = 3 . Cells were irradiated ( 2 Gy ) and processed for 53BP1 and γ-H2AX immunofluorescence as well as GFP imaging 1 hr after irradiation . ( B ) Corresponding micrographs of the experiments presented in ( A ) . Dash lines outline nucleus of the cells and scale bar represents 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 23872 . 01410 . 7554/eLife . 23872 . 015Figure 5—figure supplement 2 . Methyl resonances of LRM2 region are overlapped in 1H-13C CT-HSQC of 15N , 13C wild-type RNF169 ( UDM2 ) . Significant resonance overlap of leucine residues is observed . Data recorded at 11 . 7 T , 35°C . DOI: http://dx . doi . org/10 . 7554/eLife . 23872 . 01510 . 7554/eLife . 23872 . 016Figure 5—figure supplement 3 . NMR reveals a similar acidic patch binding mode for R689 and R700 of RNF169 ( LRM2 ) and LANA ( 1-23 ) . ( A ) Alignment of the primary sequences of RNF169 ( UDM2 ) and the LANA peptide . Common residues that have been shown to be relevant for NCP binding are shaded in gray . ( B ) Selected regions of 1H-13C HMQC spectra of free ILV-methyl labeled H2B in the context of H2AK13Cub-NCPs in black , overlaid with corresponding spectral regions as recorded on ILV-methyl labeled H2B H2AK13Cub-NCP samples with RNF169 ( UDM2 ) ( red ) , with the LANA peptide ( purple ) , with RNF169 ( R689/R690/K691A ) ( UDM2 ) ( blue ) and with RNF169 ( L699/R700/S701A ) ( UDM2 ) ( green ) . Residues with significant CSPs are labeled , and the ratio of RNF169 ( UDM2 ) or LANA to ubiquitin was 5:1 . All spectra were recorded at 14 . 1 T , 45°C . DOI: http://dx . doi . org/10 . 7554/eLife . 23872 . 016 As mentioned above , the acidic patch is a common interaction surface for nucleosome binding proteins that typically involves key contacts between an arginine sidechain and the nucleosome surface ( Kato et al . , 2011a ) . The Kaposi’s sarcoma herpes virus ( KHSV ) latency associated nuclear antigen ( LANA ) is an acidic patch binding protein with a critical LR-like motif ( Figure 5—figure supplement 3A ) . MBP pull-down assays established that a LANA-derived peptide competes with RNF169 for H2AK13/K15ub-NCP binding with an estimated IC50 of 125–500 µM ( Figure 1—figure supplement 1A ) ( Barbera et al . , 2006 ) . However , it was not clear a priori whether one or both of the basic stretches of residues in RNF169 ( UDM2 ) ( R689-R690-K691 or L699-R700-S701 ) was important for binding and whether one mimicked the critical L8-R9-S10 sequence in LANA . A comparison of CSPs in ILV-labeled H2B upon binding RNF169 ( UDM2 ) , LANA ( 1-23 ) and two RNF169 ( UDM2 ) triple mutants , ( RNF169 ( R689A/R690A/K691A ) and RNF169 ( L699A/R700A/S701A ) ) , revealed similar chemical shift perturbation patterns between LANA and both RNF169 triple mutants ( Figure 5—figure supplement 3B ) . While these results were unable to definitively identify the anchoring arginine in RNF169 ( UDM2 ) , the fact that both triple mutants bound H2AK13Cub-NCPs reinforces the importance of both the RRK and LRS regions of RNF169 ( UDM2 ) . Our experimental data indicate the importance of three interactions for the selective recognition of H2AK13/K15ub-NCPs by RNF169 ( UDM2 ) . The first , encompassing the MIU2 residues 665–682 , is of low affinity , like for many MIU-ub complexes involving the canonical hydrophobic surface of ubiquitin ( Hurley et al . , 2006 ) . This weak interaction is strengthened through two additional contacts between the nucleosome acidic patch and the RRK and LRS regions of RNF169 ( UDM2 ) that provides selectivity for NCPs ubiquitylated on K13 or K15 of H2A . As a first step to obtain the structure of the complex , we docked the MIU2 region of RNF169 , corresponding to residues 662–682 , onto ubiquitin ( PDB ID 1UBQ ) using experimental CSPs and a number of inter-molecular nuclear Overhauser effects ( NOEs ) as restraints ( see Materials and Methods ) with the HADDOCK molecular modeling program ( Dominguez et al . , 2003 ) . Methyl chemical shift changes in the MIU2 region upon binding ubiquitin are localized to L672δ1/δ2 , A673 , and L676δ1/δ2 , highly conserved MIU residues that are directly involved in the interaction ( Figure 6A ) ( Panier et al . , 2012; Penengo et al . , 2006 ) . Key NOEs connect A673 of the MIU2 to L8 and I44 of Ub , Figure 6B . The resulting HADDOCK model is shown in Figure 6C aligned with the X-ray structure of the Rabex MIU-ubiquitin complex ( Penengo et al . , 2006 ) . Aside from a slight tilt in helix orientation ( compare blue Rabex and gold RNF169 helices ) the RNF169 ( 665-682 ) MIU2-ub model is very similar to that previously published for the Rabex MIU-ub complex . 10 . 7554/eLife . 23872 . 017Figure 6 . The RNF169 MIU2-ubiquitin interaction involves the canonical binding surface of ubiquitin and a central alanine in the MIU2 . ( A ) Overlaid regions of 1H-13C CT-HSQC spectra of 13C-labeled RNF169 ( UDM2 ) upon addition of increasing amounts of free ubiquitin . Residues with significant chemical shift changes are labeled . Data recorded at 11 . 7 T , 35°C . ( B ) Selected region of 1H-1H NOESY spectrum of ILV-methyl labeled ubiquitin and 13C-labeled RNF169 ( UDM2 ) at 1:8 molar ratio ( 200 ms mixing time ) . Cross peaks between A673 of RNF169 ( UDM2 ) and I44δ1 , L8δ1/δ2 of ubiquitin are indicated . Data recorded at 14 . 1 T , 20°C . ( C ) Signature MIU primary sequence; x: any amino acid type , φ: large hydrophobic and #: acidic . Structural model of RNF169 ( MIU2 ) -ubiquitin from HADDOCK docking calculations , aligned for comparison with crystal structure of the Rabex ( MIU ) -ubiquitin complex ( 2C7N ( Penengo et al . , 2006 ) ) . Signature MIU residues within RNF169 ( MIU2 ) and Rabex and hydrophobic patch residues are shown in stick representation . Ubiquitin: magenta , RNF169 ( MIU2 ) : orange , Rabex ( MIU ) : cyan . DOI: http://dx . doi . org/10 . 7554/eLife . 23872 . 017 Having elucidated the structure of the MIU2-ub component of the complex , we next used cryo-EM in an attempt to obtain structural models of the ub-NCP particles in both the free and RNF169-bound states . Recent advances in single particle cryo-EM instrumentation and data processing have facilitated the calculation of high-resolution 3D-maps of biomolecules as small as 60 kDa ( Bai et al . , 2015; Smith and Rubinstein , 2014; Khoshouei et al . , 2016 ) . Figure 7A–B show cryo-EM density maps of free and RNF169-bound NCPs , determined at 8 . 1 and 6 . 6 Å resolution , respectively ( Figure 7—figure supplement 1 ) . In both cases the rigid NCP forms the symmetrical discoid shape expected from NCP x-ray structures ( Clapier et al . , 2008a ) . A striking difference between the two maps is the density corresponding to ubiquitin , which at the threshold level used , is only visible in the RNF169-bound state . In the absence of RNF169 ( UDM2 ) ubiquitin is highly dynamic , occupying a variety of positions with respect to the NCP due to its conjugation to the flexible N-terminal tail of H2A . The density in this portion of the map is thus low in comparison to the NCP core itself . These results are illustrated in Figure 7C using equivalent lateral slices through free and RNF169-bound NCP cryo-EM maps , with the raw map density colored according to local resolution estimates ( Kucukelbir et al . , 2014 ) . As expected , the NCP core in both maps exhibits the highest resolution , while the resolution of the ubiquitin region of the map is significantly lower in the absence of RNF169 ( UDM2 ) . The restricted motion of ubiquitin within the complex can also be established by measuring NMR spin relaxation rates of ubiquitin methyl probes in H2AK13Cub NCPs in the presence and absence of RNF169 . Here we have quantified intra-methyl 1H-1H dipolar cross-correlated relaxation rates ( Sun et al . , 2011 ) , focusing on ILV residues . In the macromolecular limit these rates are proportional to the product S2τC , where S2 is the square of an order parameter describing the amplitude of motion for the 3-fold symmetry axis of the methyl group , and τc is the molecular tumbling time that in the present case provides a measure for how rigidly attached ubiquitin is to the NCP . Residue-specific S2τc values are plotted in Figure 7D for ub-NCP ( blue ) and for RNF169 ( UDM2 ) ub-NCP ( pink ) with an average 2-fold increase , reflecting a reduction in conformational space available to ubiquitin upon addition of RNF169 ( UDM2 ) . 10 . 7554/eLife . 23872 . 018Figure 7 . Ubiquitin is highly dynamic in the absence of RNF169 ( UDM2 ) . ( A ) H2AK13Cub-NCP cryo-EM map at 8 . 1 Å resolution and ( B ) RNF169 ( UDM2 ) bound H2AK13Cub-NCP cryo-EM map at 6 . 6 Å resolution including the drosophila NCP crystal structure ( 2PYO ) ( Clapier et al . , 2008a ) fit within the map as a rigid body using UCSF chimera . ( C ) Indicated equivalent lateral slices through free H2AK13Cub-NCP ( left panels ) and RNF169 ( UDM2 ) bound H2AK13Cub-NCP maps ( right panels ) , showing the raw map density and colored according to local resolution estimates ( Kucukelbir et al . , 2014 ) . ( D ) Histogram comparison of S2τC values obtained for ILV-methyl labeled ubiquitin in free ( blue ) and RNF169 ( UDM2 ) bound H2AK13Cub-NCPs ( pink ) fit to a normal distribution . DOI: http://dx . doi . org/10 . 7554/eLife . 23872 . 01810 . 7554/eLife . 23872 . 019Figure 7—figure supplement 1 . Cryo-EM data and processing . ( A ) Representative cryo-EM micrographs . Several individual particle projections are boxed . ( B ) Examples of 2D class average images from cryo-EM of the RNF169 ( UDM2 ) -ubNCP complex . ( C ) 3D classification of the 90 200 selected particle images obtained after 2D classification . All classes show varying detail , location and volume of density for the covalently attached ubiquitin . The most populated high-resolution class was refined with C2 symmetry to yield the final RNF169 ( UDM2 ) -ubNCP map . ( D ) Fourier shell correlation curve after a gold-standard map refinement of the ubNCP structure obtained during preliminary complex optimization , corrected for the effects of map masking . ( E ) Fourier shell correlation curve after a gold-standard map refinement of the final RNF169 ( UDM2 ) -ubNCP structure , corrected for the effects of map masking . ( F ) Euler angle distribution plot of all particle images used for the symmetrized ubNCP map . Bar length and color ( blue low , red high ) corresponds to number of particle images contributing to each view . ( G ) Euler angle distribution plot of all particle images used for the symmetrized RNF169 ( UDM2 ) -ubNCP map . Bar length and color ( blue low , red high ) corresponds to number of particle images contributed to each view . DOI: http://dx . doi . org/10 . 7554/eLife . 23872 . 019 While the cryo-EM map defined aspects of the overall topology of the complex , higher resolution information is required to obtain an atomic level description of key interactions that provide specificity and , in particular , to resolve the role of the important arginine residues , R689 and R700 . We used replica-averaged molecular dynamics simulations to develop a structural model consistent with our experimental NMR and mutagenesis data ( Cavalli et al . , 2013a; Kukic et al . , 2014a , 2016 ) . In this method , experimental data are incorporated during the simulations as replica-averaged restraints whereby back-calculated parameters are compared with their experimentally measured values to evolve the system in a manner such that the agreement with the experimental restraints increases over time . This approach is described in detail in Materials and methods and illustrated schematically in Figure 8—figure supplement 1 . Figure 8A shows an overlay of ten members from the calculated ensemble of approximately 600 structures . Enlarged views of a pair of structures , focusing on the region of contact between the LRM2 ( blue ) and acidic patch residues ( yellow and red ) , are highlighted in Figure 8B . Notably , in the great majority of structures the LRM2 backbone remains highly disordered within the complex and does not form regular secondary structure . In less than 5% of the conformers the LRM2 region contains either a small antiparallel β sheet encompassing residues Y697 to M704 , or a small 3-residue helix involving residues between Y697 and M704 . Importantly , in all members of the ensemble R700 is consistently located in the position where it makes contacts with one or more of the key acidic patch residues E60 , D89 and E91 . R689 also contacts the acidic patch in all structures through interactions involving at least one of E60 ( H2A ) , E63 ( H2A ) or D48 ( H2B ) . As shown in Figure 8B , the position of R700 within the ensemble is relatively fixed , while R689 alternates between positions that permit electrostatic contacts with E63 or D48 . Although R700 and R689 were shown by mutagenesis to be almost equally important for formation of the RNF169 ( UMD2 ) -ubNCP complex ( see above ) , our replica-averaged MD results indicate that R700 , rather than R689 , is the critical anchoring arginine . Mutagenesis and NMR studies also indicate a significant role for L699 and , to a lesser extent for Y697 in complex formation ( Figure 5A ) . For a significant subset of the structural ensemble members , Y697 and L699 sample positions which direct their sidechains toward a hydrophobic pocket formed by H2A Y49 , V53 and Y56 , much like the positions of M6 and L8 in the LANA peptide – NCP structure ( Barbera et al . , 2006 ) . In addition , there are alternative conformations of L699 where it appears to make no direct contact to the NCP surface , but instead interacts with regions of the LRM2 module itself . For example , L699 is observed to interact with M704 or A705 in some members of the ensemble . 10 . 7554/eLife . 23872 . 020Figure 8 . Structural model of the RNF169 ( UDM2 ) -ubNCP complex . ( A ) Alignment of ten representative members of the RNF169 ( UDM2 ) -ubNCP structural ensemble obtained from replica-averaged MD simulations constrained by CSPs and mutagenesis data . Histones H2A and H2B were used to align the structures with only one copy of the histones and DNA shown for simplicity . Ubiquitin: magenta , RNF169 ( 662-682 ) ( MIU2 ) : orange , RNF169 ( 683-708 ) ( LRM2 ) : blue , H2A: yellow , H2B: salmon , H3: light blue , H4: light green , DNA: gray . ( B ) Enlarged view focusing on specific contacts between R689 and R700 and the nucleosome acidic patch in two separate structures . ( C ) Two viewpoints of ten aligned members of the RNF169 ( UDM2 ) ub-NCP structural ensemble fit into the RNF169 ( UDM2 ) bound H2AK13Cub-NCP cryo-EM map . Molecular graphics images were produced using the UCSF Chimera package ( Pettersen et al . , 2004 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23872 . 02010 . 7554/eLife . 23872 . 021Figure 8—figure supplement 1 . Schematic outline of replica-averaged MD protocol used . Restraints used in each stage are indicated in boxes; COM: centre of mass , CS: chemical shifts , NOEs: Nuclear Overhauser effects , contacts: CSPs and mutagenesis restraints . Temperature and timescale of each step is indicated above arrows . Explicit details can be found in Materials and Methods . DOI: http://dx . doi . org/10 . 7554/eLife . 23872 . 02110 . 7554/eLife . 23872 . 022Figure 8—figure supplement 2 . Initial position of ub in RNF ( UDM2 ) -ubNCP starting structures does not influence the final ub orientation . ( A ) Alignment of RNF ( UDM2 ) -ubNCP starting conformations for replica-averaged MD simulations . Histones H2A and H2B were used to align the structures with only one copy of the histones and DNA shown for simplicity . The starting position of ubiquitin used in a pair of simulations is shown in black and red . RNF169 ( 662-682 ) ( MIU2 ) : orange , RNF169 ( 683-708 ) ( LRM2 ) : blue , H2A: light yellow , H2B: salmon , H3: light blue , H4: light green , DNA: gray . ( B ) Alignment of RNF ( UDM2 ) -ubNCP final conformations from replica-averaged MD simulations , as in A . DOI: http://dx . doi . org/10 . 7554/eLife . 23872 . 022 We have carried out further simulations to verify the important role of R700 by performing a series of 50 simulated annealing cycles per replica in the absence of experimental restraints and cycling the simulation temperature from 300 K to 500 K and back to 300 K in each cycle ( see Materials and methods ) . In all cases R700 remained in the canonical anchoring position in the acidic patch . As a further test to examine the possibility that R689 can replace R700 in the acidic patch we started from a structure in which the sidechain of R689 was forced into the anchoring position and subsequently carried out a series of simulated annealing cycles . After approximately eight cycles R689 moved out of the anchoring pocket and after an additional 10 cycles R700 inserted into the pocket where it remained for the final 12 cycles of the simulation . Although the resolution of the cryo-EM model ( Figure 7A–C ) is not sufficient to obtain atomic insights , it can be used to cross-validate some aspects of our ensemble averaged restrained MD models , in particular the location of ubiquitin . In Figure 8C , 10 replica-averaged MD derived structures are superimposed on the cryo-EM envelope and the positions of the ubiquitin molecules ( magenta ) fit well to the cryo-EM map . Notably , the final position of ubiquitin does not depend on its initial position in MD simulations . Figure 8—figure supplement 2A highlights a pair of H2AK13Cub-NCPs with ubiquitin in different starting positions , relative to the NCP , with convergence to similar final positions obtained in both cases ( Figure 8—figure supplement 2B ) . We have presented a structural ensemble of the RNF169 ( UDM2 ) -H2AK13Cub-NCP complex based on replica-averaged MD simulations that included CSPs and mutagenesis data as experimental restraints . This structural ensemble illustrates how RNF169 is able to discriminate between numerous ubiquitylated chromatin species to specifically bind nucleosomes monoubiquitylated at H2AK13/K15 . The multicomponent complex includes a three-pronged interaction involving a MIU2-ubiquitin component and a pair of electrostatic contacts between key arginine residues within the LRM2 module of RNF169 and the acidic patch region of the nucleosome surface . Our model reveals that the MIU2 helix interacts with ubiquitin via a hydrophobic surface centered about A673 of RNF169 ( UDM2 ) that involves the canonical binding interface on ubiquitin , comprised of I44 , L8 and V70 , similar to a previously published MIU-ubiquitin crystal structure ( Penengo et al . , 2006 ) . Notably , the LRM2 module remains highly disordered in the complex , with electrostatic contacts between R689 and one or more of H2A E60/E63 and H2B D48 , and between R700 and H2A E63 , D89 and E91 observed in all members of the calculated structural ensemble . The good agreement between the position of ubiquitin within the structural ensemble and in the cryo-EM map of RNF169-bound ubNCPs provides cross-validation for the position of ubiquitin in the multicomponent complex . Moreover , the cryo-EM data establishes that within the complex ubiquitin is held more rigidly to the nucleosome core , through the combined action of the MIU2 and LRM2 , in agreement with NMR based spin relaxation measurements . Finally , because the critical C-terminal MIU2-LRM2 module of RNF169 , that confers specificity , is conserved in RNF168 the proposed model for the RNF169 ( UDM2 ) -H2AK13Cub-NCP complex is likely a good proxy for the interaction of K13ub-NCPs with RNF168 . The size of the RNF169 ( UDM2 ) -ubNCP complex and the intrinsic dynamics of both the histone H2A tail and the LRM2 module pose significant challenges for traditional structural biology techniques which often assume that the experimental restraints define a single conformer . The integrative approach taken here , including the use of methyl-TROSY NMR , mutagenesis , replica-averaged MD simulations and cryo-EM was essential in defining the molecular details of the complex . NMR has evolved in recent years to become a powerful tool for the study of the structure , and in particular the dynamics , of large proteins and protein complexes through the development of optimized protein labeling methods ( Gardner and Kay , 1997; Goto et al . , 1999; Tugarinov and Kay , 2004; Kainosho et al . , 2006 ) and TROSY-based experimental approaches that enhance both spectral resolution and sensitivity ( Tugarinov et al . , 2003; Fiaux et al . , 2002 ) . Despite the prevalence of methyl bearing residues within protein cores and at interfaces in complexes ( Janin et al . , 1988 ) , their scarcity on exposed protein surfaces and within highly charged regions can be limiting in the study of electrostatically driven interactions . Pertinent to this case is the lack of ILV residues on the surface of the NCP . While some well-positioned methyl probes , including V45 and L103 of H2B , were instrumental in identifying the acidic patch as the binding partner for the critical arginine regions of the LRM2 , methyl-based NMR strategies alone were insufficient to ascertain the structural details of this complex . Distance restraints are key elements in the elucidation of detailed structures , with short ( ~5 Å ) to long-range ( < ~30 Å ) distances available from NOE and paramagnetic relaxation enhancement ( PRE ) NMR techniques , respectively ( Battiste and Wagner , 2000; Wüthrich , 1986 ) . However , the dynamic nature of disordered protein segments can challenge the measurement of intermolecular distances ( Mittag and Forman-Kay , 2007 ) . In our hands , intermolecular methyl NOEs between RNF169 ( UDM2 ) and NCP histones were difficult to observe , reflecting both the dilute samples used ( 100 μM ) and the paucity of methyl groups at the binding interface . Furthermore , in dynamic systems , such as the one studied here , intermolecular PREs provide at best upper distance bounds that we have found to be too large to be useful for straightforward protein-protein docking approaches . To circumvent these difficulties replica-averaged MD simulations were used to account for the inherent flexibility of both the ubiquitin attachment and the LRM2 module so as to produce a structural ensemble consistent with our experimental data . Improvements in both the accuracy of the force fields that describe molecular interactions and the available computational power have increased the robustness of MD simulations and the complexity of the systems amenable to them ( Vendruscolo and Dobson , 2011; Chung et al . , 2015 ) . The inclusion of experimental NMR data in MD simulations , as ensemble-averaged restraints , provides a means to more extensively sample the conformational space of interest and to generate ensembles of structures consistent with experimental measurements ( Torda et al . , 1989; Lindorff-Larsen et al . , 2005; Cavalli et al . , 2013b ) . The complementarity of the techniques used here and the continuous effort toward improving their capabilities will facilitate future studies of dynamic macromolecules , similar to the RNF169-ubNCP complex described here . The present study reveals the molecular details of the interaction between a new class of ubiquitin reader and ubiquitylated NCPs . Recently the cryo-EM derived structure of a complex of a ubiquitylated histone reader , 53BP1 , and H2AK15 ubiquitylated-NCP was reported , revealing the basis for a separate class of interaction ( Wilson et al . , 2016 ) . In the case of 53BP1 , the interaction is established via the involvement of a pair of arginine residues flanking the ubiquitin conjugation position in the H2A N-terminal tail , R11 and R17 , that straddle the DNA to optimally position ubiquitin for contact with the ubiquitylation dependent recruitment motif ( UDR ) of 53BP1 ( Wilson et al . , 2016 ) . Modeling of the UDR region also revealed contact with the acidic patch via a single arginine residue ( R1627 ) ( Wilson et al . , 2016 ) . The acidic patch surface is a critical component of all chromatin factors bound to the NCP ( Barbera et al . , 2006; McGinty et al . , 2014; Makde et al . , 2010; Armache et al . , 2011; Morgan et al . , 2016; Kato et al . , 2013 ) , where , at a minimum , one anchoring arginine binds a cavity generated by H2A E60 , D89 and E91 side chains . In the case of RNF169 , R700 acts as the anchoring arginine , with R689 making contacts of approximately equal importance to the overall stability of the complex . While the topology of the acidic patch can accommodate interactions with a variety of structural types , RNF169 , similar to HMGN2 and CENP-C ( Kato et al . , 2011a , Kato et al . , 2013 ) , lacks defined secondary structure when bound to the NCP surface . Ubiquitylation of H2A on K13 and K15 by RNF168 is necessary for the recruitment of downstream repair factors such as RNF169 , 53BP1 and BRCA1 ( Jackson and Durocher , 2013 ) . The ability of RNF168 to specifically bind its own mark provides both a mechanism for facilitating the amplification of these marks as well as spatial and temporal control over its catalytic activity , ensuring the correct , stepwise execution of the DSB response . Notwithstanding the similarities of the UDM2 domains of RNF168 and RNF169 , the affinity of RNF169 for K13 or K15 ubNCPs is higher ( Figure 1—figure supplement 1 ) . This may provide an additional level of control on the catalytic activity of RNF168 since as RNF169 accumulates at DSBs , in an RNF168-dependant manner , it can outcompete RNF168 for H2AK13/K15 ubiquitylated NCPs to maintain the ubiquitin signal within a defined region of chromatin surrounding DSBs . A complete understanding of the interplay between the RNF168/169 ubiquitin readers and downstream repair factors 53BP1 and BRCA1 , and related implications for the cellular response to DNA damage , are important outstanding questions . In this report , we have presented an ensemble of RNF169 ( UDM2 ) -H2AK13Cub-NCP structures illustrating the inherent dynamics of the complex , while revealing the residue-specific contacts that impart selectivity . Our model shows how relatively weak interactions work synergistically to enable the selection of a specific type of ub-NCP among the diverse array of ubiquitylated chromatin sites in the nucleus . LANA1-23 peptide , ( Biotin-LC-MAPPGMRLRSGRSTGAPLTRGSY ) and the non-binding LANA1-23 LRS mutant peptide , ( Biotin-LC-MAPPGMRAAAGRSTGAPLTRGSY ) ( Figure 1—figure supplement 1 ) were synthesized by BioBasic . 153 bp 601 DNA ( Lowary and Widom , 1998 ) was prepared from a 32-copy plasmid ( a gift from Dr . Tom Muir’s lab ) , transformed into DH5α Escherichia coli cells and grown overnight in LB/ampicillin media at 37°C . Following harvest by centrifugation , Giga prep plasmid purification kits ( Qiagen , cat . 12191 ) were used to purify the 32-copy plasmid . Digestion using EcoRV ( 0 . 5units/μg plasmid ) in NEB buffer3 was carried out for 20 hr at 37°C . The resulting solution of liberated 153 bp fragments was treated with 0 . 3375 vol equivalents of fresh 40% PEG-600 and 0 . 15 vol equivalents of 5 M NaCl solution and incubated at 4°C for 1 hr to precipitate the vector backbone . Following centrifugation at 14 , 000 rpm for 30 min at 4°C , 2 . 5 vol equivalents of 100% ethanol was added to the supernatant and left at −20°C overnight to precipitate the 153 bp 601 DNA fragments . Following centrifugation at 14 , 000 rpm for 30 min at 4°C , the precipitated DNA pellet was washed once with cold 70% ethanol and resuspended in 10 mM Tris pH 8 . The 153 bp 601 DNA fragments were further purified using a HiTrap DEAE-FF column ( GE Healthcare ) equilibrated in 10 mM Tris pH 8 and eluted using a linear gradient up to 1M KCl over 9 column volumes . Fractions containing DNA were pooled , concentrated and the concentration of KCl was adjusted to 2 M for subsequent NCP reconstitution . Purified ubiquitin was prepared for conjugation by the addition of fresh 1 , 4-dithiothreitol ( DTT ) at 5 mM concentration to ensure all intermolecular disulfide bonds were reduced . Subsequent removal of DTT was achieved using PD-10 desalting cartridges ( GE Healthcare ) . The eluate was flash frozen with liquid nitrogen and subsequently lyophilized . Lyophilized H2AK13C ( ~5 mg ) was resuspended in 1 mL water with 5 mM tris ( 2-carboxyethyl ) phosphine ( TCEP ) and subsequently activated by the addition of 10-fold molar excess of 2 , 2’-dithiobis ( 5-nitropyridine ) ( DTNP ) dissolved in 2 mL of acetic acid . The reaction was allowed to proceed overnight at room temperature and verified by ESI-MS . The activated product ( H2A K13C-DTNP ) was dialyzed extensively into water to remove unused DTNP , followed by gel filtration using a HighLoad 10/300 superdex S75 gel filtration column ( GE Healthcare ) equilibrated with 6 M guanidine hydrochloride , 150 mM NaCl , 50 mM TRIS pH 6 . 9 . Following degassing of the H2A K13C-DTNP solution lyophilized UbG76C was added at a 2:1 molar excess to the histone solution and gently agitated for 1 hr to allow for the completion of the reaction . The final product was verified with ESI-MS . The procedure is outlined schematically in Figure 3—figure supplement 1A . Purified H2B , H3C110S , H4 and H2A K13C-Ub were combined at equimolar ratios and refolded into octamers , that were subsequently purified and reconstituted into NCPs as previously described ( Dyer et al . , 2004 ) with the exception that β-mercaptoethanol was not added to any buffers . Microscale test NCP reconstitution reactions ( 50 μL , 7 μM DNA ) were used to determine the optimal stoichiometry of Ub-octamers and DNA . The quality and purity of the resulting NCPs were checked using 5% native PAGE and ESI-MS . Nucleosomes were ubiquitinated by incubating 2 . 5 µg recombinant mononucleosomes with 30 nM E1 ( Uba1 ) , 1 . 4 mM UBCH5a , 4 mM RNF168 ( 1–113 ) ( Figure 1C and D , Figure 1—figure supplement 1 , Figure 3—figure supplement 1C ) or BMI1–RING1B complex ( Figure 1D ) , 11 mM ubiquitin ( Boston Biochem ) and 4 mM ATP in a buffer containing 50 mM Tris-HCl , pH7 . 5 , 100 mM NaCl , 10 mM MgCl2 , 1 mM ZnOAc and 1 mM DTT at 30°C for 2 hr . For Western blotting shown in Figure 1C and D , Figure 1—figure supplement 1 and Figure 3—figure supplement 1C we used the following primary antibodies: rabbit anti-H3 ( Ab1791 , Abcam – RRID:AB_302613 ) , mouse anti-MBP ( E8032 , NEB – RRID:AB_1559732 ) , rabbit anti-H2A ( raised against amino acid residues 100–130 ) ( Fradet-Turcotte et al . , 2013 ) , and rabbit anti-H2A ( targeting 719 the acidic patch , 07–146 , Millipore – RIDD:AB_310394 ) . Peroxidase-affiniPure goat anti-rabbit IgG ( 111 035 144 , Jackson Immuno Research – RRID:AB_2307391 ) and HRP-linked sheep anti-mouse IgG ( NA931 , GE Healthcare – RIDD:AB_772210 ) were used as secondary antibodies . For immunofluorescence and FACS analyses , as shown in Figure 5—figure supplement 2 , cells were stained for mouse anti-γ-H2AX ( clone JBW301 , Millipore– RIDD:AB_309864 ) and rabbit anti-53-BP1 ( sc-22760 , Santa Cruz – RIDD:AB_2256326 ) . The following antibodies were used as secondary antibodies in immunofluorescence microscopy: Alexa Fluor 555 anti-rabbit and AlexaFluor 647 goat anti-mouse ( Molecular Probes – RIDD:AB_141784 and RIDD:AB_141725 , respectively ) . DNA was counterstained with DAPI to trace the outline of nuclei . Human cell culture media were supplemented with 10% fetal bovine serum ( FBS ) and maintained at 37°C and at a 5% CO2 atmosphere . U-2-OS ( U2OS , RRID:CVCL_0042 ) WT were purchased from ATCC and cultured in McCoy’s medium ( Gibco ) . The cell line was tested to be negative for mycoplasma contamination and authenticated by STR DNA profiling . Plasmid transfections were carried out using Lipofectamine 2000 Transfection Reagent ( Invitrogen ) ( Figure 5—figure supplement 1 ) . Cells were grown on coverslips , irradiated with 2Gy and fixed with 2% ( w/v ) paraformaldehyde in PBS 1 hr post-irradiation . Cells were then processed for immunostaining as described previously ( Panier et al . , 2012; Escribano-Díaz et al . , 2013; Orthwein et al . , 2015 ) . Confocal images were taken using a Zeiss LSM780 laser-scanning microscope and a Leica SP5-II confocal microscope in standard scanning mode . ( Figure 5—figure supplement 1 ) NCP pull-downs ( Figure 1C and D , Figure 1—figure supplement 1 and Figure 3—figure supplement 1C ) were performed in a total volume of 100 µL by using 15–20 µL ubiquitination reaction ( see Catalytic ubiquitylation of NCPs above ) , 2 , 4 or 8 µg MBP-protein coupled to amylose resin ( New England Biolabs ) , in pull-down buffer ( 50 mM Tris-Cl pH 7 . 5 , 150 mM NaCl , 1 mM DTT , 0 . 05% NP-40 , 0 . 1% BSA ) . Pull-down reactions were incubated for 2 hr at 4°C . Pull-downs were then washed three times with 0 . 75 mL of the pull-down buffer plus 0 . 1% BSA and eluted in Laemmli SDS–PAGE sample buffer for analysis by immunoblotting . Pull-downs presented in Figure 3—figure supplement 2B were eluted in Laemmli SDS–PAGE without DTT to preserve the integrity of chemically labeled ubK13C- or ubK15C-H2A NCPs . For the competition assay ( Figure 1—figure supplement 1 ) the pull-down was performed as normal , with the addition of LANA peptide , MBP-RNF168 ( UDM2 ) or MBP-RNF169 ( UDM2 ) during the incubation with ubiquitylated NCP . Circular dichroism ( CD ) spectra and temperature melts of RNF169 ( UDM2 ) without and with the DAAA N-terminal helix extension ( Figure 2—figure supplement 1B ) were acquired using a Jasco J-815 CD Spectrometer ( Jasco , Inc . ) and a 0 . 1 cm path length cuvette . CD spectra were collected as an average of 3 scans between 190 and 240 nm using a 20 nm/min scanning rate , 8 s response time and protein concentrations in the range of 50–65 μM . For temperature melts elipticity at 222 nm was monitored over a temperature range of 10°C to 70°C with a temperature slope of 1°C/min and protein concentrations were in the 25–35 μM range . Raw data sets were corrected for buffer contributions and converted to percent helicity as previously described ( Sommese et al . , 2010; Chen et al . , 1974 ) . All NMR experiments on NCPs were performed at 45°C using 14 . 0 T Varian Inova or Bruker Avance III HD spectrometers equipped with cryogenically cooled pulse-field gradient triple-resonance probes . The assignment of RNF169 ( UDM2 ) and titrations of free ubiquitin with RNF169 ( UDM2 ) were performed at 35°C and 45°C , respectively , on a 11 . 7 T Varian Inova spectrometer with a room temperature pulse-field gradient triple-resonance probe . The NMR buffer for H2A K13Cub-NCP samples contained 100 mM NaCl , 0 . 05% azide , 0 . 5% trifluoroethanol and 20 mM sodium phosphate pH 6 , 99 . 9% D2O , with sample concentrations in the range of 50–100 uM in NCP , as determined by A260 measurement of DNA . The NMR buffer used for the assignment of RNF169 ( UDM2 ) contained 100 mM NaCl , 0 . 05% azide , and 20 mM sodium phosphate pH 6 , 90% H2O/10% D2O with 0 . 62 mM protein concentration . All NMR data were processed and analyzed using the suite of programs provided in NMRPipe/NMRDraw and NMRviewJ software packages ( Delaglio et al . , 1995; Johnson and Blevins , 1994 ) , with backbone assignments carried out using CCPnmr ( Vranken et al . , 2005 ) . Translational diffusion coefficients ( Figure 3—figure supplement 1B ) were measured by recoding a series of 1D 13C-edited spectra at 25°C using a pulse sequence analogous to an 15N-edited experiment published previously ( Choy et al . , 2002 ) , with the 15N pulses exchanged for 13C pulses . After initial gradient encoding of the magnetization a constant-delay diffusion element of 150 ms for free ubiquitin ( 8 . 9 kDa ) or 200 ms for ¼ proteasome ( 180 kDa ) and H2AK13Cub-NCP ( 235 kDa ) was employed . The resulting 1H methyl signal was integrated to quantify intensities as a function of gradient strength . Diffusion constants were obtained by nonlinear least square fits of peak intensities as a function of the square of the gradient strength to the relation I = Ioexp ( -aDG2 ) where I and Io are the integrated peak intensities in the presence and absence of the gradient G , respectively , D is the diffusion constant and a is a constant comprised of experimental parameters . Backbone resonance assignments of RNF169 ( UDM2 ) ( Figure 2 ) were completed using 2D 1H-15N HSQC and 3D HNCACB , HNCACO , HNCO and CBCA ( CO ) NH experiments ( Sattler et al . , 1999; Cavanagh , 2007; Muhandiram and Kay , 1994 ) , with side-chain assignments obtained using 3D H ( C ) ( CO ) NH-TOCSY and ( H ) C ( CO ) NH-TOCSY experiments ( Logan et al . , 1993; Grzesiek and Bax , 1992 ) . Stereospecific assignments of leucine and valine residues were achieved as previously described ( Neri et al . , 1989 ) . Briefly , RNF169 ( UDM2 ) was prepared in 100% H2O M9 minimal media supplemented with 10% [1H , 13C]-glucose/90% [1H , 12C]-glucose; subsequent analysis of CT-HSQC spectra ( Vuister and Bax , 1992; Santoro and King , 1992 ) produced assignments of leucine δ1/δ2 and valine γ1/γ2 resonances . 1H-13C HMQC spectra of nucleosomes were recorded exploiting a methyl-TROSY effect to obtain high quality spectra ( Tugarinov et al . , 2003; Wand et al . , 1996 ) . Assignments of all ILV methyl resonances within the histones and ubiquitin were transferred from those previously published ( Kato et al . , 2011b; Wand et al . , 1996 ) . Chemical shift perturbations were measured using 1H-13C HMQC spectra of free H2AK13Cub-NCP and H2AK13Cub-NCP saturated with 2 . 5-fold excess RNF169 ( UDM2 ) . Weighted CSPs of ILV residues were calculated according to: ( 1 ) CSP= ΔδH , i2+ΔδC , i2⋅wi where Δδi is the difference in chemical shift between the free and bound states ( ppm ) for a given isoleucine , leucine or valine resonance i , and the chemical shift weighting factor wi was set to σH , iσC , i ( ~0 . 16–0 . 18 ) , where σi is the standard deviation of deposited chemical shifts for isoleucine , leucine and valine methyl resonances in the Biological Magnetic Resonance Data Bank ( BMRB , http://www . bmrb . wisc . edu ) for atom i ( Kato et al . , 2011b ) . Titrations ( Figure 3A and B and Figure 3—figure supplement 2 ) were carried out by increasing the ratios of [RNF169]/[Ub] ( [RNF169]/[H2AK13Cub-NCP] ) from 0 to 42 in a series of 10 1H-13C CT-HSQC experiments ( from 0 to 2 . 6 over a series of 16 1H-13C HMQC data sets ) . The titrations were followed via 13C Ub that was either uniformly labeled ( titration of Ub ) or ILV-methyl labeled ( titration of K13Cub-NCP ) . For the titration of 13C-labeled RNF169 ( UDM2 ) ( Figure 3—figure supplement 3 ) the ratio of [ub]/[RNF169] was increased from 0 to 6 in a series of 1H-13C CT-HSQC experiments at 35°C . Note that there are two equivalents of Ub for each NCP and we have assumed independent binding of RNF169 to each Ub site . KD values were extracted from nonlinear least square fits of the resulting binding isotherms that were obtained by extracting chemical shifts in either 13C or 1H dimensions of 1H-13C correlation plots via: ( 2 ) Δδ′= ΔδMAX ′ [L]T+[P]T+Kd− ( [L]T+[P]T+ Kd ) 2−4[P]T[L]T2[P]T where [P]T and [L]T are the total concentration of Ub and RNF169 , respectively , Δδ′ is the chemical shift change ( relative to the RNF169-free spectrum ) at each titration point , and ΔδMAX′ is the difference between free and bound chemical shifts . Positions of individual peaks in fast exchange on the NMR chemical shift timescale were fit and reported errors in KD correspond to one standard deviation of these values . Kinetic parameters for the RNF169 + H2AK13Cub-NCP binding reaction ( Figure 3C ) were obtained via line-shape analysis using a KD value of 24 ± 7 μM obtained from chemical shift titration data . Experimental line-shapes ( 13C dimension ) were extracted for I44δ1 of ubiquitin for a range of [RNF169]/[H2AK13Cub-NCP] values and fit using scripts written in MATLAB ( MathWorks Inc . ) , as previously described ( Tugarinov and Kay , 2003 ) . Intrinsic transverse relaxation rates were estimated from linewidths in 1H-13C HMQC spectra and were not used as fitting parameters . Peak intensities for each titration point ( each [RNF169]/[H2AK13Cub-NCP] ratio ) were scaled to account for differential line broadening in the 1H dimension that can affect intensities of extracted 13C traces . Fitted parameters included kon and an adjustable coefficient for each titration point , as described above . ILV-methyl group dynamics ( Figure 7D ) have been measured for ubiquitin in H2AK13Cub-NCP whereby the build-up of methyl 1H triple quantum coherence is quantified as described previously ( Sun et al . , 2011 ) . Data sets were recorded in the presence and absence of RNF169 ( UDM2 ) , at 45°C . Relaxation delay values of 1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 , 9 , 10 , 12 and 14 ms were used and peak intensities corresponding to the build-up of triple quantum coherences ( Ia ) and the evolution of single quantum coherences ( Ib ) quantified from 2D spectra at each relaxation delay . Values of the product of the square of the methyl axis order parameter , S2 , and the tumbling time of the complex , τC , ( S2τC ) were extracted from nonlinear least square fits of build up curves of Ia/Ib as a function of relaxation delay ( Sun et al . , 2011 ) . Distances connecting ubiquitin and RNF169 in an RNF169 ( UDM2 ) – ubiquitin complex were obtained by recording a 1H-1H NOESY data set ( mixing time of 200 ms , 20°C ) as a first step in generating the structure of the RNF169 ( UDM2 ) - H2AK13Cub-NCP complex ( Figure 6B ) . The sample was ILV-methyl labeled in ubiquitin and uniformly 13C labeled in RNF169 ( UDM2 ) and was prepared using an approximate 8-fold excess of RNF169 over ubiquitin . NOEs between A673 of RNF169 and ubiquitin are particularly important as A673 has been shown to be a central player in the interaction in a crystal structure of the Rabex MIU-ub complex ( Penengo et al . , 2006 ) . Holey EM grids were prepared by nanofabrication with arrays of 500–800 nm holes as described ( Marr et al . , 2014 ) , with the alteration of gold evaporation onto the grids as a specimen support , rather than carbon ( Russo and Passmore , 2014 ) . For initial screening H2AK13Cub-NCP and RNF169 ( UDM2 ) were incubated at a 1:2 . 5 molar ratio in 10 mM Tris-Cl pH 7 . 5 , 200 mM KCl , 1 mM EDTA and differentially PEG precipitated as described ( Wilson et al . , 2016 ) ; the sample was diluted to 50 mM KCl immediately prior to grid preparation . The low-salt complexes were applied to grids and allowed to equilibrate for 5 s in a FEI Vitrobot and blotted for 10 s prior to freezing in a liquid ethane/propane mixture ( 1:1 v/v ) . Grids were stored in liquid nitrogen , prior to transfer to a Gatan 626 cryotransfer specimen holder . Samples were imaged with a FEI F20 electron microscope , equipped with a field emission gun and operating at 200 kV . Movies were acquired manually in counting mode with a Gatan K2 Summit direct detector device camera using a calibrated magnification of 34 , 483× , resulting in a physical pixel size corresponding to 1 . 45 Å . During movie acquisition the sample was exposed to 1 . 2 electrons/Å2/frame and a total exposure of 36 electrons/Å2 on the specimen . Image acquisition and data analysis revealed that only a subset of RNF169 ( UDM2 ) was bound to H2AK13Cub-NCP under the preliminary screening conditions . 175 movies were acquired for this preliminary dataset . The formation of RNF169 ( UDM2 ) -ubNCP complexes was optimized by incubating a 1:5 molar ratio of H2AK13Cub-NCP to RNF169 ( UDM2 ) in 10 mM tris pH 6 . 8 , 30 mM KCl and 1 mM EDTA followed by differential PEG precipitation as described ( Wilson et al . , 2016 ) , with the exception that a higher final PEG-4000 concentration of 9% ( v/v ) was used . A total of 867 movies were acquired for this sample consisting of RNF169 ( UDM2 ) -bound ubNCPs . Individual frames in a movie stack were aligned and averaged using the programs alignframes_lmbfgs and shiftframes ( Rubinstein and Brubaker , 2015 ) . The contrast transfer function ( CTF ) was calculated from the averaged frames using CTFFIND4 ( Rohou and Grigorieff , 2015 ) . Manual inspection of micrographs and their corresponding power spectra was performed in Relion 1 . 3 ( Zhao et al . , 2015 ) . Poor micrographs with ice contamination were discarded . Particle selection , based on manually selected templates , was performed in Relion 1 . 3 . A total of 42 , 343 and 301 , 275 particle images were selected for the ubNCP and RNF169 ( UDM2 ) -ubNCP maps , respectively . After particle image extraction , beam induced particle motion between frames was corrected with alignparts_lmbfgs ( Rubinstein and Brubaker , 2015 ) . A previously measured magnification anisotropy from the Toronto F20 electron microscope was corrected ( Zhao et al . , 2015 ) . Extracted particle images were subject to 2D classification in Relion 1 . 3 and high-resolution class averages were selected for 3D classification ( Figure 7—figure supplement 1B ) . A low pass filtered model of NCP based on PDB: 1KX5 ( Davey et al . , 2002 ) was used as an initial template for 3D classification . For the preliminary dataset of free ubNCP , 3D classification was performed using four classes . One class , populated with 60% of the total particles ( 11 , 063 particles ) was refined further to yield the 8 . 1 Å H2AK13Cub-NCP structure ( Figure 7A and C ) . For the dataset of RNF169 ( UDM2 ) -bound NCPs , 3D classification was performed with five classes . Particle images from one class populated with 35% of the total particles ( 31 , 760 particles ) showed high-resolution features and was refined further ( Figure 7—figure supplement 1C ) . The RNF169 ( UDM2 ) -ubNCP refined map was sharpened in Relion 1 . 3 with a B-factor of −200 . Global resolution estimates were determined using the FSC = 0 . 143 criterion after a gold-standard refinement ( Figure 7—figure supplement 1D and E ) ( Rosenthal and Henderson , 2003 ) . Local resolution was estimated with ResMap ( Kucukelbir et al . , 2014 ) . Calculations with Relion 1 . 3 were performed using the Hospital for Sick Children high performance computing facility . An atomic resolution model of the MIU2-ubiquitin complex ( extending from K662 to N682 of RNF169 , see Figure 2A ) was generated by using the HADDOCK modeling program ( Figure 6C ) ( Dominguez et al . , 2003 ) . The structure of ( isolated ) ubiquitin that was used in the docking procedure was taken from an X-ray model of a complex of the Rabex MIU with ubiquitin ( PDB 2C7M , chain B ) ( Penengo et al . , 2006 ) . The structure of the MIU2 motif for the HADDOCK calculations was built as a homology model based on the Rabex MIU , corresponding to chain A of PDB 2C7M using the program MODELLER ( Fiser and Sali , 2003 ) . Since the C-terminal LRM region of RNF169 is largely disordered ( see below ) only the N-terminal MIU2 element was included at this stage . In order to carry out the docking study , a list of ‘active’ and ‘passive’ residues was defined as required by HADDOCK . Residues L672 , A673 , L676 of RNF169 and residues L8 , I44 , V70 of ubiquitin were defined as ‘active’ residues based on NMR titration results ( Figure 6A , Figure 3A , Figure 3—figure supplement 3 ) and solvent accessibility criteria , while ‘passive’ residues were calculated using the standard procedure in HADDOCK . It is worth noting that NMR titration data indicate that several additional ubiquitin residues ( I30 , I36 , I61 , L67 ) have CSPs upon titration with MIU2 and these could potentially be involved in binding as well . They were not included as ‘active’ residues , however , since the methyl groups of these Ile/Leu probes were buried inside the hydrophobic core , with low solvent accessibilities . We employed a minimum of 20% relative solvent accessibility as a cut off for inclusion as an ‘active’ residue . This is reduced from the recommended value of 40% used for backbone atoms in Haddock calculations , as methyl groups are less exposed . Additional restraints , measured from a 200 ms mixing time NOE spectrum of an RNF169 ( UDM2 ) – ubiquitin complex , were used . These were added in a qualitative manner as upper bound distances of 7 Å between methyl protons of A673 ( RNF169 ) and I44δ1 , L8δ1 , L8δ2 ( ubiquitin ) . Docking was performed using a standard HADDOCK protocol , and the number of structures selected after the it0 , it1 and itw stages were 1000 , 200 , and 200 respectively . The final 200 structures were subject to cluster-analysis . Notably , the largest cluster has the lowest average energy and the lowest HADDOCK score in this cluster ( also lowest score of all final structures in any cluster ) . The HADDOCK model reproduces the binding interface in the X-ray structure of the Rabex MIU-ubiquitin complex ( PDB 2C7M ) , with the key feature that A673 of MIU2 is located within the hydrophobic binding pocket formed by L8 , I44 and V70 of ubiquitin . This structure was selected for further MD simulation studies , as described below . A list of synthetic NOE restraints was constructed based on this representative structure , and these were imposed during the MD simulations in order to maintain the HADDOCK generated MIU2-ubiquitin complex . The total number of restraints was 331 , which included atomic pairs in the MIU2-ubiquitin complex less than 5 Å . It is worth noting that we have repeated the calculations described above by modifying the upper distance to 10 Å between methyl protons of A673 and neighboring protons on ubiquitin . The structures within the lowest energy cluster were the same as those obtained from the original calculation using 7 Å . The simulations described here , summarized schematically in Figure 8—figure supplement 1 , were performed using the Parmbsc1 ( Ivani et al . , 2016 ) and Amberff99SB* ( Best and Hummer , 2009 ) force fields for modeling the DNA and the protein component of the RNF169 ( UMD2 ) - H2AK13ub-NCP complex , respectively , along with the TIP3P water model ( Jorgensen et al . , 1983 ) . All simulations were carried out using the software package GROMACS ( Pronk et al . , 2013 ) modified with PLUMED2 , an open source library for free energy calculations ( Tribello et al . , 2014 ) and Almost ( Fu et al . , 2014 ) , a plugin for NMR chemical shift restraints . A time step of 2 fs was used together with LINCS constraints ( Hess et al . , 2008 ) . The van der Waals interactions were implemented with a cutoff of 1 . 2 nm and long-range electrostatic effects were treated with the particle mesh Ewald method and a cut-off of 0 . 9 nm ( Darden et al . , 1993 ) .	Inside cells , genetic information is encoded by molecules of DNA . It is important for a cell to quickly identify and repair any damage to DNA to prevent harmful changes in the genetic information . In humans and other animals failures in DNA repair can lead to cancer and other diseases . A molecule of DNA is made of two strands that twist together to form a double helix . Most of the DNA in an animal cell is organised by proteins called histones . Groups of eight histones are wrapped with DNA to form structures called nucleosomes . If both strands of a DNA double helix break in the same place , this leads to a molecule called ubiquitin being attached to a histone called H2A within a nucleosome to mark the position of the damage . This promotes DNA repair by attracting another protein called RNF169 to bind to the nucleosome . The precise location of the ubiquitin molecule on histone H2A is important because ubiquitin molecules act as signals for a variety of different processes when attached to specific positions on histones and other proteins . For example , ubiquitin molecules attached to some sites on histones can alter how the cell uses the genetic information contained within the nucleosome . However , it is not clear how the number and precise locations of ubiquitins on histones can produce such different signals . Kitevski-LeBlanc , Fradet-Turcotte et al . investigated why RNF169 is only attracted to nucleosomes when ubiquitin is attached to a particular site on histone H2A following damage to DNA . The experiments reveal that two regions of RNF169 known as arginine motifs play an important role in controlling when the protein binds to nucleosomes . These arginine motifs – which are next to the region of the protein that binds to ubiquitin – identify the position of the ubiquitin on H2A by making contact with an “acidic” patch on the surface of the nucleosome . These findings show that the combination of RNF169 binding to both the ubiquitin on H2A and an acidic patch on the nucleosome ensure that this protein only promotes DNA repair when and where it is needed . This acidic patch is involved in regulating the binding of various other proteins to nucleosomes . Understanding how cells interpret the signals produced by ubiquitin binding to proteins will help us to understand how disrupting these signals can contribute to cancer and other diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2017	The RNF168 paralog RNF169 defines a new class of ubiquitylated histone reader involved in the response to DNA damage
Following fertilization , the genomes of the germ cells are reprogrammed to form the totipotent embryo . Pioneer transcription factors are essential for remodeling the chromatin and driving the initial wave of zygotic gene expression . In Drosophila melanogaster , the pioneer factor Zelda is essential for development through this dramatic period of reprogramming , known as the maternal-to-zygotic transition ( MZT ) . However , it was unknown whether additional pioneer factors were required for this transition . We identified an additional maternally encoded factor required for development through the MZT , GAGA Factor ( GAF ) . GAF is necessary to activate widespread zygotic transcription and to remodel the chromatin accessibility landscape . We demonstrated that Zelda preferentially controls expression of the earliest transcribed genes , while genes expressed during widespread activation are predominantly dependent on GAF . Thus , progression through the MZT requires coordination of multiple pioneer-like factors , and we propose that as development proceeds control is gradually transferred from Zelda to GAF . Pronounced changes in cellular identity are driven by pioneer transcription factors that act at the top of gene regulatory networks . While nucleosomes present a barrier to the DNA binding of many transcription factors , pioneer factors can bind DNA in the context of nucleosomes . Pioneer-factor binding establishes accessible chromatin domains , which serve to recruit additional transcription factors that drive gene expression ( Zaret and Mango , 2016; Iwafuchi-Doi and Zaret , 2014; Zaret and Carroll , 2011 ) . These unique characteristics of pioneer factors enable them to facilitate widespread changes in cell identity . Nonetheless , cell-fate transitions often require a combination of pioneering transcription factors to act in concert to drive the necessary transcriptional programs . Indeed , the reprogramming of a specified cell type to an induced pluripotent stem cell requires a cocktail of transcription factors , of which Oct4 , Sox2 , and Klf4 function as pioneer factors ( Takahashi and Yamanaka , 2006; Takahashi et al . , 2007; Chronis et al . , 2017; Soufi et al . , 2015; Soufi et al . , 2012 ) . Despite the many examples of multiple pioneer factors functioning together to drive reprogramming , how these factors coordinate gene expression changes within the context of organismal development remains unclear . Pioneer factors are also essential for the reprogramming that occurs in the early embryo . Following fertilization , specified germ cells must be rapidly and efficiently reprogrammed to generate a totipotent embryo capable of differentiating into all the cell types of the adult organism . This reprogramming is initially driven by mRNAs and proteins that are maternally deposited into the oocyte . During this time , the zygotic genome remains transcriptionally quiescent . Only after cells have been reprogrammed is the zygotic genome gradually activated . This maternal-to-zygotic transition ( MZT ) is broadly conserved among metazoans and essential for future development ( Schulz and Harrison , 2019; Vastenhouw et al . , 2019 ) . Activators of the zygotic genome have been identified in a number of species ( zebrafish – Pou5f3 , Sox19b , Nanog; mice – Dux , Nfy; humans – DUX4 , OCT4; fruit flies – Zelda ) , and all share essential features of pioneer factors ( Schulz and Harrison , 2019; Vastenhouw et al . , 2019 ) . Early Drosophila development is characterized by 13 , rapid , synchronous nuclear divisions . Zygotic transcription gradually becomes activated starting about the eighth nuclear division , and widespread transcription occurs at the 14th nuclear division when the division cycle slows and the nuclei are cellularized ( Schulz and Harrison , 2019; Vastenhouw et al . , 2019 ) . The transcription factor Zelda ( Zld ) is required for transcription of hundreds of genes throughout zygotic genome activation ( ZGA ) and was the first identified global genomic activator ( Liang et al . , 2008; Harrison et al . , 2011; Nien et al . , 2011 ) . Embryos lacking maternally encoded Zld die before completing the MZT ( Liang et al . , 2008; Harrison et al . , 2011; Nien et al . , 2011; Fu et al . , 2014; Staudt et al . , 2006 ) . Zld has the defining features of a pioneer transcription factor: it binds to nucleosomal DNA ( McDaniel et al . , 2019 ) , facilitates chromatin accessibility ( Schulz et al . , 2015; Sun et al . , 2015 ) and this leads to subsequent binding of additional transcription factors ( Yáñez-Cuna et al . , 2012; Xu et al . , 2014; Foo et al . , 2014 ) . By contrast to the essential role for Zld in flies , no single global activator of zygotic transcription has been identified in other species . Instead , multiple transcription factors function together to activate zygotic transcription ( Schulz and Harrison , 2019; Vastenhouw et al . , 2019 ) . Work from our lab and others has implicated additional factors in regulating reprogramming in the Drosophila embryo . Specifically , the enrichment of GA-dinucleotides in regions of the genome that remain accessible in the absence of Zld and at loci that gain accessibility late in ZGA , suggest that a protein that binds to these loci functions with Zld to define cis-regulatory regions during the initial stages of development ( Schulz et al . , 2015; Sun et al . , 2015; Blythe and Wieschaus , 2016 ) . Three proteins , GAGA factor ( GAF ) , chromatin-linked adaptor for MSL proteins ( CLAMP ) , and Pipsqueak ( Psq ) are known to bind to GA-dinucleotide repeats and are expressed in the early embryo , implicating one or all these proteins in reprogramming the embryonic transcriptome ( Rieder et al . , 2017; Soruco et al . , 2013; Kuzu et al . , 2016; Biggin and Tjian , 1988; Bhat et al . , 1996; Soeller et al . , 1993; Lehmann et al . , 1998 ) . CLAMP was first identified based on its role in targeting the dosage compensation machinery , and it preferentially localizes to the X chromosome ( Larschan et al . , 2012; Soruco et al . , 2013 ) . Psq is essential for oogenesis ( Horowitz and Berg , 1996 ) , and it has been suggested to have a role as a repressor and in chromatin looping ( Gutierrez-Perez et al . , 2019; Huang et al . , 2002 ) . GAF , encoded by the Trithorax-like ( Trl ) gene , has broad roles in transcriptional regulation , including functioning as a transcriptional activator ( Farkas et al . , 1994; Bhat et al . , 1996 ) , repressor ( Mishra et al . , 2001; Busturia et al . , 2001; Bernués et al . , 2007; Horard et al . , 2000 ) , and insulator ( Ohtsuki and Levine , 1998; Wolle et al . , 2015; Kaye et al . , 2017 ) . Through interactions with chromatin remodelers , GAF is instrumental in driving regions of accessible chromatin both at promoters and distal cis-regulatory regions ( Okada and Hirose , 1998; Tsukiyama et al . , 1994; Xiao et al . , 2001; Tsukiyama and Wu , 1995; Fuda et al . , 2015; Judd et al . , 2021 ) . Analysis of hypomorphic alleles has suggested an important function for GAF in the early embryo in both driving expression of Ultrabiothorax ( Ubx ) , Abdominal B ( Abd-B ) , engrailed ( en ) , and fushi tarazu ( ftz ) and in maintaining normal embryonic development ( Bhat et al . , 1996; Farkas et al . , 1994 ) . Given these diverse functions for GAF , and that it shares many properties of a pioneer transcription factor , we sought to investigate whether it has a global role in reprogramming the zygotic genome for transcriptional activation . Investigation of the role of GAF in the early embryo necessitated the development of a system to robustly eliminate GAF , which had not been possible since GAF is essential for maintenance of the maternal germline and is resistant to RNAi knockdown in the embryo ( Bhat et al . , 1996; Bejarano and Busturia , 2004; Rieder et al . , 2017 ) . For this purpose , we generated endogenously GFP-tagged GAF , which provided the essential functions of the untagged protein , and used the deGradFP system to deplete GFP-tagged GAF in early embryos expressing only the tagged construct ( Caussinus et al . , 2012 ) . Using this system , we identified an essential function for GAF in driving chromatin accessibility and gene expression during the MZT . Thus , at least two pioneer-like transcription factors , Zld and GAF , must cooperate to reprogram the zygotic genome of Drosophila following fertilization . To investigate the role of GAF during the MZT , we used Cas9-mediated genome engineering to tag the endogenous protein with super folder Green Fluorescent Protein ( sfGFP ) at either the N- or C-termini , sfGFP-GAF ( N ) and GAF-sfGFP ( C ) , respectively ( Pédelacq et al . , 2006 ) . There are two protein isoforms of GAF ( Benyajati et al . , 1997 ) . Because the N-terminus of GAF is shared by both isoforms , the sfGFP tag on the N-terminus labels all GAF protein ( Figure 1—figure supplement 1A ) . By contrast , the two reported isoforms differ in their C-termini , and thus the C-terminal sfGFP labels only the short isoform ( Figure 1—figure supplement 1B ) . Whereas null mutants in Trithorax-like ( Trl ) , the gene encoding GAF , are lethal ( Farkas et al . , 1994 ) , both sfGFP-tagged lines are homozygous viable and fertile . Additionally , in embryos from both lines sfGFP-labeled GAF is localized to discrete nuclear puncta and is retained on the mitotic chromosomes in a pattern that recapitulates what has been previously described for GAF based on antibody staining in fixed embryos ( Figure 1A , Figure 1—figure supplement 1C; Raff et al . , 1994 ) . Together , these data demonstrate that the sfGFP tag does not interfere with essential GAF function and localization . To begin to elucidate the role of GAF during early embryogenesis , we determined the genomic regions occupied by GAF during the MZT . We hand sorted homozygous GAF-sfGFP ( C ) stage 3 ( nuclear cycle ( NC ) 9 ) and stage 5 ( NC14 ) embryos and performed chromatin immunoprecipitation coupled with high-throughput sequencing ( ChIP-seq ) using an anti-GFP antibody . While alternative splicing generates two GAF protein isoforms that differ in their C-terminal polyQ domain , in the early embryo only the short isoform is detectable ( Benyajati et al . , 1997 ) . We confirmed the expression of the short isoform in the early embryo by blotting extract from 0 to 4 hr AEL ( after egg laying ) N- and C-terminally tagged GAF embryos with an anti-GFP antibody ( Figure 1—figure supplement 1D ) . The long isoform was undetectable in extract from embryos of both sfGFP-tagged lines harvested 0–4 hr AEL , but was detectable at low levels in extract from sfGFP-GAF ( N ) embryos harvested 13–16 hr AEL . Thus , we conclude that during the MZT the C-terminally sfGFP-labeled short isoform comprises the overwhelming majority of GAF present in the GAF-sfGFP ( C ) embryos . We identified 3391 GAF peaks at stage 3 and 4175 GAF peaks at stage 5 ( Supplementary file 1 ) . To control for possible cross-reactivity with the anti-GFP antibody , we performed ChIP-seq on w1118 stage 3 and stage 5 embryos in parallel . Since there are no GFP-tagged proteins in the w1118 embryos , any peaks identified with the anti-GFP antibody would be the result of non-specific interactions and excluded from further analysis . No peaks were called in the w1118 dataset for either stage , confirming the specificity of the peaks identified in the GAF-sfGFP ( C ) embryos ( Figure 1—figure supplement 2A , B ) . Further supporting the specificity of the ChIP data , the canonical GA-rich GAF-binding motif was the most highly enriched sequence identified in ChIP peaks from both stages , and these motifs were centrally located in the peaks ( Figure 1B ) . Peaks were identified in the regulatory regions of several previously identified GAF-target genes , including the heat shock promoter ( hsp70 ) , even-skipped ( eve ) , Krüppel ( Kr ) , Ubx , and en ( Figure 1C; Biggin and Tjian , 1988; Gilmour et al . , 1989; Soeller et al . , 1988; Lee et al . , 1992; Read et al . , 1990; Kerrigan et al . , 1991 ) . Peaks were enriched at promoters , which fits with the previously defined role of GAF in establishing paused RNA polymerase ( Figure 1—figure supplement 2C; Lee et al . , 2008; Fuda et al . , 2015; Judd et al . , 2021 ) . There was a substantial degree of overlap between GAF peaks identified at both stage 3 and stage 5 . A total of 2955 peaks were shared between the two time points , representing 87% of total stage 3 peaks and 71% of total stage 5 peaks ( Figure 1C , D , Figure 1—figure supplement 2D ) . This demonstrates that GAF binding is established prior to widespread ZGA and remains relatively unchanged during early development , similar to what has been shown for Zld , the major activator of the zygotic genome ( Harrison et al . , 2011 ) . We compared GAF-binding sites we identified in the early embryo to previously identified GAF-bound regions in S2 cells , which are derived from 20 to 24 hr old embryos ( Fuda et al . , 2015 ) . Despite the difference in cell-type and antibody used , 56% ( 1906 ) of stage 3 peaks and 43% ( 1811 ) of stage 5 peaks overlapped peaks identified in S2 cells ( Figure 1C , D , Figure 1—figure supplement 2D ) . It was previously noted that GAF binding in 8–16 hr embryos was highly similar to GAF occupancy in the wing imaginal disc harvested from the larva ( Slattery et al . , 2014 ) , and our data indicate that this binding is established early in development prior to activation of the zygotic genome . Nonetheless , peaks unique to each tissue likely represent GAF-binding events as they are centrally enriched for GA-rich GAF-binding motifs ( Figure 1—figure supplement 3 ) . Thus , while a majority of GAF-binding sites are maintained in both the embryo and cell culture , a subset of GAF-binding sites is likely tissue specific . Early embryonic GAF is maternally deposited in the oocyte , and this maternally deposited mRNA can sustain development until the third instar larval stage ( Farkas et al . , 1994 ) . To investigate the role of GAF during the MZT therefore necessitated a system to eliminate this maternally encoded GAF . RNAi failed to successfully knockdown GAF in the embryo ( Rieder et al . , 2017 ) , and female germline clones cannot be generated as GAF is required for egg production ( Bhat et al . , 1996; Bejarano and Busturia , 2004 ) . To overcome these challenges , we leveraged our N-terminal sfGFP-tagged allele and the previously developed deGradFP system to target knockdown at the protein level ( Caussinus et al . , 2012 ) . The deGradFP system uses a genomically encoded F-box protein fused to a nanobody recognizing GFP , which recruits GFP-tagged proteins to a ubiquitin ligase complex . Ubiquitination of the GFP-tagged protein subsequently leads to efficient degradation by the proteasome . To adapt this system for efficient use in the early embryo , we generated transgenic flies in which the deGradFP nanobody fusion was driven by the nanos ( nos ) promoter for strong expression in the embryo 0–2 hr after fertilization ( Wang and Lehmann , 1991 ) . In embryos laid by nos-deGradFP; sfGFP-GAF ( N ) females all maternally encoded GAF protein is tagged with GFP and thus subject to degradation by the deGradFP nanobody fusion . These embryos are hereafter referred to as GAFdeGradFP . We verified the efficiency of the deGradFP system by imaging living embryos in which nuclei were marked by His2Av-RFP . GAFdeGradFP embryos lack the punctate , nuclear GFP signal identified in control embryos that do not carry the deGradFP nanobody fusion , indicating efficient depletion of sfGFP-GAF ( N ) ( Figure 2A ) . This knockdown was robust , as we failed to identify any NC10 - 14 embryos with nuclear GFP signal , and none of the embryos carrying both the deGradFP nanobody and the sfGFP-tagged GAF hatched ( Figure 2B ) . Based on live embryo imaging , the majority of embryos died prior to NC14 , indicating that maternal GAF is essential for progression through the MZT . We identified a small number of GFP-expressing , gastrulating escapers . It is unclear if these embryos had an incomplete knockdown of maternally encoded sfGFP-GAF ( N ) , or if a small percentage of embryos survived until gastrulation in the absence of GAF and that the GFP signal was the result of zygotic gene expression . Nonetheless , none of these embryos survived until hatching . Despite being able to maintain a strain homozygous for the N-terminal , sfGFP-tagged GAF , quantitative analysis revealed an effect on viability . Embryos homozygous for sfGFP-GAF ( N ) had only a 30% hatching rate ( Figure 2B ) . By contrast , embryos homozygous for GAF-sfGFP ( C ) hatched at a rate of 66% . To determine whether the differences in hatching rates were due to differences in protein expression levels , we performed in vivo fluorescent quantification of the GFP signal in nuclei of sfGFP-GAF ( N ) and GAF-sfGFP ( C ) homozygous stage 5 embryos ( 2–2 . 5 hr AEL ) . There was no statistically significant difference in GFP signal between the genotypes ( Figure 2—figure supplement 1 ) , suggesting that the lower hatching rate for sfGFP-GAF ( N ) embryos is unlikely to be caused by reduced protein levels . Nonetheless , all future experiments controlled for the effect of the N-terminal tag by using sfGFP-GAF ( N ) homozygous embryos as paired controls with GAFdeGradFP embryos . Having identified a dramatic effect of eliminating maternally expressed GAF on early embryonic development , we used live imaging to investigate the developmental defects in our GAFdeGradFP embryos . GAFdeGradFP embryos in which nuclei were marked by a fluorescently labeled histone ( His2Av-RFP ) were imaged through several rounds of mitosis . We observed defects such as asynchronous mitosis , anaphase bridges , disordered nuclei , and nuclear dropout ( Figure 2C , D; Videos 1 and 2 ) . Nevertheless , embryos were able to complete several rounds of mitosis without GAF before arresting . Nuclear defects became more pronounced as GAFdeGradFP embryos developed and approached NC14 . The arrested/dead embryos were often arrested in mitosis or had large , irregular nuclei ( Figure 2C ) , similar to the nuclear ‘supernovas’ in GAF-deficient nuclei reported in Bhat et al . , 1996 . Our live imaging allowed us to detect an additional nuclear defect: that in the absence of maternal GAF nuclei become highly mobile and , in some cases , adopt a ‘swirling’ pattern ( Video 2 ) . While it is unclear what causes in this phenotype , GAF is suggested to have a role in maintaining genome stability . Thus , loss of GAF may lead to activation of the DNA-damage response pathway , which can result in similar nuclear phenotypes ( Bhat et al . , 1996; Sibon et al . , 2000; Takada et al . , 2003 ) . As a control , we imaged sfGFP-GAF ( N ) homozygous embryos through several rounds of mitosis . These embryos proceeded normally through NC10-14 , demonstrating that the mitotic defects in the GAFdeGradFP embryos were caused by the absence of GAF and not the sfGFP tag ( Figure 2D; Video 3 ) . The defects in GAFdeGradFP embryos are reminiscent of the nuclear-division defects previously reported for maternal depletion of zld ( Liang et al . , 2008; Staudt et al . , 2006 ) and identify a fundamental role for GAF during the MZT . During the MZT , there is a dramatic change in the embryonic transcriptome as developmental control shifts from mother to offspring . Having demonstrated that maternal GAF was essential for development during the MZT , we investigated the role of GAF in regulating these transcriptional changes . We performed total-RNA seq on bulk collections of GAFdeGradFP and control ( sfGFP-GAF ( N ) ) embryos harvested 2–2 . 5 hr AEL , during the beginning of NC14 when the widespread genome activation has initiated . Our replicates were reproducible ( Figure 3—figure supplement 1 ) , allowing us to identify 1452 genes that were misexpressed in GAFdeGradFP embryos as compared to controls . Importantly , by using sfGFP-GAF ( N ) homozygous embryos as a control we have excluded from our analysis any genes misexpressed as a result of the sfGFP tag on GAF . Of the misexpressed genes 884 were down-regulated and 568 were up-regulated in the absence of GAF ( Figure 3A , Figure 3—figure supplement 2A ) . The gene encoding GAF , Trithorax-like , was named because it was required for expression of the homeotic genes Ubx and Abd-B ( Farkas et al . , 1994 ) . Our RNA-seq analysis identified 7 of the 8 Drosophila homeotic genes ( Ubx , Abd-B , adb-A , pb , Dfd , Scr , and Antp ) down-regulated in GAFdeGrad embryos . Additionally , many of the gap genes , essential regulators of anterior-posterior patterning , are down-regulated: giant ( gt ) , knirps ( kni ) , huckebein ( hkb ) , Krüppel ( Kr ) , and tailless ( tll ) ( Supplementary file 2 ) . Gene ontology ( GO ) -term analysis showed down-regulated genes were enriched for functions in system development and developmental processes as would be expected for essential genes activated during ZGA ( Figure 3—figure supplement 2B ) . GO-term analysis of the up-regulated genes showed weak enrichment for response to stimulus and metabolic processes ( Figure 3—figure supplement 2C ) . To determine whether GAF is functioning predominantly in transcriptional activation or repression , we used our stage 5 ChIP-seq data to determine the likely direct targets of GAF by assigning GAF-ChIP peaks to the nearest gene . We found that 45% ( 397 ) of the down-regulated genes were proximal to a GAF peak . By contrast , only 17% ( 99 ) of up-regulated genes are near GAF peaks , similar to the 15% of genes with unchanged expression levels that were proximal to GAF sites ( Figure 3A ) . The significant enrichment for GAF-binding sites proximal to down-regulated genes as compared to up-regulated supports a role for GAF specifically in transcriptional activation ( p<2 . 2×10−16 , log2 ( odds ratio ) =1 . 95 , two-tailed Fisher’s exact test ) . Genes activated late in NC14 , during widespread genome activation , with a proximal GAF-binding site are significantly more down-regulated than genes activated at the same time point but lacking a GAF-binding site ( Figure 3—figure supplement 2D ) . This analysis suggests that the down-regulated genes identified by RNA-seq are unlikely to be due to a general failure in activating zygotic gene expression , but rather specifically identify genes that depend on GAF for expression . Because zygotic gene expression is required for degradation of a subset of maternal mRNAs , if the genome is not activated , maternal transcripts are not properly degraded and therefore are increased in RNA-seq data ( Harrison et al . , 2011; Hamm et al . , 2017; Liang et al . , 2008 ) . Indeed , down-regulated transcripts were enriched for zygotically expressed genes , and up-regulated transcripts were largely maternally contributed ( p<2 . 2×10−16 , log2 ( odds ratio ) =4 . 21 , two-tailed Fisher’s exact test; Figure 3B ) . Thus , GAF is essential for transcriptional activation during the MZT and likely functions along with Zld to drive zygotic genome activation . Previous data defined a role for the additional GA-dinucleotide binding protein , CLAMP , in activating zygotic gene expression ( Larschan et al . , 2012; Rieder et al . , 2017; Urban et al . , 2017b; Soruco et al . , 2013 ) . Therefore , we compared genes down-regulated in GAFdeGradFP embryos to genes down-regulated in clamp-RNAi embryos 2–4 hr AEL ( Rieder et al . , 2017 ) . A total of 174 genes are down-regulated in both datasets , comprising 19 . 7% of total down-regulated GAF targets and 50 . 1% of total down-regulated CLAMP targets ( Figure 3—figure supplement 3A ) . While this demonstrates that a subset of genes requires both GAF and CLAMP for proper expression during ZGA , a majority of GAF-regulated genes only depend on GAF for expression , independent of CLAMP . GAF and CLAMP have similar , but not identical binding preferences ( Kaye et al . , 2018 ) , which is reflected in their partially , but not completely overlapping genome occupancy ( Figure 3—figure supplement 3B ) . These binding site differences likely explain their differential requirement during ZGA . Having identified that GAF was required for ZGA , we wanted to ensure that the effects were not due to changes in the levels of Zld , the previously identified activator of the zygotic genome . Immunoblots for Zld on extract from GAFdeGradFP and control ( sfGFP-GAF ( N ) ) 2–2 . 5 hr AEL ( stage 5 ) embryos confirmed that Zld levels were consistent between extracts ( Figure 3—figure supplement 3C ) . Therefore , the effects of GAF on genome activation are not due to a loss of Zld . Reciprocally , we performed immunoblots on 2–2 . 5 hr AEL embryo extract for sfGFP-GAF ( N ) in a background in which maternal zld was depleted using RNAi driven by matα-GAL4-VP16 ( Sun et al . , 2015 ) . We found that sfGFP-GAF ( N ) levels were unchanged upon zld knockdown compared to controls ( Figure 3—figure supplement 3D ) , demonstrating that neither GAF nor Zld protein levels are affected by the knockdown of the other factor . Based on the roles for Zld and GAF in activating the zygotic genome , we investigated whether these proteins were required to activate distinct or overlapping target genes . We compared genes down-regulated in GAFdeGradFP embryos to genes down-regulated when Zld was inactivated optogenetically throughout zygotic genome activation ( NC10-14 ) ( McDaniel et al . , 2019 ) . We identified 135 genes down-regulated in both datasets , comprising 42% of the total number of down-regulated genes dependent on Zld and 15% of the total down-regulated genes dependent on GAF ( Figure 3C ) . An even lower degree of overlap is observed when only direct targets are considered with 49 down-regulated targets shared between Zld and GAF , which have 232 and 397 direct targets , respectively . By contrast only 29 up-regulated genes were shared between the two datasets ( Figure 3—figure supplement 3E ) . Genes that required both factors for activation include the gap genes previously mentioned ( gt , kni , hkb , Kr , tll ) as well as genes involved in cellular blastoderm formation such as slow as molasses ( slam ) and bottleneck ( bnk ) . While Zld and GAF share some targets , they each are required for expression of hundreds of individual genes . Activation of the zygotic genome is a gradual process that initiates with transcription of a small number of genes around NC8 . Transcripts can therefore be divided into categories based on the timing of their initial transcription ( Li et al . , 2014; Lott et al . , 2011 ) . Previous data suggested that while Zld was required for activation of genes throughout the MZT , early genes were particularly sensitive to loss of Zld and that GAF might be functioning later ( Schulz et al . , 2015; Blythe and Wieschaus , 2016 ) . To determine when GAF-dependent genes were expressed during ZGA , we took all the genes that could be classified based on their timing of activation during the MZT ( as determined in Li et al . , 2014 ) and divided them based on their dependence on Zld and GAF for activation: those down-regulated in both GAFdeGradFP and ZldCRY2 embryos ( Both ) , those down-regulated only in ZldCRY2 embryos ( ZldCRY2 ) , and those down-regulated only in GAFdeGradFP embryos ( GAFdeGradFP ) ( Figure 3D ) . Genes activated by Zld , both those regulated by Zld alone ( ZldCRY2 ) and those regulated by both Zld and GAF ( Both ) were enriched for genes expressed early ( NC10-11 ) and Mid ( NC12-13 ) . By contrast , 73% of genes activated by GAF alone ( GAFdeGradFP ) were expressed either late ( early NC14 ) or later ( late NC14 ) ( p=1 . 3×10−14 , log2 ( odds ratio ) =3 . 13 , two-tailed Fisher’s exact test ) . This analysis supports a model in which GAF and Zld are essential activators for the initial wave of ZGA and that GAF has an additional role , independent of Zld , in activating widespread transcription during NC14 . To further delineate the relationship between Zld- and GAF-mediated transcriptional activation during the MZT , we determined whether GAF binding was dependent on Zld . Not only are a substantial subset of genes dependent on both Zld and GAF for wild-type levels of expression ( Figure 3C ) , but 42% of GAF-binding sites at stage 5 are also occupied by Zld ( Harrison et al . , 2011; Figure 4A , B ) . The GAF peaks that are co-bound with Zld include many of the strongest GAF peaks identified at stage 5 ( Figure 4—figure supplement 1A ) . Zld is known to facilitate the binding of multiple different transcription factors ( Twist , Dorsal , and Bicoid ) , likely by forming dynamic subnuclear hubs ( Yáñez-Cuna et al . , 2012; Xu et al . , 2014; Foo et al . , 2014; Mir et al . , 2018; Dufourt et al . , 2018; Yamada et al . , 2019 ) . Indeed , 19% of Dorsal sites depend on Zld for occupancy ( Sun et al . , 2015 ) . We therefore examined GAF localization upon depletion of maternal zld using RNAi driven in the maternal germline by matα-GAL4-VP16 in a background containing sfGFP-GAF ( N ) ( Sun et al . , 2015 ) . Immunoblot confirmed a nearly complete knockdown of Zld ( Figure 4C ) , and we verified that RNAi-treated embryos failed to hatch . To assess the depletion of Zld on the subnuclear localization of GAF , we imaged sfGFP-GAF ( N ) ; zld-RNAi and control ( sfGFP-GAF ( N ) ) embryos using identical acquisition settings . We observed no difference in puncta formation of GAF at the beginning of ZGA ( NC10 ) or late ZGA ( NC14 ) ( Figure 4D ) . We conclude that Zld is not required for GAF to form subnuclear puncta during the MZT . To more specifically determine the impact of loss of Zld on GAF chromatin occupancy , we performed ChIP-seq with an anti-GFP antibody on embryos expressing zld-RNAi and sfGFP-GAF ( N ) along with paired sfGFP-GAF ( N ) controls at 2–2 . 5 hr AEL ( stage 5 ) . Mouse H3 . 3-GFP chromatin was used as a spike-in to normalize for immunoprecipitation efficiency between samples . In control sfGFP-GAF ( N ) embryos , we identified 6373 peaks , and these largely overlapped with the peaks identified in the GAF-sfGFP ( C ) embryos; 91% of the 4175 peaks identified in the GAF-sfGFP ( C ) embryos overlap with the peaks identified for the N-terminally tagged GAF ( Figure 4—figure supplement 1B ) . This high degree of overlap indicates that our ChIP-seq experiments identified a robust set of high-confidence GAF-bound regions . To determine whether GAF requires Zld for binding , we analyzed GAF occupancy upon zld knockdown at this set of 3800 high-confidence peaks . The majority ( 3674 ) of these high-confidence GAF peaks were maintained in the zld-RNAi background and were bound at roughly equivalent levels when normalized to the spike-in control ( Figure 4E and Figure 4—figure supplement 2A ) . Using DESeq2 , we identified 126 GAF-bound regions that were significantly different between the sfGFP-GAF ( N ) controls and the zld-RNAi embryos ( Figure 4E ) : 101 sites that were decreased and 25 sites that were increased . To assess the impact of the loss of GAF on Zld binding , we completed the reciprocal experiment and performed ChIP-seq for Zld on GAFdeGradFP and control homozygous ( sfGFP-GAF ( N ) ) embryos at 2–2 . 5 hr AEL ( stage 5 ) . We identified a set of high-confidence Zld peaks , by overlapping the Zld peaks from our control ( sfGFP-GAF ( N ) ) embryos with previously published data for Zld at NC14 ( Harrison et al . , 2011 ) . Similar to GAF binding when Zld is depleted , the majority of the 6003 high-confidence Zld peaks were maintained when GAF was depleted ( Figure 4F ) . However , in the GAFdeGradFP ChIP data we observed a global reduction in Zld peak heights as compared to the control ( Figure 4—figure supplement 2B ) . Therefore , it is possible that GAF knockdown causes a global reduction in Zld binding . However , technical , rather than biological differences , may account for this overall decrease ( see Materials and methods ) . Among the high-confidence Zld peaks , DESeq2 identified 105 Zld-binding sites that differed in occupancy between GAFdeGradFP embryos and controls: 86 sites decreased and 19 sites increased . Because of the limited number of significantly different peaks identified upon removal of either Zld ( 3 . 3% of GAF peaks changed ) or GAF ( 1 . 7% of Zld peaks changed ) , we sought to determine whether there were global changes in the distribution of these factors . For this purpose , we ranked peaks in each dataset based on the number of reads per kilobase per million mapped reads ( RPKM ) and determined if the peak ranks were correlated between the mutant and control . We identified a high degree of correlation in peak rank for both comparisons ( Pearson correlation , r = 0 . 93 for GAF ChIP in sfGFP-GAF ( N ) and sfGFP-GAF ( N ) ; zld-RNAi embryos and r = 0 . 88 for Zld ChIP in sfGFP-GAF ( N ) and GAFdeGradFP embryos; Figure 4—figure supplement 2C , D ) . Thus , the overall distribution of binding sites is maintained for each factor in the absence of the other . At some loci , pioneer factors have been shown to function together to stabilize binding ( Donaghey et al . , 2018; Chronis et al . , 2017 ) . If either GAF or Zld stabilized genomic occupancy of the other factor , we would expect to identify a loss of binding specifically at loci that are co-bound by both factors . We identified a set of GAF and Zld co-occupied sites by overlapping our high-confidence GAF peaks and high-confidence Zld peaks . Peaks with at least 100 bp overlap were considered to be shared and used for our set of co-bound regions . To test likely direct effects of GAF on Zld occupancy , we determined the number regions that showed significant changes in Zld binding in GAFdeGradFP embryos that were bound by GAF . One out of the 86 regions at which Zld binding decreased in the GAF knockdown and 5 out of 19 that increased overlapped with regions bound by GAF ( Figure 4F ) . Contrary to our expectation if GAF was directly promoting Zld occupancy , there was no enrichment for co-bound regions and those that lost Zld binding . We investigated the reciprocal effect of Zld on GAF occupancy by determining the number regions that showed significant changes in GAF binding in zld-RNAi embryos that were also bound by Zld . Fifty-one of the 101 regions that had decreased GAF binding were also bound by Zld , suggesting a potential direct effect of Zld on GAF occupancy at these regions . By contrast , only 6 of the 25 regions that had increased GAF binding were also bound by Zld ( Figure 4E ) . GAF-binding regions that were decreased in the absence of Zld were significantly enriched for co-bound sites as compared to those GAF-binding sites that were maintained upon Zld depletion ( p=4 . 8×10−5 , log2 ( odds ratio ) =1 . 19 , two-tailed Fisher’s exact test ) . Based on the enrichment for Zld binding at these GAF-bound decreased sites , we hypothesized that Zld might be pioneering chromatin accessibility at these regions to promote GAF binding . Indeed , these regions were enriched for loci that depend on Zld for chromatin accessibility ( p=2 . 2×10−16 , log2 ( odds ratio ) =4 . 07 , two-tailed Fisher’s exact test ) : 38 of the 51 GAF and Zld co-bound sites at which GAF requires Zld for occupancy also require Zld for accessibility ( Hannon et al . , 2017 ) . Thus , at this limited set of regions the pioneer factor Zld may function to establish accessible chromatin that is necessary to promote robust GAF binding . Nonetheless , the majority of Zld- and GAF-binding sites are occupied in the absence of the other factor . GAF interacts with chromatin remodelers and is required to maintain chromatin accessibility in tissue culture ( Okada and Hirose , 1998; Tsukiyama et al . , 1994; Tsukiyama and Wu , 1995; Xiao et al . , 2001; Judd et al . , 2021; Fuda et al . , 2015 ) . Furthermore , GAF-binding motifs are enriched at regions of open chromatin that are established at NC12 and NC13 and that are bound by Zld but do not depend on Zld for accessibility ( Schulz et al . , 2015; Sun et al . , 2015; Blythe and Wieschaus , 2016 ) . To directly test the function of GAF in determining accessible chromatin domains during the MZT , we performed the assay for transposase-accessible chromatin ( ATAC ) -seq on six replicates of single GAFdeGradFP and control ( sfGFP-GAF ( N ) ) embryos harvested 2–2 . 5 hr AEL . His2AV-RFP signal for each embryo was visually inspected prior to performing ATAC-seq to ensure that embryos did not have grossly distorted nuclear morphology . In contrast to our control , replicates from the GAFdeGradFP embryos showed higher variability ( Figure 5—figure supplement 1A ) , suggesting that developmental defects caused by the lack of GAF might have lowered our ability to detect subtle changes in accessibility . Nonetheless , we identified 1523 regions with significant changes in accessibility in GAFdeGradFP embryos as compared to controls; 607 regions lost accessibility and 916 regions gained accessibility ( Figure 5A , Figure 5—figure supplement 1B , C ) . Sites that lost accessibility were among the regions with the highest ATAC-signal ( Figure 5—figure supplement 1C ) . 32% ( 197 ) of the regions that lost accessibility were significantly enriched for GAF binding , as determined by stage 5 GAF-sfGFP ( C ) ChIP-seq , when compared to open regions with no change in accessibility ( p=2 . 2×10−16 , log2 ( odds ratio ) =2 . 4 , two-tailed Fisher’s exact test , Figure 5—figure supplement 1D ) . The enrichment for GAF-binding sites in those regions that depend on GAF for accessibility suggests that GAF may directly drive accessibility at these regions . Therefore , we focused our analysis on the subset of regions that depend on GAF for accessibility , which we define as those that lost accessibility in the absence of GAF . Consistent with the enrichment of GAF-binding sites in promoters , 45% of all regions that depend on GAF for accessibility were in promoters . Enrichment for promoters was even larger in those regions that were bound by GAF and dependent on GAF for accessibility ( Figure 5B ) . By contrast , regions that gain accessibility in the absence of GAF are not enriched for promoters , supporting that this is likely an indirect effect of GAF knockdown ( Figure 5B ) . Ubx , a previously identified embryonic GAF-target gene , is an example of a gene that requires GAF for chromatin accessibility at the promoter ( Figure 5C; Farkas et al . , 1994 ) . Ubx similarly requires GAF occupancy for gene expression as the transcript was significantly down-regulated in GAFdeGradFP embryos ( Figure 5C ) . Thus , GAF binding at the Ubx promoter is required for both chromatin accessibility and gene expression . Together , our data demonstrate that GAF is required for chromatin accessibility at hundreds of loci during the MZT , and that this activity occurs preferentially at promoters . Previous work from our lab and others suggested that during the MZT , GAF may be responsible for maintaining chromatin accessibility at Zld-bound regions in the absence of Zld ( Schulz et al . , 2015; Moshe and Kaplan , 2017; Sun et al . , 2015 ) . To more broadly investigate how GAF and Zld shape chromatin accessibility during early development , we focused on regions co-occupied by both factors as identified by ChIP-seq ( this work and Harrison et al . , 2011 ) . We compared our single embryo ATAC-seq data to ATAC-seq data for wild-type NC14 embryos and NC14 embryos lacking maternal zld ( zld- ) ( Figure 6A; Hannon et al . , 2017 ) . Only seven of the 1192 regions co-bound by both Zld and GAF decreased in accessibility upon removal of either factor . By contrast , 104 regions require GAF for accessibility and 190 require Zld . Regions that require GAF for accessibility had a higher average GAF ChIP-seq peak height than regions that require Zld . Similarly , the Zld ChIP-seq signal was higher in regions where Zld is necessary for accessibility . Thus , both GAF and Zld are individually required for accessibility at distinct genomic regions co-occupied by both factors , and this requirement is correlated with occupancy as reflected in ChIP-seq peak height ( Figure 6A ) . GAF is also required for accessibility at 83 additional regions at which GAF is bound without Zld ( Figure 6—figure supplement 1 ) . The majority of sites bound by both Zld and GAF ( 891 ) did not change in chromatin accessibility when either factor was removed , suggesting that at these locations GAF and Zld may function redundantly to facilitate chromatin accessibility or that other factors were sufficient to maintain accessibility at these sites . Thus , during the MZT GAF and Zld are individually required for chromatin accessibility at distinct co-bound regions . At very few co-bound regions are both factors individually required . Previous work demonstrated that the Zld-binding motif is enriched at accessible regions that are established by NC11 . By contrast , GAF-binding motifs are enriched at regions that dynamically gain chromatin accessibility later during the MZT at NC12 or NC13 ( Blythe and Wieschaus , 2016 ) . If Zld preferentially drives chromatin accessibility early in the MZT and GAF is preferentially required later , we would expect that regions that require GAF for accessibility would be enriched for regions that gain accessibility at NC12 and NC13 . To test this prediction , we compared our ATAC-seq data to ATAC-seq on embryos precisely staged by nuclear cycle ( Blythe and Wieschaus , 2016 ) . Of the GAF-bound , GAF dependent accessible regions that were identified in the staged ATAC-seq data , three were open by NC11 , 17 were newly opened at NC12 , and 42 were newly opened at NC13 ( Figure 6B , Figure 6—figure supplement 2A ) , demonstrating that GAF-bound , GAF-dependent regions are enriched for regions that dynamically open at NC12 or NC13 as compared to regions that are already accessible at NC11 ( p=5 . 7×10−7 , log2 ( odds ratio ) =3 . 2 , two-tailed Fisher’s exact test ) . Indeed , the Ubx promoter dynamically gained accessibility at NC13 ( Figure 5C ) . Confirming that the loss of accessibility observed at late-opening regions was not due to differences in staging between the control ( sfGFP-GAF ( N ) ) and GAFdeGradFP embryos , the vast majority of sites that gained accessibility at NC13 ( 2452 ) were unchanged in accessibility in the GAFdeGradFP embryos ( Figure 6—figure supplement 2 ) . In contrast to GAF , Zld-bound , Zld-dependent accessible regions were enriched for regions already accessible at NC11: 1182 were open by NC11 , 343 were newly opened at NC12 , and 244 were newly opened at NC13 ( p<2 . 2×10−16 , log2 ( odds ratio ) =2 . 7 , two-tailed Fisher’s exact test ) ( Figure 6C ) . Together , these analyses show that GAF is preferentially required for chromatin accessibility at regions that gain accessibility over the MZT , while Zld is required for accessibility at regions that are open early . The preferential requirement for GAF-mediated accessibility during NC12 and NC13 allowed us to test whether GAF binding preceded the establishment of chromatin accessibility at these regions or whether this accessibility was co-incident with GAF binding . We determined whether the 42 regions that depended on GAF for accessibility and that gained accessibility at NC13 were already bound by GAF at stage 3 ( NC9 ) or whether these regions were among those that were only occupied by GAF at stage 5 ( NC14 ) . All 42 of these regions were bound by GAF at stage 3 and stage 5 , demonstrating that GAF occupancy at these regions precedes accessibility and supports a role for GAF as a pioneer factor at these regions . Using the deGradFP system , we demonstrated that maternally encoded GAF is required for embryogenesis . In contrast to zygotic GAF null embryos , which survive until the third instar larval stage , GAFdeGradFP embryos do not hatch ( Farkas et al . , 1994 ) . Our imaging showed that depletion of GAF in the early embryo resulted in severe defects , including asynchronous mitosis , anaphase bridges , nuclear dropout , and high nuclear mobility ( Figure 2C , D , Videos 1 and 2 ) . Similar defects are seen when zld is maternally depleted from embryos ( Liang et al . , 2008; Staudt et al . , 2006 ) and like embryos lacking maternal zld , GAFdeGradFP embryos died before the completion of the MZT . Distinct from embryos lacking maternal zld , a subset of GAFdeGradFP embryos display more intense defects . We identified defects in a subset of GAFdeGradFP embryos as early as NC10 , which is consistent with our genomics data indicating that GAF is required for proper expression of some of the earliest genes expressed during ZGA ( Figure 3D ) . It is possible that some of the phenotypic defects observed in GAFdeGradFP embryos are the result of GAF knockdown in the female germline . However , GAF null female germline clones are unable to produce eggs ( Bejarano and Busturia , 2004 ) , which stands in contrast to the large number of GAFdeGradFP embryos produced by females expressing sfGFP-GAF ( N ) and the deGradFP nanobody fusion . Together these phenotypic differences between null mutants and our deGrad-based knockdown along with our genomics data identifying misregulation of previously identified embryonic GAF-targets in the GAFdeGradFP embryos demonstrate an essential function for GAF in the early embryo . Live imaging of GFP-tagged , endogenously encoded GAF demonstrated that GAF is mitotically retained in small foci . This is similar to what was reported for antibody staining on fixed embryos , which showed GAF localized at pericentric heterochromatin regions of GA-rich satellite repeats ( Raff et al . , 1994; Platero et al . , 1998 ) . Because this prior imaging necessitated fixing embryos , it was unclear if mitotic retention of GAF was required for mitosis . Our system enabled us to determine that in the absence of GAF nuclei can undergo several rounds of mitosis albeit with noticeable defects ( Videos 1 and 2 ) . We conclude that GAF is not strictly required for progression through mitosis . However , the nuclear defects observed in GAFdeGradFP embryos support the model that GAF is broadly required for nuclear division and chromosome stability , in addition to its role in transcriptional activation during ZGA ( Bhat et al . , 1996 ) . Our imaging also identified high nuclear mobility in GAFdeGradFP embryos as compared to control embryos; nuclei of a subset of embryos showed a dramatic ‘swirling’ pattern of movement ( Video 2 ) . This defect is potentially due a DNA damage response that inactivates centrosomes to promote nuclear fallout ( Sibon et al . , 2000; Takada et al . , 2003 ) , or general disorder in the cytoskeletal network that is responsible for nuclear migration and division in the syncytial embryo ( Sullivan and Theurkauf , 1995 ) . Altogether , our phenotypic analysis of GAFdeGradFP embryos shows for the first time that maternal GAF is required for progression through the MZT , and suggests GAF has an early , global role in nuclear division and chromosome stability . In addition to GAF and Zld , the transcription factor CLAMP is expressed in the early embryo and functions in chromatin accessibility and transcriptional activation ( Rieder et al . , 2017; Soruco et al . , 2013; Urban et al . , 2017a; Urban et al . , 2017b; Rieder et al . , 2019 ) . While CLAMP , like GAF , binds GA-dinucleotide repeats , the two proteins preferentially bind to slightly different GA-repeats ( Kaye et al . , 2018 ) . GAF and CLAMP can compete for binding sites in vitro , and , in cell culture , when one factor is knocked down the occupancy of the other increases , suggesting that CLAMP and GAF compete for a subset of binding sites and may have partial functional redundancy ( Kaye et al . , 2018 ) . We demonstrate that GAF , like CLAMP , is essential in the early embryo , indicating that these two GA-dinucleotide-binding proteins cannot completely compensate for each other in vivo during early development ( Rieder et al . , 2017 ) . Furthermore , our sequencing analysis identified that a majority of genes that require GAF for expression are distinct from those that are regulated by CLAMP . Thus , while GAF and CLAMP may have some overlapping functions , they are independently required to regulate embryonic development during the MZT . GAF is a multi-purpose transcription factor with known roles in transcriptional regulation at promoters and enhancers as well as additional suggested roles in high-order chromatin structure . Our analysis showed that during the MZT GAF acts largely as an activator , directly binding and activating hundreds of zygotic transcripts during ZGA ( Figure 3 ) . We identified thousands of regions bound by GAF throughout the MZT , and these regions were preferentially associated with genes whose transcription decreased when maternally encoded GAF was degraded . This function may be driven , in part , through GAF-mediated chromatin accessibility as we identified hundreds of regions that depend on GAF for accessibility ( Figure 5 ) . This activity , in both transcriptional activation and mediating open chromatin , is similar to Zld , the only previously identified essential activator of the zygotic genome in Drosophila , and is shared with genome activators in other species , such as Pou5f3 and Nanog ( zebrafish ) and Dux ( mammals ) ( Schulz and Harrison , 2019; Vastenhouw et al . , 2019 ) . Our data support a pioneer-factor like role for GAF in this process as we demonstrated that GAF is already bound early in the MZT to regions that require GAF for accessibility later . Thus , GAF occupancy precedes GAF-mediated accessibility at a subset of sites . In addition to a direct role in mediating chromatin accessibility through the recruitment of chromatin remodelers ( Okada and Hirose , 1998; Tsukiyama et al . , 1994; Tsukiyama and Wu , 1995; Xiao et al . , 2001; Judd et al . , 2021; Fuda et al . , 2015 ) , GAF may also indirectly affect chromatin accessibility through a role in shaping three-dimensional chromatin structure . Sixty-seven percent of the regions that lost chromatin accessibility in the absence of GAF did not overlap a GAF-binding site as identified by ChIP-seq . Thus , at these regions GAF may function indirectly through the ability to facilitate enhancer-promoter loops ( Mahmoudi et al . , 2002; Melnikova et al . , 2004; Petrascheck et al . , 2005 ) . In addition , GAF-binding motifs are enriched at TAD boundaries which form during the MZT ( Hug et al . , 2017 ) , and GAF binding is enriched at Polycomb group dependent repressive loops that form following NC14 ( Ogiyama et al . , 2018 ) . Further investigation is necessary to determine the role of GAF in establishing three-dimensional chromatin architecture in the early embryo . Reprogramming in culture requires a cocktail of transcription factors that possess pioneer factor activity ( Takahashi and Yamanaka , 2006; Soufi et al . , 2012; Soufi et al . , 2015 ) . Similarly , in zebrafish and mice multiple pioneering factors are required for the rapid and efficient reprogramming that occurs during the MZT ( Schulz and Harrison , 2019; Vastenhouw et al . , 2019 ) . Here we have shown that , in addition to the essential pioneer factor Zld , GAF is a pioneer-like factor required for both gene expression and chromatin accessibility in the early Drosophila embryo . By analyzing the individual contributions of these two factors , we have begun to elucidate the different mechanisms by which multiple pioneer factors can drive dramatic changes in cell identity . While some pioneer factors work together to stabilize their interaction on chromatin ( Chronis et al . , 2017; Donaghey et al . , 2018; Liu and Kraus , 2017; Swinstead et al . , 2016 ) our data support primarily independent genome occupancy by Zld and GAF . Our reciprocal ChIP-seq data demonstrated that at regions bound by GAF and Zld , binding of each factor was largely retained in the absence of the other ( Figure 4 ) . Analysis of chromatin accessibility further supports these independent roles . We would predict that if Zld binding were lost in the absence of GAF or vice versa that at regions occupied by both factors either Zld or GAF individually would be required for accessibility . However , we identified only seven regions that lost accessibility in the absence of either Zld or GAF . By contrast , we identified more than one hundred regions that were individually dependent on each factor alone , and 891 regions that did not change in accessibility upon loss of either Zld or GAF . These data support independent roles for Zld and GAF in establishing or maintaining accessibility . Furthermore , we propose that because each factor can retain genome occupancy in the absence of the other that this may explain the 891 regions that remain accessible in the absence of either Zld or GAF: in the absence of Zld , GAF may be able to maintain accessibility and vice versa . However , until we investigate the chromatin landscape upon the removal of both Zld and GAF we cannot rule out that at these regions other factors may be instrumental in maintaining accessibility . Indeed , Duan et al . identify an essential role for the GA-dinucleotide-binding protein CLAMP in directing Zld binding and promoting chromatin accessibility ( Duan et al . , 2020 ) . At most loci , our data support independent binding for GAF and Zld . Nonetheless , we cannot eliminate a possible global role for GAF on Zld occupancy as the Zld ChIP-seq signal is considerably lower in the GAFdeGradFP embryos as compared to controls . We would predict that if GAF is directly functioning to stabilize Zld binding then this effect would be more evident at regions where both GAF and Zld bind , but this is not what we observed . Nonetheless , because of the proposed role of GAF in regulating three-dimensional chromatin architecture , GAF may be globally required for robust Zld occupancy . Furthermore , at a small subset of loci our data support a pioneering role for Zld in promoting GAF binding . In this experiment , the endogenously tagged GAF allowed us to use antibodies directed against the GFP epitope for ChIP . The use of the epitope tag enabled us to robustly control for immunoprecipitation efficiency by using another GFP-tagged protein as a spike-in control for the immunoprecipitation of GAF . In this manner , we identified 101 GAF-bound regions that decreased upon Zld knockdown . Of these , 51 were Zld-bound regions and 38 of these depend on Zld for chromatin accessibility . Thus , at a small subset of sites the pioneering function of Zld may be required to stabilize GAF binding . We propose that there is a handoff between Zld and GAF pioneer-like activity as the embryo progresses through the MZT: Zld functions primarily at the initiation of zygotic genome activation and GAF functions primarily later during the major wave of zygotic transcription . We identified that genes regulated by Zld are enriched for those activated during NC10-13 , while genes that depend on GAF are enriched for those that initiate expression during NC14 . Similarly , we determined that the majority of regions that require Zld for accessibility are already accessible at NC11 while those that require GAF are enriched for regions that gain accessibility at NC12 and NC13 . This is supported by prior analysis that showed that the Zld-binding motif is enriched at regions that are already accessible at NC11 and the GAF-binding motif is enriched at regions that dynamically gain accessibility later during the MZT ( Blythe and Wieschaus , 2016 ) . Based on this evidence , we propose that there is a gradual handoff in the control of transcriptional activation and chromatin remodeling from Zld to GAF as the MZT progresses ( Figure 7 ) . During the initial stages of the MZT , the nuclear division cycle is an incredibly rapid series of synthesis and mitotic phases . At NC14 , this cycle slows , and these different dynamics likely influence the mechanisms by which accessibility can be established . While it is unclear how Zld establishes accessibility , GAF interacts with chromatin remodelers . It is possible that the activities of these complexes may have a more substantial impact once the division cycle slows . Together our data support the requirement for at least two pioneer-like transcription factors , Zld and GAF , to sequentially reprogram the zygotic genome following fertilization and allow for future embryonic development . It is likely that there are additional factors that function along with Zld and GAF to define accessible cis-regulatory regions and drive genome activation . Future studies will enable more detailed mechanistic insights into how multiple pioneering factors work together to reshape the transcriptional landscape and transform cell fate in the early embryo . All stocks were grown on molasses food at 25°C . Fly strains used in this study: w1118 , His2Av-RFP ( II ) ( Bloomington Drosophila Stock Center ( BDSC ) #23651 ) , mat-α-GAL4-VP16 ( BDSC #7062 ) , UAS-shRNA-zld ( Sun et al . , 2015 ) . sfGFP-GAF ( N ) and GAF-sfGFP ( C ) mutant alleles were generated using Cas9-mediated genome engineering ( see below ) . nos-deGradFP ( II ) transgenic flies were made by PhiC31 integrase-mediated transgenesis into the PBac{yellow[+]-attP-3B}VK00037 docking site ( BDSC #9752 ) by BestGene Inc . The sequence for NSlmb-vhhGFP4 was obtained from Caussinus et al . , 2012 and amplified from genomic material from UASp-Nslmb . vhhGFP4 ( BDSC #58740 ) . The NSlmb-vhhGFP4 sequence was cloned using Gibson assembly into pattB ( DGRC #1420 ) with the nanos promoter and 5’UTR . To obtain the embryos for ChIP-seq and live embryo imaging in a zld knockdown background , we crossed mat-α-GAL4-VP16 ( II ) /Cyo; sfGFP-GAF N ( III ) flies to UAS-shRNA-zld ( III ) flies and took mat-α-GAL4-VP16/+ ( II ) ; sfGFP-GAF ( N ) / UAS-shRNA-zld ( III ) females . These females were crossed to their siblings , and their embryos were collected . For controls , mat-α-GAL4-VP16 ( II ) /Cyo; sfGFP-GAF N ( III ) flies were crossed to w1118 flies and embryos from mat-α-GAL4-VP16/+ ( II ) : sfGFP-GAF ( N ) /+ ( III ) females crossed to their siblings were collected . To obtain embryos for live imaging , hatching rate assays , RNA-seq , ATAC-seq , and ChIP-seq in a GAF knockdown background we crossed nos-degradFP ( II ) ; sfGFP-GAF ( N ) /TM6c ( III ) flies to His2Av-RFP ( II ) ; sfGFP-GAF ( N ) ( III ) flies and selected females that were nos-degradFP/His2Av-RFP ( II ) ; sfGFP-GAF ( N ) ( III ) . These females were crossed to their siblings of the same genotype , and their embryos were collected . Embryos from His2Av-RFP ( II ) ; sfGFP-GAF ( N ) ( III ) females were used as paired controls . Cas9-mediated genome engineering as described in Hamm et al . , 2017 was used to generate the N-terminal and C-terminal super folder Green Fluorescent Protein ( sfGFP ) -tagged GAF . The double-stranded DNA ( dsDNA ) donor was created using Gibson assembly ( New England BioLabs , Ipswich , MA ) with 1 kb homology arms flanking the sfGFP tag and GAF N-terminal or C-terminal open reading frame . sfGFP sequence was placed downstream of the GAF start codon ( N-terminal ) or just upstream of the stop codon in the fifth GAF exon , coding for the short isoform ( C-terminal ) . Additionally , a 3xP3-DsRed cassette flanked by the long-terminal repeats of PiggyBac transposase was placed in the second GAF intron ( N-terminal ) or fourth GAF intron ( C-terminal ) for selection . The guide RNA sequences ( N-terminal- TAAACATTAAATCGTCGTGT ) , ( C-terminal- AAATGAATACTCGATTA ) were cloned into pBSK under the U63 promoter using inverse PCR . Purified plasmid was injected into embryos of yw; attP40{nos-Cas9}/CyO for the N-terminal line and y1 M{vas-Cas9 . RFP-}ZH-2A w1118 ( BDSC#55821 ) for the C-terminal line by BestGene Inc Lines were screened for DsRed expression to verify integration . The entire 3xP3-DsRed cassette was cleanly removed using piggyBac transposase , followed by sequence confirmation of precise tag integration . Embryos were dechorionated in 50% bleach for 2 min and subsequently mounted in halocarbon 700 oil . Due to the fragility of the GAFdegradFP embryos , embryos used for videos were mounted in halocarbon 700 oil without dechorionation . The living embryos were imaged on a Nikon A1R+ confocal at the University of Wisconsin-Madison Biochemistry Department Optical Core . Nuclear density , based on the number of nuclei/2500 μm2 , was used to determine the cycle of pre-gastrulation embryos . Nuclei were marked with His2AV-RFP . Image J ( Schindelin et al . , 2012 ) was used for post-acquisition image processing . Videos were acquired at 1 frame every 10 s . Playback rate is seven frames/second . A minimum of 50 females and 25 males of the indicated genotypes were allowed to mate for at least 24 hr before lays were taken for hatching rate assays . Embryos were picked from overnight lays and approximately 200 were lined up on a fresh molasses plate . Unhatched embryos were counted 26 hr or more after embryos were selected . Proteins were transferred to 0 . 45 μm Immobilon-P PVDF membrane ( Millipore , Burlington , MA ) in transfer buffer ( 25 mM Tris , 200 mM Glycine , 20% methanol ) for 60 min ( 75 min for Zld ) at 500mA at 4°C . The membranes were blocked with blotto ( 2 . 5% non-fat dry milk , 0 . 5% BSA , 0 . 5% NP-40 , in TBST ) for 30 min at room temperature and then incubated with anti-GFP ( 1:2000 , #ab290 ) ( Abcam , Cambridge , United Kingdom ) anti-Zld ( 1:750 ) ( Harrison et al . , 2010 ) , or anti-tubulin ( DM1A , 1:5000 ) ( Sigma , St . Louis , MO ) , overnight at 4°C . The secondary incubation was performed with goat anti-rabbit IgG-HRP conjugate ( 1:3000 ) ( Bio-Rad , Hercules , CA ) or anti-mouse IgG-HRP conjugate ( 1:3000 ) ( Bio-Rad ) for 1 hr at room temperature . Blots were treated with SuperSignal West Pico PLUS chemiluminescent substrate ( Thermo Fisher Scientific , Waltham , MA ) and visualized using the Azure Biosystems c600 or Kodak/Carestream BioMax Film ( VWR , Radnor , PA ) . 2–2 . 5 hr AEL embryos were dechorionated in bleach and subsequently mounted in halocarbon 700 oil . Embryos were imaged on a Nikon Ti-2e Epifluorescent microscope using ×60 magnification . Images were acquired of a single z-plane . Analysis was performed using the Nikon analysis software . Ten circular regions of interest ( ROIs ) were drawn around individual nuclei and the mean fluorescent intensity of each nucleus was calculated . Ten circular ROIs were drawn in the regions outside of the nuclei to measure the background fluorescent level of the embryo , and the mean fluorescent intensity of the background was calculated . To normalize values to the background fluorescent intensity , the final mean intensity of the nuclei was determined to be the mean fluorescent intensity of the nuclei after subtracting the mean fluorescent intensity of the background . This analysis was performed on images from 27 GAF-sfGFP ( C ) homozygous embryos and 26 sfGFP-GAF ( N ) homozygous embryos . ChIP was performed as described previously ( Blythe and Wieschaus , 2015 ) on: stage 3 and stage five hand selected GAF-sfGFP ( C ) homozygous embryos , stage 3 and stage five hand selected w1118 embryos , 2–2 . 5 hr AEL hand selected embryos from mat-α-GAL4-VP16/+ ( II ) ; sfGFP-GAF ( N ) / UAS-shRNA-zld ( III ) females and mat-α-GAL4-VP16/+ ( II ) ; sfGFP-GAF ( N ) /+ ( III ) females , 2–2 . 5 hr AEL embryos from nos-degradFP/His2Av-RFP ( II ) ; sfGFP-GAF ( N ) ( III ) females , and 2–2 . 5 hr AEL embryos from His2Av-RFP ( II ) ; sfGFP-GAF ( N ) ( III ) females . For each genotype , two biological replicates were collected . Briefly , 1000 stage 3 embryos or 400–500 stage five embryos were collected , dechorionated in 50% bleach for 3 min , fixed for 15 min in 0 . 45% formaldehyde and then lysed in 1 mL of RIPA buffer ( 50 mM Tris-HCl pH 8 . 0 , 0 . 1% SDS , 1% Triton X-100 , 0 . 5% sodium deoxycholate , and 150 mM NaCl ) . The fixed chromatin was then sonicated for 20 s 11 times at 20% output and full duty cycle ( Branson Sonifier 250 ) . Chromatin was incubated with 6 μg of anti-GFP antibody ( Abcam #ab290 ) or 8 μl of anti-Zld antibody ( Harrison et al . , 2010 ) overnight at 4°C , and then bound to 50 μl of Protein A magnetic beads ( Dynabeads Protein A , Thermo Fisher Scientific ) . The purified chromatin was then washed , eluted , and treated with 90 μg of RNaseA ( 37°C , for 30 min ) and 100 μg of Proteinase K ( 65°C , overnight ) . The DNA was purified using phenol/chloroform extraction and concentrated by ethanol precipitation . Each sample was resuspended in 25 μl of water . Sequencing libraries were made using the NEB Next Ultra II library kit and were sequenced on the Illumina Hi-Seq4000 using 50 bp single-end reads or the Illumina NextSeq 500 using 75 bp single-end reads at the Northwestern Sequencing Core ( NUCore ) . ChIP-seq data was aligned to the Drosophila melanogaster reference genome ( version dm6 ) using bowtie 2 v2 . 3 . 5 ( Langmead and Salzberg , 2012 ) with the following non-default parameters: -k 2 , --very-sensitive . Aligned reads with a mapping quality < 30 were discarded , as were reads aligning to scaffolds or the mitochondrial genome . To identify regions that were enriched in immunoprecipitated samples relative to input controls , peak calling was performed using MACS v2 ( Zhang et al . , 2008 ) with the following parameters: -g 1 . 2e8 , --call-summits . To focus analysis on robust , high-quality peaks , we used 100 bp up- and downstream of peak summits , and retained only peaks that were detected in both replicates and overlapped by at least 100 bp . All downstream analysis focused on these high-quality peaks . Peak calling was also performed for control ChIP samples performed on w1118 with the α-GFP antibody . No peaks were called in any of the w1118 controls , indicating high specificity of the α-GFP antibody . To compare GAF-binding sites at stage 3 , stage 5 and in S2 cells , and to compare GAF , Zld and CLAMP binding , we used GenomicRanges R package ( Lawrence et al . , 2013 ) to compare different sets of peaks . Peaks overlapping by at least 100 bp were considered to be shared . To control for differences in data processing and analysis between studies , previously published ChIP-seq datasets for Zld ( Harrison et al . , 2011 , GSE30757 ) , GAF ( Fuda et al . , 2015 , GSE40646 ) , and CLAMP ( Rieder et al . , 2019 , GSE133637 ) were processed in parallel with ChIP-seq data sets generated in this study . DeepTools ( Ramírez et al . , 2016 ) was used to generate read depth for 10 bp bins across the genome . A z-score was calculated for each 10 bp bin using the mean and standard deviation of read depth across all 10 bp bins . Z-score normalized read depth was used to generate heatmaps and metaplots . De novo motif discovery was performed using MEME-suite ( Bailey et al . , 2009 ) . Genome browser tracks were generated using raw bigWig files with the Gviz package in R ( Hahne and Ivanek , 2016 ) . Numbers used for all Fisher’s exact tests are included in Supplementary file 4 . Chromatin for GAF ChIP following depletion of Zld was prepared as described above . Prior to addition of the anti-GFP antibody , mouse chromatin prepared from cells expression an H3 . 3-GFP fusion protein was added to Drosophila chromatin at a 1:750 ratio . Following sequencing , reads were aligned to a combined reference genome containing both the Drosophila genome ( version dm6 ) and the mouse genome ( version mm39 ) . Only reads that could be unambiguously aligned to one of the two reference genomes were retained . To control for any variability in the proportion of mouse chromatin in the input samples , the ratio of percentage of spike in reads in the IP relative to the input were used . A scaling factor was calculated by dividing one by this ratio . Z-score normalized read depth was adjusted by this scaling factor , and the resulting spike-in normalized values were used for heatmaps . The global decrease in Zld ChIP signal in GAF-depleted embryos ( Figure 4—figure supplement 2C ) caused us to consider technical and biological factors that may have affected these experiments . Global Zld-binding signal may have been lower in the GAFdeGradFP background because GAFdeGradFP embryos die at variable times around the MZT . To attempt account for this variability , we collected embryos from a tight 2–2 . 5 hr AEL timepoint rather than sorting . Therefore , a portion of these embryos collected for ChIP-seq may have been dead . Additionally , the immunoprecipitation efficiency can vary between experiments and thus might have been lower in the GAFdeGradFP ChIP-seq as compared to the control . Alternatively , as discussed in the Results section , GAF may be broadly required for robust Zld occupancy . To control for these technical factors , we performed an analysis based on peak rank , in addition to peak intensity . The number of reads aligning within each peak was quantified using featureCounts from the Subread package ( v1 . 6 . 4 ) ( Liao et al . , 2014 ) . Peaks were then ranked based on the mean RPKM-normalized read count between replicates , allowing comparison of peak rank between different conditions . DESeq2 ( Love et al . , 2014 ) was used to identify potential differential binding sites in a more statistically rigorous way . To control for variable detection of low-intensity peaks , only high-confidence GAF stage 5 peaks identified both in homozygous GAF-sfGFP ( C ) and heterozygous sfGFP-GAF ( N ) embryos ( 3800 peaks ) were analyzed . For Zld ChIP-seq , analysis was restricted to 6003 high-confidence peaks shared between our control dataset and previously published Zld ChIP ( Harrison et al . , 2011 ) . A table of read counts for these peaks , generated by featureCounts as described above , was used as input to DESeq2 . Peaks with an adjusted p-value<0 . 05 and a fold change >2 were considered to be differentially bound . Numbers used for all Fisher’s exact tests are included in Supplementary file 4 . A total of 150–200 embryos from His2Av-RFP/nos-degradFP ( II ) ; sfGFP-GAF ( N ) ( III ) and His2Av-RFP ( II ) ; sfGFP-GAF ( N ) ( III ) females were collected from a half hour lay and aged for 2 hr . For each genotype , three biological replicates were collected . Embryos were then picked into Trizol ( Invitrogen , Carlsbad , CA ) with 200 μg/ml glycogen ( Invitrogen ) . RNA was extracted and RNA-seq libraries were prepared using the Universal RNA-Seq with NuQuant , Drosophila AnyDeplete Universal kit ( Tecan , Männedorf , Switzerland ) . Samples were sequenced on the Illumina NextSeq500 using 75 bp single-end reads at the Northwestern Sequencing Core ( NUCore ) . RNA-seq data was aligned to the Drosophila melanogaster genome ( dm6 ) using HISAT v2 . 1 . 0 ( Kim et al . , 2015 ) . Reads with a mapping quality score <30 were discarded . The number of reads aligning to each gene was quantified using featureCounts , generating a read count table that was used to analyze differential expression with DESeq2 . Genes with an adjusted p-value<0 . 05 and a fold change >2 were considered statistically significant . To identify GAF-target genes , GAF ChIP peaks were assigned to the nearest gene ( this study ) . Zygotically and maternally expressed genes ( Lott et al . , 2011 , GSE25180 ) zygotic gene expression onset ( Li et al . , 2014 , GSE58935 ) , Zld-dependent genes ( McDaniel et al . , 2019 , GSE121157 ) , CLAMP-dependent genes ( Rieder et al . , 2017 , GSE102922 ) , and Zld targets ( Harrison et al . , 2011 , GSE30757 ) were previously defined . Genome browser tracks were generated using bigWigs with the Gviz package in R ( Hahne and Ivanek , 2016 ) . Numbers used for all Fisher’s exact tests are included in Supplementary file 4 . Embryos from His2Av-RFP/nos-degradFP ( II ) ; sfGFP-GAF ( N ) ( III ) and His2Av-RFP ( II ) ; sfGFP-GAF ( N ) ( III ) females were collected from a half hour lay and aged for 2 hr . Embryos were dechorionated in bleach , mounted in halocarbon 700 oil , and imaged on a Nikon Ti-2e Epifluorescent microscope using 60x magnification . Embryos with nuclei that were not grossly disordered as determined by the His2Av-RFP marker were selected for single embryo ATAC-seq . Six replicates were analyzed for each genotype . Single-embryo ATAC-seq was performed as described previously ( Blythe and Wieschaus , 2016; Buenrostro et al . , 2013 ) . Briefly , a single dechorionated embryo was transferred to the detached cap of a 1 . 5 ml microcentrifuge tube containing 10 µl of ice-cold ATAC lysis buffer ( 10 mM Tris pH 7 . 5 , 10 mM NaCl , 3 mM MgCl2 , 0 . 1% NP-40 ) . Under a dissecting microscope , a microcapillary tube was used to homogenize the embryo . The cap was placed into a 1 . 5 ml microcentrifuge tube containing an additional 40 µl of cold lysis buffer . Tubes were centrifuged for 10 min at 500 g at 4°C . The supernatant was removed , and the resulting nuclear pellet was resuspended in 5 µl buffer TD ( Illumina , San Diego , CA ) and combined with 2 . 5 µl H2O and 2 . 5 µl Tn5 transposase ( Tagment DNA Enzyme , Illumina ) . Tubes were placed at 37°C for 30 min and the resulting fragmented DNA was purified using the Minelute Cleanup Kit ( Qiagen , Hilden , Germany ) , with elution performed in 10 µl of the provided elution buffer . Libraries were amplified for 12 PCR cycles with unique dual index primers using the NEBNext Hi-Fi 2X PCR Master Mix ( New England Biolabs ) . Amplified libraries were purified using a 1 . 2X ratio of Axygen magnetic beads ( Corning Inc , Corning , NY ) . Libraries were submitted to the University of Wisconsin-Madison Biotechnology Center for 150 bp , paired-end sequencing on the Illumina NovaSeq 6000 . Adapter sequences were removed from raw sequence reads using NGMerge ( Gaspar , 2018 ) . ATAC-seq reads were aligned to the Drosophila melanogaster ( dm6 ) genome using bowtie2 with the following parameters: --very-sensitive , --no-mixed , --no-discordant , -X 5000 , -k 2 . Reads with a mapping quality score < 30 were discarded , as were reads aligning to scaffolds or the mitochondrial genome . Analysis was restricted to fragments < 100 bp , which , as described previously , are most likely to originate from nucleosome-free regions ( Buenrostro et al . , 2013 ) . To maximize the sensitivity of peak calling , reads from all replicates of GAFdeGradFP and control embryos were combined . Peak calling was performed on combined reads using MACS2 with parameters -f BAMPE --keep-dup all -g 1 . 2e8 --call-summits . This identified 64 , 133 accessible regions . To facilitate comparison to published datasets with different sequencing depths , peaks were filtered to include only those with a fold-enrichment over background > 2 . 5 ( as determined by MACS2 ) . This resulted in 36 , 571 peaks that were considered for downstream analysis . Because greater sequencing depth can result in the detection of a large number of peaks with a lower level of enrichment over background , this filtering step ensured that peak sets were comparable across all datasets . 201 bp peak regions ( 100 bp on either side of the peak summit ) were used for downstream analysis . Reads aligning within accessible regions were quantified using featureCounts , and differential accessibility analysis was performed using DESeq2 with an adjusted p-value<0 . 05 and a fold change > 2 as thresholds for differential accessibility . Previously published ATAC-seq datasets ( Hannon et al . , 2017 ) , GSE86966; ( Blythe and Wieschaus , 2016 ) , GSE83851; ( Nevil et al . , 2020 ) , GSE137075 . were re-analyzed in parallel with ATAC-seq data from this study to ensure identical processing of data sets . Heatmaps and metaplots of z score-normalized read depth were generated with DeepTools . MEME-suite was used to for de novo motif discovery for differentially accessible ATAC peaks and to match discovered motifs to previously known motifs from databases . Genome browser tracks were generated using bigWigs with the Gviz package in R ( Hahne and Ivanek , 2016 ) . Numbers used for all Fisher’s exact tests are included in Supplementary file 4 .	Most cells in an organism share the exact same genetic information , yet they still adopt distinct identities . This diversity emerges because only a selection of genes is switched on at any given time in a cell . Proteins that latch onto DNA control this specificity by activating certain genes at the right time . However , to perform this role they first need to physically access DNA: this can be difficult as the genetic information is tightly compacted so it can fit in a cell . A group of proteins can help to unpack the genome to uncover the genes that can then be accessed and activated . While these ‘pioneer factors’ can therefore shape the identity of a cell , much remains unknown about how they can work together to do so . For instance , the pioneer factor Zelda is essential in early fruit fly development , as it enables the genetic information of the egg and sperm to undergo dramatic reprogramming and generate a new organism . Yet , it was unclear whether additional helpers were required for this transition . Using this animal system , Gaskill , Gibson et al . identified GAGA Factor as a protein which works with Zelda to open up and reprogram hundreds of different sections along the genome of fruit fly embryos . This tag-team effort started with Zelda being important initially to activate genes; regulation was then handed over for GAGA Factor to continue the process . Without either protein , the embryo died . Getting a glimpse into early genetic events during fly development provides insights that are often applicable to other animals such as fish and mammals . Ultimately , this research may help scientists to understand how things can go wrong in human embryos .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression" ]	2021	GAF is essential for zygotic genome activation and chromatin accessibility in the early Drosophila embryo
Visual processing depends on specific computations implemented by complex neural circuits . Here , we present a circuit-inspired model of retinal ganglion cell computation , targeted to explain their temporal dynamics and adaptation to contrast . To localize the sources of such processing , we used recordings at the levels of synaptic input and spiking output in the in vitro mouse retina . We found that an ON-Alpha ganglion cell's excitatory synaptic inputs were described by a divisive interaction between excitation and delayed suppression , which explained nonlinear processing that was already present in ganglion cell inputs . Ganglion cell output was further shaped by spike generation mechanisms . The full model accurately predicted spike responses with unprecedented millisecond precision , and accurately described contrast adaptation of the spike train . These results demonstrate how circuit and cell-intrinsic mechanisms interact for ganglion cell function and , more generally , illustrate the power of circuit-inspired modeling of sensory processing . Neural computations in the retina are generated by complex circuits that drive the responses of ~30 distinct ganglion cell types ( Baden et al . , 2016; Demb and Singer , 2015; Sanes and Masland , 2015 ) . Despite the complexity of retinal circuitry , many aspects of the responses of ganglion cells to visual stimuli can be predicted with a straightforward Linear-Nonlinear ( LN ) cascade model ( Shapley , 2009 ) . In this model , a linear receptive field filters the stimulus , and a nonlinear function shapes the output by implementing the spike threshold and response saturation ( Baccus and Meister , 2002; Chichilnisky , 2001; Kim and Rieke , 2001 ) . However , many aspects of ganglion cell firing deviate from LN model predictions . For example , the LN model does not capture the effects of contrast adaptation , which includes a reduced gain ( i . e . , filter amplitude ) at high contrast ( Kim and Rieke , 2001; Meister and Berry , 1999; Shapley and Victor , 1978 ) . The LN model also does not predict firing at high temporal resolution ( Berry and Meister , 1998; Butts et al . , 2016 , 2007; Keat et al . , 2001; Passaglia and Troy , 2004; Uzzell and Chichilnisky , 2004 ) , and yet precise firing likely represents an essential element of downstream visual processing ( Bruno and Sakmann , 2006; Havenith et al . , 2011; Kelly et al . , 2014; Wang et al . , 2010a ) . To improve on the LN model , several nonlinear approaches have been proposed . The first approach describes the nonlinear function between stimulus and response as a mathematical expansion , extending from the linear receptive field ( Chichilnisky , 2001 ) to 'second-order' quadratic terms , using either spike-triggered covariance ( Fairhall et al . , 2006; Liu and Gollisch , 2015; Samengo and Gollisch , 2013; Vaingankar et al . , 2012 ) or maximally informative dimension analyses ( Sharpee et al . , 2004 ) . Such expansion terms better predict the spike train , but they are difficult to interpret functionally and with respect to the underlying circuitry ( Butts et al . , 2011; McFarland et al . , 2013 ) . The second approach targets specific aspects of the response , such as spike-refractoriness ( Berry and Meister , 1998; Keat et al . , 2001; Paninski , 2004; Pillow et al . , 2005 ) , gain changes associated with contrast adaptation ( Bonin et al . , 2005; Mante et al . , 2008; Meister and Berry , 1999; Shapley and Victor , 1978 ) , the interplay of excitation and inhibition ( Butts et al . , 2016 , 2011 ) , and rectification of synaptic release , associated with nonlinear spatial processing ( Freeman et al . , 2015; Gollisch , 2013; Schwartz and Rieke , 2011 ) . However , each of these models primarily focuses on one type of nonlinear computation and does not generalize to explain a range of response properties . Here we derive a novel nonlinear modeling framework inspired by retinal circuitry . The model is constrained by recordings at two stages of processing: excitatory synaptic input and spike output , recorded in mouse retinal ganglion cells . We focused on ON-Alpha ganglion cells because they comprise a major input to lateral geniculate nucleus ( LGN ) and superior colliculus , and because they could be targeted routinely , in vitro , based on their large soma size . Furthermore , their spiking response to contrast modulation is mediated predominantly by synaptic excitation ( Murphy and Rieke , 2006 ) . We devised a tractable model of excitatory currents that incorporates a nonlinear structure based on realistic circuit elements . In particular , we allowed for divisive suppression acting on a ganglion cell’s excitatory inputs to capture the computations implemented by presynaptic inhibition ( Eggers and Lukasiewicz , 2011; Franke et al . , 2016 ) and synaptic depression ( Jarsky et al . , 2011; Ozuysal and Baccus , 2012 ) at bipolar cell terminals . Ganglion cell firing , further shaped by spike generation mechanisms , could be predicted by the model with millisecond precision . Our study establishes a unified model of nonlinear processing within ganglion cells that accurately captures both the generation of precise firing events and fast contrast adaptation . Similar circuit-inspired modeling could be applied widely in other sensory systems . Because some degree of contrast adaptation is already present in a ganglion cell’s excitatory synaptic inputs ( Beaudoin et al . , 2007; Beaudoin et al . , 2008; Kim and Rieke , 2001 ) , we hypothesized that we might uncover the source of the nonlinear processing by directly modeling the synaptic input currents . We therefore made whole-cell patch clamp recordings on the same neurons we recorded spike responses from , and performed a similar LN analysis on excitatory synaptic currents ( Figure 1D ) . The LN model of the currents ( Figure 1E ) – like that of the spike response – accurately predicted the observed response at low contrast , but performed relatively poorly at high contrast ( Figure 1D ) . To compare the precision of the LN model to the observed data , we measured the coherence between the trial-averaged response or model prediction and the responses on individual trials ( see Methods ) ; this measure captures the consistency of the response across repeats across time scales . At low contrast , the coherence of the excitatory current matched that of the LN model prediction , whereas at high contrast the coherence of the current extended to finer time scales ( i . e . , higher frequencies ) and hence exceeded the precision predicted by the LN model ( Figure 1F ) . Contrast adaptation was measured in the synaptic currents by comparing LN models at each contrast level ( Figure 1E ) . The linear filter for the current responses had a larger amplitude ( i . e . , higher gain ) in low contrast compared with high contrast ( Beaudoin et al . , 2007; Beaudoin et al . , 2008; Kim and Rieke , 2001 ) . This adaptation occurred rapidly after the contrast switch and showed a barely discernable slow component that has been observed in other ganglion cell types ( Figure 1—figure supplement 1; [Baccus and Meister , 2002; Manookin and Demb , 2006] ) . To compare the contrast-dependent gain change in the currents and spikes , we define contrast gain as the ratio between the standard deviation of the filter in low contrast over that in high contrast . The contrast gain was significantly larger for spikes ( 1 . 61 ± 0 . 23 ) than for currents ( 1 . 10 ± 0 . 14; p<10−6 , unpaired two-sample t-test , spikes: n = 11 , current: n = 13 ) ( Zaghloul et al . , 2005 ) ; in cases where both currents and spikes were recorded in the same cell , the contrast gain was larger for spikes by 30 . 1% ± 12 . 0% ( n = 3 ) . These observations suggest that both contrast adaptation and temporal precision in ON-Alpha ganglion cell spike responses are generated in large part by retinal circuitry upstream of the ganglion cell , but that further transformation occurs between currents and spikes ( Kim and Rieke , 2001 ) . In constructing a nonlinear description of the computation present in excitatory synaptic currents , we sought to emulate elements of the retinal circuit that shape these currents ( Figure 2A ) . Excitatory synaptic inputs to ganglion cells come from bipolar cells , and bipolar cell voltage responses to our stimuli are well described by an LN model ( Baccus and Meister , 2002; Rieke , 2001 ) . This suggests that mechanisms responsible for the nonlinear behavior of the postsynaptic excitatory current are localized to the bipolar-ganglion cell synapses . Possible sources of such nonlinear behavior include presynaptic inhibition from amacrine cells , which can directly gate glutamate release from bipolar cell terminals ( Eggers and Lukasiewicz , 2011; Euler et al . , 2014; Franke et al . , 2016; Schubert et al . , 2008; Zaghloul et al . , 2007 ) , and synaptic depression at bipolar terminals caused by vesicle depletion ( Jarsky et al . , 2011; Markram et al . , 1998; Ozuysal and Baccus , 2012 ) . 10 . 7554/eLife . 19460 . 006Figure 2 . The divisive suppression ( DivS ) model of synaptic currents . ( A ) Schematic of retinal circuitry . The vertical excitatory pathway , cones → bipolar cells → ganglion cell , can be modulated at the bipolar cell synapse by amacrine cell-mediated inhibition of bipolar cell release or by synaptic depression . We model both processes by divisive suppression ( bottom ) , where an LN model , representing the collective influence of amacrine cell inhibition and synaptic depression , multiplicatively modulates excitatory inputs from bipolar cells to the ganglion cell . ( B ) The excitatory ( green ) and suppressive ( cyan ) temporal filters of the DivS model for an example ON-Alpha cell . ( C ) Divisive suppression is delayed relative to excitation , demonstrated by the distributions of latencies measured for each pair of filters ( mean delay = 10 . 9 ± 2 . 2 ms , p<0 . 0005 , n = 13 ) . ( D ) Excitatory ( left ) and suppressive nonlinearities ( right ) for the DivS model . The solid line indicates model fits for the example cell , and the gray lines are from other cells in the population , demonstrating their consistent form . The distribution of the filtered stimulus is also shown as the shaded area for HC ( blue ) and LC ( red ) . The suppressive nonlinearity ( right ) falls below one for stimuli that match the kernel or are opposite , implying that divisive suppression is ON-OFF . ( E ) To validate the form of the DivS model , we compared its performance to alternative models , including a more general model where the form of the nonlinearity is not assumed ( 2-D , see below ) , a model where excitatory and suppressive terms interact additively ( AddS ) instead of divisively , and a covariance ( COV ) model similar to spike triggered covariance ( Figure 2—figure supplement 1 ) . The DivS model performed significantly better than the LN , AddS and COV models ( p<0 . 0005 , n = 13 ) , and matched the performance of the 2-D model . ( F ) We used a 2-dimensional nonlinearity to capture any more general interaction between excitatory and suppressive filters , shown with schematic ( left ) , and the resulting fits ( middle ) . Consistent with the model performance ( E ) , the form of this 2-D nonlinearity could be reproduced by the DivS model ( right ) . ( G ) Accuracy of the ability of the DivS , AddS , and COV models to reproduce the 2-D nonlinearity across neurons ( p<0 . 0005 , n = 13 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19460 . 00610 . 7554/eLife . 19460 . 007Figure 2—figure supplement 1 . Comparison to covariance-based models . ( A–F ) Covariance-based analysis of synaptic currents for the same example cell as in Figure 1 . Covariance analysis follows the intuition of spike-triggered covariance ( STC ) , but uses continuous current input rather than spikes ( see Materials and methods ) . ( A ) Left: Cross-correlation between the stimulus and current response ( the equivalent of a spike-triggered average ) for high contrast ( HC , blue ) and low contrast ( LC , red ) stimuli . Filters are scaled to have the same standard deviation , for comparisons of shape . Middle: The eigenvalue spectrum for the response-triggered covariance matrix in HC , revealing two significant eigenvalues ( color-coded ) . Right: The corresponding eigenvectors . ( B ) The locations of the cross-correlations in HC ( blue , left ) and LC ( red , right ) within the 2-D subspace spanned by the two significant eigenvectors for all neurons ( n = 13 ) . Because they are all close to the unit circle , both HC and LC cross-correlations were largely contained in the covariance ( COV ) subspace , consistent with previously reported results for spikes ( Liu and Gollisch , 2015 ) . ( C ) Model performance for the LN , DivS , and COV models ( n = 13 ) , reproduced from Figure 2E . This demonstrates that the COV filters coupled to a 2-D nonlinearity ( described below ) can nearly match the performance of the DivS model . ( D ) Left: The excitatory ( green ) and suppressive ( cyan ) filters of the DivS model , plotted in comparison to the filters identified by covariance analysis ( dashed lines ) . Middle: The DivS model filters shared the same 2-D subspace as the covariance filters , as shown by comparing the filters to optimal linear combinations of the COV filters ( black dashed ) , following previous work based on spikes ( Butts et al . , 2011 ) . Right: The DivS filters projected into the COV filters subspace across neurons , using the same analysis as in ( B ) . Their proximity to the unit circle shows they are almost completely in the covariance subspace for all neurons , again consistent with previous work with spikes ( Butts et al . , 2011 ) . ( E ) Left: The 2-D nonlinearity associated with the COV filters , for the example neuron considered . Right: The best 2-D nonlinearity reconstructed from 1-D nonlinearities operating on the COV filters . Unlike the 2-D nonlinearity associated with the DivS filters ( Figure 2F ) , this nonlinearity could not be represented as the product of two 1-D nonlinearities . ( F ) The separability of 2-D nonlinearities for the COV and DivS models , measured as the ability of the 1-D nonlinearities to reproduce the measured 2-D nonlinearity ( R2 ) across neurons ( **p<0 . 0005 , n = 13 ) . ( G–H ) STC analysis applied to an example neuron for which there was enough spiking data . ( G ) The spike-triggered average ( left ) , eigenvalue spectrum ( middle ) , and significant STC filters ( right ) . ( H ) As with the analyses of current responses above , the DivS filters ( green , cyan ) did not match those identified by STC ( left , dashed ) , but were largely contained in the subspace spanned by the STC filters ( right ) , as shown by comparing to their projections into the STC subspace ( dashed black ) . Note that there was not enough data to estimate 2-D nonlinearities for the spiking data , and so no comparison of STC model performances could be made . DOI: http://dx . doi . org/10 . 7554/eLife . 19460 . 007 To capture the computations that could be performed by such suppressive mechanisms , we constructed a 'divisive suppression' ( DivS ) model ( Figure 2A , bottom left ) . Terms simulating bipolar cell excitation and suppression are each described by a separate LN model , with a multiplicative interaction between their outputs such that the suppression impacts bipolar cell release ( Figure 2A ) . Note that the divisive gain control matches earlier models of both presynaptic inhibition ( Olsen and Wilson , 2008 ) and synaptic depression ( Markram et al . , 1998 ) . The suppressive term drops below one when the stimulus matches the suppressive filter , causing a proportional decrease in excitation of the ganglion cell . If the suppression does not contribute to the response , its nonlinearity would simply maintain a value of one , and the DivS model reduces to the LN model . The DivS model construction can be tractably fit to data using recent advances in statistical modeling ( Ahrens et al . , 2008b; McFarland et al . , 2013 ) . The DivS model fits were highly consistent across the population , with similarly shaped excitatory and suppressive filters across cells ( Figure 2B ) . For each cell , the suppressive filter was delayed relative to the excitatory filter ( 10 . 9 ± 2 . 2 ms , p<0 . 0005 , n = 13 , Figure 2C ) . The excitatory nonlinearity was approximately linear over the range of stimuli ( Figure 2D , left ) , whereas the suppressive nonlinearity decreased below one when the stimulus either matched or was opposite to the suppressive filter ( Figure 2D , right ) , resulting in ‘ON-OFF’ selectivity to both light increments and decrements . The DivS model outperformed the LN model in predicting the observed currents ( Figure 2E ) . Furthermore , it performed as well or better than models with other nonlinear interactions between the two filters . We first tested a more general form of nonlinear interaction by directly estimating a two-dimensional nonlinear function based on the filters derived from the DivS model . This 2-D nonlinearity maps each combination of the excitatory and suppressive filter outputs to a predicted current ( Figure 2F; see Materials and methods ) . While this 2-D model contains many more parameters than the DivS model , it did not perform significantly better ( Figure 2E ) ; indeed , the estimated 2-D nonlinearities for each neuron were well approximated by the separable mathematical form of the DivS model ( R2 for 2-D nonlinearity reconstruction = 0 . 94 ± 0 . 02; Figure 2G ) . We also tested an additive suppression ( AddS ) model , where suppression interacts with excitation additively ( see Materials and methods ) . The AddS model had significantly worse predictive power than the DivS model ( p<0 . 0005 , n = 13; Figure 2E ) and less resemblance to the corresponding 2-D nonlinearities compared to the DivS model ( p<0 . 0005 , n = 13; Figure 2G ) . Finally , we compared the DivS model to a form of spike-triggered covariance ( Fairhall et al . , 2006; Liu and Gollisch , 2015; Samengo and Gollisch , 2013 ) adapted to the continuous nature of the synaptic currents ( see Materials and methods ) . This covariance analysis generated different filters than the DivS model ( Figure 2—figure supplement 1 ) , although both sets of filters were within the same subspace ( Butts et al . , 2011; McFarland et al . , 2013 ) , meaning that the covariance-based filters could be derived as a linear combination of the DivS filters and vice versa . Because the filters shared the same subspace , the 2-D nonlinear mapping that converts the filter output to a predicted current had roughly the same performance as the 2-D model based on the DivS filters ( Figure 2E ) . However , because the covariance model used a different pair of filters ( and in particular the DivS filters are not orthogonal ) , its 2-D mapping differed substantially from that of the DivS model . Consequently , the 2-D mapping for the STC analysis , unlike the DivS analysis , could not be decomposed into two 1-D components ( Figure 2—figure supplement 1 ) ( Figure 2G ) . Thus , despite the ability of covariance analysis to nearly match the DivS model in terms of model performance ( Figure 2E ) , it could not reveal the divisive interaction between excitation and suppression . The DivS model therefore provides a parsimonious description of the nonlinear computation at the bipolar-ganglion cell synapse and yields interpretable model components , suggesting an interaction between tuned excitatory and suppressive elements . As we demonstrate below , the correspondingly straightforward divisive interaction detected by the DivS model on the ganglion cell synaptic input is essential in deriving the most accurate model of ganglion cell output , which combines this divisive interaction with subsequent nonlinear components related to spike generation . In addition to nearly perfect predictions of excitatory current at high contrast ( Figure 2; Figure 3C ) , the DivS model also predicted the time course of the synaptic currents at low contrast . Indeed , using a single set of parameters , the model was similarly accurate in both contrast conditions ( Figure 3A ) , and outperformed an LN model that used separate filters fit to each contrast level ( e . g . , Figure 1E ) . The DivS model thus implicitly adapts to contrast with no associated changes in parameters . 10 . 7554/eLife . 19460 . 008Figure 3 . DivS model explains temporal precision and contrast adaptation in synaptic currents . ( A ) The predictive power of models across contrasts . The DivS model is fit to both contrasts using a single set of parameters , and outperforms LN models fit separately to either high or low contrast ( LN-H and LN-L ) . As expected , the LN model fit for both contrasts ( LN-HL ) performs worse than separately fit LN models , because the LN-HL model cannot capture the filter changes without changes in model parameters . ( B ) Average coherence between model predictions and recorded synaptic currents on individual trials ( n = 13 ) , shown for high contrast ( HC ) and low contrast ( LC ) . The DivS model prediction performs almost identically to the trial-averaged response . ( C ) DivS model explains precision and contrast adaptation through the interplay of excitation and suppression . Top: comparison of predictions of synaptic current response of the LN model and the DivS model for the cell in Figure 1 . second row: normalized output of the excitatory and delayed suppressive filter . 3rd row: suppressive modulation obtained by passing the filtered output through the suppressive nonlinearity ( middle inset ) . Bottom: excitatory output of the DivS model before and after the suppressive modulation . In LC , the suppressive term ( third row ) does not deviate much from unity , and consequently the DivS model output resembles the excitatory input . ( D ) Comparison of the measured ( left ) and DivS model predicted ( right ) LN models across contrast . ( E ) The LN analysis applied to the DivS model predictions captures changes of both contrast gain ( left: R = 0 . 96 , p<10–6 ) and biphasic index ( right: R = 0 . 86 , p<0 . 0005 ) of the temporal filters across contrasts . ( F ) The DivS models predict the changes in tonic offset without any additional parameter shifts ( R = 0 . 90 , p<10–4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19460 . 008 The adaptation of the DivS model arises from the scaling of the divisive term with contrast . The fine temporal features in the synaptic currents observed at high contrast ( Figure 3C , left ) arise from the product of the output of the excitatory LN component and the output of the suppressive LN component . Because suppression is delayed relative to the excitation and has both ON and OFF selectivity , suppression increases at both positive and negative peaks of the suppressive filter output ( Figure 3C inset ) . This divisive suppression makes the DivS model output more transient compared to its excitatory component output alone; the difference between the two predictions is pronounced surrounding the times of peak excitation . At low contrast ( Figure 3C , right ) , both excitatory and suppressive filter outputs are proportionately scaled down . Because the suppression is divisive and close to one , the DivS model becomes dominated by the excitatory term and closely matches the LN model , as well as the measured excitatory current . The close match between data and DivS predictions across contrast levels suggests that the DivS model should exhibit contrast adaptation , as measured by the LN model filters ( e . g . , Figure 1E ) . Indeed , using LN analysis to describe the DivS-model-predicted currents across contrast shows that the changes in filtering properties predicted by the DivS model were tightly correlated with those from the measured data ( Figure 3D ) , including changes in contrast gain and biphasic index ( Figure 3E ) . Furthermore , a small tonic offset of the synaptic currents across contrast levels ( Figure 1—figure supplement 1 ) , which resulted in a vertical shift in the nonlinearity of the LN model ( Figure 1E; Figure 1—figure supplement 1 ) , was captured by the DivS model without any parameter changes ( Figure 3F ) . The mathematical form of the DivS model ( Figure 2A ) is consistent with two pre-synaptic mechanisms that shape temporal processing: synaptic depression ( Jarsky et al . , 2011; Ozuysal and Baccus , 2012 ) and presynaptic inhibition ( Eggers and Lukasiewicz , 2011; Schubert et al . , 2008 ) . Indeed , a model of ganglion cells that explicitly implements synaptic depression , the linear-nonlinear-kinetic model ( LNK model ) ( Ozuysal and Baccus , 2012 ) can also predict the temporal features of ganglion cell intracellular recordings across contrast . The LNK model fits a single LN filter ( analogous to the excitatory kE and fE ( . ) of the DivS model; Figure 2A ) , with additional terms that simulate use-dependent depletion of output ( Figure 4—figure supplement 1 ) . This depletion depends on the previous output of the model ( recovering over one or more time scales ) , and divisively modulates the output of the LN filter . For our data , the LNK model captured excitatory currents in response to the temporally modulated spot ( Figure 4A ) , also outperforming the LN model ( p<0 . 0005 , n = 13 ) , although not with the level of performance as the DivS model ( p<0 . 0005 , n = 13 ) . Furthermore , when data were generated de novo by an LNK model simulation , the resulting DivS model fit showed a delayed suppressive term , whose output well approximated the effect of synaptic depression in the LNK model ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 19460 . 009Figure 4 . Probing the mechanism of divisive suppression with center-surround stimuli . ( A ) For the large spot stimulus , the Linear-Nonlinear-Kinetic ( LNK ) model nearly matches the performance of the DivS model , and outperforms the LN model . ( B ) To distinguish between different sources of divisive suppression , we presented a spot-annulus stimulus ( left ) , where each region is independently modulated . Model filters can be extended to this stimulus using a separate temporal kernel for center and surround , shown for the LN and LNK model filters ( right ) , which are very similar . ( C ) After the linear filter , the LNK model applies a nonlinearity ( left ) , whose output drives the transition between resting and activated states ( middle ) , which is further governed by kinetics parameters as shown . Critical kinetics parameters for LNK models differed between the large-spot and spot-annulus stimulus ( right ) , with the spot-annulus model very quickly transitioning from Inactive back to Active states , minimizing the effects of synaptic depression . ( D ) The performance of the spatiotemporal LNK model is only slightly better than that of the LN model , and neither captures the details of the modulation in synaptic current , compared with the DivS model . ( E ) The spatiotemporal DivS model shown for an example neuron exhibits different spatial footprints for excitation and suppression , with excitation largely driven by the spot and suppression by the annulus . This divisive suppression cannot be explained exclusively by synaptic depression , which predicts overlapping sources of suppression and excitation ( Figure 4—figure supplement 1 and 2 ) . ( F ) The contribution of the center component in the DivS model for excitation ( left ) and suppression ( right ) . Excitation was stronger in the center than in the surround ( center contribution>0 . 5 , p=0 . 016 , n = 7 ) and suppression was weaker in the center ( center contribution<0 . 5 , p=0 . 016 , n = 7 ) for every neuron . ( G ) The DivS model captured temporal transients in the current response to spot-annulus stimuli better than the LN and LNK models . DOI: http://dx . doi . org/10 . 7554/eLife . 19460 . 00910 . 7554/eLife . 19460 . 010Figure 4—figure supplement 1 . DivS model localizes the suppressive components of LNK model and reproduces its simulated response . We simulated LNK models resembling the example neurons considered in Figure 4 . ( A–D ) LNK simulation in response to a temporally modulated spot . ( A ) The LNK model components consist of a temporal filter k ( left ) and static nonlinearity f ( ∙ ) ( middle ) , whose output u ( t ) =f[k∙s ( t ) ] governs the transition rate between the resting ( R ) and active ( A ) states . The current output is proportional to active state occupation , and other constants govern the transition to inactive ( I ) state and back to resting state . The parameters for this LNK simulation were derived from an LNK fit to an example neuron ( see Materials and methods ) . ( B ) A DivS model was fit to the LNK model simulated response , with components labeled as in Figure 2 . The temporal filter of suppression ( cyan ) is delayed relative to the excitation ( left ) and only results in suppression for ON stimuli , as expected given its relationship to synaptic depression . ( C ) Model performance ( R2 ) for the LN model and DivS model across all neurons demonstrates that the DivS model could reproduce LNK simulations with greater than 90% accuracy , across simulations of all LNK models of recorded neurons ( n = 13 ) . ( D ) Simulated response of the LNK model in ( A ) in response to a temporal modulated spot stimulus ( top ) . 2nd row: The output of the LNK simulation ( black ) could be reproduced better by a DivS model ( red ) fit to the simulated data , as compared to the LN model ( blue ) . 3rd row: The occupation of each internal state determined the current output in addition to the output of the LN component of the model . 4th row: The dynamics of the divisive suppression of the DivS model ( cyan ) roughly matched the occupation of the resting state of the LNK model ( 3rd row , green ) : the resting state occupancy ( and availability for transition to the active state and resulting current output in the LNK model ) was low at the same times there is suppression in the DivS model . ( E–G ) LNK simulation in response to the spot-annulus stimulus . ( E ) LNK components are labeled identically as in ( A ) , but now the filter k consists of separate components for the spot ( left , solid ) and annulus ( dashed ) regions of the stimulus . The temporal filter and nonlinearity were derived from the example cell in Figure 4B , but the kinetics parameters of the temporally modulated stimulus ( A ) were used in place of those derived for the spot-annulus stimuli , because the latter parameters did not result in nonlinear effects . ( F ) A DivS model fitted to the LNK model simulated response , with components labeled as in Figure 2 , resulting in the expected delayed ON suppression ( as with the temporally modulated spot simulations in B ) . ( G ) Simulated response using the LNK model with the spot-annulus stimulus , again with the divisive suppression of the DivS model ( 4th row , cyan ) capturing the occupancy of resting state of the LNK model ( 3rd row , green ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19460 . 01010 . 7554/eLife . 19460 . 011Figure 4—figure supplement 2 . DivS model descriptions of extended LNK models . Here we consider additional model structures involving synaptic depression . The simulations here incorporate nonlinear rectified subunits , and were limited to two components corresponding to those independently modulated in the stimulus: spot and annulus . ( A–D ) First we considered an extended LNK model with independent stimulus processing of the spot and annulus stimuli , and a shared synaptic depression stage . ( A ) Model schematic , showing that the separate 'center' and 'surround' components ( corresponding to spot and annulus stimuli ) are each rectified before being combined , and fed into the LNK model for synaptic depression , using the same kinetic parameters considered for simulations in Figure 4—figure supplement 1 . Simulated data were generated for a range of models of this form , where the weight for the ‘spot’ component wspot was fixed and the annulus component weight wannu was varied . ( B ) The DivS model components fit to an example simulated response of the extended LNK model ( with wspot = wannu ) . As with simpler circuits ( e . g . , Figure 4—figure supplement 1 ) , suppression was delayed relative to excitation . Note that the DivS model was limited to only a single rectified component to match the form used to describe experiments described in Figure 4 . ( C ) The performance of the LN ( red ) and DivS ( purple ) models across simulations with different annulus component weights . The DivS model performance was significantly better than that of the LN model over a wide range of parameters ( each point corresponds to the results of simulation with different choice of wannu ) , suggesting a large portion of the synaptic depression effect was captured by the DivS model . Note , however , that the DivS model has a more difficult time explaining this [simulated] data than the data from real ON-Alpha ganglion cells ( i . e . , Figure 4D ) . ( D ) For all simulations , the 'spatial profile' of suppression matched that of excitation , as measured by the 'center fraction' , which was given by the norm of the center component of the filter divided by the norm of the full filter . [The center fraction is one for no surround component , and zero for no center component . ] ( E–H ) We next considered an extended LNK model with both independent stimulus processing and independent kinetics . ( E ) Model schematic , showing the separate center and surround components each with independent synaptic depression — again with the same kinetic parameters previously considered . ( F ) The DivS model components fit to an example simulated response of the extended LNK model ( with wspot = wannu ) . ( G ) Performance of DivS model and LN model on simulated LNK model response . ( H ) Tight correlation of the center fractions of excitation versus divisive suppression of the DivS model components ( as in panel D ) . This and related simulations ( i . e . , with additional center-surround filtering prior to the rectification stage ) involving synaptic depression never yielded a case where DivS excitation was largely from the center and suppression was largely from the surround , which was observed in the real ON-Alpha cell data ( e . g . , Figure 4E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19460 . 011 The DivS and LNK models , however , yielded distinct predictions to a more complex stimulus where a central spot and surrounding annulus were modulated independently ( Figure 4B ) . The models described above were extended to this stimulus by including two temporal filters , one for the center and one for the surround . As expected from the center-surround structure of ganglion cell receptive fields , an LN model fit to this condition demonstrated strong ON-excitation from the center , and a weaker OFF component from the surround ( Figure 4B ) . The 'spatial' LNK model’s filter resembled that of the LN model ( Figure 4B ) . Consistent with this resemblance to the LN Model , the spatial LNK model had rate constants that minimized the time that the model dwelled in the inactivated state ( i . e . , was 'suppressed' ) ( Figure 4C ) . These rate constants were significantly different from those of the LNK model fit to the single temporally modulated spot . Correspondingly , the LNK model in the spot-annulus condition exhibited little performance improvement over the LN model ( predictive power improvement 1 . 8% ± 1 . 3% , p=0 . 016 , n = 7; Figure 4D ) . By comparison , the DivS model significantly outperformed the LN model with an improvement of 8 . 6% ± 3 . 3% ( p=0 . 016; n = 7 ) , and was 6 . 7 ± 2 . 8% better than the LNK model ( p=0 . 016; n = 7 ) . The suppressive term of the DivS model showed a very distinct spatial profile relative to excitation , with a greater drive from the annulus region , while excitation was mostly driven by the spot region ( Figure 4E , F ) . The suppressive filter processing the spot region was typically slower than the annulus filter: the peak latency for the suppressive filter was 129 ± 16 ms within the spot region compared to 120 ± 15 ms within the annulus region ( faster by 9 . 7 ± 4 . 3 ms; p=0 . 0156 , n = 7 ) . The strong suppression in the surround detected by the DivS model could not be explained by the LNK model , which cannot flexibly fit an explicit suppressive filter . Indeed , suppression in the LNK model arises from excitation , and thus the two components share the same spatial profile ( Figure 4B; Figure 4—figure supplement 1 and 2 ) . This can be demonstrated not only with simulations of the LNK model , but also more complex models with separate synaptic depression terms in center and surround ( Figure 4—figure supplement 2 ) . In all cases , application of the DivS model to data generated by these synaptic-depression-based simulations revealed that the suppressive term roughly matched the spatial profile of excitation , which is inconsistent with the observed data ( Figure 4F ) . While these analyses do not eliminate the possibility that synaptic depression plays a role in shaping the ganglion cell response ( and contributing to the suppression detected by the DivS model ) , the strength of surround suppression detected by the DivS model suggests that synaptic depression alone cannot fully describe our results . With an accurate model for excitatory synaptic currents established , we returned to modeling the spike output of ON-Alpha cells . Following previous likelihood-based models of ganglion cell spikes , we added a spike-history term , which implements absolute and relative refractory periods ( Butts et al . , 2011; McFarland et al . , 2013; Paninski , 2004; Pillow et al . , 2005 ) . The output of this spike-history term is added to the output of the DivS model for the synaptic currents , and this sum is further processed by a spiking nonlinearity ( Figure 5A ) to yield the final predicted firing rate . Using a standard likelihood-based framework , all terms of the model – including the excitatory and suppressive LN models that comprised the prediction of synaptic currents – can then be tractably fit using spike data alone . But it is important to note that this model architecture was only made clear via the analyses of synaptic currents described above . 10 . 7554/eLife . 19460 . 012Figure 5 . The extended divisive suppression model explains ganglion cell spike trains with high precision . ( A ) Model schematic for the divisive suppression model of spiking , which extends DivS model for the current data by adding an additional suppressive term for spike-history ( refractoriness ) , with the resulting sum passed through a rectifying spiking nonlinearity . ( B–E ) The model components for the same example neuron considered in Figures 1–3 . ( B ) The excitatory and suppressive filters . ( C ) . The excitatory and suppressive nonlinearities . The filters and nonlinearities were similar to the DivS model fit from current data ( shown in Figure 2B ) . ( D ) The spike-history term , demonstrating an absolute and relative refractory period . ( E ) The spiking nonlinearity , relative to the distribution of generating signals ( shaded ) . ( F ) . The predictive power of different models applied to the spike data in HC and LC . The DivS model performs better than other models ( HC: p<0 . 001; LC: p<0 . 002 , n = 11 ) , including the LN model , the LN model with spike history term ( LN+RP ) , and a divisive suppression model lacking spike refractoriness ( DivS-RP ) . Only a single set of parameters was used to fit the DivS model for both contrasts , whereas all other models shown used different parameters fit to each contrast . DOI: http://dx . doi . org/10 . 7554/eLife . 19460 . 012 When fit using spiking data alone , the resulting excitatory and suppressive filters and nonlinearities closely resembled those found when fitting the model to the synaptic currents recorded from the same neurons ( e . g . , Figure 2B , D ) . Suppression was consistently delayed relative to excitation ( Figure 5B ) , and exhibited both ON and OFF selectivity ( Figure 5C ) . The spike-history term was suppressive and had two distinct components , a strong absolute refractory period that lasted 1–2 ms and a second relative refractory period that lasted more than 15 ms ( Berry and Meister , 1998; Butts et al . , 2011; Keat et al . , 2001; Paninski , 2004; Pillow et al . , 2005 ) . The resulting model successfully captured over 90% of the predictable variance in the firing rate for all neurons in the study ( Figure 5F , median = 91 . 5% ± 1 . 0%; n = 11 ) , representing the best model performance reported in the literature for ganglion cell spike trains considered at millisecond resolution . By comparison , the standard LN model had a median predictive power of 62 . 8% ± 1 . 9% ( n = 11 ) ; which modestly increased to 68 . 8% ± 1 . 9% upon inclusion of a spike-history term ( Figure 5F ) . This suggests that ganglion cell spikes are strongly shaped by the nonlinear computations present at their synaptic input , and that the precise timing of ganglion cell spiking involves the interplay of divisive suppression with spike-generating mechanisms . To evaluate the relative contributions of divisive suppression and spike refractoriness to predicting firing , we simulated spike trains using different combinations of model components ( Figure 6A ) . We found that the parameters of the divisive suppression components could not fit without including a spike-history term , suggesting that each component predicts complementary forms of suppression . We could generate a DivS model without a spike-history term , however , by first determining the full model ( with spike-history term ) , and then removing the spike-history term and refitting ( see Materials and methods ) , resulting in the DivS–RP model . This allowed for direct comparisons between models with a selective deletion of either divisive suppression or spike refractoriness ( Figure 6A ) . 10 . 7554/eLife . 19460 . 013Figure 6 . Spike patterning is shaped by a combination of nonlinear mechanisms . ( A ) Top: Spike rasters recorded over ten repeats for an example cell ( black ) compared with simulated spikes from four models: LN , LN model with spike-history term ( LN+RP ) , the DivS model without spike-history ( DivS-RP ) , and the full DivS model ( DivS ) . Colors in the raster label separate spike events across trials ( see Materials and methods ) . Bottom: The PSTHs for each model demonstrate that suppressive terms are important in shaping the envelope of firing ( DivS prediction is shaded ) . ( B–E ) Using event labels , spike statistics across repeats were compiled to gauge the impact of different model components . ( B ) The temporal properties of events compared with model predictions , across contrast ( same as Figure 1 , with DivS-based models added ) . Both spike-history and divisive suppression contribute to reproduce the temporal scales across contrast . ( C ) The Fano factor for each event is a measure of reliability , which increased ( i . e . , Fano factor decreased ) for models with a spike-history term . DOI: http://dx . doi . org/10 . 7554/eLife . 19460 . 013 Event analyses on the resulting simulated spike trains , compared with the observed data , demonstrate that both divisive suppression ( derived from the current analyses above ) and spike refractoriness were necessary to explain the precision and reliability of ganglion cell spike trains . By comparing the two models without DivS ( LN and LN+RP ) to those with DivS ( DivS and DivS–RP ) , it is clear that divisive suppression is necessary to predict the correct envelope of the firing rate ( Figure 6B ) . Note , however , that DivS had little impact in the low contrast condition , which lacked fine-time-scale features of the spike response . By comparison , the spike-history term had little effect on the envelope of firing ( Figure 6A , bottom ) , and contributed little to the fine time scales in the ganglion cell spike train at high contrast ( Figure 6B ) . Instead , the spike-history term had the largest effect on accurate predictions of event reliability , as reflected in the event Fano factor ( Figure 6C ) . By comparison , both models without the spike-history term had much greater variability in spike counts within each event . The presence of the suppression contributed by the spike-history term following each event allows the predicted firing rate to be much higher ( and more reliable ) during a given event , resulting in reliable patterns of firing within each event ( Figure 6A ) ( Pillow et al . , 2005 ) . We conclude that a two-stage computation present in the spike-DivS model , with both divisive suppression and spike refractoriness , is necessary to explain the detailed spike patterning on ON-Alpha ganglion cells . In addition to accurate reproduction of precise spike outputs of ganglion cells , the DivS model also captured the effects of contrast adaptation observed in the ganglion cell spike trains . For both contrast conditions , the simulated spike trains , which are predicted for both contrasts using a single set of parameters , were almost indistinguishable from the data ( Figure 7A , top ) . As with the performance of the models of excitatory current ( Figure 3 ) , the DivS model outperformed LN models that were separately fit for each contrast level ( Figure 5F ) . 10 . 7554/eLife . 19460 . 014Figure 7 . Contrast adaptation in the spike output depends on both divisive suppression and spike refractoriness . ( A ) The full spike-DivS model accurately captured contrast adaptation . Top: observed PSTH and predicted firing rates of the DivS model at HC and LC . ( B ) The DivS model predicted the changes in LN filter shape and magnitude with contrast for an example cell . Predicted changes are shown for each model , demonstrating that the full effects of contrast adaptation require both divisive suppression and spike-history terms . ( C ) Measured and predicted contrast gain ( top ) and changes of biphasic index ( bottom ) . The DivS model accurately predicted a contrast gain and changes biphasic index across contrast across cells ( contrast gain: slope of regression = 0 . 75 , R = 0 . 85 , p<0 . 001; biphasic index: slope of regression = 0 . 87 , R = 0 . 87 , p<0 . 001 ) . DivS model without the spike history term underestimated contrast adaptation ( contrast gain: slope of regression = 0 . 53 , R = 0 . 51 , p = 0 . 10; biphasic index: slope of regression = 0 . 49 , R = 0 . 73 , p<0 . 05 ) , and the LN+RP model failed to predict adaptation altogether ( contrast gain: slope of regression = 0 . 06 , R = 0 . 28 , p = 0 . 41; biphasic index: slope of regression = −0 . 07 , R = −0 . 18 , p = 0 . 60 ) . ( D ) The suppressive effect from the spike-history term was amplified at HC , due to the increased precision of the spike train . Dashed lines show the onset of HC spike events , which predict the largest difference in the magnitudes of the suppression between contrasts . DOI: http://dx . doi . org/10 . 7554/eLife . 19460 . 014 The ability to correctly predict the effects of contrast adaptation depended on both the divisive suppression and spike-refractoriness of the spike-DivS model . This is shown for an example neuron by using LN filters of the simulated output of each model at high and low contrasts ( Figure 7B ) . In this case , only the DivS model ( which includes a spike-history term ) shows adaptation similar to that observed by the LN filters fit to the data . We quantified this by identifying the most prominent feature of adaptation of the LN filters , the change in filter amplitude ( i . e . , contrast gain ) . Across the population , the DivS correctly predicted the magnitude of this change ( Figure 7C , top ) , as well as the changes in biphasic index across contrasts ( Figure 7C , bottom ) , and outperformed models with either the divisive suppression or spike-history terms missing . As expected , spike refractoriness imparted by the spike-history term contributed to the stronger effects of contrast adaptation observed in spikes relative to synaptic inputs ( Beaudoin et al . , 2007; Kim and Rieke , 2001 , 2003; Rieke , 2001; Zaghloul et al . , 2005 ) . Specifically , at high contrast , spikes concentrate into relatively smaller time windows , leading to a consistently timed effect of spike refractoriness ( Figure 7D ) . As a result , despite similar numbers of spikes at the two contrasts , the effect of the spike-history term has a bigger impact at high contrast . Thus , fast contrast adaptation – and more generally the temporal shaping of ON-Alpha ganglion cell spike trains – depends on nonlinear mechanisms at two stages of processing within the retinal circuit . Both aspects of nonlinear processing originate at the level of ganglion cell synaptic inputs , shaped by divisive suppression , and become amplified by spike-refractoriness to generate the array of nonlinear properties evident in the spike train output . One important nonlinear response property of early sensory neurons is high temporal precision . Temporal precision of spike responses has been observed in the retinal pathway with both noise stimuli ( Berry et al . , 1997; Reinagel and Reid , 2000 ) and natural movies ( Butts et al . , 2007 ) . The precise spike timing suggests a role for temporal coding in the nervous system ( Berry et al . , 1997 ) , or alternatively simply suggests that analog processing in the retina must be oversampled in order to preserve information about the stimulus ( Butts et al . , 2007 ) . Temporal precision also plays an important role in downstream processing of information provided by ganglion cells ( Stanley et al . , 2012; Usrey et al . , 2000 ) . The generation of temporal precision involves nonlinear mechanisms within the retina , which may include both spike-refractoriness within ganglion cells ( Berry and Meister , 1998; Keat et al . , 2001; Pillow et al . , 2005 ) and the interplay of excitation and inhibition ( Baccus , 2007; Butts et al . , 2016; Butts et al . , 2011 ) . Such distinct mechanisms contributing to ganglion cell computation are difficult to distinguish using recordings of the spike outputs alone , which naturally reflect the total effects of all upstream mechanisms . By recording at two stages of the ganglion cell processing , we demonstrated that high temporal precision has already presented in the synaptic current inputs at high contrast , and temporal precision of both current inputs and spike outputs can be accurately explained by the DivS model . The DivS model explained fast changes in the response through the interplay of excitation and suppression . For both the spike and current models , suppression is consistently delayed relative to excitation . The same suppression mechanism also likely underlies high temporal precision of LGN responses , which can be captured by a model with delayed suppression ( Butts et al . , 2011 ) . Indeed , recordings of both LGN neurons and their major ganglion cell inputs suggest that precision of LGN responses is inherited from the retina and enhanced across the retinogeniculate synapse ( Butts et al . , 2016; Carandini et al . , 2007; Casti et al . , 2008; Rathbun et al . , 2010; Wang et al . , 2010b ) . Therefore , our results demonstrate that the temporal precision in the early visual system likely originates from nonlinear processing in the inputs to retinal ganglion cells . Note that the full spiking-DivS model did not incorporate any form of direct synaptic inhibition onto the ON-Alpha cell , consistent with findings that the impact of such inhibition is relatively weak and that excitation dominates the spike response in the stimulus regime that we studied ( Kuo et al . , 2016; Murphy and Rieke , 2006 ) . Our results show that the contribution of spike-history term to precision – as measured by the time scale of events and first-spike jitter – seems minor , consistent with earlier studies in the LGN ( Butts et al . , 2016; Butts et al . , 2011 ) . Nevertheless , the spike-history term does play an important role in spike patterning within the event ( Pillow et al . , 2005 ) and the resulting neuronal reliability ( Berry and Meister , 1998 ) . In fact , we could not fit the divisive suppression term robustly without the spike-history term in place , suggesting that both nonlinear mechanisms are important to explain the ganglion cell firing . Here we modeled contrast adaptation at the level of synaptic currents and spikes from the same ganglion cell . We found contrast adaptation in synaptic inputs to ganglion cells , consistent with previous studies ( Beaudoin et al . , 2007; Kim and Rieke , 2001; Rieke , 2001; Zaghloul et al . , 2005 ) . Such adaptation could be explained by divisive suppression , which takes a mathematical form similar to previously proposed gain control models ( Heeger , 1992; Shapley and Victor , 1979 ) . Because the suppressive nonlinearity has a very different shape than the excitatory nonlinearity , divisive suppression has a relatively strong effect at high contrast and results in a decrease in measured gain . Moreover , the same divisive suppression mechanism may also explain nonlinear spatial summation properties of ganglion cells ( Shapley and Victor , 1979 ) , because suppression generally has broader spatial profiles than excitation . Contrast adaptation is amplified in the spike outputs mostly due to spike refractoriness and changes of temporal precision across contrast . At high contrast , the response had higher precision and occurred within shorter event windows ( Butts et al . , 2010 ) . As a result , the accumulated effect of spike refractoriness was stronger within each response event . Note that the effect of the spike-history term was highly dependent on the ability of the model to predict fine temporal precision at high contrast , which largely originates from the divisive suppression term as discussed earlier . Therefore , the two nonlinear properties of retinal processing , contrast adaptation and temporal precision , are tightly related and can be simultaneously explained by the DivS model . Divisive suppression has been observed in the invertebrate olfactory system ( Olsen and Wilson , 2008 ) , the lateral geniculate nucleus ( Bonin et al . , 2005 ) , the primary visual cortex ( Heeger , 1992 ) , and higher visual areas such as area MT ( Simoncelli and Heeger , 1998 ) . A number of biophysical and cellular mechanisms for divisive suppression have been proposed , including shunting inhibition ( Abbott et al . , 1997; Carandini et al . , 1997; Hao et al . , 2009 ) , synaptic depression ( Abbott et al . , 1997 ) , presynaptic inhibition ( Olsen and Wilson , 2008; Zhang et al . , 2015 ) and fluctuation in membrane potential due to ongoing activity ( Finn et al . , 2007 ) . We evaluated different mechanistic explanations of the divisive suppression identified in this study . Divisive suppression underlying synaptic inputs to ganglion cells cannot be attributed to fluctuations in membrane potential or shunting inhibition since we recorded synaptic currents under voltage-clamp conditions that minimize inhibitory inputs . Although synaptic depression could also explain fast transient responses and contrast adaptation ( Ozuysal and Baccus , 2012 ) , synaptic depression will generally result in excitation and suppression that have the same spatial profiles ( Figure 4—figure supplement 1 and 2 ) , whereas we show that excitation and suppression have distinct spatial profiles ( Figure 4 ) . Therefore , the divisive suppression in our model likely depends partly on presynaptic inhibition from amacrine cells , which can extend their suppressive influence laterally ( Euler et al . , 2014; Franke et al . , 2016; Schubert et al . , 2008 ) . This was somewhat surprising , because earlier studies had shown that contrast adaptation persisted in the presence of inhibitory receptor antagonists , suggesting that adaptation depended primarily on mechanisms intrinsic to the bipolar cell ( e . g . , synaptic depression ) , independent of synaptic inhibition ( Beaudoin et al . , 2007; Brown and Masland , 2001; Rieke , 2001 ) . Indeed , synaptic depression was likely the primary mechanism for adaptation in conditions with inhibition blocked , but the present results suggest that lateral inhibitory mechanisms also play a role in generating adaptation under conditions with inhibition intact , at least for some retinal circuits . Models for contrast adaptation based on synaptic depression rely on a change in the average level of synaptic activity with contrast . This condition is met for synapses with lower rates of tonic release ( Jarsky et al . , 2011 ) . However , the ON-Alpha cell receives a relatively high rate of tonic release from presynaptic type 6 bipolar cells ( Borghuis et al . , 2013; Schwartz et al . , 2012 ) . Consequently , the average excitatory synaptic input would change less during contrast modulation compared to a ganglion cell that received a lower rate of glutamate release . An inhibitory mechanism for contrast adaptation thus may play a relatively prominent role for retinal circuits , such as the ON-Alpha cell , driven by high rates of glutamate release . Detailed anatomical studies suggest that each ganglion cell type receives inputs from a unique combination of bipolar and amacrine cell types , contributing to a unique visual computation ( Baden et al . , 2016 ) . By focusing on a single cell type , the ON-Alpha cell , we identified a particular computation consistent across cells . We expect that other ganglion cell types will perform different computations , and likewise have different roles in visual processing . This could include additional contrast-dependent mechanisms , including slow forms of adaptation ( Baccus and Meister , 2002; Manookin and Demb , 2006 ) , sensitization ( Kastner and Baccus , 2014 ) and complex changes in filtering ( Liu and Gollisch , 2015 ) . Thus , further applications of the approach described here will uncover a rich diversity of computation constructed by retinal circuitry to format information for downstream visual processing . Data were recorded from ON-Alpha ganglion cells from the in vitro mouse retina using the procedures described previously ( Borghuis et al . , 2013; Wang et al . , 2011 ) . Spikes were recorded in the loose-patch configuration using a patch pipette filled with Ames medium , and synaptic currents were recorded using a second pipette filled with intracellular solution ( in mM ) : 110 Cs-methanesulfonate; 5 TEA-Cl , 10 HEPES , 10 BAPTA , 3 NaCl , 2 QX-314-Cl , 4 ATP-Mg , 0 . 4 GTP-Na2 , and 10 phosphocreatine-Tris2 ( pH 7 . 3 , 280 mOsm ) . Lucifer yellow was also included in the pipette solution to label the cell using a previously described protocol ( Manookin et al . , 2008 ) . The targeted cell was voltage clamped at ECl ( −67 mV ) to record excitatory currents after correcting for the liquid junction potential ( −9 mV ) . Cells in the ganglion cell layer with large somas ( 20–25 μm diameter ) were targeted . Cells were confirmed to be ON-Alpha cells based on previously established criteria ( Borghuis et al . , 2013 ) : ( 1 ) an ON response; ( 2 ) high rate of spontaneous firing; and a high rate of spontaneous excitatory synaptic input; ( 3 ) a low input resistance ( ~40–70 MΩ ) . In some cases , we imaged the morphology of recorded cells and confirmed ( 4 ) a relatively wide dendritic tree ( 300–400 μm diameter ) and ( 5 ) stratification on the vitreal side of the nearby ON cholinergic ( starburst ) amacrine cell processes . We made recordings from 27 ON-Alpha cells total , each in one or more of the experimental conditions described . Of the 15 cells recorded in cell-attached configuration ( spike recordings ) , 4 cells were excluded where low reliability across trials indicated an unstable recording , as indicated by much higher spike event Fano Factors ( >0 . 2 , see below ) . All procedures were conducted in accordance with National Institutes of Health guidelines under protocols approved by the Yale University Animal Care and Use Committee . The temporally modulated spot stimulus was described previously ( Wang et al . , 2011 ) . The retina was stimulated by UV LEDs ( peak , 370 nm; NSHU-550B; Nichia America ) to drive cone photoreceptors in the ventral retina . UV LEDs were diffused and windowed by an aperture in the microscope’s fluorescence port , with intensity controlled by pClamp 9 software via a custom non-inverting voltage-to-current converter using operational amplifiers ( TCA0372; ON Semiconductor ) . The stimulus was projected through a 4X objective lens ( NA , 0 . 13 ) . The stimulus was a flickering spot ( 1-mm diameter ) , with intensity generated from low pass Gaussian noise with a 30 Hz cutoff frequency . We used a contrast-switching paradigm ( Baccus and Meister , 2002; Kim and Rieke , 2001; Zaghloul et al . , 2005 ) , in which the temporal contrast alternately stepped up or down every 10 s . The contrast of the stimulus is defined by the SD of the Gaussian noise and was either 0 . 3 times ( high contrast ) or 0 . 1 times ( low contrast ) the mean . Note that this is only a three-fold difference in contrast versus the seven-fold difference considered in Ozuysal and Baccus ( 2012 ) , but sufficient to see clear contrast effects . The stimulus comprised 10 cycles of 10 s for each contrast . The first 7 s were unique in each cycle ( used for fitting model parameters ) , and the last 3 s were repeated across cycles ( used for cross-validation of model performance ) . The center-surround stimuli ( Figure 4B ) were generated in Matlab ( Mathworks , Natick ) using the Psychophysics Toolbox ( Brainard , 1997 ) and presented with a video projector ( M109s DLP; Dell , or identical HP Notebook Companion; HP ) , modified to project UV light ( single LED NC4U134A , peak wavelength 385 nm; Nichia ) as previously described ( Borghuis et al . , 2013 ) . The center and surround stimuli were independently modulated with Gaussian noise ( 60-Hz update rate ) . A spot covered the receptive field center ( e . g . , 0 . 3 mm ) , and an annulus extended into the surround ( e . g . , inner/outer diameters of 0 . 35/1 . 0 mm ) . We recorded 7 ON-Alpha cells in this condition . For a subset of the recordings ( n=5 ) , we explored a range of inner/outer diameters , and selected the diameters that maximized the difference between the spatial footprints of excitatory and suppressive terms of the DivS model ( see below ) . The mean luminance of the stimulus was calculated to evoke ~4×104 photoisomerizations cone−1 sec−1 , under the assumption of a 1 μm2 cone collecting area . For all methods of stimulation , the gamma curve was corrected to linearize output , and stimuli were centered on the cell body and focused on the photoreceptors . We verified that the relatively short stimulus presentation did not result in significant bleaching , as the response ( and model parameters ) had no consistent trends from the beginning of the experiment to the end ( Figure 1—figure supplement 1–2 ) . We modeled the synaptic current response of neurons using the traditional linear-nonlinear ( LN ) cascade model ( Paninski , 2004; Truccolo et al . , 2005 ) , the Linear-Nonlinear-Kinetic model ( Ozuysal and Baccus , 2012 ) , a 2-D nonlinear model ( ‘2-D’ ) , and the Divisive Suppression model ( ‘DivS’ ) introduced in this paper . In all cases ( with the exception of the LN analyses of contrast adaptation effects described below ) , we optimized model parameters to minimize the mean-squared error ( MSE ) between the model-predicted and observed currents: ( 1 ) MSE=∑t[c ( t ) −cobs ( t ) ]2 To limit the number of model parameters in the minimization of MSE , we represented temporal filters by linear coefficients weighting a family of orthonormalized basis functions ( Keat et al . , 2001 ) : ( 2 ) ζ ( t ) =sin⁡[πn ( 2t/tF− ( t/tF ) 2 ) ] , where tF=200 ms . We applied several statistical models to describe the spike response of ganglion cells . We first considered the generalized linear model ( GLM ) ( Paninski , 2004; Truccolo et al . , 2005 ) . We assumed that spike responses are generated by an inhomogeneous Poisson process with an instantaneous rate . The GLM makes prediction of the instantaneous firing rate of the neuron r ( t ) based on both the stimulus s ( t ) and the recent history of the spike train R ( t ) : ( 8 ) r ( t ) =Fspk[klin⋅s ( t ) +hspk⋅R ( t ) −θ] , where klin is a linear receptive field , hspk is the spike-history term and θ is the spiking threshold . Note that for fitting the model parameters of the GLM ( as well as those of other models with a spike-history term hspk ) , we used the observed spikes for predicting firing rates ( i . e . , R ( t ) was derived from the observed spike train ) . However , to generate cross-validated predictions , we did not use the observed spikes , and instead calculated R ( t ) iteratively as the history of a simulated spike train that was generated using a Poisson process ( see Butts et al . 2011 for more details ) . The parameters of the GLM are all linear functions inside the spiking nonlinearity Fspk[⋅] . The LN model consists of only the linear receptive field and the spiking threshold; the full GLM further includes the spike-history term ( denoted as LN+RP in figures ) . The spiking nonlinearity had a fixed functional form Fspk[g]=log[1+exp ( g ) ] , satisfying conditions for efficient optimization ( Paninski , 2004 ) . The choice of this particular parametric form of spiking nonlinearity was verified with standard non-parametric estimation of the spiking nonlinearity ( Figure 1B , E ) ( Chichilnisky , 2001 ) . The model parameters were estimated using maximum likelihood optimization . The log-likelihood ( LL ) of the model parameters that predict a firing rate r ( t ) given the observed neural response robs ( t ) is ( Paninski , 2004 ) : ( 9 ) LL=∑t[robs ( t ) log r ( t ) −r ( t ) ] The optimal model parameters could then be determined using gradient-descent based optimization of LL . Although this formulation of the model assumes probabilistic generation of spikes , this fitting procedure was able to capture the parameters of the equivalent integrate-and-fire neuron , and thus was not impacted by the form of noise related to spike generation ( Butts et al . , 2011; Paninski et al . , 2007 ) . To capture nonlinear properties of the spike response , we extended the Nonlinear Input Model ( NIM ) ( McFarland et al . , 2013 ) to include multiplicative interactions . The predicted firing rate of the NIM is given as: ( 10 ) r ( t ) =Fspk[∑iζi[s ( t ) ] +hspk⋅R ( t ) −θ] , where each ζi[⋅] represents a component of [potentially nonlinear] upstream processing . The original formulation of the NIM assumed this upstream processing had the form of an LN model ( making the NIM an LNLN cascade ) . However , based on knowledge of nonlinear processing in the synaptic current response , here we assumed that the upstream components ζi[s ( t ) ] take the form of a DivS model: ( 11 ) ζ[s ( t ) ]=fe[ke⋅s ( t ) ] × fs[ks⋅s ( t ) ] . Similar to parameter estimation of the DivS models of synaptic current response , we alternately estimated the filters and nonlinearities until they converged . The set of constraints on excitatory and suppressive nonlinearities was the same as with DivS model of synaptic currents . We performed a more traditional LN model analysis to gauge the adaptation to contrast of both the observed data as well as the predictions of nonlinear models , following ( Chander and Chichilnisky , 2001; Baccus and Meister , 2002 ) . We first separately performed LN analysis for each contrast level , establishing the filter by reverse correlation and then using the filter output at each contrast to separately estimate each nonlinearity . The nonlinearities were then aligned by simultaneously estimating a scaling factor for the x-axis and an offset for the y-axis to minimize the mean squared deviation between them . The associated scaling factor was incorporated into the linear filters such that contrast gain was attributable entirely to changes in the linear filter ( Chander and Chichilnisky , 2001 ) . While the offset parameter was not used in ( Chander and Chichilnisky , 2001 ) , we found it was necessary to fully describe the changes in the nonlinearities associated with the synaptic current ( but not spiking ) data ( Figure 1—figure supplement 1 ) . This additional offset is shown in the comparison between the two nonlinearities across contrast ( e . g . , Figure 1E ) . Once the linear filters at both contrasts were obtained , we calculated contrast gain as the ratio of standard deviations of the filters at low and high contrast conditions . To make more detailed comparisons about the filter shape , we also calculated a biphasic index , based on the ratio of the most negative to the most positive amplitude of the LN filter k , i . e . , \|min ( k ) /max ( k ) \| . We fit all models on the 7 s segments of unique stimuli within each 10 s block , and cross-validated model performance on the 3 s repeat trials . We calculated the predictive power , or percent of explainable variance ( Sahani and Linden , 2003 ) , to quantify how well the model captured the trial-averaged response for both intracellular and extracellular recordings . This metric is based on the fraction of explained variance ( R2 ) but corrects for noise-related bias due to a limited number of trials . For validation of spike-based models , we simulated individual instances of spike trains using a non-homogeneous Poisson process , and the model predictions were based on measuring a PSTH to 500 simulated repeats in response to the cross-validation stimulus . All measures of model performance compared predicted to measured responses using 1 ms bins , which was necessary to measure how accurately the different models captured temporal precision ( Butts et al . , 2011 , 2007 ) . The general model performance metrics such as predictive power and cross-validated likelihood do not reveal which aspects of the response are not captured by the model . We thus devised a new coherence-based metric to quantify how well the model performs across temporal frequencies . The coherence between the model predicted current response c ( t ) and the recorded current response on the ith trial cobsi ( t ) is ( Butts et al . , 2007 ) : ( 12 ) γi2 ( ω ) =⟨\|Cobsi ( ω ) C ( ω ) ¯\|2⟩⟨\|Cobsi ( ω ) \|2⟩⟨\|C ( ω ) \|2⟩ where C ( ω ) and Cobsi ( ω ) are the Fourier transforms of c ( t ) and cobsi ( t ) respectively , and the bar denotes complex conjugate . We used angular frequency ω=2πf instead of f to be consistent with common conventions . The coherence measure on individual trials was averaged across repeats for each cell . Because the observed response on each trial contains noise , a coherence of one throughout the frequency is not a realistic target . To correct for this bias , we calculated the coherence between the trial-averaged current response ( i . e . , the ideal predictor of response ) and the recorded current on each trial . This noise corrected coherence metric represents an upper bound of coherence that can be achieved by any stimulus-processing model . It also reflects the consistency of current response at each frequency . For example , in the low contrast condition , the response contained little high frequency components ( Figure 7A–B ) , and consequently the measured coherence was close to zero above 30 Hz . We modified a previously established method to identify short firing episodes ( events ) in the spike train ( Berry et al . , 1997; Butts et al . , 2010; Kumbhani et al . , 2007 ) . Specifically , events were first defined in the peristimulus time histogram ( PSTH ) as times of firing interspersed with periods of silence lasting ≥ 8 ms . Each resulting event was further analyzed by fitting the PSTH with a two-component Gaussian mixture model . An event was broken into two events if the differences of means of the two Gaussian components exceed two times the sum of standard deviations . Event boundaries were defined as the midpoint between neighboring event centers and were used when assigning event labels to simulated spikes . Events were excluded from further analysis if no spike was observed on more than 50% of the trials during the event window . This criterion excluded spontaneous spikes that occurred on few trials . Event analysis was first performed on responses at high contrast . Events at low contrast were defined using the event boundaries obtained from high contrast data . These particular methods were chosen because they gave the most reasonable results with regards to visual inspection , but the results presented here do not qualitatively depend on the precise methods . Once the events were parsed , we measured several properties associated with each event relating to their precision and reliability ( Figures 1 , 6 ) . First , we measured the jitter in the timing of the first-spike , using the SD of the first spike of the event on each trial . Then , the event time scale was estimated as the SD of all spike times in each event , which is related to the duration of each event . Finally , the event Fano factor measured the ratio between the variance of spike count and the mean spike count in each event . All statistical tests performed in the manuscript were non-parametric Wilcoxon signed rank tests , unless otherwise stated . All significant comparisons were also significant using t-tests .	Visual processing begins in the retina , a layer of light-sensitive tissue at the back of the eye . The retina itself is made up of three layers of excitatory neurons . The first comprises cells called photoreceptors , which absorb light and convert it into electrical signals . The photoreceptors transmit these signals to the next layer , the bipolar cells , which in turn pass them on to the final layer , the retinal ganglion cells . The latter are responsible for sending the signals on to the brain . Other cells in the retina inhibit the excitatory neurons and thereby regulate their signals . While the basic structure of the retina has been described in detail , we know relatively little about how retinal ganglion cells represent information from visual scenes . Existing models of vision fail to explain several aspects of retinal ganglion cell activity . These include the exquisite timing of ganglion cell responses , and the fact that retinal ganglion cells adjust their responses to suit different visual conditions . In the phenomenon known as contrast adaptation , for example , ganglion cells become more sensitive during small variations in contrast ( differences in color and brightness ) and less sensitive during high variations in contrast . To understand how ganglion cells process visual stimuli , Cui et al . recorded the inputs and outputs of individual ganglion cells in samples of tissue from the mouse retina . By feeding these data into a computer model , Cui et al . were able to identify the mathematical calculations that take place at each stage of the retinal circuit . The findings suggest that a key element shaping the response of ganglion cells is the interaction between two visual processing pathways at the level of the bipolar cells . The resulting model can predict the responses of ganglion cells to specific inputs from bipolar cells with millisecond precision . Future studies should extend the model to more complex visual stimuli . The approach could also be adapted to study different types of ganglion cells in order to obtain a more complete picture of the workings of the retina .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Divisive suppression explains high-precision firing and contrast adaptation in retinal ganglion cells
The perception of visceral pain is a complex process involving the spinal cord and higher order brain structures . Increasing evidence implicates the gut microbiota as a key regulator of brain and behavior , yet it remains to be determined if gut bacteria play a role in visceral sensitivity . We used germ-free mice ( GF ) to assess visceral sensitivity , spinal cord gene expression and pain-related brain structures . GF mice displayed visceral hypersensitivity accompanied by increases in Toll-like receptor and cytokine gene expression in the spinal cord , which were normalized by postnatal colonization with microbiota from conventionally colonized ( CC ) . In GF mice , the volumes of the anterior cingulate cortex ( ACC ) and periaqueductal grey , areas involved in pain processing , were decreased and enlarged , respectively , and dendritic changes in the ACC were evident . These findings indicate that the gut microbiota is required for the normal visceral pain sensation . Accumulating evidence indicates that the gut microbiota communicates with the central nervous system ( CNS ) in a bidirectional manner thereby influencing brain function and behavior ( Sampson and Mazmanian , 2015; Dinan and Cryan , 2012; Mayer , 2011 ) . Although the majority of studies investigating the effects of the microbiota on brain function involve animal models of anxiety , depression , and cognitive dysfunction , it is now becoming clear that the gut microbiota may also have a role in other CNS-related conditions , such as visceral pain ( O'Mahony et al . , 2014; Gareau et al . , 2007; McKernan et al . , 2010 ) . Abdominal pain , often characterized by visceral hypersensitivity , is a common , and at times , dominant symptom of several gastrointestinal disorders , including functional dyspepsia and irritable bowel syndrome ( IBS ) ( Enck et al . , 2016 ) . There is also a high comorbidity among visceral pain and psychiatric disorders such as depression and anxiety ( Felice et al . , 2015 ) . These painful events are often recurring and unpredictable , which can have a debilitating impact on a person’s daily life ( Quigley , 2006 ) . Moreover , many gastrointestinal disorders with visceral pain as a component lack an identifiable pathology and can be difficult to treat with current pharmaceuticals , many of which are associated with undesirable side effects ( Wood , 2013; Moloney et al . , 2016 ) . The perception of visceral pain is a complex process involving peripheral sensory nerves , and , in the CNS , spinal and cortical pathways as well as areas associated with integration of the experience of pain ( Apkarian et al . , 2005 ) . Pathological pain states have been associated with altered neuroimmune signaling and glial activation in the spinal cord ( Ji et al . , 2013; Grace et al . , 2014 ) . In the brain , there is a significant overlap in areas regulating the affective component of visceral pain and those mediating psychological stress , a major predisposing factor for visceral hypersensitivity ( Larauche et al . , 2012 ) . Imaging studies in humans with IBS ( Tillisch et al . , 2011; Mertz et al . , 2000 ) and in animal models of visceral hypersensitivity ( Gibney et al . , 2010; Felice et al . , 2014; Bliss et al . , 2016 ) have revealed increased activation in the medial prefrontal cortex ( mPFC ) in response to both viscerally painful and stressful stimuli . Numerous studies now indicate that the perception of visceral pain can by influenced by the intestinal microbiota . In rodent studies , specific bacterial strains ( probiotics ) have been shown to ameliorate visceral pain induced by stress ( Gareau et al . , 2007; McKernan et al . , 2010; Ait-Belgnaoui et al . , 2006 ) or antibiotic administration ( Verdú et al . , 2006 ) , and many probiotics have been shown to benefit humans with abdominal pain ( Clarke et al . , 2012 ) . Intriguingly , visceral hypersensitivity can be transferred via the microbiota of IBS patients to animals previously lacking microbes ( Crouzet et al . , 2013 ) . However , the mechanisms underlying the effects of the microbiota on visceral pain perception remain to be elucidated . Integrating these observations , this study is based on the hypothesis that the microbiota is required for the normal processing of visceral pain stimuli . To this end , we used germ-free mice ( GF; mice raised without any exposure to microorganisms ) to study pain-related behavior , genes , and brain structures . We first studied visceral sensitivity , expression of cytokines , Toll-like receptors ( TLRs ) , and glial markers in the spinal cord , and mPFC morphology in these animals . Secondly , we determined if postnatal microbial colonization could normalize visceral hypersensitivity and immune status within the spinal cord of GF mice . Four cohorts of animals were used to perform the experiments in this paper . Animals in the first cohort ( CC = 10 , GF = 9; Figure 1 ) underwent CRD and then were euthanized . Animals in the second cohort ( CC and GF = 9–10; Figure 2 ) were sacrificed without anesthesia and spinal cord tissue was collected to perform gene expression analyses . Animals in the third cohort ( CC = 20 , GF = 20 , GFC = 10; Figures 5 , 6 and 7 ) underwent CRD and spinal cord tissue was collected . Animals in the fourth cohort ( CC = 17 , GF = 18; Figures 3 and 4 , and Figure 4—figure supplement 1 ) were euthanized without undergoing any procedures . See Materials and methods for more details . 10 . 7554/eLife . 25887 . 003Figure 1 . Visceral hypersensitivity in GF mice . Visceral sensitivity to colorectal distention ( CRD; ascending paradigm from 10 to 80 mm Hg ) was assessed as the number of visceromotor responses ( VMR ) over pressures . ( A ) GF mice displayed increased visceral pain responses compared to controls . ( B ) The pain threshold was lower in GF compared to CC mice . In this and subsequent figures , p≤0 . 05 , p<0 . 01 , and **p<0 . 001 versus CC animals . CC , n = 10; GF , n = 9 . DOI: http://dx . doi . org/10 . 7554/eLife . 25887 . 00310 . 7554/eLife . 25887 . 004Figure 1—figure supplement 1 . Schematic illustration showing the ascending phasic distension paradigm ( 10–80 mmHg ) ( A ) and representative CRD traces at the pressure of 40 mmHg and 65 mmHg for conventional ( B , C ) or germ-free ( D , E ) mice . The paradigm consists of three repeated pulses at each pressure level , with a pulse duration of 20 s at 5-min intervals . DOI: http://dx . doi . org/10 . 7554/eLife . 25887 . 00410 . 7554/eLife . 25887 . 005Figure 2 . Increased Toll-like receptor and cytokine gene expression in the spinal cord of GF mice . ( A–H ) Gene expression levels of the Toll-like receptors TLR1 ( A ) , TLR2 ( B ) , TLR3 ( C ) , TLR4 ( D ) , TLR5 ( E ) , TLR7 ( F ) , TLR9 ( G ) , and TLR12 ( H ) were significantly elevated in the spinal cord of GF versus CC mice . ( I–M ) When compared to controls , GF mice showed increased gene expression levels of the cytokines IL6 ( I ) , TNFα ( J ) , IL10 ( K ) , IL1α ( M ) , and IL1β ( N ) . There was no change in the gene expression of the cytokines IFNγ ( L ) , IL12α ( O ) , and IL12β ( P ) . CC , n = 9–10; GF , n = 9–10 . DOI: http://dx . doi . org/10 . 7554/eLife . 25887 . 00510 . 7554/eLife . 25887 . 006Figure 3 . Reduction in ACC and increase in PAG volume in GF mice . ( A , C , E ) , Representative thionin-stained section of the mPFC ( A ) , cortex ( C ) , and PAG ( E ) . The volumes of the defined ( black lines ) subregions of interest were estimated using Cavalieri’s principle . Scale bars = 0 . 5 mm . ( B ) In GF mice , the volume of the ACC was reduced . ( D ) Cortical volume did not differ between CC and GF mice . ( F ) The PAG was larger in GF versus CC mice . CC , n = 5; GF , n = 6–7 . DOI: http://dx . doi . org/10 . 7554/eLife . 25887 . 00610 . 7554/eLife . 25887 . 007Figure 4 . Basilar dendritic elongation in ACC pyramidal neurons of GF mice . ( A , B , C ) Representative images of Golgi-stained layer II/III pyramidal neurons of CC and GF mice ( A ) . Neurons were reconstructed in 3D using morphometric software ( B ) . Sholl analysis was performed on 2D renderings of the neurons ( incremental radii from the soma are indicated by the color gradient; C ) . Scale bars = 50 μm . ( D ) There was no group difference in the topographical location ( i . e . cell layer ) of the neurons in the ACC . ( E ) When compared to controls , the dendrites of ACC pyramidal neurons of GF mice were longer . This elongation was principally localized to the basilar dendritic arbor . ( F ) There was no change in the number of branch points of ACC pyramidal neurons between groups . ( G ) Sholl analysis of the total dendritic arbor revealed no difference in dendritic complexity between CC and GF mice . ( H ) In GF mice , the apical dendritic arbor showed altered dendritic complexity; however , post hoc comparisons revealed no statistically significant distances in which this change occurred . ( I ) There was no group difference in the complexity of basilar dendrites . For both CC and GF mice , n = 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 25887 . 00710 . 7554/eLife . 25887 . 008Figure 4—figure supplement 1 . Increase of thin and stubby spines on ACC pyramidal neurons of GF mice . ( A ) Image of a dendritic segment with examples of thin , stubby , and mushroom spines . ( B , C ) Representative images of dendritic spine density from ACC pyramidal neurons of CC ( B ) and GF mice ( C ) . Scale bars = 10 μm . ( D , E ) There was no group difference in spine width ( D ) or length ( E ) . ( F ) In GF mice , spine density was increased over the entire dendritic arbor . ( G ) When compared to controls , thin spine density was higher in GF mice , with a trend toward an increase in thin spines on basilar dendrites . ( H ) In GF mice , stubby spine density was increased overall and on basilar dendrites . ( I ) There was no significant difference in mushroom spine density in CC compared to GF mice . For both CC and GF mice , n = 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 25887 . 008 This study adds proof-of-principle evidence to the growing body of support for an association between the gut microbiota and visceral nociception ( Eisenstein , 2016 ) . We have previously shown that exposure to stressors or antibiotic treatment in early life can have enduring effects on visceral pain in rodents , and these effects are accompanied by a change in microbial composition ( O'Mahony et al . , 2014 , 2009 ) . Moreover , in rodents , probiotics can ameliorate visceral pain induced by stress ( Gareau et al . , 2007; McKernan et al . , 2010; Ait-Belgnaoui et al . , 2006 ) or antibiotic administration ( Verdú et al . , 2006 ) , and even exert an inhibitory influence on visceral sensitivity in healthy rats ( Kamiya et al . , 2006 ) . Clinical studies have repeatedly found that the composition of the gut microbiota is altered in patients with IBS ( Collins , 2014; Jeffery et al . , 2012 ) and that patients suffering from abdominal pain can benefit from probiotic treatment ( Clarke et al . , 2012 ) . Although these studies report an association between altered microbial composition and visceral pain processing , the impact of the microbiota cannot be separated from other factors , such as stress or antibiotic side effects . In this set of experiments , we use GF as a tool to show that visceral sensitivity is heightened in mice in the absence of any microbiota . Although it is now clear that the gut microbiota plays a role in visceral nociception , the exact mechanism by which the microbiota exerts its influence on the CNS remains unknown . It is likely that many systems simultaneously contribute to the altered nociception in GF mice , including immune , neural , endocrine , and metabolic processes ( Sampson and Mazmanian , 2015; Dinan and Cryan , 2012; Mayer , 2011; Luczynski et al . , 2016; Cryan and Dinan , 2012 ) . Indeed , GF mice have blunted immune function and altered TLR expression in the gut ( Shanahan , 2002; Clarke et al . , 2013 ) . Interestingly , the development in early life and function of the enteric nervous system ( ENS ) in adulthood is altered in GF mice ( Collins et al . , 2014; McVey Neufeld et al . , 2013 ) . Enteric sensory neurons synapse with vagal nerve endings in the gut ( Perez-Burgos et al . , 2014 ) , providing a direct route whereby the intestinal bacterial status can be communicated to the brain . Growing up germ-free heightens hypothalamic-pituitary-adrenal axis responsivity , which is thought to occur via a neuronal mechanism ( Sudo et al . , 2004 ) . GF rodents also have a reduced production of short chain fatty acids ( den Besten et al . , 2013 ) , which are metabolites produced by gut bacteria thought to be key mediators of gut-brain signaling . Moving forward , it would be interesting to determine if is it the cell surface proteins on intact bacteria or their metabolites and other products that are involved in changes along the brain-gut axis . Within the gastrointestinal tract , nociceptors respond to many stimuli including stretch , pH , bacterial products , immune signaling molecules , and neurotransmitters released from the ENS or from the bacteria themselves ( Sengupta , 2009 ) . From the gut , nociceptive signals are transmitted to the spinal cord , and subsequently the brain . Recent evidence has implicated neuroimmune and spinal microglial mechanisms in chronic visceral pain ( Ji et al . , 2013; Grace et al . , 2014; Bradesi , 2010 ) . TLR receptor signaling is also involved in visceral nociception and IBS pathology ( Tramullas et al . , 2014 , Tramullas et al . , 2016 ) . Moreover , microglial structure and homeostasis are disrupted in both GF and antibiotic-treated mice ( Erny et al . , 2015 ) , indicating that the microbiota is required for normal microglial function in the CNS . Microglia can be activated through TLR signaling , and once activated , increase the secretion of various cytokines ( Bradesi , 2010 ) . In the present study , we report that glial activation , TLR expression , and cytokine signaling are all increased in the lumbosacral spinal cord of GF mice , an area associated with signals from the colon . Indeed , microglial activation ( Saab et al . , 2006 ) , reduced glutamate-reuptake by astrocytes ( Gosselin et al . , 2010 ) , and increased TLR signaling ( Tramullas et al . , 2014 ) in the spinal cord have been previously shown in rodent models with visceral hypersensitivity . TRPV1 receptors are broadly expressed in the gastrointestinal tract and many areas of the CNS , including the spinal cord and is well-recognized as a transducer of noxious stimuli ( Nagy et al . , 2004 ) . We found no changes in the mRNA expression of the TRPV1 receptor in the GF spinal cord . This result is in line with previous data from our group showing that the colon of GF mice exhibits the same responsiveness to capsaicin ( a TRPV1 agonist ) relative to controls ( Lomasney et al . , 2014 ) , suggesting no change in the expression of the TRPV1 receptor . In contrast , rats exposed to vancomycin in early in life resulting in an altered gut microbiome , show enhanced visceral pain perception and a decrease in spinal cord TRPV1 expression ( O'Mahony et al . , 2014 ) . In the gut , TRPV1 receptors play a key role in pain perception ( Holzer , 2011 ) ; however , TRPV1 signaling does not appear to be responsible for the visceral hypersensitivity observed in GF mice . Further studies are required to confirm changes in the expression of TRPV1 receptor in other areas of the CNS involved in visceral pain perception . With regard to what influences the changes in cytokines and TLRs in the spinal cord of GF mice we can speculate that lack of certain metabolites and or short chain fatty acids which reduce visceral pain or exaggerated sensation in a conventional mouse are not present to gate the pain signals which can lead to altered signaling in the spinal cord . Moreover , GF are known to have an altered immune system and hence changes in spinal cord immune players fits well with the literature . The affective or emotional component of pain is mediated by the ACC ( Apkarian et al . , 2005 ) . Studies in humans with IBS ( Tillisch et al . , 2011; Mertz et al . , 2000 ) and in animal models ( Gibney et al . , 2010; Felice et al . , 2014; Bliss et al . , 2016 ) have revealed increased activation in the mPFC in response to visceral pain . In addition , imaging studies have consistently observed reduced cortical grey matter in patients with IBS ( Davis et al . , 2008 ) . Although cortical structure has yet to be investigated in animal models of visceral pain , rodents with long-lasting neuropathic pain show a reduction of ACC volume ( Seminowicz et al . , 2009 ) and basilar dendritic hypertrophy of pyramidal neurons ( Metz et al . , 2009 ) . We report a remarkably similar reduction in ACC volume in GF mice as well as elongation of single ACC pyramidal neurons . Importantly , GF mice showed no difference in mushroom spines , which represent mature , long-lasting postsynaptic connections ( Matsuzaki et al . , 2004 ) . Instead , the density of ‘immature’ thin and stubby spines was higher in these animals . These results can be interpreted as a microbiota-induced deficit in synaptic pruning resulting in the hyperactivity of ACC neurons . Such changes in ACC signaling would likely impact visceral sensitivity , as this area receives and sends projections to many pain-relevant brain areas including the limbic system , ventral tegmental area and the PAG ( Hoover and Vertes , 2007 ) . The neurons in the PAG , once activated are involved in an endogenous pain inhibitory system ( Apkarian et al . , 2011 ) . It is currently unclear what the relative contribution of the altered PAG volume is to the visceral hypersensitivity behavior observed in GF mice . The PAG receives signals from cortical areas such as the ACC where this connection is thought to mediate expectation and placebo analgesia ( Petrovic et al . , 2002 ) . Hence , input changes from the ACC may have a reciprocal effect on the PAG . What is evident , however , is that both cortical and subcortical pain pathways are altered in GF mice . Taken together , our results suggest that gross morphological and ultrastructural changes in the ACC and PAG could underlie the visceral hypersensitivity observed in GF mice . Our results show that the gut microbiota dramatically impacts visceral sensitivity and affects signaling in key structures involved in the processing and integration of painful stimuli . These results further reinforce the idea that the microbial composition of mice in conventional facilities may affect the responsiveness to CRD ( Verdú et al . , 2006 ) . Importantly , this maladaptive pain responsivity is amenable to microbial-based interventions in later life . This is noteworthy as numerous postnatal colonization experiments have been performed in GF mice , with certain behavioral and physiological changes proving reversible ( Clarke et al . , 2013; Sudo et al . , 2004; Diaz Heijtz et al . , 2011 ) and others irreversible ( Desbonnet et al . , 2014; Stilling et al . , 2015 ) . Hence the addition of colonization experiments with regard to morphological measures would be of interest in future studies . Moreover , we find across a number of different domains that the effects of the microbiome on brain and behavioural are much more robust in males than females ( Clarke et al . , 2013; Desbonnet et al . , 2015; Hoban et al . , 2016 ) . Therefore , future analysis of visceral sensitivity in female animals is indeed worthy of investigation . Improving our understanding of the impact of microbiota on visceral pain sensitivity ( O'Mahony et al . , 2017 ) may ultimately inform novel therapies for the treatment of gastrointestinal pain disorders . Swiss Webster breeding pairs for both GF and CC mice were supplied by Taconic ( Germantown , New York , USA ) and first-generation male offspring were studied in all experiments . In the University College Cork GF Unit , GF mice were group housed in flexible film gnotobiotic isolators maintained at 21 ± 1°C , with 55–60% relative humidity , under a 12 hr light/dark cycle . A group of GF mice were colonized ( GFC ) on postnatal day 21 with microbiota obtained from CC mice . Briefly , GF mice were removed from the GF facility and allowed to grow to adulthood in the conventional animal facility in cages with bedding and fecal matter from CC mice , a method which has previously been shown to effectively restore a normal microbiota ( Clarke et al . , 2013; Desbonnet et al . , 2014; O'Connell Motherway et al . , 2011 ) . CC and GFC mice were group housed in the standard animal facility which was held at the same controlled conditions and light/dark cycle as the GF mice . GF , CC , and GFC mice were age-matc and fed the same autoclaved pelleted diet ( Special Diets Services , product code 801010 ) and were housed 2–5 per cage ( only one group of CC mice was housed two per cage ) . Animals were tested or euthanized at 5–10 weeks of age . All experiments were performed in accordance with the guidelines of European Directive 86/609/EEC and the Recommendation 2007/526/65/EC and were approved by the Animal Experimentation Ethics Committee of University College Cork . Four cohorts of animals were used to perform the experiments in this paper . Animals in the first cohort ( CC = 10 , GF = 9; Figure 1 ) underwent CRD and then were euthanized . No samples were collected from this cohort . Animals in the second cohort ( CC and GF = 9–10; Figure 2 ) were sacrificed without anesthesia and spinal cord tissue was collected to perform gene expression analyses . These mice were not exposed to CRD . After data analysis , we decided that it would be of interest to determine if bacterial colonization could rescue the visceral hypersensitivity and associated changes noted in GF mice from the first and second cohorts . With the third cohort of mice ( CC = 20 , GF = 20 , GFC = 10; Figures 5 , 6 and 7 ) ) , we repeated the CRD and collected the spinal cord tissue . Due to the limited number of GFC mice available , most GFC animals used in the spinal cord analysis had been exposed to CRD . However , we verified that there was no statistical difference between those animals that underwent CRD and those that did not in the CC and GF groups . Animals in the fourth cohort ( CC = 17 , GF = 18; Figures 3 and 4 , and 4 supplement 1 ) were euthanized without undergoing any procedures . Brain tissue from 10 animals from each group was processed for Golgi-Cox staining and the remainder of the brains were processed as per the stereology protocol . CRD was carried out as previously described ( O'Mahony et al . , 2012; Tramullas et al . , 2012 ) . The CRD-system was composed of a barostat ( Distender Series II , G and J Electronics , Toronto , ON , Canada ) and a transducer amplifier ( LabTrax 4 , World Precision Instruments , Sarasota , FL ) . A custom-made balloon ( 2 cm length x 1 cm inflated diameter ) prepared from a polyurethane plastic bag ( GMC Medical , Denmark ) was tied over a PE60 catheter with silk 4 . 0 . Before securing the balloon to the catheter , several holes were punched in the distal 20 mm of the tubing with a 27-gauge needle to allow the balloon to inflate . On the experimental day , mice were lightly anesthetized with isoflurane ( 2% vapor in oxygen; IsoFlo , Abbott , UK ) and a lubricated balloon with a connecting catheter was inserted into the colon , 0 . 5 cm proximal to the anus . The catheter was fixed to the base of the tail with tape to avoid any displacement . Unrestrained mice were allowed to recover for 10 min before starting the CRD procedure . The balloon was connected to the barostat system and subsequent pressure changes within the distending balloon , observed in response to a distension paradigm , were monitored and recorded using Data Trax two software ( World Precision Instruments , Sarasota , FL ) . The ascending phasic distension ( from 10 to 80 mmHg ) paradigm , consisting of three 20 s pulses at each pressure and 5-min inter-pulse intervals , was used . The visceromotor responses ( VMR ) were quantified as pressure changes observed within the colonic distending balloon during the colorectal distension procedure . VMR were calculated as the average of the three consecutive pulses for each pressure . Additionally , for each animal , pain threshold was defined as the pressure which exceeded the mean baseline activity plus three times the standard deviation ( see Figure 1-supplement 1A 4 for phasic stages ) . At the end of the experiments , the balloon was carefully removed and the animals were returned to their home cages . The volumes of the mPFC , total cortex , and periaqueductal grey ( PAG ) were estimated using Cavalieri’s principle ( Gundersen et al . , 1988 ) . Brain structures were characterized using the Paxinos and Franklin , 2001 atlas as a guide . The sample size was determined by a power calculation and aimed at detecting differences between groups at the 0 . 05 level ( Lomasney et al . , 2014; Matsuzaki et al . , 2004 ) . Data are expressed as means + or ±1 SEM . The unpaired Student’s t-test ( α = 0 . 05 ) was used to compare two independent groups ( CC vs GF; right vs left hemisphere ) . Comparisons of more than two groups were performed by one-way ANOVA . Group differences in the CRD and Sholl analyses were tested for significance with a two-way mixed design ANOVA , with pressure and radial distance from the soma as within group factors . Post-hoc comparisons were made using a Bonferroni’s correction ( p<0 . 05 ) . The Grubbs method ( Grubbs , 1950 ) was used to test for outliers for the CRD and spinal cord analyses .	The human gut is home to over 100 trillion microbes collectively known as the gut microbiota . These microbes help us to digest food and absorb the nutrients effectively . A diverse and stable community of gut microbes is believed to be important for good health . Recently , it has also become clear that the microbiota can also influence the brain and how we behave . For example , many studies suggest that gut microbiota can alter how an individual perceives pain , but it is not clear how this works . Rodents are often used in experiments as models of human biology . One of the most frequently used rodent models in studies of gut microbes is the “germ-free” mouse . These mice grow up in laboratory environments that are completely free of microbes , making it possible to study how having no gut microbes affects the health and behaviour of the mice . Luczynski , Tramullas et al . used germ-free mice to study how the gut microbiota influences an animal’s sensitivity to pain . The experiments show that , compared to mice with normal gut microbiota , the germ-free mice were more sensitive to pain from internal organs especially the gut . These mice also produced larger amounts of specific proteins involved in immune responses , which contributed to the animal’s increased sensitivity to pain . Allowing the germ-free mice to be colonised with gut microbes could reverse these changes . The experiments also show that the germ-free mice had changes in the size of two areas of the brain involved in sensing pain: an area called the anterior cingulate cortex was smaller , while the periaqueductal grey region was enlarged . There were also differences in individual nerve cells within the anterior cingulate cortex compared to normal mice . The findings of Luczynski , Tramullas et al . reinforce the idea that the gut microbiota is involved in the sensation of pain from internal organs , and show that hypersensitivity to this form of pain can be reversed later in life by colonising the gut with microbes . Continuing to study the impact of microbes on this type of pain could aid the development of new therapies for the treatment of pain disorders in humans .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience", "microbiology", "and", "infectious", "disease" ]	2017	Microbiota regulates visceral pain in the mouse
The dynein-2 microtubule motor is the retrograde motor for intraflagellar transport . Mutations in dynein-2 components cause skeletal ciliopathies , notably Jeune syndrome . Dynein-2 contains a heterodimer of two non-identical intermediate chains , WDR34 and WDR60 . Here , we use knockout cell lines to demonstrate that each intermediate chain has a distinct role in cilium function . Using quantitative proteomics , we show that WDR34 KO cells can assemble a dynein-2 motor complex that binds IFT proteins yet fails to extend an axoneme , indicating complex function is stalled . In contrast , WDR60 KO cells do extend axonemes but show reduced assembly of dynein-2 and binding to IFT proteins . Both proteins are required to maintain a functional transition zone and for efficient bidirectional intraflagellar transport . Our results indicate that the subunit asymmetry within the dynein-2 complex is matched with a functional asymmetry between the dynein-2 intermediate chains . Furthermore , this work reveals that loss of function of dynein-2 leads to defects in transition zone architecture , as well as intraflagellar transport . Cytoplasmic dyneins are minus-end directed motors that use the energy of ATP hydrolysis to move along microtubules . Two cytoplasmic dyneins have been identified . The better-characterized dynein-1 is involved in the transport of cargos in the cytoplasm , organelle dynamics and in mitotic spindle organization during mitosis ( Roberts et al . , 2013 ) . Dynein-2 is responsible for retrograde transport in cilia and flagella . Primary ( non-motile ) cilia are hair-like extensions present on almost all animal cells that act as antennae for extracellular signals and are fundamental to proper metazoan development and ongoing health . They integrate signals in key pathways including sonic hedgehog ( Shh ) , Wnt and platelet-derived growth factor signaling and participate in metabolic control and autophagy ( Reiter and Leroux , 2017 ) . Cilia are particularly important to ensure correct Shh signaling during embryonic development ( Goetz and Anderson , 2010; He et al . , 2017 ) . Defects in cilia are linked to many human diseases , known collectively as ciliopathies , including developmental disorders , neurodegeneration and metabolic diseases ( Reiter and Leroux , 2017; Yee and Reiter , 2015 ) . Ciliogenesis is initiated in non-dividing cells by the docking of pre-ciliary vesicles with the mother centriole . The pre-ciliary vesicles fuse and then surround the mother centriole concomitant with the assembly of a series of protein modules that form a diffusion barrier separating the distal end of the mother centriole from the rest of the cell cytoplasm ( Garcia-Gonzalo et al . , 2011 ) . A microtubule bundle , the axoneme , then extends from the centriole to allow cargo transport by the process of intraflagellar transport ( IFT ) ( Yee and Reiter , 2015 ) . A ‘transition zone’ ( TZ ) that separates the mother centriole from the main length of the axoneme forms a diffusion barrier for both soluble and membrane proteins at the base of the cilium ( Garcia-Gonzalo et al . , 2011; Garcia-Gonzalo and Reiter , 2017 ) . Once established , cilia are maintained and operated by the process of IFT ( Hou and Witman , 2015 ) . IFT-B complexes ( comprising of a core subcomplex of nine subunits ( IFT88 , −81 , –74 , −70 , –52 , −46 , –27 , −25 , and −22 ) with five additional , peripherally-associated subunits ( IFT172 , −80 , –57 , −54 , and −20 ) ) undergo kinesin-2-driven motility from base to tip where the complexes are then reorganized prior to retrograde transport of IFT-A ( comprising six subunits ( IFT144 , -140 , -139 , -122 , -121/WDR35 , and -43 ) ) driven by dynein-2 ( Hou and Witman , 2015; Jensen and Leroux , 2017 ) . This process ensures the correct localization of receptors and signaling molecules within cilia and directs transduction of signals from the cilium to the rest of the cell . The genes encoding subunits of the dynein-1 and dynein-2 motors are largely distinct . Some light chain subunits are common to both motors but the major subunits ( heavy , intermediate and light intermediate chains ) are different between the two holoenzyme complexes . Dynein-2 is built around a heavy chain dimer of DHC2/DYNC2H1 ( Criswell et al . , 1996; Mikami et al . , 2002 ) . This associates with two intermediate chains ( ICs ) , WDR34 and WDR60 , first identified as dynein-2 subunits named FAP133 and FAP163 in Chlamydomonas ( Patel-King et al . , 2013; Rompolas et al . , 2007 ) and subsequently shown to be components of metazoan dynein-2 ( Asante et al . , 2013; Asante et al . , 2014 ) . This asymmetry distinguishes dynein-2 from dynein-1 where two identical IC subunits form the holoenzyme . The reason for this asymmetry is unclear . In addition , a dynein-2-specific light intermediate chain ( LIC3/DYNC2LI1 ) has been identified ( Hou and Witman , 2015; Mikami et al . , 2002 ) as well as a specific light chain , TCTEX1D2 ( Asante et al . , 2014; Schmidts et al . , 2015 ) . Mutations in genes encoding dynein-2 subunits are associated with skeletal ciliopathies , notably short rib-polydactyly syndromes ( SRPSs ) and Jeune asphyxiating thoracic dystrophy ( JATD , Jeune syndrome ) . These are recessively inherited developmental disorders characterized by short ribs , shortened tubular bones , polydactyly and multisystem organ defects ( Huber and Cormier-Daire , 2012 ) . In recent years , whole exome-sequencing technology has enabled the identification of new mutations involved in skeletal ciliopathies , notably a range of mutations affecting DYNC2H1 ( DHC2 , [Chen et al . , 2016; Cossu et al . , 2016; Dagoneau et al . , 2009; El Hokayem et al . , 2012; Mei et al . , 2015; Merrill et al . , 2009; Okamoto et al . , 2015; Schmidts et al . , 2013a] ) . Additionally , mutations in WDR34 ( Huber et al . , 2013; Schmidts et al . , 2013b ) , WDR60 ( Cossu et al . , 2016; McInerney-Leo et al . , 2013 ) , LIC3/DYNC2LI1 ( Kessler et al . , 2015; Taylor et al . , 2015 ) and TCTEX1D2 ( Schmidts et al . , 2015 ) have also been reported . The role of the dynein-2 heavy chain has been extensively studied in Chlamydomonas , C . elegans , and mice . In all cases , loss of dynein heavy chain results in short , stumpy cilia that accumulate IFT particles at the tip , consistent with the role of dynein-2 in retrograde ciliary transport ( Hou and Witman , 2015 ) . Recently , more interest has been focused on the role of the subunits associated with DHC2/DYNC2H1 . Two studies in Chlamydomonas and in human patient-derived fibroblasts revealed that LIC3/DYNC2LI1 ( D1bLIC in Chlamydomonas ) plays an important role in ciliogenesis and stabilization of the entire dynein-2 complex ( Li et al . , 2015; Taylor et al . , 2015 ) . Similarly , loss of Tctex2b ( TCTEX1D2 ) destabilizes dynein-2 and reduces IFT in Chlamydomonas ( Schmidts et al . , 2015 ) . Previous work from our lab and others has shown that loss of function of dynein-2 intermediate chains , WDR34 and WDR60 , is associated with defects in ciliogenesis . Knockdown of WDR60 or WDR34 in hTERT-RPE1 cells results in a reduction of ciliated cells , with an increase in length of the remaining cilia , likely depending on depletion efficiency ( Asante et al . , 2014 ) . Mutations in WDR34 have also been shown to result in short cilia with a bulbous ciliary tip in patient fibroblast cells affected by SRP ( Huber et al . , 2013 ) . Consistent with the results obtained in patient cells , loss of WDR34 in mice also results in short and stumpy cilia with an abnormal accumulation of ciliary proteins and defects in Shh signaling ( Wu et al . , 2017 ) . Similarly , mutations in WDR60 patient fibroblasts are associated with a reduction in cilia number , although the percentage of ciliated cells was variable in different affected individuals ( McInerney-Leo et al . , 2013 ) . These findings are all consistent with roles for WDR34 and WDR60 in IFT . Moreover , further recent data found that WDR60 plays a major role in retrograde ciliary protein trafficking ( Hamada et al . , 2018 ) . In this study , we sought to better understand the role of dynein-2 in human cells using engineered knockout ( KO ) cell lines for WDR34 and WDR60 . We define a functional asymmetry within the complex , where WDR34 is absolutely required for cilia extension , while WDR60 is not . Loss of either IC results in defects in ciliary transition zone assembly and/or maintenance . We also find that loss of WDR60 leads to reduced assembly of the dynein-2 holocomplex and reduced interactions with IFT particles . Surprisingly , WDR34 is not required for the other dynein-2 subunits to assemble , instead loss of WDR34 results in delocalization of dynein-2 from the ciliary base and an accumulation of IFT proteins at this site . These data are consistent with a model in which WDR34 configures the dynein-2 complex for dynamic assembly and disassembly with IFT proteins to facilitate axoneme extension . To understand the function of WDR34 and WDR60 , we generated KO human telomerase-immortalized RPE1 ( hTERT-RPE1 ) cells using CRISPR-Cas9 . We derived two WDR34 KO clones ( 1 and 2 ) using guide RNAs ( gRNAs ) targeting exons 2 and 3 , and one KO clone for WDR60 , targeting exon 3 . Genomic sequencing of these clones identified insertion/deletion mutations on the targeted sequences ( Figure 1—figure supplement 1 ) . All cell clones were analyzed for protein expression by immunoblot using polyclonal antibodies against multiple epitopes to exclude the possibility of downstream initiation sites being used . Neither WDR34 nor WDR60 was detected in the respective KO cells compared to the controls ( Figure 1—figure supplement 2 ) . To mitigate against the possibility of any off-target effects , we grew KO cells alongside control CRISPR cells which had been transfected with Cas9 and gRNA in the same way as the KO lines but showed no mutation at the target site . These cells ( WDR34 KO CTRL and WDR60 KO CTRL ) did not present any cilia defects when stained with Arl13b or IFT88 ( Figure 1—figure supplement 2 ) and no differences were seen between these cells and wild type ( WT ) RPE-1 in any assay . Images in all figures show WT cells where indicated but indistinguishable results were also obtained using the control cell lines . Defects in ciliogenesis in both WDR34 and WDR60 KO cells were rescued by overexpressing WT proteins , confirming that the phenotypes we observed were not due to off-target mutations ( described below ) . Loss of WDR34 severely impaired the ability of the KO cells to extend a microtubule axoneme ( Figure 1A , B ) , although Arl13b localized within those few cilia that did form . In contrast , loss of WDR60 did not significantly affect the ability of cells to extend an axoneme ( Figure 1B ) but did result in a change in localization of Arl13b to a more pronounced enrichment at the base and tip of many cilia compared to the more uniform ciliary distribution in control cells . Cilia were shorter in both WDR60 and WDR34 KO cells ( Figure 1C ) . Next , we examined the assembly and structure of primary cilia in WDR34 and WDR60 KO cells by transmission electron microscopy ( EM ) . After 24 hr of serum starvation , WT RPE1 cells extend a defined axoneme surrounded by a ciliary membrane ( Figure 1D ) . In contrast , WDR34 KO cells failed to extend an axoneme ( Figure 1E ) but showed a large docked pre-ciliary vesicle , consistent with the small Arl13b-positive structures seen by immunofluorescence . WDR60 KO cells showed apparently normal cilia ( Figure 1F ) with normal basal body structures and axoneme extension . However , when an entire cilium was captured in WDR60 KO serial sections ( Figure 1G ) , we observed a bulged cilium tip containing accumulated electron dense particles ( Figure 1H ) . To our surprise , we also observed the ciliary membrane bulged at a second point along the axoneme and this region contained intraciliary vesicular structures ( Figure 1Hi ) . The abnormal structure of cilia in the KO cells led us to analyze the steady-state localization of the IFT machinery . After 24 hr serum starvation , IFT88 ( part of IFT-B ) was found almost exclusively at the base of the cilia in WT RPE1 cells but in WDR60 KO and WDR34 KO cells IFT88 was found throughout the cilia and accumulated at the tips ( Figure 2A and B , quantified in Figure 1Ai ) . A similar phenotype was observed for IFT-B components , IFT54 ( Figure 2C ) and IFT57 ( Figure 2D , quantified in Figure 2Di ) . In those WDR34 KO cells that did extend cilia , IFT-B proteins were seen to accumulate at the tip of cilia . The limited number of ciliated WDR34 KO cells precluded further quantification . Another IFT-B protein , IFT20 ( Figure 2E ) , was enriched at both the tip and the base of cilia in WDR60 KO cells ( quantification in Figure 2F ) . IFT20 is the only IFT component found to localize to the Golgi until now . As expected , in WT cells IFT20-GFP was found at the ciliary base with a ribbon-like distribution , consistent with its known localization to the Golgi . This pool of IFT20-GFP was largely absent from WDR60 KO cells ( Figure 2G ) . We have tried to measure IFT in these cells using a variety of markers including Arl13b- and IFT88-fusions but have not been able to derive reliable quantitative data . Next , we analyzed the localization of IFT-A proteins . We found that both IFT140 ( Figure 3A ) and IFT43 ( Figure 3B ) were accumulated along the length of the axoneme within cilia in WDR60 KO cells , as well as at the tips in the few cilia present in WDR34 KO cells , while they were found only at the base of the cilia in WT cells ( quantification in Figure 3Ai–iii and 3Bi-iii ) . Further quantification showed that both IFT140 ( Figure 3C ) and IFT43 ( Figure 3D ) are enriched in WDR60 KO cilia compared to controls . In addition , we determined the localization of a subunit of the anterograde kinesin-2 motor , KAP3 which was also accumulated at the ciliary tip in WDR60 KO cells ( Figure 3E ) . To study whether defects in dynein-2 affect the transport of membrane proteins we used GFP-fusions with Arl13b , somatostatin receptor type 3 ( SSTR3 ) , 5-hydroxytryptamine receptor type 6 ( 5HT6 ) and Rab8a . We found that in live cells Arl13b-GFP and EGFP-SSTR3 localize at the ciliary tip in WDR60 KO cells , which appear enlarged and bulbous in these cells , consistent with EM data ( Figure 4A and B ) . We also noticed a consistent reduction in the amount of Arl13b-GFP within cilia in WDR60 KO cells compared to WT cells ( Figure 4C and Ci ) . The same observation was made with EGFP-SSTR3 ( Figure 4D and Di ) and EGFP-5HT6 ( Figure 4E and Ei ) . In contrast , GFP-Rab8a localization in the cilia was indistinguishable between WT and WDR60 KO cells ( Figure 4F and Fi ) . Previous studies have shown that the protein content of cilia is maintained by a diffusion barrier formed by the transition zone . Changes in transition zone composition have been associated with the mislocalization of ciliary proteins , including the membrane marker Arl13b ( Li et al . , 2015; Shi et al . , 2017 ) . To test if the reduction in Arl13b seen in our WDR60 KO was caused by a defect in the transition zone , we labeled KO and WT cilia with known transition zone markers . We found that the core transition zone marker , RPGRIP1L ( also known as MKS5 ) , is no longer restricted to an area adjacent to the mother centriole in WDR60 KO cells ( Figure 5A , quantified in 5Ai ) . Conversely , TMEM67 ( also known as MKS3 ) , which in WT cilia extends from the basal body through a more distal region , becomes much more tightly restricted to the base of the cilium in WDR60 KO cells ( Figure 5B , quantified in 5Bi ) . We also determined the transition zone organization in WDR34 KO cells . The few cilia found in the WDR34 KO cells recapitulate the same phenotype observed in the WDR60 KO cilia with an expansion of RPGRIP1L to a more distal position and a reduction of the TMEM67 domain ( Figure 5A and Figure 5B ) . In contrast to TMEM67 and RPGRIP1L , no changes were observed for the transition zone marker TCTN1 in both WDR34 KO and WDR60 KO cells with respect to the control ( Figure 5C ) . Defects in the dynein-2 motor have been previously associated with deregulation of the Shh pathway ( May et al . , 2005 ) . Smoothened ( Smo ) , a key component of Shh signaling , localizes to the cilia in response to Shh stimulation but is normally excluded from cilia in cells that have not been treated with Shh or an equivalent agonist such as Smoothened agonist ( SAG ) . Unexpectedly , we found that Smo was localized to cilia in WDR60 KO cells even in the absence of SAG stimulation ( Figure 6A and Ai ) . This localization did not increase upon agonist treatment . In contrast , Smo was indeed excluded from cilia in WT cells at steady state ( Figure 6A and Ai ) but was readily detected within cilia following SAG stimulation ( Figure 6B , Bi , and Bii ) . Many mutations in WDR34 and WDR60 have been associated with SRPs and JATD syndromes . We engineered selected patient mutations into a WDR60 construct and expressed the proteins in WDR60 KO cells to see how well the mutated proteins rescue KO phenotypes compared to expression of WT proteins . Two WDR60 mutations were selected from one SRPS patient with compound heterozygosity for WDR60 ( WDR60[T749M] , WDR60[Q631] ) ( McInerney-Leo et al . , 2013 ) . The first mutation is located in the WD repeat region ( WDR60[T749M] ) and the second ( WDR60[Q631] ) is located just before the WD repeat domain ( Figure 7A ) . Protein expression of stably transfected WT and HA-WDR60 mutants in WDR60 KO cells is shown in Figure 7—figure supplement 1 . Both WT HA-WDR60 and HA-WDR60[T749M] efficiently rescued defects in the localization of IFT88 ( Figure 7B ) . However , expression of HA-WDR60[Q631] was unable to restore the basal body localization of IFT88 ( Figure 7B and Bi ) . Similar data were obtained for the localization of IFT140 ( Figure 7C and Ci ) ; expression of both WT and the HA-WDR60[T749M] mutant of WDR60 restored the localization of IFT140 to the base of the cilia , but IFT140 persisted throughout the cilium following expression of HA-WDR60[Q631] as in WDR60 KO cells . Next , we tried to mimic the compound heterozygosity described in patient cells by generating a stable cell line expressing both WDR60 [T749M] and [Q631] mutants ( Figure 7—figure supplement 1B ) . When the two mutants were co-expressed in the same WDR60 KO cells we saw no additive or dominant negative effects , but cilia appeared normal with IFT88 only localized to the base ( Figure 7—figure supplement 1 ) , as was seen with the HA-WDR60[T749M] mutant rescue . In parallel , we found that expression of WT mGFP-WDR34 restored ciliogenesis and axoneme extension in WDR34 KO cells ( Figure 7D , quantified in Figure 7Di ) . Moreover , WT WDR34 was able to rescue IFT88 localization to the basal body ( Figure 7E ) . To better understand the function of WDR60 we analyzed how the dynein-2 complex is assembled in the presence of WT and mutant WDR60 proteins by performing immunoprecipitation . Immunoblotting using an anti-HA antibody showed that WT HA-WDR60 and HA-WDR60[T749M] ( 125 kDa ) were expressed at similar levels to the truncated HA-WDR60[Q631] mutant ( 75 kDa ) ( Figure 7F ) . We found that immunoprecipitation of HA-WDR60 expressed in WDR60 KO cells effectively pulls down the chaperone NudCD3 , known to interact with dynein-2 via its WD repeat domains ( Asante et al . , 2014 ) . As expected , NudCD3 did not bind to HA-WDR60[Q631] lacking the WD repeat domain ( Figure 7F ) . WT WDR60 also bound effectively to WDR34 and , notably , this interaction was very similar with the HA-WDR60[T749M] or [Q631] mutants ( Figure 7G ) . In contrast , LIC3/DYNC2LI1 was readily detected with WT WDR60 but less so with WDR60[T749M] and only a very small amount of LIC3/DYNC2LI1 was detected bound to WDR60[Q631] ( Figure 7G ) . TCTEX1/DYNLT1 was found to bind effectively to both WT and HA-WDR60[T749M] but less well to HA-WDR60[Q631] ( Figure 7G ) . Next , we tested the interactions with IFT proteins , the primary cargo of the ciliary motors . We found that WT WDR60 can bind to IFT140 , IFT88 , and IFT57; WDR60[T749M] binds to all 3 IFT subunits tested but binds less well to IFT140 ( Figure 7H ) . In contrast , WDR60[Q631] pulled down reduced levels of IFT88 and IFT57 and did not interact with IFT140 . It has been reported that loss of some components of dynein-2 modifies the stability of the whole dynein-2 complex . In Chlamydomonas depletion or loss of LIC3/DYNC2LI1 ( D1bIC2 ) causes a reduction of DHC2/DYNC2H1 in whole cell lysate , whereas the expression level of the intermediate chain is less affected ( Reck et al . , 2016 ) . Similar results were obtained analyzing expression levels of DHC2/DYNC2H1 in patients cells with LIC3/DYNC2LI1 mutations ( Taylor et al . , 2015 ) . We have shown previously that siRNA depletion of WDR34 affects the stability of WDR60 and vice versa ( Asante et al . , 2014 ) . To determine whether loss of one intermediate chain had an effect on the stability of the other , we analyzed levels of WDR34 and WDR60 in whole cells lysate of serum-starved KO cells . Notably , we found that in serum-starved cells , expression levels of WDR34 were reduced in WDR60 KO cells , although not completely lost . Correspondingly , there was a reduction of WDR60 expression levels in WDR34 KO whole cell lysate ( Figure 8A ) . Next , we sought to determine the effect of WDR60 and WDR34 loss on the localization of other dynein-2 subunits . We found that LIC3/DYNC2LI1 localized in cilia of WT cells , but this localization was lost in WDR34 or WDR60 cells ( Figure 8B ) . DHC2/DYNC2H1 was detected at the base of the cilia in WT and WDR60 KO cells , but not along the ciliary axoneme . Interestingly , DHC2/DYNC2H1 localization at the ciliary base was reduced in WDR34 KO cells ( Figure 8C and 8 Ci ) . TCTEX1 was enriched at the base of the cilium in all cell lines ( DYNLT1 , Figure 8D , and 8Di ) . To test how the loss of one dynein-2 intermediate chain affected the localization of the other we overexpressed HA-WDR34 and HA-WDR60 in WDR60 and WDR34 KO cells . Both HA-WDR34 and HA-WDR60 were enriched at the base and in the ciliary axoneme in WT cells ( Figure 8—figure supplement 1A ) . We observed no changes in the localization of HA-WDR34 in WDR60 KO cells ( Figure 8—figure supplement 1B ) however , surprisingly , HA-WDR60 was greatly enriched in the stumpy cilia of WDR34 KO cells compared to the cilia in WT cells ( Figure 8—figure supplement 1A1 ) . Overexpression of HA-WDR60 could not rescue axoneme elongation in WDR34 KO cells ( Figure 8—figure supplement 1C ) . Additionally , overexpression of HA-WDR34 could not rescue abnormal IFT88 localization in WDR60 KO cells ( Figure 8—figure supplement 1D ) . The results described above suggest a defect in the assembly of the dynein-2 holoenzyme and therefore we used a proteomic approach to define the assembly of dynein-2 . We stably expressed HA-WDR34 in WT and WDR60 KO cells and HA-WDR60 in WT and WDR34 KO cells and performed immunoprecipitations using HA-GFP as a control . Confirming that both WDR34 and WDR60 exist in the same complex , immunoprecipitation analysis showed that in WT cells HA-WDR60 pulls down WDR34 , while HA-WDR34 pulls down WDR60 ( Figure 9—figure supplement 1 ) . Multiplex tandem-mass-tag ( TMT ) labeling enabled us to define the interactome of WDR34 in the presence and absence of WDR60 and of WDR60 in the presence and absence of WDR34 ( Figure 9A and Supplemental data file S6 ) . No obvious candidate proteins emerged from this analysis that could explain the axoneme extension defect in WDR34 KO cells . Therefore we focussed on changes in interactions with known dynein-2 and IFT proteins . We found that the interactions of HA-WDR34 with dynein-2 components are reduced but not lost in WDR60 KO cells compared to WT cells ( up to three log2-fold change , Figure 9B ) . Notably , loss of WDR60 caused a reduction in the amount of DHC2/DYNC2H1 and TCTEX1/DYNLT1 associated with HA-WDR34 ( Figure 9B and C ) . Moreover , interactions between HA-WDR34 and LIC3/DYNC2LI1 , TCTEX3/DYNLT3 , and the molecular chaperone NudCD3 were also reduced in WDR60 KO cells . Furthermore , we found a reduction in the amount of several IFT-B components ( IFT57 , IFT74 , and IFT88 ) associated with HA-WDR34 . Two IFT-A components ( WDR19/IFT144 and WDR35/IFT121 ) and two BBSome components ( BBS7 and BBS9 ) were identified to interact with HA-WDR34 and their binding to HA-WDR34 was reduced in the absence of WDR60 . Among these components , the binding of WDR19/IFT144 with HA-WDR34 was the most affected by the loss of WDR60 . This suggests a more loosely assembled dynein-2 motor that also shows reduced , but importantly still readily detectable , binding to IFT and BBSome proteins . In contrast , and to our surprise , in the absence of WDR34 , interactions of HA-WDR60 with other dynein-2 components , including DHC2/DYNC2H1 , were only slightly reduced ( <1 log2-fold change , Figure 9B and D ) . LIC3/DYNC2LI1 , TCTEX1/DYNLT1 , TCTEX3/TCTEX1L/DYNLT3 , LC8-type 1/DYNLL1 , LC8-type 2/DYNLL2 and NudCD3 bound with similar efficiency to HA-WDR60 in WT and WDR34 KO cells . This shows that the dynein-2 complex is largely intact in WDR34 KO cells . The only exception was DYNLRB1 which was found to bind to HA-WDR60 >1 log2-fold change less efficiently in WDR34 KO compared to WT cells . Furthermore , interaction of IFT-B and IFT-A with HA-WDR60 was only slightly reduced in the absence of WDR34 . Notably , we found that several key components of the ciliary machinery , including the BBSome component , BBS7 , and the IFT-B protein , IFT54 , bound to WDR60 more tightly in the WDR34 KO cells than in WT cells . This suggests that loss of WDR34 leads to defects in axoneme extension by stalling assembled complexes and preventing dynamic assembly/disassembly of these vital IFT machineries . Our data show that the structural asymmetry within the dynein-2 motor is matched by functional asymmetry . Perhaps most strikingly , WDR34 is essential for axoneme extension during the early steps of ciliogenesis , whereas WDR60 is not required for cilium extension ( the latter finding being validated recently by others [Hamada et al . , 2018] ) . Depletion of WDR34 using RNAi is also associated with ciliary defects ( Asante et al . , 2013 ) , whilst patient fibroblasts have shorter cilia with a bulbous tip ( Huber et al . , 2013 ) and fibroblasts from WDR34 knockout mice have stumpy cilia and defects in Shh signaling ( Wu et al . , 2017 ) . It is intriguing that in our experiments some cells missing WDR34 can still extend a rudimentary cilium . However , even here , ciliary protein localization is severely disrupted . Thus , both subunits are necessary to maintain proper ciliary protein composition . Since WDR60 cells can extend an axoneme , WDR34 and WDR60 have clearly distinct but overlapping functions in cells . Our proteomics and imaging data show that , in WDR34 KO cells , the system is stalled at the point of axoneme extension , after docking of the ciliary vesicle , despite apparently efficient co-assembly of the remaining dynein-2 motor and necessary IFT proteins . DHC2/DYNC2H1 and LIC3/DYNC2LI1 levels are , however , reduced at the ciliary base in WDR34 KO cells compared to WT cells . In contrast , several IFT proteins are shown to accumulate there . We suggest that in the absence of WDR34 , dynein-2 can associate with the relevant large multimeric complex that includes IFT proteins and the BBSome , but cannot form the functional assemblies necessary for axoneme extension . One possibility is that WDR34 is required to target dynein-2 to the base of the nascent cilium , which is consistent with our immunofluorescence data . We did not identify any obvious components of the system that are missing in the absence of WDR34 , suggesting that this is a mechanochemical defect . For example , the presence of WDR34 within the dynein-2 complex may induce a conformational change necessary for its concentration at the base of the cilium and to enable the functional assembly of IFT trains required for axoneme extension . Further structural work would be required to define this in molecular terms . Paradoxically , our data also show that in the absence of WDR60 , the dynein-2 holocomplex cannot form as effectively as it does in WT cells or in cells lacking WDR34 , yet axoneme extension occurs normally . Unlike in WDR34 KO cells , we do observe DHC2/DYNC2H1 concentration at the base of the cilium in WDR60 KO cells , suggesting in this case targeting is not affected . Inefficient assembly of dynein-2 reduces interactions with IFT and BBS proteins but is sufficient and correctly positioned to permit axoneme extension . Thus , the WDR34-dependent , WDR60-independent , localization of DHC2/DYNC2H1 at the base of the cilium is required for axoneme extension . Meanwhile , reduced binding to IFT complexes and less efficient assembly of the functional dynein-2 holoenzyme both explain the IFT defects seen in WDR60 KO cells . Our data do not eliminate the possibility that WDR34 is itself required outside of the context of the dynein-2 complex . While we cannot rule out dynein-2-independent functions of WDR34 and WDR60 , all of our data provide strong evidence that they co-exist in the dynein-2 holoenzyme . Interaction of WDR60 with WDR34 in our pull-down experiments indicates that dynein-2 complexes contain both WDR34 and WDR60 , in agreement with our own previously reported data ( Asante et al . , 2014 ) . Moreover , we found that stably expressed WDR60 could not rescue ciliogenesis defects in WDR34 KO cells , neither could WDR34 rescue IFT88 localization defects observed in WDR60 KO cells . Therefore WDR34 and WDR60 are not functionally redundant , at least in this regard . This supports a model where WDR34 and WDR60 play different roles in ciliogenesis and in IFT within the context of the dynein-2 complex , likely through different interactions with distinct components . Loss of either WDR34 or WDR60 leads to IFT particle accumulation at the base of as well as within cilia . We found that loss of WDR60 results in an increase of the IFT-B proteins , IFT20 , IFT57 , and IFT88 , not only at the tip but also close to the base of the cilium . This suggests that IFT-B proteins could be retained at the basal body or around the transition zone . Consistent with these results , mutations in IFT-A or dynein-2 in mice also result in abnormal accumulation of IFT particles near the base of the cilium ( Goggolidou et al . , 2014; Liem et al . , 2012; Ocbina et al . , 2011 ) . This has been linked to defects in the export of ciliary cargo across the transition zone ( He et al . , 2017 ) . Similar defects are seen following disruption of the heavy chain , DHC2/DYNC2H1 ( Hou and Witman , 2015 ) . In addition to these defects , we show that KAP3 , a subunit of kinesin-2 , accumulates at the ciliary tip of WDR60 KO cells . This would decrease the levels of kinesin-2 available to load onto departing anterograde trains and further cause accumulation of IFT particles at the base . This might also reflect some functional coupling of dynein-2 and kinesin-2 in the assembly of anterograde IFT particles . Notably , like WDR34 KO , KIF3B knockouts in both mice ( Nonaka et al . , 1998 ) and cultured cells ( Funabashi et al . , 2018 ) fail to form cilia . WDR34 might therefore be required to assemble functional IFT trains but we did not detect kinesin-2 in our WDR34 pull-down experiments . Our data are also consistent with models where , in metazoa , kinesin-2 motors are returned to the ciliary base by dynein-2-dependent retrograde IFT ( Broekhuis et al . , 2013; Chien et al . , 2017; Mijalkovic et al . , 2017; Prevo et al . , 2015; Signor et al . , 1999; Williams et al . , 2014 ) but in contrast with Chlamydomonas studies showing kinesin-2 diffuses back to the ciliary base ( Engel et al . , 2012 ) . Overall , our data support models where dynein-2 acts both in loading of cargo into cilia and in exit from cilia . These findings suggest a complex interplay between IFT particles and IFT motors to control entry to and exit from cilia . Notably , interfering with either dynein-2 or IFT results in perturbation of transition zone organization , IFT , and ciliary membrane protein localization ( Garcia-Gonzalo and Reiter , 2017 ) . The transition zone is the ciliary region most proximal to the mother centriole and it functions as a gate , acting as a diffusion barrier to prevent unregulated entry of high molecular weight proteins and maintaining the protein composition of the ciliary membrane ( Garcia-Gonzalo and Reiter , 2017; Jensen and Leroux , 2017 ) . Our results show that loss of either dynein-2 intermediate chain disrupts the organization of the ciliary transition zone . This could be significant in the context of Jeune syndrome where several cases result from mutations in WDR34 that are predicted to be complete loss of function mutations ( e . g . c . 472C > T ( p . Gln158∗ ) ( Schmidts et al . , 2013b ) ) . The same is true for WDR60 ( McInerney-Leo et al . , 2013 ) . However , no defects in transition zone structure have been described in cells derived from these patients . The defect in the transition zone could also explain the presence of intraciliary vesicles within the cilia of WDR60 KO cells that would normally be excluded by the diffusion barrier . The presence of intraciliary vesicles could also indicate a failure in the formation of ectosomes to remove the excess membrane from the cilia . Such vesicles have been described in motile cilia ( Shah et al . , 2008 ) and in mouse photoreceptor cells ( Gilliam et al . , 2012 ) from BBS mutants , as well as in wild-type zebrafish ( Goetz et al . , 2014 ) . However , we always see swollen ciliary tips in WDR60 KO cells when imaging living cells expressing ciliary membrane markers and have not detected any shedding of vesicles during such experiments . This suggests that any mechanisms to reduce membrane accumulation might not be able to overcome any defect in retrograde IFT . The role of dynein-2 in maintaining transition zone composition is also consistent with the fact that LIC3/DYNC2LI1 and overexpressed WDR34 and WDR60 localize at the transition zone in RPE1 cells . Moreover , in a previous study , active phosphorylated TCTEX1/DYNLT1 was detected at the transition zone of neural progenitors ( Li et al . , 2011 ) . Changes in transition zone composition are associated with a reduced localization of soluble and membrane proteins in the cilium ( Berbari et al . , 2008b; Chih et al . , 2011 ) . 5HT6 , SSTR3 , and Arl13B are all reduced in abundance within cilia of WDR60 KO cells , suggesting that these proteins might not be effectively retained within cilia , but leak out through the diffusion barrier . An alternative possibility is that these proteins are less effectively loaded into cilia . We did not find a difference in the intensity levels of overexpressed Rab8a in WDR60 KO compared to WT cells suggesting that at least some proteins can enter normally , likely reflecting a differential requirement for dynein-2 function . Our data do not discriminate between direct or indirect roles for dynein-2 in either building or maintaining the ciliary transition zone . Overall , our data suggest that the reduced interaction of dynein-2 with IFT and BBS proteins likely underpins these defects in the transition zone structure in WDR60 KO cells . Indeed , it is quite likely that the role of dynein-2 in transition zone structure is a result of its established role in IFT . Recent data from the Blacque lab using IFT-A mutants ( Scheidel and Blacque , 2018 ) and from the Leroux labs using a temperature sensitive allele of the dynein-2 heavy chain ( Jensen et al . , personal communication ) , both using C . elegans , provides support for this interpretation . The list of mutations causing disease in genes associated with primary cilia is continuously expanding . Our results show that Smo localization is deregulated in the absence of WDR60 . Smo is thought to enter cilia continuously but then rapidly exits in the absence of ligand . Our data are consistent with this , with loss of WDR60 resulting in aberrant accumulation of Smo in cilia , likely leading to perturbed hedgehog signaling ( also recently validated by others [Hamada et al . , 2018] ) . Interestingly , abnormal accumulation of Smo in the cilia has also been observed in mouse fibroblasts with mutations in DHC2/DYNC2H1 . In these mutant mice , inactivation of dynein-2 causes loss of Shh signaling and midgestation lethality ( Ocbina et al . , 2011 ) . Given the links between hedgehog signaling and skeletogenesis , this is likely to be a major cause of the phenotypes seen on loss-of-function of dynein-2 in animal models and in patients ( Dagoneau et al . , 2009; Li et al . , 2015; May et al . , 2005; Wu et al . , 2017 ) . In this study , we also characterized the function of dynein-2 using disease-causing mutations found in SRPs and JATD syndromes ( McInerney-Leo et al . , 2013 ) . Using mutagenesis to recreate patient mutations WDR60[Q631] and WDR60[T749M] , we show that the N-terminal region of WDR60 is sufficient to bind to WDR34 and TCTEX1/DYNLT1 but not to LIC3/DYNC2LI1 . Further analysis of patient mutation [Q631] reveals that the C-terminal β-propeller domain of WDR60 is required for binding of the IFT-B proteins , including IFT88 . The data do not let us conclude whether there is a direct interaction between WDR60 and IFT-B or whether WDR60 is required to assemble intact dynein which in turn can bind to IFT-B . The observed reduction in binding between the WDR60[Q631] mutant and NudCD3 is expected as the NudC family act as co-chaperones with Hsp90 to fold β-propellers such as WD repeat domains ( Taipale et al . , 2014 ) . In contrast , WDR60[T749M] binds to other dynein-2 proteins , IFT proteins , and NudCD3 . The reduction in binding to LIC3/DYNC2LI1 seen with this mutation suggests that less efficient dynein-2 assembly might contribute to the patient phenotypes . While these biochemical data are clear , a caveat here is that we are overexpressing these mutants which could overcome subtle defects . Our data show that not only is dynein-2 required for retrograde IFT but also to build and maintain a functional diffusion barrier at the base of the cilium . Our data do not discriminate between roles in assembly versus maintenance of the transition zone . Other recent work has shown that Joubert syndrome is caused by disruption of the transition zone ( Shi et al . , 2017 ) . Joubert syndrome leads to severe neurological effects that are underpinned by developmental defects in hedgehog signaling . Jeune syndrome is also attributed to disrupted hedgehog signaling but in this case this manifests as skeletal defects . Together our work and that of Shi et al . , 2017 suggests that disruption of transition zone architecture and resulting defects in developmental signaling might define a common root cause of both Joubert and Jeune syndromes , and indeed perhaps other ciliopathies . The human WDR34 gene was obtained from the Origene ( SC319901 , Cambridge Bioscience ) , human WDR60 was generated by gene synthesis ( Life Technologies , Paisley , UK ) . An HA tag or Myc tag for WDR34 and WDR60 was added by PCR and both proteins were subcloned in the pLVX-puro vector . Mutant [T749M WDR60] was generated by site-directed mutagenesis PCR using the primers: Fw: 5’- CAGAACCGCCatgTTCTCCACC-3’ and Rv 5’-GGTGGAGAACATGGCGGTTCTG-3’ changing codon ACG ( Threonine ) to ATG ( Methionine ) . WDR60 [Q631*] mutant was constructed by site-directed mutagenesis using the primers: Fw: 5’-GATAGCAGCTCCtagCTGAATACC-3’ and 5’-GGTATTCAGCTAGGAGCTGCTATC-3’ changing codon CAG ( Glutamine ) to TAG ( STOP codon ) . All constructs were validated by DNA sequencing . Mouse L13-Arl13b-GFP was a gift from Tamara Caspary ( Addgene plasmid # 40879 , [Larkins et al . , 2011] ) , IFT20-GFP ( [Follit et al . , 2006] , plasmid JAF2 . 13 ) was a gift from Gregory Pazour ( Addgene plasmid # 45608 ) . pEGFPN3-SSTR3 and pEGFPN3-5HT6 were gifts from Kirk Mykytyn ( Addgene plasmid #35624 and #35623 , [Berbari et al . , 2008a] ) , EGFP-Rab8 was a gift from Johan Peränen ( University of Helsinki ) . Cell lines were purchased from ATCC and not verified further other than to confirm no mycoplasma contamination . Human telomerase-immortalized retinal pigment epithelial cells ( hTERT-RPE-1 , ATCC CRL-4000 ) were grown in DMEM-F12 supplemented with 10% FBS ( Life Technologies , Paisley , UK ) at 37°C under 5% CO2 . Cells were not validated further after purchase from ATCC . Transient transfections of Arl13b-GFP , SSTR3-GFP , 5HT6-GFP , and Rab8-GFP were performed using Lipofectamine 2000 ( Life Technologies , Paisley , UK ) according to the manufacturer’s protocol . Lentiviral particles for each of the stable RPE-1 cell lines were produced in HEK293T cells using the Lenti-XTM HTX Packaging System ( Clontech , Saint-Germain-en-Laye , France ) . Low passage hTERT-RPE1 cells were transduced with the resultant viral supernatant , strictly according to the manufacturer’s directives and at 48 hr post-transduction , cells were subcultured in presence of 5 µg/ml puromycin . RPE-1 cells were incubated in serum-free medium for 24 hr to induce ciliogenesis . Confluent cells were placed in serum-free media and treated with Shh agonist SAG ( Selleckchem ( from Stratech Scientific , Ely , UK ) Catalog No . S7779 ) at the final concentration of 100 nM for 24 hr . The guide RNAs ( gRNA ) targeting WDR34 were designed using ‘chop chop’ software ( Labun et al . , 2016 ) or using CRISPR design http://crispr . mit . edu/ for designing WDR60 gRNA ( Hsu et al . , 2014 ) . pSpCas9 ( BB ) −2A-GFP ( Addgene plasmid , #PX458 ) was used as the vector to generate a gRNA . The gRNA sequences ( 5'- A GCC TTT CTT CGG AGA GTG G-3'; and 5'-CA GGT GTC TTG TCT GTA TAC −3' ) were designed to target Exon2 and Exon3 of human WDR34 . Similarly , the gRNA ( 5'-AG GTG CAG GGA TCC CGA CCA-3' ) was designed to target exon 3 of WDR60 . RPE-1 cells were transfected with 1 µg of pSpCas9 ( BB ) −2A-GFP . After 48 hr GFP-positive cells were sorted , and singles cells were plated in a 96 well plate . To check the WDR34 and WDR60 genes , genomic DNA was extracted and the target sequences subjected to PCR . Subsequently , the PCR products were cloned in the pGEM T Easy vector according to the manufacturer’s instructions and sequenced . In three cells clones , identified as WDR34 KO#1 , WDR34 KO#2 and WDR60 KO , small deletions/insertions causing a frameshift were detected in both alleles ( Supplementary Figure 1 for details ) . On the contrary , the cell clones identified as CRISPR CTRL WDR34 and WDR60 cells , transfected and treated in the same conditions of our knock out clones , did not show any mutation in the targeted genomic DNA region . The antibodies used , and their dilutions for western blotting ( WB ) and immunofluorescence ( IF ) are as follows: Acetylated tubulin ( Sigma ( Poole , UK ) T6793 1:2000 for IF ) , rabbit anti-HA ( Cell Signaling Technologies ( New England Biolabs , Hitchin , UK ) 1:2000 WB , 1:1000 IF ) , rabbit IFT88 ( Proteintech ( Manchester , UK ) 13967–1-AP , 1:200 WB , 1:300 IF ) , rabbit anti-IFT140 ( Proteintech 17460–1-AP , 1:200 , WB 1:100 IF ) , rabbit anti-IFT57 ( Proteintech 11083–1 , -AP 1:200 , WB 1:100 IF ) , rabbit anti-IFT43 ( Proteintech 24338–1-AP , 1:50 IF ) , anti-IFT20 ( Proteintech 13615–1-AP , 1:200 IF ) , anti-TMEM67 ( proteintech13975-1-AP , 1:50 IF ) , anti-RPGRIP1L ( Proteintech 55160–1-AP , 1:100 IF ) , anti-DYNC2HC1 ( Proteintech 55473–1-AP , 1:100 IF ) , rabbit anti-LIC3 ( Proteintech 15949–1-AP , 1:250 WB , 1:100 IF ) , rabbit anti-TCTEX1 ( Santa Cruz Biotechnology ( from Insight Biotechnology , Wembley , UK ) sc-28537 , 1:200 WB , 1:100 IF ) , rabbit anti-Arl13B ( Proteintech 17711-1AP , 1:1000 IF ) , rabbit anti-TCTN1 ( Proteintech 15004–1-AP , 1:100 IF ) , rabbit anti-Smo ( Abcam ( Cambridge , UK ) ab38686 , 1:100 IF ) , Sheep anti-Myc ( [Fan et al . , 2010] kindly provided by Harry Mellor , University of Bristol ) , rabbit anti-WDR60 ( Novus Biologicals ( from Bio-Techne Abingdon , UK ) NBP1-90437 1:300 WB in Figure 9 ) , rabbit anti-WDR60 ( Sigma HPA021316 , 1:300 WB in Figure 1—figure supplement 2 and Figure 9—figure supplement 1 ) , rabbit anti-WDR34 ( Novus NBP188805 , 1:300 WB ) , mouse anti-GAPDH ( Abcam ab9484 , 1:1000 WB ) , p150 glued ( BD 6127009 , city 1:1000 WB ) , LIS1 ( Bethyl A300-409A , ( Montgomery , TX ) 1:1000 WB ) , dic74 ( MAB1618 Millipore ( Watford , UK ) , 1:1000 WB ) , NUDCD3 ( Sigma HPA019136 , 1:350 WB ) . Cells grown on 0 . 17 mm thick ( #1 . 5 ) coverslips ( Fisher Scientific , Loughborough , UK ) were washed in PBS , and then fixed for 10 min in PFA and permeabilized with PBS containing 0 . 1% Triton X-100 for 5 min . Alternatively , cells were fixed in ice-cold methanol at −20°C for 5 min for TMEM67 , RPGRIP1L , TCTN1 , DYNC2H1 and IFT20 staining . For TCTEX1 immunolabelling , cells were washed twice with pre-warmed cytoskeletal buffer ( CB , containing 100 mM NaCl 300 mM sucrose , 3 mM MgCl2 , and 10 mM PIPES ) and fixed for 10 min in CB-PFA , as described previously ( Hua and Ferland , 2017 ) . Subsequently , cells were blocked using 3% BSA in PBS for 30 min at room temperature . The coverslips were incubated with primary antibodies for 1 hr at room temperature , washed in PBS and then incubated with secondary antibodies for another 1 hr at room temperature . Nuclear staining was performed using DAPI [4 , 6-diamidino-2-phenylindole ( Life Technologies ) , diluted at 1:5000 in PBS] for 3 min at room temperature , and the cells were then rinsed twice in PBS . Cells were imaged using an Olympus IX- 71 or IX-81 widefield microscope with a 63x objective , and excitation and emission filter sets ( Semrock , Rochester , NY ) controlled by Volocity software ( version 4 . 3 , Perkin-Elmer , Seer Green , UK ) . Alternatively , cells in Figure 4 and Figure 7 were imaged using Leica SP5 confocal microscope ( Leica Microsystems , Milton Keynes , UK ) . Live images in Figure 3 were imaged using Leica SP8 . All images were acquired as 0 . 5 µm z-stacks . All graphs show mean and standard deviation . For ‘rescue’ experiments , stable WDR60 KO cell lines overexpressing wild-type and HA-tagged WDR60 mutants were generated . Cells were serum starved for 24 hr , fixed and processed for immunofluorescence analysis . Cells were serum starved 24 hr and fixed in 2 . 5% glutaraldehyde for 20 min . Next , the cells were washed for 5 min in 0 . 1 M cacodylate buffer then post-fixed in 1% OsO4/0 . 1 M cacodylate buffer for 30 min . Cells were washed 3x with water and stained with 3% uranyl acetate for 20 min . After another rinse with water , cells were dehydrated by sequential 10 min incubations with 70 , 80 , 90 , 96 , 100% and 100% ethanol before embedding in Epon at 70°C for 48 hr . Thin ( 70 nm ) serial sections were cut and stained with 3% uranyl acetate then lead citrate , washing 3x with water after each . Once dried , sections were imaged using an FEI ( Cambridge , UK ) Tecnai12 transmission electron microscope . Cells were lysed in buffer containing 50 mM Tris pH7 . 5 , 150 mM NaCl , 1% Igepal and 1 mM EDTA . Samples were separated by SDS-PAGE followed by transfer to nitrocellulose membranes . Membranes were blocked in 5% milk-TBST . Primary antibodies diluted in blocking buffer were incubated with membrane overnight and detected using HRP-conjugated secondary antibodies ( Jackson ImmunoResearch , West Grove , PA ) and enhanced chemiluminescence ( GE Healthcare , Cardiff , United Kingdom ) . Quantification of fluorescence intensity was performed using original images . Measurement of intensity was performed using the average projections of acquired z-stacks of the area of the ciliary marker acetylated tubulin . Fluorescence intensity along the ciliary axoneme was measured using ImageJ plot profile tool . Fluorescence intensity at the ciliary base was measured drawing same diameter circles at the ciliary base . RPE-1 cells expressing the indicated cDNA constructs were washed with PBS and incubated with crosslinker solution ( 1 mM DSP , Thermo Fisher Scientific #22585 ) for 30 min on ice . The reaction was quenched by adding 500 mM Tris-HCl pH 7 . 5 for 15 min . Cells were washed twice with PBS and lysed in a buffer containing 50 mM Tris/HCl , pH 7 . 4 , 1 mM EDTA , 150 mM NaCl , 1% Igepal and protease inhibitors ( 539137 , Millipore ) . Subsequently , cells were incubated on a rotor at 4 ˚C for 30 min and then lysates were centrifuged at 13 , 000 g at 4 ˚C for 10 min . Cell lysates were added to the equilibrated anti-HA-Agarose beads ( Sigma A2095 , batch number 026M4810V ) and incubated on a rotor at 4 ˚C . Next , the beads were washed three times by centrifuging at 2000 g for 2 min at 4 ˚C with 1 ml of washing buffer ( 50 mM Tris-HCl , 150 mM NaCl , 0 . 5 mM EDTA , Triton X-100 0 . 3% SDS 0 . 1% ) containing protease inhibitors ( 539137 , Millipore ) . Samples used for SDS-PAGE and immunoblotting were resuspended in 50 µl of LDS sample buffer ( NP007 , Life Technologies ) containing sample reducing agent ( NP007 , Life Technologies ) and boiled at 95 ˚C for 10 min . For TMT Labelling and high pH reversed-phase chromatography , the samples were digested from the beads with trypsin ( 2 . 5 µg trypsin , 37°C overnight ) , labeled with Tandem Mass Tag ( TMT ) six-plex reagents according to the manufacturer’s protocol ( Thermo Fisher Scientific , Loughborough , UK ) and the labeled samples pooled . The pooled sample was then desalted using a SepPak cartridge according to the manufacturer’s instructions ( Waters , Milford , Massachusetts , USA ) ) . Eluate from the SepPak cartridge was evaporated to dryness and resuspended in buffer A ( 20 mM ammonium hydroxide , pH 10 ) prior to fractionation by high pH reversed-phase chromatography using an Ultimate 3000 liquid chromatography system ( Thermo Fisher Scientific ) . In brief , the sample was loaded onto an XBridge BEH C18 Column ( 130 Å , 3 . 5 µm , 2 . 1 mm X 150 mm , Waters , UK ) in buffer A and peptides eluted with an increasing gradient of buffer B ( 20 mM ammonium hydroxide in acetonitrile , pH 10 ) from 0–95% over 60 min . The resulting fractions were evaporated to dryness and resuspended in 1% formic acid prior to analysis by nano-LC MSMS using an Orbitrap Fusion Tribrid mass spectrometer ( Thermo Fisher Scientific ) . High pH RP fractions were further fractionated using an Ultimate 3000 nano-LC system in line with an Orbitrap Fusion Tribrid mass spectrometer ( Thermo Fisher Scientific ) . In brief , peptides in 1% ( vol/vol ) formic acid were injected onto an Acclaim PepMap C18 nano-trap column ( Thermo Fisher Scientific ) . After washing with 0 . 5% ( vol/vol ) acetonitrile 0 . 1% ( vol/vol ) formic acid , peptides were resolved on a 250 mm ×75 μm Acclaim PepMap C18 reverse phase analytical column ( Thermo Fisher Scientific ) over a 150 min organic gradient , using seven gradient segments ( 1–6% solvent B over 1 min . , 6–15% B over 58 min . , 15–32% B over 58 min . , 32–40% B over 5 min . , 40–90% B over 1 min . , held at 90% B for 6 min and then reduced to 1% B over 1 min . ) with a flow rate of 300 nl min−1 . Solvent A was 0 . 1% formic acid and Solvent B was aqueous 80% acetonitrile in 0 . 1% formic acid . Peptides were ionized by nano-electrospray ionization at 2 . 0 kV using a stainless steel emitter with an internal diameter of 30 μm ( Thermo Fisher Scientific ) and a capillary temperature of 275°C . All spectra were acquired using an Orbitrap Fusion Tribrid mass spectrometer controlled by Xcalibur 2 . 0 software ( Thermo Fisher Scientific ) and operated in data-dependent acquisition mode using an SPS-MS3 workflow . FTMS1 spectra were collected at a resolution of 120 000 , with an automatic gain control ( AGC ) target of 400 000 and a max injection time of 100 ms . Precursors were filtered with an intensity range from 5000 to 1E20 , according to charge state ( to include charge states 2–6 ) and with monoisotopic precursor selection . Previously interrogated precursors were excluded using a dynamic window ( 60 s ± 10 ppm ) . The MS2 precursors were isolated with a quadrupole mass filter set to a width of 1 . 2 m/z . ITMS2 spectra were collected with an AGC target of 10 000 , max injection time of 70 ms and CID collision energy of 35% . For FTMS3 analysis , the Orbitrap was operated at 30 000 resolution with an AGC target of 50 000 and a max injection time of 105 ms . Precursors were fragmented by high energy collision dissociation ( HCD ) at a normalized collision energy of 55% to ensure maximal TMT reporter ion yield . Synchronous Precursor Selection ( SPS ) was enabled to include up to 5 MS2 fragment ions in the FTMS3 scan . The raw data files were processed and quantified using Proteome Discoverer software v2 . 1 ( Thermo Fisher Scientific ) and searched against the UniProt Human database ( 140000 entries ) and GFP sequence using the SEQUEST algorithm . Peptide precursor mass tolerance was set at 10 ppm , and MS/MS tolerance was set at 0 . 6 Da . Search criteria included oxidation of methionine ( +15 . 9949 ) as a variable modification and carbamidomethylation of cysteine ( +57 . 0214 ) and the addition of the TMT mass tag ( +229 . 163 ) to peptide N-termini and lysine as fixed modifications . Searches were performed with full tryptic digestion and a maximum of 1 missed cleavage was allowed . The reverse database search option was enabled and the data was filtered to satisfy false discovery rate ( FDR ) of 5% .	Almost all cells in the human body are covered in tiny hair-like structures known as primary cilia . These structures act as antennae to receive signals from outside the cell that regulate how the body grows and develops . The cell has to deliver new proteins and other molecules to precise locations within its cilia to ensure that they work properly . Each cilium is separated from the rest of the cell by a selective barrier known as the transition zone , which controls the movement of molecules to and from the rest of the cell . Dynein-2 is a motor protein that moves other proteins and cell materials within cilia . It includes two subunits known as WDR34 and WDR60 . The genes that produce these subunits are mutated in Jeune and short rib polydactyly syndromes that primarily affect how the skeleton forms . However , little is known about the roles the individual subunits play within the motor protein . Vuolo et al . used a gene editing technique called CRISPR-Cas9 to remove one or both of the genes encoding the dynein-2 subunits from human cells . The experiments show that the two subunits have very different roles in cilia . WDR34 is required for cells to build a cilium whereas WDR60 is not . Instead , WDR60 is needed to move proteins and other materials within an established cilium . Unexpectedly , the experiments suggest that dynein-2 is also required to maintain the transition zone . This work provides the foundations for future studies on the role of dynein-2 in building and maintaining the structure of cilia . This could ultimately help to develop new treatments to reduce the symptoms of Jeune syndrome and other diseases caused by defects in cilia .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2018	Dynein-2 intermediate chains play crucial but distinct roles in primary cilia formation and function
It is unclear that how subcellular organelles respond to external mechanical stimuli . Here , we investigated the molecular mechanisms by which mechanical force regulates Ca2+ signaling at endoplasmic reticulum ( ER ) in human mesenchymal stem cells . Without extracellular Ca2+ , ER Ca2+ release is the source of intracellular Ca2+ oscillations induced by laser-tweezer-traction at the plasma membrane , providing a model to study how mechanical stimuli can be transmitted deep inside the cell body . This ER Ca2+ release upon mechanical stimulation is mediated not only by the mechanical support of cytoskeleton and actomyosin contractility , but also by mechanosensitive Ca2+ permeable channels on the plasma membrane , specifically TRPM7 . However , Ca2+ influx at the plasma membrane via mechanosensitive Ca2+ permeable channels is only mediated by the passive cytoskeletal structure but not active actomyosin contractility . Thus , active actomyosin contractility is essential for the response of ER to the external mechanical stimuli , distinct from the mechanical regulation at the plasma membrane . Mechanical factors are known to play crucial roles in both development and tissue regeneration from stem cells . However , it remains unclear how these factors , such as mechanical forces , are converted into biochemical signals in stem cells to regulate regeneration processes . Calcium ion ( Ca2+ ) as one of the most important biochemical signals , is involved in many cellular processes , including muscle contraction , differentiation , proliferation , gene expression , and apoptosis ( Berridge et al . , 2000; West et al . , 2001; McKinsey et al . , 2002; Landsberg and Yuan , 2004; Clapham , 2007; Maeda et al . , 2007; Rong and Distelhorst , 2008 ) . Various mechanical stimulations can affect cytosolic Ca2+ signals as well as Ca2+ dynamics in organelles or subcellular compartments , such as mitochondria and focal adhesion sites ( Balasubramanian et al . , 2007; Belmonte and Morad , 2008; Hayakawa et al . , 2008; Horner and Wolfner , 2008 ) . To apply mechanical force precisely , we utilize optical laser tweezers to generate force in a bead coupled to the cell surface through the ligation of adhesion receptors and transmit it into the cell at subcellular locations to trigger signal transduction ( Berns , 2007; Botvinick and Wang , 2007 ) . Most cells mobilize their Ca2+ signals via the Ca2+ entry across the plasma membrane and/or the Ca2+ release from intracellular stores such as endoplasmic reticulum ( ER ) or sarcoplasmic reticulum ( SR ) ( Wehrens et al . , 2005; Clapham , 2007 ) . Biochemical signals , such as inositol 1 , 4 , 5-trisphosphate ( IP3 ) , are known to regulate Ca2+ release from ER . But direct regulation of ER Ca2+ signals by mechanical force is unknown . The human mesenchymal stem cells ( HMSCs ) displaying Ca2+ oscillations provide a model system to study that ( Kawano et al . , 2002; Sun et al . , 2007 ) . As Ca2+ influx at the plasma membrane and release from ER are the only two sources for Ca2+ oscillations in HMSCs ( Kawano et al . , 2002; Kim et al . , 2009 ) , we dissected the effect of mechanical force on each process by monitoring the calcium signals at subcellular locations . To visualize Ca2+ signal with high spatiotemporal resolutions , we employed a fluorescence resonance energy transfer ( FRET ) -based Ca2+ biosensor and its variants anchored at subcellular organelles ( Miyawaki et al . , 1997; Ouyang et al . , 2008 ) . Here , we combined optical laser tweezers to deliver local mechanical force and FRET biosensors to investigate how mechanical force regulates Ca2+ signals at different subcellular locations in HMSCs . We first investigated how force regulates Ca2+ release from ER with the FRET-based Ca2+ biosensor ( Figure 1A ) to dissect the effect of mechanical force on each of the two Ca2+ mobilization processes ( Figure 1B ) . The experiments were done in the absence of extracellular Ca2+ so that there was no Ca2+ influx through plasma membrane and the oscillations were mainly from ER release ( Figure 1B ) . When fibronectin ( Fn ) -coated beads were seeded onto the HMSCs and 300 pN of mechanical force was applied by optical laser tweezers as described previously ( Wang et al . , 2005 ) , Ca2+ oscillations were induced in HMSCs ( Figure 1C-F , Video 1 ) but not in bovine aortic endothelial cells ( BAECs ) ( Figure 1—figure supplement 1 ) . In contrast , laser-tweezer-traction on bovine serum albumin ( BSA ) -coated beads did not cause any oscillations ( Figure 1C–F ) . These results indicate that without extracellular Ca2+ , mechanical force can induce Ca2+ oscillations by triggering Ca2+ release from ER . Further experiments showed that depletion of ER Ca2+ by Thapsigargin ( TG ) or inhibition of Ca2+ release from the ER by 2-Amino-ethoxydiphenylborate ( 2-APB ) entirely abolished the force-induced oscillations ( Figure 1—figure supplement 2A , B ) . These results confirmed that the ER Ca2+ store is the main source for the force-induced oscillations in HMSCs without extracellular Ca2+ . 10 . 7554/eLife . 04876 . 003Figure 1 . Intracellular Ca2+ oscillations in response to mechanical force in HMSCs with Ca2+-free medium . ( A ) A schematic drawing of the activation mechanism of the Ca2+ FRET biosensor . ( B ) Beads coated with Fn or BSA were seeded onto the cell and mechanical force was applied by pulling a Fn-coated bead using optical laser tweezers . Both Ca2+ influx and ER Ca2+ release can contribute to force-induced Ca2+ signals . ( C ) Color images represent the YPet/ECFP emission ratio of the cytoplasmic Ca2+ biosensor . The color scale bars represent the range of emission ratio , with cold and hot colors indicating low and high levels of Ca2+ concentration , respectively . ( D ) The time courses represent the YPet/ECFP emission ratio averaged over the cell body outside of nucleus upon seeding of Fn or BSA-coated beads and force application . ( E–F ) Bar graphs represent the frequency or ratio of the intracellular Ca2+ oscillations evoked by mechanical force . Error bars indicate standard error of mean; p < 0 . 05 , n = 14 . ( Scale bar: 10 µm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04876 . 00310 . 7554/eLife . 04876 . 004Figure 1—figure supplement 1 . Laser-tweezer pulling of a Fn-coated bead on a BAEC in Ca2+-free medium . Color images ( upper panels ) represent the YPet/ECFP emission ratio of the cytoplasmic Ca2+ biosensor . The color scale bars represent the range of emission ratio , with cold and hot colors indicating low and high levels of Ca2+ concentration , respectively . The time courses of the YPet/ECFP emission ratio averaged over the cell body outside of nucleus is shown in the lower graph . DOI: http://dx . doi . org/10 . 7554/eLife . 04876 . 00410 . 7554/eLife . 04876 . 005Figure 1—figure supplement 2 . Mechanical force doesn’t induce any increase in IP3 level . The time courses represent the YPet/ECFP emission ratio of cytoplasmic Ca2+ in HMSCs pretreated with ( A ) Thapsigargin ( TG , 10 µM ) , a SERCA pump blocker , or ( B ) 2-APB ( 100 µM ) , an IP3R blocker . ( C ) IP3 production is monitored by a FRET-based IP3 biosensor , LIBRAvIIs . ATP treatment induces IP3 increase , which can be clearly detected by an IP3 biosensor LIBRAvIIs ( upper panels ) . However , laser-tweezer pulling of a Fn-coated bead to produce the mechanical force did not cause any increase in IP3 ( lower panels ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04876 . 00510 . 7554/eLife . 04876 . 006Video 1 . A HMSC transfected with cytosolic Ca2+ biosensors before and after mechanical force application by optical laser tweezers on a Fn-coated bead attached to the cell ( Duration of Video: 2700 s ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04876 . 006 Two possible mechanisms can regulate Ca2+ release from ER upon mechanical laser-tweezer-traction . External force can either 1 ) transmit deep inside the cell and mechanically alter the channels on ER for Ca2+ release ( Himmel et al . , 1993; Missiaen et al . , 1996; Rath et al . , 2010 ) ; or 2 ) trigger biochemical signaling cascades to produce IP3 that diffuses inside to activate IP3-sensitive Ca2+ channels ( Lehtonen and Kinnunen , 1995; Matsumoto et al . , 1995 ) . To distinguish them , we directly monitored the IP3 level using a FRET-based biosensor , LIBRAvIIs . Mechanical force did not induce any change of IP3 while ATP increased IP3 production in HMSCs ( Figure 1—figure supplement 2C ) , suggesting that the second mechanism is unlikely . Therefore , laser-tweezer-traction should transmit deep inside the cells to mechanically release Ca2+ from ER . Cytoskeleton is known to transmit mechanical forces and conduct mechanotransduction ( Hamill and Martinac , 2001; Orr et al . , 2006; Schwartz and DeSimone , 2008 ) , so we investigated the role of cytoskeleton and its associated proteins in the force-induced ER Ca2+ release . The disruption of cytoskeletal actin filaments by cytochalasin D ( Cyto D ) or microtubules by nocodazole ( Noc ) completely eliminated the force-induced Ca2+ oscillations ( Figure 2A–B ) . In addition , the inhibition of actomyosin contractility by ML-7 or blebbistatin had the same effect ( Figure 2C–D ) . Thus , the deep penetration and transmission of force inside the cell and the induction of Ca2+ release from ER depend on both cytoskeletal support and actomyosin contractility . This matches with the previous reports that long-distance force propagation to the deep cytoplasm depends on cytoskeleton tension ( Hu et al . , 2003 ) as well as the pivotal role of myosin light chain kinase ( MLCK ) and myosin II in regulating force development ( Olsson et al . , 2004; Herring et al . , 2006; Fajmut and Brumen , 2008; He et al . , 2008 ) . There are two possible signals in ER that may trigger the Ca2+ release in response to mechanical force . 1 ) IP3R channels on the ER membrane are mechanosensitive and can be directly opened by transmitted mechanical force . Several lines of evidence supported this hypothesis . First , IP3R is coupled to cytoskeleton and associated proteins allowing mechanical coupling . A direct binding between IP3R and myosin II was discovered in C . elegans ( Walker et al . , 2002 ) . In addition , IP3R has linkage to actin mediated by an adaptor 4 . 1N protein ( Fukatsu et al . , 2004; Turvey et al . , 2005 ) . IP3R also binds to ankyrins , which are adaptor proteins coupled to the spectrin-based cytoskeleton ( Bourguignon et al . , 1993; Joseph and Samanta , 1993 ) . Second , IP3R channel has an α-helix bundle at the pore forming region , similar to voltage-gated potassium and calcium channels ( Schug et al . , 2008 ) , which are generally found to be mechanosensitive ( Morris , 2011 ) . The mechanism for their mechanosensitivity is possibly that the α-helix tilt angle tends to change when the membrane thins upon mechanical tension , in order to do proper hydrophobic matching with the interfacial region of the membrane , which leads to channel opening ( Cheng et al . , 2004; Kim and Im , 2010 ) . Therefore , it is likely that the IP3R channel is also mechanosensitive . 2 ) Other mechanosensitive channels on ER , for example , transient receptor potential ( TRP ) family , may also contribute to this force-induced Ca2+ release . A number of TRP channels have been found to express at ER membranes , such as TRPC1 ( Berbey et al . , 2009 ) , TRPV1 ( Gallego-Sandin et al . , 2009 ) , TRPM8 ( Bidaux et al . , 2007 ) , and TRPP2 ( Koulen et al . , 2002 ) . As some TRP channels have been shown to be mechanosensitive and have linkage to cytoskeleton ( Barritt and Rychkov , 2005 ) , it is likely that TRP channels located at ER may mediate , at least in part , the force-induced ER calcium release . Notably , these two possibilities are not mutually exclusive as more than one type of channels can be responsible for the force-induced ER calcium release . 10 . 7554/eLife . 04876 . 007Figure 2 . Cytoskeletal support , actomyosin contractility , and TRPM7 channels mediate the force-induced intracellular Ca2+ oscillations . The time courses represent the YPet/ECFP emission ratio of cytoplasmic Ca2+ in HMSCs in the absence of extracellular Ca2+ when these cells were pretreated with ( A ) 2 μM Cyto D ( n = 8 ) , ( B ) 1 μM Noc ( n = 8 ) , ( C ) 5 μM ML-7 ( n = 8 ) , and ( D ) 5 μM Bleb ( n = 8 ) . ( E ) Color images represent the YPet/ECFP emission ratio of the cytoplasmic Ca2+ biosensor in HMSCs transfected with NT or TRPM7 siRNA . The color scale bars represent the range of emission ratio , with cold and hot colors indicating low and high levels of Ca2+ concentration , respectively . ( F ) The time courses represent the YPet/ECFP emission ratio averaged over the cell bodies outside of nucleus treated with siRNA as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 04876 . 00710 . 7554/eLife . 04876 . 008Figure 2—figure supplement 1 . Stretch-activated or store-operated channels at the plasma membrane mediate force-induced Ca2+ release from ER . The time courses represent the YPet/ECFP emission ratio of cytoplasmic Ca2+ in HMSCs pretreated with ( A ) GdCl3 ( 5 µM ) , a broad spectrum Ca2+ channel inhibitor , ( B ) LaCl3 ( 100 µM ) , a broad spectrum Ca2+ channel blocker , ( C ) streptomycin ( 200 µM ) , a mechanosensitive channel inhibitor , or ( D ) Nifedipine ( 10 µM ) , an L-type Ca2+ channel inhibitor , all in the absence of extracellular Ca2+ . DOI: http://dx . doi . org/10 . 7554/eLife . 04876 . 00810 . 7554/eLife . 04876 . 009Figure 2—figure supplement 2 . TRPM7 channels as well as cytoskeletal support and actomyosin contractility mediate the force-induced intracellular Ca2+ oscillations . ( A ) HMSCs expressing Ca2+ biosensor and transfected with non-targeting ( NT ) or TRPM7 siRNA were immunostained ( upper images ) or immunoblotted ( lower panels ) with polyclonal TRPM7 antibody to assess the amount of TRPM7 . TRPM7 specific siRNA induced knockdown of TRPM7 . ( B ) Bar graphs represent the percentile of HMSCs showing intracellular Ca2+ oscillations . Three groups of cells ( 1: Ca2+ biosensor only , 2: biosensor and non-targeting ( NT ) siRNA , and 3: biosensor and TRPM7 siRNA ) were measured and compared . The number of cells displaying Ca2+ oscillations in both control group ( 55% , 11 of 20 cells ) and NT-siRNA group ( 45% , 9 of 20 cells ) was approximately 9–11 fold higher than that of TRPM7 siRNA group ( 4 . 76% , 1 of 21 cells ) . ( C–D ) Bar graphs represent the frequency or ratio of the intracellular Ca2+ oscillations induced by mechanical force in the presence of different inhibitors . Error bars indicate standard error of mean; p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 04876 . 009 Surprisingly , blocking stretch-activated or store-operated channels at the plasma membrane by Gd3+ , La3+ , or streptomycin , but not L-type voltage-operated Ca2+ channels by Nifedipine , abolished the mechanical force-induced Ca2+ release from ER ( Figure 2—figure supplement 1 ) . These results suggest a possible coupling between force-induced Ca2+ release at ER and stretch-activated and store-operated channels at the plasma membrane . As TRPM7 is one of the major Ca2+ permeable mechanosensitive channels ( Wei et al . , 2009 ) , we knocked down TRPM7 with targeting small interfering RNA ( siRNA ) to examine its role in the force-induced Ca2+ oscillations . The decreased expression of TRPM7 and the lower percentile of HMSCs with Ca2+ oscillations confirmed the effect of siRNA ( Figure 2—figure supplement 2A , B ) TRPM7 siRNA further abrogated the force-induced oscillations ( Figure 2E–F ) . It is intriguing that the inhibition of TRPM7 at the plasma membrane can block the force transmission into ER in regulating calcium signals . Several possibilities may contribute to the observed results . 1 ) TRPM7 is functionally coupled to integrin , actomyocin contractility and cytoskeleton . As such , it may mediate and facilitate the force transmission to ER . Indeed , it has been shown that TRPM7 kinase can phosphorylate myosin II heavy chain ( Clark et al . , 2006 ) and regulate actomyocin contractility . 2 ) TRPM7 activity may have some downstream effect on IP3R in ER . For example , TRPM7 can control the protease calpain ( Su et al . , 2006 ) , which can regulate IP3R degradation in ER ( Diaz and Bourguignon , 2000 ) . 3 ) TRPM7 may also be directly coupled to IP3R in the ER through adaptor proteins . Indeed , another TRP channel , TRPC1 has been shown to directly couple to IP3R in the ER through an adaptor protein Homer ( Yuan et al . , 2003 ) . It becomes clear that membrane channels are not isolated entities floating in the plasma membrane . Instead , they are intimately coupled to integrins , cytoskeleton , actomyocin contractility , and ER membrane channels ( Cantiello et al . , 2007; Matthews et al . , 2007; Deng et al . , 2009 ) . Therefore , these structural and physical couplings enable membrane channels to participate in direct force transmission to ER . We further visualized the Ca2+ release from ER directly , by generating and employing an improved ER-targeting Ca2+ biosensor ( D3ER ) ( Palmer et al . , 2006; Ouyang et al . , 2008 ) . Mechanical force could directly induce the Ca2+ release from ER without extracellular Ca2+ , as evidenced by the decrease in ER Ca2+ concentration ( Figure 3A-B , Video 2 ) . The measurements of direct Ca2+ release from ER in the presence of inhibitors or siRNA of TRMP7 were also consistent with our observations when cytosolic Ca2+ biosensor was used ( Figure 3C ) . 10 . 7554/eLife . 04876 . 010Figure 3 . The visualization of force-induced Ca2+ release from ER using a FRET-based ER Ca2+ biosensor . ( A ) The time course and ( B ) the color images of YPet/ECFP emission ratio in HMSCs expressing the D3ER biosensor before and after force application . The red arrows indicated episodes of Ca2+ release from ER . ( C ) The bar graphs represent the normalized changes of YPet/ECFP emission ratio of the D3ER in HMSCs upon force application without extracellular Ca2+ in the untreated cells as the control group ( n = 3 ) or cells pretreated with CytoD ( n = 5 ) , Noc ( n = 5 ) , ML-7 ( n = 6 ) , Bleb ( n = 5 ) , or TRPM7 siRNA ( n = 9 ) as indicated . * represents p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 04876 . 01010 . 7554/eLife . 04876 . 011Video 2 . A HMSC transfected with ER Ca2+ biosensors before and after mechanical force stimulation by optical laser tweezers on a Fn-coated bead attached to the cell ( Duration of Video: 450 s ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04876 . 011 To gain more insights , we then examined the effect of force on Ca2+ influx at the plasma membrane . ER Ca2+ release was blocked by an IP3Rs inhibitor , 2-APB , in the presence of extracellular Ca2+ so that the only source of intracellular Ca2+ change is the calcium influx from extracellular space ( Bootman et al . , 2002 ) . Nifedipine did not have any effect on the force-induced Ca2+ influx , while Gd3+ , La3+ , streptomycin , and TRPM7 siRNA significantly inhibited this Ca2+ influx ( Figure 4A–B ) , confirming the involvement of mechanosensitive channels , especially TRPM7 , which was shown to be directly activated by a membrane stretch or shear stress in various cell types ( Numata et al . , 2007a , 2007b; Wei et al . , 2009 ) . The disruption of actin filaments or microtubules also inhibited the force-induced Ca2+ influx ( Figure 4C–D ) , suggesting that cytoskeletal integrity is essential for the membrane channels to respond to force , consistent with previous reports ( Hayakawa et al . , 2008 ) . Interestingly , ML-7 or blebbistatin did not affect this force-induced Ca2+ influx , suggesting that active actomyosin contractility may not be involved in the mechano-regulation of membrane channels ( Figure 4C–D ) , different from their indispensable roles in the force-induced ER Ca2+ release ( Figure 4E ) . 10 . 7554/eLife . 04876 . 012Figure 4 . Ca2+ influx in response to mechanical force in Ca2+-containing medium . Ca2+ release from ER in all the HMSCs was blocked by pretreatment with 2-APB . ( A , C ) Color images represent the YPet/ECFP emission ratio of the cytoplasmic Ca2+ biosensor in control cells treated by 2-APB only ( n = 5 ) or those co-treated by Nifedipine ( n = 5 ) , Gd3+ ( n = 3 ) , La3+ ( n = 6 ) , STM ( n = 8 ) , or TRPM7 siRNA ( n = 9 ) , CytoD ( n = 8 ) , Noc ( n = 4 ) , ML-7 ( n = 9 ) or blebbistatin ( n = 6 ) . Arrows in DIC images point to the direction of applied force . ( B , D ) Bar graphs represent the normalized change of YPet/ECFP emission ratio of the cytoplasmic Ca2+ biosensor under different conditions as indicated in ( A , C ) . Error bars indicate standard errors of mean; * represents p < 0 . 05 . ( E ) The models depicting the mediators of mechanical force-induced Ca2+ influx or ER Ca2+ release . DOI: http://dx . doi . org/10 . 7554/eLife . 04876 . 01210 . 7554/eLife . 04876 . 013Figure 4—figure supplement 1 . Src , FAK or PI3K has no effect on the mechanical force-induced Ca2+ signals . ( A ) The inhibition of neither Src by PP1 ( n = 4 ) , FAK by PF228 ( n = 4 ) , nor PI3K by LY294002 ( n = 3 ) abolished the force-induced cytosolic Ca2+ oscillations in HMSCs without extracellular Ca2+ . ( B ) The Ca2+ release from ER in all the HMSCs was blocked by pretreatment with 2-APB . Color images represent the YPet/ECFP emission ratio of the cytoplasmic Ca2+ biosensor in cells pretreated by 2-APB together with PP1 ( n = 5 ) , PF228 ( n = 4 ) , or LY294002 ( n = 6 ) . Arrows in DIC images point to the direction of applied force . Bar graphs represent the normalized change of YPet/ECFP emission ratio of the cytoplasmic Ca2+ biosensor under different conditions as indicated in ( A ) . ( C ) The average delay time between force application and the first Ca2+ signals in three groups . First two groups were ER Ca2+ release monitored by either cytoplasmic or ER Ca2+ biosensor . The last group was Ca2+ influx in the presence of 2-APB monitored by cytoplasmic Ca2+ biosensor . ( *** , p < 0 . 001 ) Error bars indicate standard errors of mean . DOI: http://dx . doi . org/10 . 7554/eLife . 04876 . 013 Since mechanical stimulation such as stretch affects integrin adhesion and integrin-associated signaling , including Src and focal adhesion kinase ( FAK ) ( Wang et al . , 2005; Orr et al . , 2006 ) , we investigated whether Src and FAK mediate the force-induced Ca2+ signals . Our results indicated that neither PP1 , an inhibitor of Src family kinases , nor PF228 , an inhibitor of FAK , had a significant effect on the force-induced Ca2+ influx or ER release ( Figure 4—figure supplement 1 ) . Inhibition of phosphoinositide 3-kinases ( PI3Ks ) with LY294002 did not have any effect either ( Figure 4—figure supplement 1 ) . Although integrin-mediated Src phosphorylation and activation can regulate L-type Ca2+ channel functions ( Gui et al . , 2010 ) , the force-induced Ca2+ oscillations were clearly independent of L-type Ca2+ channels in HMSCs ( Figure 2—figure supplement 1 ) . Mechanical stretch can induce the phosphorylation of FAK and nitric oxide formation in cardiomyocytes ( Pinsky et al . , 1997; Khan et al . , 2003 ) , which can modulate ryanodine receptor 2 ( RYR2 ) and Ca2+ release at the SR ( Vila Petroff and Mattiazzi , 2001 ) . However , there is minimal expression of RYRs in HMSCs ( Kawano et al . , 2002 ) , which may explain the lack of FAK involvement in the mechano-regulation of Ca2+ signals . Together , the biochemical activities above have no detected involvement in the force-induced Ca2+ signals . Further analysis of the duration between force application and the first Ca2+ signal indicated that in the absence of extracellular Ca2+ , the average delay time of ER calcium release monitored by both cytosolic Ca2+ biosensor and ER Ca2+ biosensor are around 100 s ( Figure 4—figure Supplement 1C ) . Meanwhile , the average delay time of the calcium influx through plasma membrane in the presence of 2-APB and extracellular calcium is much shorter , around 20 s ( Figure 4—figure supplement 1C ) , indicating faster response to mechanical force stimulation . Although the delay time of ER calcium release is longer than that of calcium influx across the plasma membrane , it does not contradict mechanical signal-mediated mechanism . Several reasons may contribute to longer time delay of ER calcium release . 1 ) The mechanical coupling machinery may need time for reinforcement to allow sufficient force transfer between the plasma membrane and the ER membrane . It is likely that mechanical force transmits through integrins and cytoskeleton ( Wang et al . , 1993; Perozo and Rees , 2003 ) as only Fn-coated , but not BSA-coated beads can induce ER calcium release ( Figure 1C–F ) . However , the discrete network linkage of the existing cytoskeleton at the time of force application may not be sufficient for force focusing and transmission to ER , particularly because the apical integrins bound to the Fn-coated beads had not previously experienced force and therefore may have limited connection with cytoskeleton ( Matthews et al . , 2006 ) . As force can unfold proteins , change their conformations to expose cryptic binding sites to allow the assembly of new/stronger network linkages ( del Rio et al . , 2009; Johnson et al . , 2007 ) , mechanical force can modulate the structural coupling of molecules and mechanical properties of the cells through cytoskeletal remodeling ( Kaazempur Mofrad et al . , 2005; Matthews et al . , 2006; Rosenblatt et al . , 2007 ) . These modulation processes of mechanical coupling and reinforcement may need time to reach the threshold for sufficient mechanotransduction of ER calcium release; 2 ) the channels on ER membrane could have different kinetics and mechanosensitivity ( Moe et al . , 1998 ) , therefore they may need larger focused and reinforced force to be developed at the site of ER to reach the threshold for physical opening . All these factors may contribute to the longer delay time for force-induced ER calcium release . Again our model does not rule out biochemical signal-mediated mechanisms . As a matter of fact , biochemical signal-mediated mechanisms , such as protein–protein interaction and cytoskeletal remodeling under mechanical tension are important to mediate mechanical signals as discussed above . As such , our results have provided molecular insights on how mechanical force triggers intracellular Ca2+ oscillations through two mechanisms in HMSCs: Ca2+ influx at the plasma membrane and Ca2+ release from ER ( Figure 1B ) . Our results showed that the deep penetration and transmission of mechanical force to regulate ER functions is dependent on not only the passive cytoskeletal support of actin filaments and microtubules , but also the active actomyosin contractility controlled by MLCK and myosin II . In contrast , the passive cytoskeletal support , but not active actomyosin contractility , is needed for the mechanotransduction at the plasma membrane levels , including the mechanoactivation of channels and Ca2+ influx across the plasma membrane ( Figure 4E ) . These results hence provide direct evidence that the mechanotransduction at different depths of cell body is mediated by differential sets of mechanosensing elements . The construct of FRET-based Ca2+ biosensor has been described well in our previous articles ( Ouyang et al . , 2008; Kim et al . , 2009 ) . In brief , the fragment containing enhanced cyan fluorescent protein ( ECFP ) , calmodulins ( CaMs ) , and M13 was fused to YPet and subcloned into pcDNA3 . 1 ( Invitrogen , Carlsbad , CA ) for mammalian cell expression by using BamHI and EcoRI sites . The ECFP/YPet pair has allowed a higher sensitivity of FRET biosensors than those based on ECFP/Citrine pair . To generate an improved ER-targeting Ca2+ biosensor , the mutant peptide and CaMs regions were replaced with those regions of D3cpv and cloned between a truncated ECFP and YPet . For the ER targeting motifs , the calreticulin signal sequence MLLPVLLLGLLGAAAD was added 5′ to ECFP , and an ER retention sequence KDEL to the 3′ end of YPet . The construct of a FRET-based IP3 biosensor , LIBRAvIIs was kindly provided by Professor Akihiko Tanimura at University of Hokkaido , Japan ( Tanimura et al . , 2009 ) . Human mesenchymal stem cells ( HMSCs ) and bovine aortic endothelial cells ( BAECs ) were obtained from the American Type Culture Collection ( ATCC , Rockville , MD ) . HMSCs and BAECs were cultured in human mesenchymal stem cell growth medium ( MSCGM , PT-3001 , Lonza Walkersville , Inc . , Walkersville , MD ) and in Dulbecco's modified Eagle's medium ( DMEM ) , respectively , supplemented with 10% fetal bovine serum ( FBS ) , 2 mM L-glutamine , 100 U/ml penicillin , and 100 µg/ml streptomycin . The cells were cultured in a humidified incubator of 95% O2 and 5% CO2 at 37°C . The DNA plasmids were transfected into the cells by using Lipofectamine 2000 ( Invitrogen , Carlsbad , CA ) according to the product instructions . Double-stranded small interfering RNA ( siRNA ) sequences targeting human TRPM7 ( ON-TARGETplus SMARTpool siRNA ) and non-targeting control sequences were designed by Dharmacon RNAi Technology ( Dharmacon Inc . , Lafayette , CO ) . HMSCs were transfected with 1–2 µg siRNA specific for TRPM7 or a non-silencing control sequence according to the product instructions . The cells transfected with TRPM7 or non-targeting siRNA were washed twice with cold phosphate buffered saline ( PBS ) and then lysed in lysis buffer containing 50 mM Tris , pH 7 . 4 , 150 mM NaCl , 1 mM EDTA , 1% Triton X-100 , and a mix of serine and cysteine protease inhibitors . Lysates were centrifuged at 10 , 000×g at 4°C for 10 min . Cell lysates were then applied to 15% SDS-polyacrylamide gel electrophoresis , transferred to nitrocellulose , blocked with 5% non-fat milk , and detected by Western blotting using polyclonal goat anti-TRPM7 antibody ( 1:100; Abcam Inc . , Cambridge , MA ) . A polyclonal antibody against TRPM7 was used in both normal HMSCs and TRPM7-knockdown HMSCs . After being washed in cold phosphate buffered saline ( PBS ) , the samples were fixed by 4% paraformaldehyde in PBS at room temperature for 15 min . The samples were incubated with a goat polyclonal antibody against TRPM7 ( 1:100; Abcam Inc . , Cambridge , MA ) at room temperature for 2 hr , followed by the incubation with TRITC-conjugated anti-goat IgG ( 1:100 , Jackson ImmunoResearch Lab . , Inc . , West Grove , PA ) at room temperature for 1 hr before the mounting of anti-photobleaching reagent ( Vector Lab . , Inc . , Burlingame , CA ) . Imaging experiments were conducted with Ca2+ free Hanks balanced salt solution ( HBSS , Invitrogen ) containing 20 mM HEPES , 1 mM D-glucose , 0 . 5 mM EGTA , 1 mM MgCl2 , and 1 mM MgSO4 ( pH 7 . 4 ) . During imaging experiments , the solution was kept in streptomycin free condition to prevent a possible effect on mechanosensitive ion channels . The chemical reagents 2-Amino-ethoxydiphenyl borate ( 2-APB ) , Nifedipine , Thapsigargin ( TG ) , LaCl3 , GdCl3 , streptomycin , nocodazole , cytochalasin D , blebbistatin , and ML-7 were purchased from Sigma–Aldrich ( Sigma , St . Louis , MO ) . PP1 and LY294002 were commercially obtained from Calbiochem ( San Diego , CA ) . PF228 was obtained from Tocris Bioscience ( Ellisville , MO ) . The amount of drug administration was based on previous publications ( Kim et al . , 2009; Lu et al . , 2011; Seong et al . , 2011 ) . A fiber-coupled IR ( infra-red ) laser ( 1064 nm , 5W , 5 mm diameter , YLD-5-1064-LP , IPG Photonics ) was used for the experiment . We used a piezoelectric system for a steering mirror and the piezo-mirror system was designed with a closed loop , and an automated shutter ( LS6ZM2 , Uniblitz , Rochester , NY ) with a shutter controller ( VCM-D1 , Uniblitz ) . Mirrors ( designed for IR ) , lenses ( BK7 , plano-convex ) , and other basic optics were purchased from Thorlabs ( Newton , New Jersey ) . A hot mirror ( FM01 , wide band , Thorlabs ) was installed inside a microscope to block the IR scattering . The piezo-mirror ( S-334 . 2 , ( PI ) Physik instrumente ) was installed together with the computer interface module of ( E-516 . I3 , PI ) . The interface module set up with the other drivers ( E-503 . 00 and E-509 . S3 , PI ) makes it possible to control the piezo-mirror system by a computer . The laser beam passes through a laser-beam expander , a steering mirror , and a dichroic long-pass beam splitter to enter the microscope side port . Beads coated with fibronectin ( Fn; 50 µg/ml ) or BSA as the control were prepared as previously reported ( Wang et al . , 2006 ) . The size of the beads is 10 µm and the beads were incubated for 10–20 min to allow them to adhere to cell membrane surface . Single-beam gradient optical laser tweezers with controlled 300 pN of mechanical force were applied to pull the adhered beads . A similar optical trapping system has been described in our previous report ( Botvinick and Wang , 2007 ) . Cells expressing various exogenous proteins were starved with 0 . 5% FBS for 36–48 hr before imaging experiments . All images were obtained by using Zeiss Axiovert inverted microscope equipped with a charge-coupled device ( CCD ) camera ( Cascade 512B , Photometrics ) and a 420DF20 excitation filter , a 450DRLP dichroic mirror , and two emission filters controlled by a filter changer ( 480DF30 for ECFP and 535DF25 for YPet ) . Time lapse fluorescence images were acquired at 10 s interval by MetaFluor 6 . 2 software ( Universal Imaging , West Chester , PA ) . The emission ratio of YPet/ECFP was directly computed and generated by the MetaFluor software to represent the FRET efficiency before they were subjected to quantification and analysis by Excel ( Microsoft , Redmond , WA ) . The results were expressed as the mean standard error of the mean ( S . E . M ) . Statistical analysis of the data was performed by the unpaired Student's t-test to determine the statistical differences between the two mean values . The statistically significant level was determined by p < 0 . 05 .	Cells receive many signals from their environment , for example , when they are compressed or pulled about by neighboring cells . Information about these ‘mechanical stimuli’ can be transmitted within the cell to trigger changes in gene expression and cell behavior . When a cell receives a mechanical stimulus , it can activate the release of calcium ions from storage compartments within the cell , including from a compartment called the endoplasmic reticulum . Calcium ions can also enter the cell from outside via channels located in the membrane that surrounds the cell ( the plasma membrane ) . Kim et al . investigated how mechanical forces are transmitted in a type of human cell called mesenchymal stem cells using optical tweezers to apply a gentle force to the outside of a cell . These tweezers use a laser to attract tiny objects , in this case a bead attached to proteins in the cell's outer membrane . The cell's response to this mechanical stimulation was measured using a sensor protein that fluoresces a different color when it binds to calcium ions . With this set-up , Kim et al . found that mesenchymal stem cells are able to transmit mechanical forces to different depths within the cell . The forces can travel deep to trigger the release of calcium ions from the endoplasmic reticulum . This process involves a network of protein fibers that criss-cross to support the structure of a cell—called the cytoskeleton—and also requires proteins that are associated with the cytoskeleton to contract . However , calcium ion entry through the plasma membrane due to a mechanical force does not require these contractile proteins—only the cytoskeleton is involved . These results demonstrate that the transmission of mechanical signals to different depths within mesenchymal stem cells involves different components . Future work should shed light on how these mechanical signals control gene expression and the development of mesenchymal stem cells .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "physics", "of", "living", "systems" ]	2015	Distinct mechanisms regulating mechanical force-induced Ca2+ signals at the plasma membrane and the ER in human MSCs
This study focuses on computational and theoretical investigations of neuronal activity arising in the pre-Bötzinger complex ( pre-BötC ) , a medullary region generating the inspiratory phase of breathing in mammals . A progressive increase of neuronal excitability in medullary slices containing the pre-BötC produces mixed-mode oscillations ( MMOs ) characterized by large amplitude population bursts alternating with a series of small amplitude bursts . Using two different computational models , we demonstrate that MMOs emerge within a heterogeneous excitatory neural network because of progressive neuronal recruitment and synchronization . The MMO pattern depends on the distributed neuronal excitability , the density and weights of network interconnections , and the cellular properties underlying endogenous bursting . Critically , the latter should provide a reduction of spiking frequency within neuronal bursts with increasing burst frequency and a dependence of the after-burst recovery period on burst amplitude . Our study highlights a novel mechanism by which heterogeneity naturally leads to complex dynamics in rhythmic neuronal populations . Mixed-mode oscillations ( MMOs ) represent rhythmic behaviors of dynamical systems characterized by an alternation between large amplitude ( LA ) and small amplitude ( SA ) oscillations ( Desroches et al . , 2012 ) and have been observed in many physical , chemical , and biological systems , including a variety of neural structures . The latter include populations of neurons in the entorhinal cortex ( Dickson et al . , 2000; Yoshida and Alonso , 2007 ) , hippocampal neurons ( Winson , 1978 ) , dopaminergic neurons ( Medvedev et al . , 2003; Medvedev and Cisternas , 2004 ) , neurons of the medullary pre-Bötzinger complex ( Del Negro et al . , 2002c ) vibrissa motoneurons ( Golomb , 2014 ) and spinal motoneurons ( Iglesias et al . , 2011 ) in rats . Theoretical investigations of MMOs typically focus on the mechanisms by which MMOs emerge from a complex interplay of multiple distinct time scales in the nonlinear processes governing a system’s activity ( Desroches et al . , 2012 ) . In this work , we introduce and explain a novel alternative paradigm for the generation of MMOs . The key element in the mechanism that we present is that a network of coupled oscillators can generative repetitive MMOs based on heterogeneity within the network . The importance of this paradigm for neural systems relates to central pattern generators ( CPGs ) that can intrinsically generate rhythmic activity controlling different motor behaviors such as breathing and locomotion . Heterogeneity in the quantitative features of the neurons involved is likely a ubiquitous property of such circuits ( Butera et al . , 1999b; Marder , 2011; Buzsáki and Mizuseki , 2014 ) , and thus our work predicts that MMO patterns should be attainable in a wide range of brain structures with rhythmic activity depending on mechanisms for neuronal synchronization . Furthermore , predictions that follow from the existence of this MMO-generation mechanism should be of similarly widespread relevance . For concreteness , the present study focuses on computational models of a neuron population in a particular brain area , the pre-Bötzinger complex ( pre-BötC ) , where MMOs have been previously observed ( Del Negro et al . , 2002c ) . The pre-BötC is a medullary region representing an excitatory kernel circuit of the respiratory CPG in mammals that is critically involved in generating the inspiratory phase of respiration ( Smith et al . , 1991; Smith et al . , 2007; Smith et al . , 2009; Smith et al . , 2013 ) . The pre-BötC can generate rhythmic bursting activity in vitro , in medullary slices containing this structure ( Koshiya and Smith , 1999; Del Negro et al . , 2001; Rigatto et al . , 2001; Thoby-Brisson and Ramirez , 2001 ) and even in isolated 'islands' extracted from these slices ( Johnson et al . , 2001; Figure 1A ) . This rhythmic activity is typically induced by elevating the extracellular concentration of potassium ( [K+]out ) up to 7–9 mM , which putatively increases neuronal excitability ( Koshiya and Smith , 1999; Lieske et al . , 2000; Del Negro et al . , 2001; Johnson et al . , 2001; Thoby-Brisson and Ramirez , 2001 ) . Pre-BötC neurons , through a pre-motor population , project to the hypoglossal nuclei containing motor neurons , the activity of which can be recorded in rhythmically active slices from the hypoglossal ( XII ) nerve ( see Figure 1 , panels A , B , and C1 ) . Simultaneous optical recordings from individual neurons and XII output have shown that bursts in the XII root represent the synchronized activity of pre-BötC neurons ( Koshiya and Smith , 1999; Figure 1C1 , C2 ) and the amplitude of XII bursts clearly depends on the number of pre-BötC neurons involved . Interestingly , a progressive increase in [K+]out in slices containing the pre-BötC evokes complex population MMOs characterized by amplitude modulation , with large amplitude ( LA ) bursts alternating with a series of small amplitude ( SA ) bursts ( Koshiya and Smith , 1999; Del Negro et al . , 2002c; Kam et al . , 2013 ) ( see Figure 1A , bottom ) . An amplitude irregularity similar to the MMOs recorded from the pre-BötC in vitro has also been observed during acute intermittent hypoxia simulated in vitro ( Zanella et al . , 2014 ) . Similar pathological patterns of breathing have been observed in vivo in association with different diseases , such as myocardial infarcts , obstructive sleep apneas , apneas of prematurity , Rett syndrome , and sudden infant death syndrome ( Zanella et al . , 2014 ) . 10 . 7554/eLife . 13403 . 003Figure 1 . Mixed-mode and endogenous oscillations in the pre-Bötzinger complex in vitro . ( A ) Top: medullary slice showing 'pre-BötC island' ( shaded dark gray ) and labeled structures: XII , hypoglossal motor nucleus; NTS , nucleus tractus solitarius; SP 5 , spinal trigeminal tract . Bottom: Excised pre-BötC island with extracellular recording from the pre-BötC that demonstrates MMOs ( i . e . , interleaved large and small amplitude bursts ) . Modified from Johnson et al . ( 2001 ) . ( B ) Intracellular recording from pre-BötC neuron with baseline membrane potential of -54 mV ( top trace ) and -49 mV ( bottom trace ) . The corresponding integrated hypoglossal motor output ( ∫XII ) is shown below each neuronal recording . In the top trace , each neuronal burst coincided with the activity in the hypoglossal motor output . At the more depolarized baseline potential , bursting occurred at higher frequency and several ectopic bursts did not correspond to ∫XII output . ( C1 ) Optical recording from pre-BötC neuron activity ( Ca2+ imaging ) . Left: three inspiratory neurons ( 1–3 ) show synchronized Ca2+ activities ( ΔF/F ) and corresponding ∫XII output ( synchronization marked with dotted lines ) . Right: Application of CNQX ( 6-cyano-7-nitroquinoxaline-2 , 3-dione , blocking fast glutamatergic synaptic transmission , 50 µM ) caused a loss of bursting in ∫XII and neurons 1 and 2 showed desynchronized bursting activity ( see dotted lines ) . ( C2 ) Cross-correlograms for neurons 1 and 2 in C1 . The loss of a peak at 0 time lag after CNQX indicates loss of synchronization . B , C1 , and C2 were adapted from Koshiya and Smith ( 1999 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13403 . 003 To theoretically investigate the mechanisms underlying these MMOs , we developed and analyzed two models: ( a ) a computational model of a network of 100 neurons , described in the Hodgkin-Huxley style , with bursting properties defined by the persistent ( slowly inactivating ) sodium current ( INaP ) incorporated in each neuron , with sparse excitatory synaptic interconnections , and with randomly distributed neuronal parameters , and ( b ) a simplified model consisting of three mutually excitatory non-spiking neurons that allowed us to apply qualitative analytical methods for understanding key system behaviors . Our simulations and analysis suggest that neurons with low excitability , which generate low frequency bursting with high intra-burst spike frequency , recruit LA bursts by synchronizing the activity of many neurons in the network and therefore play a critical role in the generation of MMOs . Our simulations and analysis of these models provide important insights into how heterogeneity of neural excitability and other network features contribute to the generation of rhythmic activities in neuron populations that are key components of central pattern generators in vertebrates . Intracellular recordings from individual pre-BötC neurons in rhythmically active slices show a range of resting membrane potentials and other quantitative properties among individual neurons ( Del Negro et al . , 2001; Del Negro et al . , 2002a; Peña et al . , 2004; Koizumi and Smith , 2008 ) . Neurons with more negative resting membrane potentials usually generate bursting activity that is fully consistent with , and reflected in , XII output activity , whereas neurons with less negative resting membrane potentials demonstrate higher burst frequencies and often generate 'ectopic' busts not reflected in the XII output ( see example in Figure 1B ) . Pharmacological blockade of synaptic transmission within the pre-BötC by 6-cyano-7-nitroquinoxaline-2 , 3-dione ( CNQX ) results in a reduction and desynchronization of neuronal activity within the pre-BötC , with no activity in the hypoglossal output ( see example in Figure 1C1 , C2 ) . In light of these experimental findings , we modeled the pre-BötC as an excitatory network consisting of 100 neurons described in the Hodgkin-Huxley style , with sparse excitatory synaptic interconnections between neurons . The intrinsic bursting properties of these neurons were based on the persistent ( slowly inactivating ) sodium current , INaP ( Butera et al . , 1999a; Butera et al . , 1999b; Del Negro et al . , 2001; Rybak et al . , 2003a; Rybak et al . , 2003b; Rybak et al . , 2004; Rybak et al . , 2014; Dunmyre and Rubin , 2010; Jasinski et al . , 2013; see Materials and methods ) . To account for neuronal heterogeneity , we distributed the reversal potential of the leak current , EL , across the population ( see Materials and methods and Table 1 ) . We also included mild variability in the maximal conductance of INaP ( g¯NaP , Table 1 ) , with a range of values that allowed all neurons to be conditional bursters . In the absence of coupling ( when all weights of connections were set to zero ) , the population contained silent neurons , as well as neurons with bursting and tonic activities ( Figure 2A1 ) . Figure 2A2 presents the raster plot of neuronal activity in the same population , in which neurons were sorted in order of increasing ( from bottom to top ) excitability ( defined by the assigned EL ) . This figure shows that neurons with the most negative EL values were silent ( neurons with ID numbers from 1 to 49 ) , neurons with intermediate EL exhibited bursting activity with burst frequency increasing with EL ( neurons 50–94 ) , and neurons with greatest EL displayed tonic spiking ( neurons 95–100 ) . The lack of network interactions resulted in asynchronous neuronal activity and the corresponding integrated population histogram lacked phasic modulation ( Figure 2A3 ) . 10 . 7554/eLife . 13403 . 004Figure 2 . Distribution of neural excitability in a sparsely connected network causes mixed-mode oscillations . ( A1-A3 ) Network simulation when excitatory interactions between neurons were removed ( w=0 , 'uncoupled' network ) . ( B1-B3 ) Simulation of network activity with sparse excitatory synaptic interconnections at w=2 . 5 and p=0 . 15 . ( A1 , B1 ) Unsorted raster plots depicting the timing of action potentials in neurons with randomly distributed EL values . ( A2 , B2 ) Raster plots with neurons sorted by EL values such that the lowest Neuron ID was assigned to the neuron with the most negative EL . ( A3 , B3 ) Histogram of population activity . No phasic component is observed in A3 due to the desynchronized bursting in the uncoupled population . B3 shows a typical MMO pattern including LA bursts alternating with SA bursts of varying amplitudes . DOI: http://dx . doi . org/10 . 7554/eLife . 13403 . 004 The patterns of population activity and integrated output dramatically changed when relatively weak and sparse excitatory synaptic connections among neurons were incorporated in the model ( Figure 2B1–B3 ) . The raster plot of the same sorted neurons in this coupled case ( Figure 2B2 ) shows the presence of overlapping clusters ( sub-populations ) of neurons with synchronized bursting , which generate MMOs characterized by alternating LA and SA population bursts ( Figure 2B3 ) . Figure 3 shows another example of our simulations , including 'uncoupled' ( panel A1 ) and coupled ( panel B1 ) cases for sorted neurons of the same populations and the integrated population activity for the coupled case ( panel C1 ) . In both A2 and B2 panels we plotted the membrane potentials ( V ) of four selected representative neurons that in the uncoupled case exhibited ( bottom-up ) : silence ( trace 1 ) , bursting with low burst frequency ( trace 2 ) , bursting with higher burst frequency ( trace 3 ) , and tonic spiking ( trace 4 ) . Also in these figures , the time course of the INaP inactivation variable ( hNaP ) of each neuron , which defined the burst recovery period , was superimposed onto its V time course ( red trace ) . An important feature of all neurons operating in bursting mode is illustrated in Figure 3A3 ( uncoupled case ) : while the burst frequency ( blue curve ) increased with the neuronal excitability ( bottom-up ) , the spike frequency within the burst ( red curve ) changed in an inverse manner , i . e . , decreased with increasing neuronal excitability . This reduction of spike frequency within the bursts in more excited neurons limited their ability to synchronize and recruit other neurons' activity in the coupled case ( see below ) . 10 . 7554/eLife . 13403 . 005Figure 3 . Neurons with similar excitabilities activate in clusters within a heterogeneous network with sparse connectivity . ( A1–A3 ) Simulation results for uncoupled network , w=0 . ( A1 ) Sorted raster plot showing silent ( most negative EL , lowest Neuron IDs ) , bursting , and tonic ( least negative EL , highest Neuron IDs ) neurons . ( A2 ) Endogenous activity of four individual neurons ( Neuron IDs: 30 , 65 , 85 and 97 ) showing membrane potential ( black ) and inactivation , hNaP , of the persistent sodium current ( red ) . ( A3 ) Burst frequencies ( blue ) and intra-burst spike frequencies ( red ) were calculated for each neuron in the uncoupled case . Boundaries separating bursting from silent and tonic neurons are marked ( black , dot-dashed lines ) . ( B1–B3 ) Simulation results for w=1 . 8 and p=0 . 15 . ( B1 ) Raster plot sorted by EL ( Neuron ID ) . Two SA bursts ( blue rectangle ) and one LA burst ( red rectangle ) were selected for the inset in C2 . ( B2 ) Membrane potential ( black ) and hNaP ( red ) are shown for the four neurons originally selected in A2 . ( B3 ) Spike frequency of neurons sorted by excitability in the coupled ( w=1 . 8 ) case ( the dashed red curve shows spike frequencies for the uncoupled ( w=0 ) case in A3 , for comparison ) . ( C1 ) Histogram of population activity corresponding to B1 . ( C2 ) Insets depicting magnified raster plots from the selected bursts in B1 . Different color families were used to identify neurons that belong to different clusters , with each cluster defined as a group of neurons that participated in the same set of bursts . The clusters of neurons with lowest excitability ( LE ) , contributing to only LA bursts , were highlighted by red; on the other side of the 'color spectrum' , the neurons with the highest excitability , exhibiting sustained activity , were colored yellow . ( D , right ) The color-coding scheme from C2 was used in conjunction with a histogram depicting the number of active neurons within a 100 ms window . The vertical dot-dashed black line marks the time of onset of LE neuron activation in an LA burst and the horizontal dot-dashed line intersects this onset time to show the total number of neurons already active at the time of LE activation . The horizontal dot-dashed line is extended to the two SA bursts and demonstrates that LE activation failed despite the presence of a sufficient number of active neurons in the network . ( D , left ) Comparison of number of neurons active over time from two SA bursts ( purple and blue curves ) and one LA burst ( red curve ) . The intersection of the two dashed , black lines compares the SA and LA burst amplitudes when the LE neurons ( red bars ) first start to activate in an LA burst . DOI: http://dx . doi . org/10 . 7554/eLife . 13403 . 005 The pattern of population activity in a coupled network is shown in Figure 3B1–B3 , C1 . Several clusters of neurons with synchronous bursting activity emerged dynamically in the population . Clusters differed by the number of the population bursts in which they participated ( panel B1 ) , which in turn defined the amplitude of integrated population bursts ( panel C1 ) . The same panels also show that several relatively small , distinct or partly overlapping clusters with synchronous bursts were formed by neurons with relatively high ( less negative ) EL . These clusters generated a series of high-frequency SA bursts . Generation of low-frequency LA bursts involved synchronization of many neurons and included those with low excitability ( most negative EL ) ( Figure 3B1 , B2 , C1 ) . Figure 3C2 shows two insets from the raster plot in Figure 3B1 that correspond to two SA bursts ( left ) and one LA burst ( right ) . The neuronal clusters in these insets are colored as follows: spikes of neurons with default tonic spiking - yellow; spikes of neurons involved in SA bursts - light and dark green , light and dark blue , and purple , arranged in order of increasing excitability; spikes of neurons involved only in LA bursts - red . The left inset ( within the blue rectangle ) in Figures 3C2 and 3D depicts spikes in the raster plot corresponding to two SA bursts . Two clusters of high excitability neurons , colored by yellow and green , participated in both of these bursts . In addition , the blue cluster participated in the first , but not the second , SA burst , and a purple cluster participated in the second , but not the first , SA burst . The neurons belonging to the red cluster were only active during LA bursts ( see right inset within the red rectangle ) . To evaluate the role of different clusters in SA and LA bursts in both insets , we built integrated histograms showing the number of neurons , from each colored cluster , that were active within a 100 ms bin ( Figure 3D ) . Note that the sub-population of low excitability neurons , colored red , do not contribute to SA bursts . Activation of this sub-population during the LA burst is marked by a black , dot-dashed , vertical line at about 32 . 7 s . This vertical line intersects with a black , dashed , horizontal line indicating a threshold for the activation of red neurons . This line intersects the two SA bursts demonstrating that , although the amplitudes of both SA bursts rose above the marked threshold for activation of the red sub-population in the LA burst , the latter neurons were not recruited in SA bursts ( note the absence of the red neuron cluster in SA bursts ) and hence the full LA burst did not develop . We further find that the sub-population of neurons with low excitability ( colored red ) cannot be recruited by other sub-populations ( participating in SA bursts ) , and hence cannot generate LA bursts , until sufficient recovery of bursting capability in the low excitability neurons ( defined by the INaP inactivation variable hNaP ) has occurred . This observation suggests that with fixed parameter values , even though the low excitability neurons do not burst when uncoupled , the generation of LA bursts and the durations of their interburst intervals ( IBIs ) are mostly defined by the operation of an intrinsic burst-supporting mechanism in the less excitable neurons , rather than by variations in the intensity of their recruitment by the activity of highly excitable neurons involved in SA bursts . To study the dependence of MMOs on neuronal interactions within the network , we observed changes in the network activity when the weights and/or probability of synaptic connections were varied across simulations . Figure 4A1 , A2 , A3 shows three heat maps that demonstrate quantal changes in MMO regimes defined by ratios of LA to SA bursts ( e . g . 1:5 , 1:4 , etc . ) as several key parameters were varied ( Figure 4A1 , A2 , A3 ) . When either weights ( Figure 4A1 , A2 , B1 ) or probability of connections ( Figure 4A1 , A3 , B2 ) were increased , the frequency of LA bursts increased and the number of SA bursts between successive LA bursts decreased . This corresponded to a progressive change in the quantal state of the network toward regimes with high LA to SA burst ratios . 10 . 7554/eLife . 13403 . 006Figure 4 . Parameter dependence of mixed-mode oscillations . ( A1–A3 ) Heat maps depicting quantal changes in the ratio of LA to SA bursts , representing quantal MMO regimes calculated with variation of the connection weights ( w ) , probability of connections ( p ) , and maximal conductance of the persistent sodium channel ( g¯NaP ) . In A1 , w and p were iteratively varied at g¯NaP= 5 nS . In A2 , w and g¯NaP were varied at p=0 . 24 , and in A3 , p and g¯NaP were varied at w=3 . ( B1–B3 ) Histograms of population activity ( spikes/10ms ) were calculated as a parameter of interest was varied . In B1 , w was varied between 1 . 0 and 4 . 5 at p=0 . 15 and g¯NaP= 5 nS; these changes correspond to the horizontal red , dashed line in A1 . Progressive increase of w caused the frequency of LA bursts to increase and the number of SA bursts between LA bursts to decrease . In B2 , p was varied from 0 . 09 to 0 . 4 at w=1 . 8 and g¯NaP= 5 nS; these changes correspond to the vertical blue , dashed line in A1 . Similarly to changes of w , increasing p caused an increase in frequency of LA bursts and decrease in the number of SA bursts between LA bursts . In B3 , g¯NaP was decreased from 5 . 0 to 3 . 0 nS , with fixed values w=3 . 0 and p=0 . 24 , corresponding to the black , dashed lines in A2 and A3 , respectively . This progressive decrease caused a decline in LA burst frequency , and an emergence of SA bursts , until all network activity stopped at g¯NaP=3 . 0 nS . DOI: http://dx . doi . org/10 . 7554/eLife . 13403 . 006 Figure 4B1 shows regimes observed when the probability of connections was fixed ( p=0 . 15 ) and only the weights of connections were varied . At the lowest weights ( w = 1 . 0 ) , only irregular SA bursts were observed because of insufficient neuronal synchronization ( top trace , Figure 4B1 ) . Weights between 1 . 0 and 1 . 8 caused regimes characterized by low-frequency irregular LA bursts with irregular patterns of SA bursts ( not shown ) . At a weight of 1 . 8 , each LA burst emerged regularly following five SA bursts ( second trace ) ; no parameter sets produced stable regimes with more than five SA bursts per one LA burst . Further increases in weights caused a quantal increase of LA frequency and the corresponding reduction in the number of SA bursts between LA bursts ( traces 2–4 ) , until strong enough weights yielded LA bursts only ( trace 5 ) . A similar trend is seen in Figure 4B2 with increases in the probability of connections at a fixed value of synaptic weights ( w = 1 . 8 ) . Overall , for fixed connection weights , the availability of INaP in low excitability neurons still selects the cycles on which LA bursts occur during MMOs . Furthermore , our simulations showed increased IBIs following the LA bursts , relative to IBIs observed after SA bursts , in all instances of MMOs ( Figures 2B3 , 3C1 , and 4B1 , B2 ) . In the next section , Reduced model analysis of interburst intervals ( IBIs ) , we use a reduced model to explain these effects . Finally , to study the dependence of MMOs on INaP , we varied the average maximal conductance for INaP ( g¯NaP ) and either weights ( Figure 4A2 ) or probability of connections ( Figure 4A3 ) . The resulting heat maps show a qualitatively similar pattern where the ratio of LA to SA bursts decreases as g¯NaP is reduced . Activity traces corresponding to g¯NaP changes at fixed weights and probability of connections are shown in Figure 4B3 ( w = 3 . 0 , p=0 . 24 ) . At the typical value of g¯NaP ( 5 . 0 nS ) , network activity consisted entirely of LA bursts ( Figure 4B3 , top trace ) . When g¯NaP was reduced , a decrease in LA burst frequency and an increase in SA burst count between LA bursts were observed ( traces 2–4 ) until busting fully stopped at g¯NaP= 3 . 2 nS ( trace 5 ) . Thus , while raising the weights or probability of synaptic connections can enhance the rate of LA burst generation in some parameter regimes , if there is insufficient availability of burst-supporting current , then the recruitment of low excitability neurons is precluded . A reduced model was developed to allow qualitative mathematical analysis of the MMOs that we observed . The model consisted of three neurons with mutual excitatory synaptic interactions ( see Figure 5A1 ) . It was considered that each model neuron represented a sub-population of spiking neurons with a particular level of excitability . Each neuron was described using a non-spiking , activity-based model ( Rubin et al . , 2009b; Rubin et al . , 2011; Molkov et al . , 2015; see Materials and methods ) . The behavior of each neuron was defined by two dynamical variables , the membrane voltage , V , and INaP inactivation , hNaP . For each neuron we calculated a nonlinear output function , f ( V ) , which approximated the aggregate activity of a cluster of neurons in the original 100-neuron model . EL values were distributed such that in the absence of coupling , neuron 1 ( high excitability , HE ) engaged in high frequency bursting , neuron 2 ( moderate excitability , ME ) engaged in low frequency bursting with no special frequency relation to the bursting of the HE neuron , and neuron 3 ( low excitability , LE ) was silent; the three neurons’ summed activity provided a representation of network output ( Figure 5B1 , C1 , D1 , E1 ) . For each simulation , in addition to voltage and summed activity time courses , we visualized the network trajectory as it evolved in ( hNaP1 , V1 , V3 ) -space . Without coupling , this trajectory was cyclic , corresponding to the oscillations of the HE neuron 1 ( i . e . , of hNaP1 , V1 ) without changes in V3 ( Figure 5B2 ) . 10 . 7554/eLife . 13403 . 007Figure 5 . Reproduction and analysis of mixed-mode oscillations in a reduced model . ( A1 ) Schematic of reduced model with mutual excitatory connections between all neurons . Indices correspond to ( 1 ) high-excitability ( HE ) , ( 2 ) medium-excitability ( ME ) , and ( 3 ) low-excitability ( LE ) neurons . ( A2 ) 1:4 regime represented in toroidal state space ( product of two cyclical variables ) . Four rotations around the larger cycle , corresponding to SA bursts , occur during a single rotation in the smaller cycle , corresponding to an LA burst . Adapted from Rubin et al . , 2011 . ( B1–B3 ) Simulation results when w=0 . ( B1 ) Output activity , f ( Vi ) , was calculated for each neuron . The 'Sum' trace depicts aggregate network output and is asynchronous when w=0 ( uncoupled network ) . ( B2 ) A trajectory ( red trace ) in the ( hNaP1 , V1 , V3 ) -plane depicts endogenous HE oscillations ( cyclical movement in the ( hNaP1 , V1 ) -plane ) , and a silent LE neuron ( no movement in V3 ) . ( B3 ) In the ( hNaP , V ) -plane an endogenously bursting neuron’s trajectory ( red trace ) travels around the local minima and maxima of a V-nullcline ( blue curve ) that intersects the hNaP-nullcline ( black , dotted curve ) . A band of V-nullclines was calculated for the range of EL ∈ [-59 . 0 , -53 . 8] mV where endogenous bursting occurred ( gray band ) . EL values above and below this range caused tonic activity and silence , respectively . ( C1–C3 ) Simulation results when w=2 . ( C1 ) Output activity showed a pattern of three SA bursts between two LA bursts ( 1:4 quantal regime ) . LA bursts occurred when all three neurons were active , low amplitude SA bursts occurred when only the HE neuron was active , and higher amplitude SA bursts occurred when both HE and ME neurons were synchronously active . ( C2 ) The system’s trajectory ( red curve ) projected into ( hNaP1 , V1 , V3 ) . Four rotations in ( hNaP1 , V1 ) occurred along with only a single rotation in ( V1 , V3 ) , denoting an LA burst . ( C3 ) The LE neuron’s trajectory ( red curve ) is projected into the ( V3 , hNaP3 ) -plane . The hNaP3-nullcline ( black , dotted curve ) intersects three V3-nullclines: the black nullcline curve corresponds to LE neuron's resting state ( no excitatory input ) , and the blue and green nullcline curves correspond to excitatory inputs from the HE neuron and both HE and ME neurons , respectively . The LE neuron receives four inputs , marked ( i ) - ( iv ) , while at rest . Only input ( iv ) results in a successful LE activation , and therefore an LA burst . ( D1–D3 ) Simulation results when w=3 . ( D1 ) Two SA bursts occurred between pairs of LA bursts ( 1:2 quantal regime ) . ( D2 ) In ( hNaP1 , V1 , V3 ) the trajectory makes two rotations in ( hNaP1 , V1 ) during one rotation in ( V1 , V3 ) . ( D3 ) In ( V3 , hNaP3 ) , the LE neuron receives two excitatory inputs , at points marked ( i ) and ( ii ) . Nullcline colors are consistent with B3 . ( E1–E3 ) Simulation results when w=4 . ( E1 ) Only LA bursts were observed ( 1:1 quantal regime ) . ( E2 ) In ( hNaP1 , V1 , V3 ) , one rotation occurs in ( hNaP1 , V1 ) for each rotation in ( V1 , V3 ) . ( E3 ) The LE trajectory is projected into ( V3 , hNap3 ) for the 1:1 regime . The LE neuron activates when it receives an excitatory input from the other neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 13403 . 007 In subsequent simulations , neurons in this model interacted through excitatory synaptic interconnections with the weights of connections increasing top-down in Figure 5 from panels B1-B3 to panels E1-E3 . Similarly to the previous model , when connection weights were progressively increased , the network underwent a series of regime transitions progressing from only SA bursts ( Figure 5B1 , B2 ) to only LA bursts ( Figure 5E1 , E2 ) . The intermediate regimes ( Figures 5C1 , C2 and 5D1 , D2 ) are referred to as 'quantal' and labeled as 1:N regimes if there were N-1 SA bursts between each pair of LA bursts; these correspond to the MMOs in the 100-neuron model described above . The periods of oscillations were calculated for all neurons as weights of connections were gradually increased ( Figure 6A ) , and these clearly distinguished the different quantal states observed . As in the previous model , LA bursts involved activation of all neurons and occurred exactly on the cycles when the LE neuron activated ( Figure 5C1 , C2 , D1 , D2 , E1 , E2 ) . 10 . 7554/eLife . 13403 . 008Figure 6 . Emergence of quantal regimes and analysis of interburst intervals . ( A ) The burst period of each neuron was continuously calculated as the connection weights ( w ) were increased and neuronal periods on each cycle were plotted . LA bursts occurred at w>1 . 4 ( LE emergence , red dots ) . The quantal regime was determined by the ratio of LE and HE periods . Transitions between stable regimes , i . e . bifurcations , occurred when the LE period 'jumped' to progressively lower integer ratios of the HE period . The ME and HE neurons had longer periods following LA bursts than SA bursts . This phenomenon creates multiple branches in the ME and HE periods for a given quantal regime ( see the pair of HE period branches at w=3 in the 1:2 quantal regime , for example ) . ( B ) The HE neuron's trajectory ( red curve ) is projected into the ( hNaP1 , V1 ) -plane when w=3 . 0 ( 1:2 regime ) . Distinct oscillations arise in the HE neuron's trajectory for SA and LA bursts . The black V1-nullcline governs HE activity when it is endogenously bursting during an SA burst . The green V1-nullclines govern HE activity during network-wide activation ( LA burst ) and are depicted as a band because of the progressive decay of output from LE and ME neurons ( resulting from the decrease in f ( V ) as their voltages decreased , see Equation 14 , following LA burst onset . ( C ) Projection of 1:1 trajectory ( grey , w =3 . 4 ) and trajectory at the transition to the 1:2 regime ( blue , w=3 . 2 ) to the ( input2 , hNaP2 ) plane . The latter exhibits a tangency to the curve of knees ( black dashed ) of the V2-nullcline , where it fails to activate and thus 1:1 regime is lost . ( D ) Projection of 1:2 trajectory ( grey , w =2 . 4 ) and trajectory at the transition to the 1:4 regime ( blue , w =2 . 1 ) to the ( input3 , hNaP3 ) plane . The latter exhibits a tangency to the curve of knees ( black dashed ) of the V3-nullcline , where it fails to activate and thus 1:2 regime is lost . The curve of fixed points , where the V3-nullcline and h3-nullcline intersect , is also shown ( black dotted ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13403 . 008 The reduced model provided an explanation for the emergence of quantal MMOs . A key point was that for each neuron , when it was silent , there was a level of synaptic input that caused its activation . This level depended on the degree of INaP deinactivation in the neuron , quantified by hNaP , as well as on its excitability . When one neuron was activated , it excited the other two neurons , and each of these could be activated if and only if the input it received was sufficiently large ( cf . Rubin and Terman , 2002 ) . For the LE neuron , there were therefore discrete windows of opportunity for activation , corresponding to activation times of the other neurons . This idea can be visualized by considering the trajectory of the full system projected to the ( V3 , hNaP3 ) -plane ( Figure 5B3 , C3 , D3 , E3; see Materials and methods , Time-scale decomposition in the reduced model ) . When the LE neuron is not active , the trajectory evolves along the left branch of the cubic V3-nullcline , corresponding to low V3 . The LE neuron is activated if the trajectory rises above the left knee , or local maximum , of the V3-nullcline ( analogously to the sample trajectory in Figure 5B3 ) . Incoming synaptic excitation lowers the V3-nullcline ( Figure 5C3 , D3 , E3 ) , a well-known effect known as fast threshold modulation ( Somers and Kopell , 1993 ) ; the amount of lowering depends on the input strength . In Figure 5C3 , three V3-nullclines are shown: black corresponds to no input , blue to input from the HE neuron only , and green to input from the HE and ME neurons . If a synaptic input lowers the left knee below the current value of hNaP3 , then the LE neuron is activated ( e . g . , Figure 5C3 , marked with 'iv' ) . Therefore , the activation of the LE neuron depends on the recovery of hNaP3 when input arrives , and hence on the rate of recovery of hNaP3 relative to the frequency of input arrival . For example , in Figure 5C3 , an SA burst involving only the HE neuron occurs when the trajectory is at position 'i' . Since the trajectory is below the knee of the blue nullcline , the LE neuron does not activate . An SA burst involving the HE and ME neurons occurs when the trajectory is at 'ii' . Again , LE neuron activation fails , because the trajectory is below the knee of the green nullcline . A failure similar to the first occurs at 'iii' . Finally , when the HE and ME neurons activate with the trajectory at 'iv' , the green nullcline becomes relevant , the trajectory is above the knee , and the LE neuron activates , yielding an LA burst . When synaptic weights were increased , the correspondingly larger excitatory input moved the V3-nullcline to lower hNaP3 values , allowing activation of the LE neuron with less recovery time ( increase of hNaP3 ) and hence with fewer input cycles . Figure 5D3 shows one SA burst without LE neuron activation ( 'i' ) and one cycle with LE neuron activation ( 'ii' ) , while in Figure 5E3 , the LE neuron can activate the first time it receives excitation . In all cases , a discrete number of activations of the HE and ME neurons is needed before hNaP3 recovers to a level from which the LE neuron can activate ( Figure 5C3 , D3 , E3 ) , which gives rise to the quantal nature of the MMO patterns ( Figure 6A ) . The same idea , that activation of a neuron on a specific cycle depends on whether it rises above the knee corresponding to the input it receives , can be used to pinpoint the events associated with transitions between regimes as shown in Figure 6A . Since the hNaP value of a knee for a neuron depends on the input level to that neuron ( Figure 5C3 , D3 , E3 ) , a curve of knees for the i-th neuron can be drawn in the ( inputi , hNaPi ) plane ( see also Materials and methods ) . Critical connection weights that separate regimes correspond to tangencies to such curves . For example , the transition from 1:1 to 1:2 regimes as w is decreased occurs when the ME neuron no longer can activate on every cycle ( Figure 5D1 ) . At the transitional weights , the trajectory , projected to the ( input2 , hNaP2 ) plane , exhibits a tangency to the curve of knees for the ME neuron ( Figure 6C ) . Similarly , the next transition , from 1:2 to 1:4 , occurs when the LE neuron no longer can activate on every second cycle . Thus , at the weights for this transition , the projection of the trajectory to the ( input3 , hNaP3 ) plane exhibits a tangency to the curve of knees for the LE neuron ( Figure 6D ) . In contrast to changes in connection weights , a change in the excitability of the LE neuron alone could alter the V3-nullclines ( for all input levels ) and hence change the frequency of the LA cycles within each quantal MMO rhythm without any change in the overall oscillation frequency of the 3-neuron population ( data not shown ) . On the other hand , an increase in the excitability of the HE neuron alone caused an increase of the SA burst frequency . Since the time between SA cycles became shorter , there was less recovery of the LE neuron per cycle , such that more SA cycles occurred between LA cycles and the overall LA frequency remained approximately constant ( data not shown ) . Another feature of the MMOs observed in our large-scale model is that IBIs were longer after LA bursts than after SA bursts . This property was seen in the reduced model as well ( Figure 5C1 , D1 , and see the multiple values of the period for the HE neuron within each quantal regime in Figure 6A ) . The reduced framework elucidates the mechanism underlying this feature . When some neurons are activated , the active neurons excite each other . Each active neuron’s variables evolve along the right branch of its V-nullcline , and activation ends when they reach the right knee , or local minimum , of this nullcline ( see Figure 5B3 , red trace ) . Stronger excitation pushes a neuron’s V-nullcline , including its right knee , to lower hNaP values and hence causes the active phase to end with more INaP inactivation ( i . e . , lower hNaP-coordinate ) . Thus , a longer recovery period is needed before subsequent activation of the leading neuron . On LA cycles , all neurons excite each other , which causes a maximal lowering of V-nullclines and subsequently yields the longest IBIs . The difference in post-burst recovery times is evident in the HE neuron’s trajectory when the 1:2 regime is simulated ( w=3 . 0 , see Figure 6B ) . The different-size loops shown in ( V1 , hNaP1 ) correspond to SA and LA bursts , respectively , and therefore have different maximal V1 and minimal hNaP1 values , defined by positions of the V1-nullcline during HE activation . The SA bursts occur due to the HE neuron’s intrinsic rhythmicity . When the ME and LE neurons excite the HE , the V1-nullcline moves to lower hNaP1 and V1 values ( lowest green nullcline , Figure 6B ) . This movement extends the active phase by pushing the right knee of the V1-nullcline down . As ME and LE neuron activity adapts , excitation gradually decreases ( green band , Figure 6B ) but nonetheless , when excitation from ME and LE neurons is removed , the HE neuron returns to the left branch of the V1-nullcline at much lower hNaP1 values than following an SA burst . Therefore , the time it takes the HE neuron to recover , following an LA burst , is longer than the recovery following an SA burst . To investigate the dependence of MMO regimes on excitability ( EL ) we proportionally reduced excitability in all neurons . Quiescence could be induced in the LE and ME neurons after decreasing all excitabilities by 8% ( Figure 7A ) . The frequency of the HE neuron decreased , and this produced low frequency SA bursts with no LA bursts . A similar regime of only SA bursts could be produced by decreasing weights of neuronal interconnections ( Figure 7B ) . In the example shown , both the HE and ME neurons participated in the SA bursts . No change occurred in the frequency of the HE and ME neurons ( Figure 7B ) . 10 . 7554/eLife . 13403 . 009Figure 7 . Modulation of excitability and connection weights alters reduced model activity pattern . ( A ) Output levels , f ( Vi ) , for all three neurons with w=1 . 7 . A transient decrease of 10% in EL was implemented between 4 and 8 s ( blue shaded region ) , causing the 1:5 quantal regime to transition to a regime with only HE active . ( B ) Output levels , f ( Vi ) , for all three neurons with w=2 . 0 , producing the 1:4 quantal regime . A transient reduction of w by 50% between 4 and 8 s ( blue shaded region ) caused a loss of LA bursts . Resulting SA bursts featured activation of HE alone ( lower amplitude SA bursts ) or synchronized HE and ME activity ( higher amplitude SA bursts ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13403 . 009 The phase diagram in Figure 5B3 can be used to explain the effects of reduction in neuronal excitability and connection weights . Changing excitability moved the V-nullclines corresponding to the unexcited , or resting , state of a neuron . For an uncoupled neuron , increasing EL caused progressive transitions from silence , to bursting , to tonic behavior . The transitions between these behaviors occurred when the fixed point ( intersection of the neuron’s V- and hNaP-nullclines ) moved from the V-nullcline’s left branch ( silence ) , to its middle branch ( bursting ) , to its right branch ( tonic ) . When excitability was decreased in a coupled network ( Figure 7A ) , the fixed points of the ME neuron moved to the left branch of the V-nullcline ( the LE neuron’s fixed point was already on the left branch , corresponding to the quiescence of the LE neuron in the uncoupled case , see Figure 5B1 ) . This decreased excitability increased the amplitude of excitation required to induce bursting in these neurons , and thus the low amplitude HE neuron’s phasic excitation was insufficient . When synaptic weights were changed ( Figure 5B3 , C3 , E3 ) only the V-nullclines corresponding to the presence of phasic excitation ( from other neurons in the network ) were altered . Thus , the intrinsic dynamics of each neuron stayed the same under changes in weights , such that the HE and ME neurons both remained able to activate . With decreased synaptic weights , however , we again found that synaptic excitation could no longer recruit the LE neuron ( cf . Analysis of the quantal nature of MMOs with the reduced model ) . We have presented and explored a novel , network-based mechanism for the emergence of MMOs , featuring repetitive alternations of SA and LA bursts of activity , in a heterogeneous population of neurons coupled via sparse excitatory synaptic interactions . In this form of MMOs , the time intervals between bursts are on a similar time scale regardless of whether an SA or an LA burst has just occurred , yielding quantal patterns of SA and LA events , although precise IBI durations actually depend on the amplitude of preceding bursts , and hence IBIs following LA bursts are longer than those following SA bursts ( Figure 6A ) . These MMOs appear to be a natural , perhaps inevitable , behavior of heterogeneous neural networks with excitatory coupling that can be expected to emerge widely in the nervous system , in which the rate of recovery of high excitability neurons dictates the period of subsequent events , while the recovery of low excitability neurons determines which subsequent events become LA bursts . LA bursts correspond to synchronous activation of most neurons in the network and occur when the least excitable neurons in the network can be recruited . Furthermore , feedback from these least excitable to the more excitable neurons is essential for synchronizing the network during LA bursts . The substantial mathematical analysis of synchronization and phase relations in small neural networks with polyrhythmic or multiphase activity has been performed previously without an explicit connection to MMOs ( e . g . , Shilnikov et al . , 2008; Rubin and Terman , 2012; Schwabedal et al . , 2014 ) . MMOs have been reported in a variety of neural systems ( Winson , 1978; Dickson et al . , 2000; Medvedev et al . , 2003; Medvedev and Cisternas , 2004; Yoshida and Alonso , 2007; Iglesias et al . , 2011; Golomb , 2014 ) . The computational and mathematical analysis of these patterns has largely focused on mechanisms that emerge from the separation of time scales typically found within neural dynamics , between voltages and fast gating and synaptic kinetics on one hand and slower gating , synaptic , and ionic concentration kinetics on the other . Within the corresponding MMOs , SA oscillations occur during a delayed transition between two different attractors for the fast dynamics and are often relatively high frequency events that emerge after a quiescent period , whereas the actual transitions between attractors yield LA events ( Desroches et al . , 2012 ) . Our present work deals with a very different form of MMOs where different oscillation amplitudes correspond to the participation of different numbers of neurons from within a network . In these MMOs , even within SA events , there is a complete transition between different attracting states ( hyperpolarized and depolarized ) for the fast voltage dynamics , although only some variables in the network are involved in this transition . The MMOs that we studied here depend critically on the synaptic interactions leading to the emergence of neuronal clusters with synchronous bursting activity , whereas the other described classes of MMOs mainly arise from intrinsic dynamics even in single neurons . Therefore , we observed a transition through a range of quantal MMO regimes as synaptic parameters were varied ( Figures 4A , B , 5 , 6 ) . Furthermore , LA bursts are gained , as parameters are varied , by conversion of particular SA bursts , arising roughly evenly between pairs of LA bursts , into LA bursts ( reminiscent of period-doubling ) , whereas in time-scale-based MMOs , transitions involve the less radical loss or gain of individual SA oscillations occurring just before each LA burst . The previous analyses closely related to this novel form of MMO were presented in two earlier papers , both motivated by the pre-BötC in the respiratory brainstem . In one study , synchrony could emerge in a group of modeled neurons with heterogeneous excitability , coupled with synaptic excitation ( Rubin and Terman , 2002 ) . It was noted that , starting in a 1:1 regime , weakening synaptic strengths could cause less excitable neurons to skip some cycles . In the other previous work , the reduced neuron models were used to investigate quantal recruitment of normally-silent late-expiratory neurons under hypercapnia ( Rubin et al . , 2011 ) . However , the model was not a heterogeneous excitatory network but rather consisted of several distinct neuronal populations coupled with a combination of excitation and inhibition , and the quantal effects observed involved only the single expiratory population , without any clustering or other alterations in other neurons’ behaviors . The other previous study closely related to the present work focused on the dynamic cycle-by-cycle variability in the assembly of neurons contributing to population bursts in the pre-BötC ( Carroll and Ramirez , 2013 ) . The authors extended previous deterministic models ( Butera et al . , 1999b; Rybak et al . , 2003b; Rybak et al . , 2004 ) by incorporating stochastic drive to all neurons and random , sparse neuronal interconnections . This model could qualitatively reproduce the patterns seen in spike rasters from in vitro records . The authors demonstrated the importance of sparse connections in these networks and showed that intrinsically bursting neurons within a sparse network topology play a stochastic , dynamic , and flexible role in the assembly of respiratory rhythms on a cycle-by-cycle basis , which is consistent with our present study . Despite many years of studies , the exact cellular mechanisms ( and ionic currents ) responsible for rhythmic bursting in the pre-BötC in vitro remain poorly understood and represent a subject of ongoing debate in the literature ( Thoby-Brisson and Ramirez , 2001; Del Negro et al . , 2002b , Del Negro et al . , 2005; Peña et al . , 2004; Pace et al . , 2007; Koizumi and Smith , 2008; Krey et al . , 2010; Dunmyre and Rubin , 2010; Beltran-Parrazal et al . , 2012; Ben-Mabrouk et al . , 2012; Jasinski et al . , 2013; Kam et al . , 2013; Feldman and Kam , 2015; Rybak et al . , 2014; Rubin et al . , 2009a ) . There are many ionic currents that can be present in pre-BötC neurons and can potentially be involved in population activity . These currents include the persistent ( slowly inactivating ) sodium current , INaP , ( Butera et al . , 1999a; Butera et al . , 1999b; Del Negro et al . , 2001; Del Negro et al . , 2002a; Rybak et al . , 2003a; Rybak et al . , 2003b; Rybak et al . , 2004; Koizumi and Smith , 2008 ) , a calcium-activated , non-specific cation current , ICAN , and various Ca+ currents ( Thoby-Brisson and Ramirez , 2001; Peña et al . , 2004; Del Negro et al . , 2005; Pace et al . , 2007 ) , a transient potassium current , IA ( Hayes et al . , 2008 ) , and Ih ( Picardo et al . , 2013 ) . INaP and ICAN have been considered to be the main candidates for currents that are critically involved in pre-BötC bursting . INaP has been found in pre-BötC neurons and the rhythmic bursting activity in the pre-BötC could be abolished by pharmacological blockade of this current ( Del Negro et al . , 2002a; Rybak et al . , 2003a; Rybak et al . , 2003b; Hayes et al . , 2008 ) but its critical role in the pre-BötC bursting has been debated ( Del Negro et al . , 2002b ) . In turn , a series of recent studies of ICAN-based bursting in the pre-BötC ( Peña et al . , 2004; Krey et al . , 2010; Beltran-Parrazal et al . , 2012; Ben-Mabrouk et al . , 2012 ) also produced inconsistent results . Del Negro et al . , 2005 suggested that INaP may be the primary rhythm-generating current up to postnatal day 4 or 5 ( P4 or P5 ) , after which ICAN is also expressed and strongly contributes to rhythm generation . However , all currently available data on MMOs , including the early data ( Koshiya and Smith , 1999; Johnson et al . , 2001 ) illustrated here ( see Figure 1 ) and the recent data presented by Kam and Feldman ( Kam et al . , 2013; Feldman and Kam , 2015 ) , were obtained , respectively , in slices from the neonatal animals of P0-P2 , P0-P3 , and P0-P5 , i . e . within the developmental range in which INaP is considered to be the primary rhythm-generating current , supporting the inclusion of INaP in our models . In the present work , we studied MMOs in a large-scale neuron population consisting of 100 neurons modeled in the Hodgkin-Huxley style , which were coupled through sparse excitatory synaptic connections . All neurons in the model were capable of endogenous generation of rhythmic bursting activity ( Figure 3A2 ) within a particular range of excitability ( their resting membrane potential , defined by EL; see Figures 2A1 , A2 and 3A1 , A2 ) . Following the previous computational models of pre-BötC neurons ( Butera et al . , 1999a; Butera et al . , 1999b; Rybak et al . , 2003a; Rybak et al . , 2003b; Rybak et al . , 2004; Dunmyre and Rubin , 2010; Jasinski et al . , 2013 ) , the INaP inactivation variable , hNaP , evolved with a large time constant and its slow dynamics defined a slow neuronal 'recovery' , i . e . , gradual depolarization in the post-activity phase ( red traces in Figure 3A2 ) . The reversal potential of the leak current ( EL ) was randomly distributed across neurons in the network to provide a range of excitabilities and subsequent behaviors . This combination of distributed neuronal excitability with slow voltage-dependent recovery provided two important characteristics of neurons within the population: These two key features of the large-scale model were preserved in our reduced model , in which the spike frequency within the burst was explicitly represented by the amplitude of neuronal output . Therefore , this amplitude decreased with the increasing neuronal excitability ( from LE to HE neurons ) , and the slow recovery of LE neurons ( defined by the voltage-dependent time constant for hNaP3 ) , was greater than the recovery of HE neurons , and prevented the LE neuron from participation in higher frequency synchronized bursts ( Figure 5 ) . A limitation of our study is that we did not consider burst-generating currents other than INaP . However , although these key features in both of our models result directly from INaP kinetics , they actually are not specific to the INaP-dependent bursting mechanism analyzed herein . Instead , they represent a common feature of most known cellular bursting mechanisms , in which the post-burst recovery time depends on the neuronal activity within the bursts and vice versa . For example , in the case of intrinsic bursting mechanisms based on Ca2+-dependent potassium ( IK ( Ca2+ ) ) , Ca2+-activated nonspecific ( ICAN ) , or Na+-dependent potassium ( IK ( Na+ ) ) currents , involving intracellular accumulation of Ca2+ or Na+ ions , a functionally similar slow recovery is usually connected with operation of either the Ca2+ or Na+/K+ pumps ( Ekeberg et al . , 1991; el Manira et al . , 1994; Wallén et al . , 2007; Rubin et al . , 2009a; Ryczko et al . , 2010; Dunmyre and Rubin , 2010; Jasinski et al . , 2013; Rybak et al . , 2014 ) . Therefore the two key features formulated above , which are critical for generation of network-based MMOs , appear to represent common properties of populations of intrinsically bursting neurons with distributed excitability that extend across many different bursting mechanisms . This conclusion clearly contradicts a recently published opinion ( Feldman and Kam , 2015 ) that previous computational models reproducing the MMOs observed in the pre-BötC ( e . g . , Butera et al . , 1999b; Rybak et al . , 2004 ) are not valid because the neuronal bursting in these models is critically dependent on slow deactivation kinetics of INaP . To evaluate the potential role of INaP in the considered MMOs , we used our large-scale model to investigate the transition of the population activity pattern during progressive suppression of INaP in all neurons ( Figure 4A2 , A3 , B3 ) . A regime with only LA bursts was selected as a starting point for this study ( top trace ) . When INaP conductance ( g¯NaP ) was suppressed , the frequency of LA bursts decreased and an MMO regime emerged ( Figure 4B3 , traces 2 and 3 ) until eventually only SA bursts remained and then activation completely ceased . We consider this result as a prediction for future experimental study , suggesting that a progressive suppression of INaP in the pre-BötC in vitro by its specific blocker , riluzole , should cause a transitional MMO regime before abolishing rhythmicity completely . When the weights of excitatory connections were progressively increased in our large-scale model , a succession of stable network rhythms , or 'regimes' , were observed ( Figure 4A1 , B1 ) . Low weights of connections produced only SA bursts in the network’s activity ( top trace in Figure 4B1 ) , intermediate weights caused MMOs ( traces 2–4 ) , and strong weights produced regimes with only LA bursts ( bottom trace ) . Similar regimes emerged when the probability of connections was increased at fixed weights of connections ( Figure 4B2 ) . In all of these cases , the overall frequency of burst events remained similar; what changed was the frequency with which those bursts were of large amplitude . Similar transformations in the integrated pattern occurred when weights of interconnections were increased in the reduced model ( Figure 5 ) . In contrast , reduction of either the general neuronal excitability ( Figure 7A ) or weights of connections ( Figure 7B ) could cause LE neurons to remain silent , leading to an integrated pattern with only SA bursts present ( Figure 7A , B ) . These simulation results may provide a reasonable explanation for the transformation of MMOs observed during application of cadmium ( Cd2+ ) in a medullary slice exhibiting MMOs ( Kam et al . , 2013 ) . In these experiments , Cd2+ application abolished LA bursts whereas SA oscillations persisted . We therefore suggest that the effects of Cd2+ , a blocker of calcium currents , could either attenuate neuronal excitability or reduce excitatory synaptic interconnections within the pre-BötC , as seen in our simulations ( Figure 7A , B ) . However , more experimental investigations , particularly regarding frequency changes following Cd2+ exposure , are needed to distinguish these possibilities . The analysis of neuronal 'clustering' of our large-scale model showed that groups of neurons with different excitability participated either in SA and LA bursts or only in LA bursts ( see Figure 3B1 , B2 , C1 , C2 , D ) . Specifically , neurons with relatively high excitability ( EL ) , and therefore with the high burst frequency ( HE neurons ) , participated in some SA and all LA bursts , whereas neurons with the lowest excitability and the lowest burst frequency ( LE neurons ) participated only in LA bursts . Importantly , since LE neurons had the highest spike frequency ( Figure 3B3 ) within the bursts , they could provide the strongest excitatory synaptic inputs to other neurons , resulting in the network-wide synchronization underlying the generation of LA bursts . It is also interesting to note that LE neurons could fail to activate even when receiving excitatory inputs of sufficient strength ( see intersection of dashed lines in Figure 3D ) , if the time from the last LA burst was insufficient for the recovery of LE neurons . This suggests that a mechanism intrinsic to the LE neurons and connected with their slow recovery is critically involved in the generation of LA bursts , defining their IBIs and the output burst frequency . Our reduced model exhibited a similar dependence on LE neuron recovery , which could be confirmed by analysis using time-scale decomposition in the ( V , hNaP ) -plane ( Figure 5C3 , D3 , E3 ) . This analysis showed that whether or not an excitatory input could recruit the LE neuron and induce an LA burst depended on the relative sizes of two quantities: ( a ) the hNaP-coordinate of the LE neuron at the time of input ( longer periods of recovery , or inactivity , led to higher hNaP-coordinates ) and ( b ) the hNaP-coordinate of the left knee of the V3-nullcline corresponding to the excitatory input ( stronger inputs induced lower hNaP-coordinates ) . Successful LE neuron activation occurred when ( a ) was greater than ( b ) , as at point ( iv ) in Figure 5C3 , and activation failed when ( b ) was greater than ( a ) , as at ( i ) - ( iii ) in Figure 5C3 . When weights were increased , the V3-nullcline was shifted to lower hNaP values , which allowed the LE neuron to activate with less recovery . Interestingly , based on this analysis and previous work ( Dunmyre and Rubin , 2010 ) we can infer that the strong mutual excitation , that occurs during an LA burst , is responsible for the pause in activity of the tonic spiking neurons after an LA burst in the large-scale model ( Figure 3B2 ) . Both the prolonged IBI and the pause in tonic spiking after LA bursts rely on the synaptic excitation from the full collection of neurons in the network , and thus their presence can be taken as evidence that the least excitable neurons in the network are not recipients of feed-forward inputs but rather participate in the recurrent network structure . The emergence of MMOs in the pre-BötC has been recently studied in vitro in medullary slices from neonatal mice ( Kam et al . , 2013 ) . These MMOs were artificially evoked at a moderate level of neuronal excitability produced by elevation of [K+]out to 5–6 mM and were characterized by a series of SA bursts ( 'burstlets' ) alternating with single LA bursts that , in contrast to the burstlets , were able to trigger the rhythmic bursts in the hypoglossal motor output and hence defined the frequency of output oscillations . This study established the quantal nature of MMOs emerging in the pre-BötC in these conditions ( e . g . , Figure 2 of Kam et al . , 2013 ) . The emergence of these MMOs in the pre-BötC allowed Feldman and Kam to propose a novel 'burstlet concept' of inspiratory rhythm generation that 'fundamentally breaks with the burst hypothesis' ( Feldman and Kam , 2015 ) . According to this concept , 'rhythm- and pattern-generating functions common to all CPGs are assumed to be segregated' so that the rhythm and the pattern are generated by 'separable microcircuits' and 'distinct mechanisms' ( Kam et al . , 2013; Feldman and Kam , 2015 ) , similar to that in a previous model of the spinal locomotor CPG suggesting the existing separate circuits for rhythm generation and pattern formation ( Rybak et al . , 2006a; Rybak et al . , 2006b; McCrea and Rybak , 2008 ) . In this interpretation , the role of intrinsic bursting mechanisms in neurons generating the LA bursts in the pre-BötC is fully disregarded , and the lack of these bursts on the top of each burstlet ( SA bursts ) is considered as equivalent to the non-resetting spontaneous deletions ( missing bursts ) observed during fictive locomotion in the spinal cord . Our computational study does not support the interpretation of MMOs in the pre-BötC as indicative of separate rhythm- ( burstlets ) and pattern- ( bursts ) generating sub-networks . The results of our present modeling study instead suggest that a single , inseparable population of coupled excitatory neurons incorporating endogenous neuronal oscillators with distributed excitability can reproduce , and is sufficient to explain , the coexistence of burstlets and bursts in population rhythmic activity ( i . e . , the MMOs described in this work ) . We implemented a sparse network connectivity pattern that reflects experimental data more completely than previous models ( Rybak et al . , 2004; Jasinski et al . , 2013 ) and precludes the existence of separable sub-networks . In the models of the locomotor CPG in the spinal cord mentioned above , the pattern formation circuits did not affect the rhythm generator circuits , but just responded 1:1 to the rhythm-generating input , unless accidental perturbations happened , changing the excitability of the pattern formation network and producing deletions ( Rybak et al . , 2006a; McCrea and Rybak , 2008 ) . In contrast , in the interconnected single network considered here , the activity of low-excitable neurons involved in generation of low frequency LA bursts ( attributed by Kam et al . to the 'pattern generating circuits' ) synchronize the entire population activity , explicitly defining its output frequency ( 'rhythm' ) . Therefore , the intrinsic properties of these low-excitable neurons , specifically the temporal characteristics of their recovery ( see Figures 5D1 , D2 and 6B ) , but not deletions of unknown origin , define the output frequency of a rhythm generator that interacts with other circuits to shape the CPG activity pattern . In the large-scale population model all neurons were modeled in the single-compartment Hodgkin-Huxley style , in accordance with our previous models ( Rybak et al . , 2003b , Rybak et al . , 2004 , Rybak et al . , 2007; Smith et al . , 2007; Jasinski et al . , 2013 ) . For each neuron , the membrane potential , V , was described by the following differential equation: ( 1 ) CdVdt=−INa−INaP−IK−IL−ISynE , where C is membrane capacitance . The following ionic currents were included in the model: fast sodium ( INa ) ; persistent , slowly inactivating sodium ( INaP ) ; delayed-rectifier potassium ( IK ) ; leak ( IL ) ; and excitatory synaptic ( ISynE ) . These currents were described as follows: ( 2 ) INa=g¯Na⋅mNa3⋅hNa⋅ ( V−ENa ) ; ( 3 ) INaP=g¯NaP⋅mNaP⋅hNaP⋅ ( V−ENa ) ; ( 4 ) IK=g¯K⋅mK4⋅ ( V−EK ) ; ( 5 ) IL=g¯L⋅ ( V−EL ) ; ( 6 ) ISynE=gSynE⋅ ( V−ESynE ) , where g¯x terms ( with index x denoting the particular current ) represent maximal conductances; gSynE denotes the conductance of the excitatory synaptic current to the neuron ( see below ) ; Ex is the current’s reversal potential; and mx and hx are dynamic variables describing current x activation and inactivation , respectively . Activation and inactivation kinetics obey the following equations: ( 7 ) τmx ( V ) dmxdt=mx∞ ( V ) −mx , ( 8 ) τhx ( V ) dhxdt=hx∞ ( V ) −hx , where mx∞ ( V ) and hx∞ ( V ) define steady-state voltage-dependent activation and inactivation , respectively , and τmx ( V ) and τhx ( V ) are the corresponding voltage-dependent time constants ( see Table 1 ) . Equations 1–8 were used for each neuron in the population , with all variables indexed by a numerical subscript specifying the identity of each neuron . 10 . 7554/eLife . 13403 . 010Table 1 . Steady-state functions for voltage-dependent activation and inactivation of ionic channels and other parameter values of the large-scale model . DOI: http://dx . doi . org/10 . 7554/eLife . 13403 . 010Ionic channelsFast sodium ( Na ) mNa∞ ( V ) =1/ ( 1+exp ( - ( V+43 . 8 ) /6 . 0 ) ) ; τmNa ( V ) =0 . 25/cosh ( − ( V+43 . 8 ) /14 . 0 ) ms; hNa∞ ( V ) =1/ ( 1+exp ( ( V+67 . 5 ) /10 . 8 ) ) ; τhNa ( V ) =8 . 46/cosh ( ( V+67 . 5 ) /12 . 8 ) ms; g¯Na=170 . 0 nS . Persistent sodium ( NaP ) mNaP∞ ( V ) =1/ ( 1+exp ( − ( V+47 . 1 ) /3 . 1 ) ) ; τmNaP ( V ) =1/cosh ( - ( V+47 . 1 ) /6 . 2 ) ms; hNaP∞ ( V ) =1/ ( 1+exp ( ( V+60 . 0 ) /9 . 0 ) ) ; τhNaP ( V ) =6000/cosh ( ( V+60 . 0 ) /9 . 0 ) ms; g¯NaP=5 . 0 ± 0 . 5 nS . Delayed-rectifier potassium ( K ) mK∞ ( V ) =αK∞/ ( αK∞+βK∞ ) ; τmK ( V ) =1/ ( αK∞+βK∞ ) ms; where αK∞=0 . 01⋅ ( V+45 . 0 ) / ( 1-exp ( - ( V+45 . 0 ) /5 . 0 ) ) ; βK∞=0 . 17⋅exp ( - ( V+49 . 0 ) /40 . 0 ) ; g¯K=180 . 0 nS . Leak ( L ) g¯L=2 . 5 nS . Neuron parametersReversal potentialsENa=60 . 0 mV , EK=-94 . 0 mV , ESynE=-10 . 0 mV , EL=-62 . 0 ± 0 . 93 mV . Membrane capacitanceC=36 . 2 pF . Synaptic/network parametersSynaptic connectionsg¯SynE=0 . 05 nS , τSynE=5 . 0 ms , wij=w ∈ [1 . 0 , 5 . 0] , p ∈ [0 . 09 , 0 . 40]; Spike threshold=-35 . 0 mV . We considered only excitatory synaptic connections between neurons . The excitatory synaptic conductance was zero at rest and was increased when each excitatory input occurred , such that ( 9 ) gSynEi=g¯SynE⋅∑jwji⋅∑tkj<texp[− ( t−tkj ) /τSynE] , where wji is the synaptic weight from neuron j to neuron i , g¯SynE is the maximal synaptic conductance , τSynE is the synaptic time constant , tkj is the time of the k-th spike from neuron j , and each term in the sum is evaluated for t > tkj . . That is , each new spike from neuron j increases the excitatory synaptic conductance of neuron i by g¯SynE⋅wji . The probability of each connection ( p ) was set a priori , where in a network of N neurons , pN represents the mean number of neurons with which an individual neuron would form synapses . To form a network , a random number generator was used to determine whether or not each possible synaptic connection among neurons was actually present . Neuronal heterogeneity within the population was generated with Gaussian distributions for the leak reversal potential ( EL ) and the maximal conductance of the persistent sodium current ( g¯NaP ) . The means and variances of these parameter distributions , as well as all other parameters used in the large-scale model , are provided in Table 1 . Initial conditions for neuronal membrane potentials and variables defining currents' activation and inactivation were randomly distributed within physiologically realistic ranges for each variable . To rule out chaotic behaviors , simulations were repeated with redistributed initial conditions for each parameter set . Finally , results were only considered following an initial simulation period of 20 s to minimize the likelihood of transient dynamics . Integrated population activity was represented by a histogram showing the number of spikes in all neurons per 10ms bin . Maximal values of these histograms during synchronized population bursts , in spikes/bin , were considered as population burst amplitudes . Bursts with amplitude more than 50 spikes/bin were considered to be LA bursts and bursts with amplitude less than 50 spikes/bin were classified as SA bursts . All simulations were performed using the simulation package NSM 3 . 0 , developed at Drexel University by SN Markin , IA Rybak , and NA Shevtsova . Differential equations were solved using the exponential Euler integration method with a step size of 0 . 1 ms . To study the effects of changing the connection weights , probability of connections , and the maximal conductance of INaP on MMO pattern , we calculated the ratio of LA and SA burst frequencies as these parameters were varied ( Figure 4A1 , A2 , A3 ) . 50-second simulations were performed , and the final 40 s were extracted for processing . LA population bursts were defined by histogram activity above 20 spikes in a 100 ms window , and the remaining bursting events were categorized as SA bursts . The ratio of LA and SA bursts was color coded so that boundaries could be visualized in various parameter spaces . Mathematical analysis of the large-scale model was prevented by its high dimensionality ( 100 neurons , each with several differential equations per neuron ) . However , a preliminary analysis of the simulation results suggested that a minimal neural network could be used to reproduce the development of MMOs caused by the clustering of neurons with similar excitabilities . We therefore developed a reduced network consisting of three neurons simulated by an 'activity-based , ' non-spiking model with different excitability defined by the EL value for each neuron . In this reduced formalization , a neuron’s activity represents the aggregate activity of a distinct cluster in the large-scale model . Similar reduced three-neuron models were previously considered in other contexts ( e . g . , Shilnikov et al . , 2008; Rubin and Terman , 2012; Schwabedal et al . , 2014 ) . The simplified neuron models have been also previously used to simulate and analyze the behavior of larger models of respiratory networks , including the pre-BötC ( Rubin et al . , 2009b; Rubin et al . , 2011 ) . Each neuron is described by one 'fast' dynamic variable , V , that governs changes in a neuron’s membrane potential and obeys the following differential equation: ( 10 ) C⋅dVidt=−INaPi−ILi−ISynEi , where i ∈ {1 , 2 , 3} is the index corresponding to the neuron’s number shown in Figure 5A1 and C is membrane capacitance . This reduced model excluded the fast sodium ( INa ) and potassium ( IK ) currents included in the large-scale model . However , similar formalizations of the persistent ( slowly inactivating ) sodium ( INaP ) , leak ( IL ) , and excitatory synaptic ( ISynE ) currents were used: ( 11 ) INaPi=g¯NaP⋅mNaP∞ ( Vi ) ⋅hNaPi⋅ ( Vi−ENa ) ; ( 12 ) ILi=g¯L⋅ ( Vi−ELi ) ; ( 13 ) ISynEi=∑j=1j≠i3 ( wji⋅f ( Vj ) ) ⋅g¯SynE⋅ ( Vi−ESynE ) , where for x ∈ {NaP , L , SynE} , g¯x is the maximal conductance and Ex is the channel’s reversal potential , respectively . EL was uniformly distributed across the 3 neurons in the range [-54 . 5 , -63 . 5] mV to produce one neuron that was intrinsically quiescent and two that were intrinsically oscillating at different frequencies ( Figure 5B1 ) ; we labeled these as low excitability ( LE ) , medium excitability ( ME ) , and high excitability ( HE ) neurons . The excitatory synaptic current in Equation 13 includes inputs to neuron i from neurons j , each of which is the product of fixed connection weights , wji = w , and a piecewise linear function , f ( V ) : ( 14 ) f ( V ) ={0 , if V<Vmin ( V−Vmin ) / ( Vmax−Vmin ) , if Vmin≤V<Vmax1 , if V≥Vmax , where Vmin and Vmax define the voltages at which threshold and saturation occur , respectively . An activity level ( or normalized firing rate ) for each neuron is implicitly associated with the value of its voltage , and the function f ( V ) represents an output signal corresponding to that activity level . The activation of the persistent sodium current , INaP , is described by the voltage-dependent steady state gating variable , mNaP∞: ( 15 ) mNaP ∞ ( V ) = ( 1+exp{ ( V−VmNaP ) /kmNaP} ) −1 . INaP activation is considered instantaneous . The 'slow' dynamical variable in the reduced model , hNaP , represents inactivation of the persistent sodium current and is governed by the following equation: ( 16 ) τhNaP ( V ) ⋅dhNaPdt=hNaP∞ ( V ) −hNaP , where hNaP∞ and τNaP∞ describe the voltage-dependent steady-state and time constant for inactivation , respectively: ( 17 ) hNaP ∞ ( V ) = ( 1+exp{ ( V−VhNaP ) /khNaP} ) −1; ( 18 ) τhNaP∞ ( V ) =τhNaPmax/cosh { ( V−VτhNaP ) /kτhNaP} . The parameters VxNaP and kxNaP for x∈{m , h , τh} in Equations 15 , 17 , 18 represent each gating variable’s half-activation voltage and slope , respectively . All parameters of the reduced model were taken from previous works ( Rubin et al . , 2009b; Rubin et al . , 2011 ) and are specified in Table 2 . The distribution of EL was first set manually to match the large-scale model and then optimized by calculating a series of iterative one-dimensional bifurcation diagrams . The robustness of a given regime ( for example , the LE period branches marked '1:X' in Figure 6A ) was determined by the range of connection weights across which the LE period maintained an integer ratio to the HE period . Simulations were performed and visualized using custom C++ scripts and gnuplot , respectively . 10 . 7554/eLife . 13403 . 011Table 2 . Parameter values for the reduced model . DOI: http://dx . doi . org/10 . 7554/eLife . 13403 . 011Ionic channelsPersistent sodium ( NaP ) VmNaP=-40 . 0 mV , kmNaP=-6 . 0 mV; VhNaP=-59 . 0 mV , khNaP=10 . 0 mV; VτNaP=-59 . 0 mV , kτhNaP=20 . 0 mV , τhNaPmax=5000 ms; g¯NaP=5 . 0 nS . Leak ( L ) g¯L=2 . 8 nS . Synaptic Current ( SynE ) g¯SynE=0 . 1 nS . Neuron parametersPotentialsENa=50 . 0 mV; EL1=-54 . 5 mV , EL2=-59 . 0 mV , EL3=-63 . 5 mV; ESynE=-10 . 0 mV . Membrane capacitanceC=20 . 0 pF . Synaptic/network parametersSynaptic connectionswji=w ∈ [0 . 0 , 5 . 0] . Parameters of output function , f ( V ) Vmin=-50 . 0 mV , Vmax=0 . 0 mV . The complete range of a neuron’s dynamics , as a function of EL , was investigated with time-scale decomposition in the ( V , hNaP ) -plane ( Figure 5B3 ) . When projected into the ( V , hNaP ) -plane , the dynamical variables , V and hNaP , had steady states or 'nullclines' ( sets of points for which the right-hand sides of Equations 10 , 16 , respectively , were set to zero ) . Some possible positions of the cubic V-nullclines are depicted by a gray band in Figure 5B3 . The upper and lower boundaries of the band correspond to the lowest and highest values of EL that produced bursting , respectively . That is , the intersection of the V- and hNaP-nullclines created a fixed point for the system that , when stable , denotes the point where solutions converge . There were two possible stable fixed points in our model for each neuron: ( i ) along the left branch of the V-nullcline ( silence ) , and ( ii ) on the right branch of the V-nullcline , creating a state of constant depolarization ( the activity-based analog to tonic spiking ) . When EL was intermediate to values that produced silence and tonic behavior , the hNaP-nullcline intersected the V-nullcline’s middle branch , creating an unstable fixed point with a stable periodic orbit , or oscillation ( Figure 5B3 , red trace ) , that encompassed the local maximum and minimum of the V-nullcline ( Figure 5B3 , blue curve ) . The presence of a stable periodic orbit corresponded to endogenous bursting in these neurons . Each periodic orbit has two 'slow' components located close to the neuron’s V-nullcline and governed by the neuron’s hNaP ( slow ) dynamics , and two 'fast components' connecting between V-nullcline branches and governed by the neuron’s V ( fast ) dynamics . . During the slow components , the neuron could be silent or at rest when its trajectory was traveling up the left branch of its V-nullcline corresponding to an absence of spike generation , and it could be active or depolarized when its trajectory was traveling down the right branch of its V-nullcline , corresponding to spike generation . While at rest , a neuron in the bursting regime slowly 'recovered , ' with its trajectory rising to higher hNaP-coordinates until it reached the left knee ( or fold ) of the V-nullcline ( a bursting neuron , shown in Figure 5B3 , red trace ) . At the left knee , a neuron’s trajectory moved rightward in the ( V , hNaP ) -plane under the fast dynamics to approach the right branch of the V-nullcline , corresponding to activation of the neuron . Once active , the neuron’s trajectory traveled downward , to lower hNaP–coordinates , along the right branch of the V-nullcline until it reached the right knee ( fold ) of the V-nullcline , which caused a leftward jump in the ( V , hNaP ) -plane corresponding to burst termination ( Figure 5B3 , red trace ) . Similarly , a neuron with a stable fixed point could have slow transient dynamics and be in a rest ( active ) state as its trajectory traveled along the left ( right ) branch of its V-nullcline . When a neuron became more excitable , either by an increase in EL or in its excitatory inputs , the right-hand side of its voltage equation was altered , causing a change in the position of its V-nullcline , to a location downward and to the right of the original in the ( V , hNaP ) -plane . Such a change could cause the neuron’s fixed point to switch from one branch of its V-nullcline to another , yielding a transition from silence to bursting to tonic spiking , depending on fixed point location . This change would also change knee locations; correspondingly , plots can be made showing the hNaP-coordinate of the left knee as a function of a parameter or as a function of the input to a neuron in Equation 13 . The executable files and scripts used to generate the simulations presented in this manuscript may be downloaded from: http://neurobio . drexelmed . edu/rybakweb/software . htm .	Each breath we take removes carbon dioxide from the body and exchanges it for oxygen . A structure called the brainstem , which connects the brain with the spinal cord , generates the breathing rhythm and controls its rate . While this process normally occurs automatically , we can also control our breathing voluntarily , such as when singing or speaking . Within the brainstem , a group of neurons in the area known as the pre-Bötzinger complex is responsible for ensuring that an animal breathes in at regular intervals . Recordings of the electrical activity from slices of brainstem show that pre-Bötzinger neurons display rhythmic activity with characteristic patterns called “mixed-mode oscillations” . These rhythms consist of bursts of strong activity ( “large amplitude bursts” ) , essential for triggering regular breathing , separated by a series of bursts of weak activity ( “small amplitude bursts” ) . However , it is not clear how mixed-mode oscillations arise . Bacak , Kim et al . now provide insights into this process by developing two computational models of the pre-Bötzinger complex . The first model consists of a diverse population of 100 neurons joined by a relatively small number of weak connections to form a network . The second model is a simplified version of the first , consisting of just three neurons . By manipulating the properties of the simulated networks , and analysing the data mathematically , Bacak , Kim et al . identify the properties of the neurons that allow them to generate mixed-mode oscillations and thus rhythmic breathing . The models suggest that mixed-mode oscillations result from the synchronization of many neurons with different levels of activity ( excitability ) . Neurons with low excitability have low bursting frequencies , but generate strong activity and recruit other neurons , ultimately producing large amplitude bursts that cause breathing . Many parts of the nervous system are also made up of networks of neurons with diverse excitability . A challenge for future studies is thus to investigate whether other networks of neurons similar to the pre-Bötzinger complex generate rhythms that control other repetitive actions , such as walking and chewing .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "neuroscience" ]	2016	Mixed-mode oscillations and population bursting in the pre-Bötzinger complex
TMEM16A forms calcium-activated chloride channels ( CaCCs ) that regulate physiological processes such as the secretions of airway epithelia and exocrine glands , the contraction of smooth muscles , and the excitability of neurons . Notwithstanding intense interest in the mechanism behind TMEM16A-CaCC calcium-dependent gating , comprehensive surveys to identify and characterize potential calcium sensors of this channel are still lacking . By aligning distantly related calcium-activated ion channels in the TMEM16 family and conducting systematic mutagenesis of all conserved acidic residues thought to be exposed to the cytoplasm , we identify four acidic amino acids as putative calcium-binding residues . Alterations of the charge , polarity , and size of amino acid side chains at these sites alter the ability of different divalent cations to activate the channel . Furthermore , TMEM16A mutant channels containing double cysteine substitutions at these residues are sensitive to the redox potential of the internal solution , providing evidence for their physical proximity and solvent accessibility . Transmembrane proteins of the TMEM16 family include a number of calcium-dependent ion channels , such as the TMEM16A ( Caputo et al . , 2008; Schroeder et al . , 2008; Yang et al . , 2008b ) and TMEM16B ( Schroeder et al . , 2008; Stöhr et al . , 2009 ) calcium-activated chloride channels ( CaCCs ) and the TMEM16F ( Yang et al . , 2012 ) small conductance calcium-activated nonselective cation ( SCAN ) channel . These proteins have been implicated in a wide range of physiological activities ( Hartzell et al . , 2009; Kunzelmann et al . , 2011; Berg et al . , 2012; Scudieri et al . , 2012; Huang et al . , 2012a ) including the control of fluid secretion in epithelia ( Romanenko et al . , 2010; Huang et al . , 2012b ) , the regulation of membrane excitability in neurons ( Billig et al . , 2011; Huang et al . , 2012c ) , the scrambling of membrane lipids in platelets ( Suzuki et al . , 2010; Yang et al . , 2012 ) , and the metastasis of certain cancers ( Katoh , 2003; Duvvuri et al . , 2012 ) . Despite their important physiological roles , the mechanism by which calcium activates these proteins is still extensively debated . Based on studies of calcium-activated potassium channels ( KCas ) ( Maylie et al . , 2004; Salkoff et al . , 2006 ) , two models have been proposed to account for the calcium sensitivity of TMEM16A-CaCC ( Figure 1A ) . In one model resembling that of the small-conductance KCa ( SK ) channels ( Xia et al . , 1998; Schumacher et al . , 2001 ) , calcium ions interact with calmodulin ( CaM ) , and the physical association of calmodulin with TMEM16A is required to transduce the free energy of ion–protein interaction into movements of the channel gate ( Tian et al . , 2011; Vocke et al . , 2013 ) . In a second model similar to that of the large-conductance KCa ( BK ) channels ( Schreiber and Salkoff , 1997; Shi et al . , 2002; Xia et al . , 2002; Wu et al . , 2010; Yuan et al . , 2010 ) , calcium binds directly to TMEM16A , and this interaction drives channel activation ( Yu et al . , 2012 , 2014; Terashima et al . , 2013 ) . 10 . 7554/eLife . 02772 . 003Figure 1 . Calmodulin ( CaM ) is not responsible for the calcium-dependent activation of TMEM16A calcium-activated chloride channels ( CaCC ) . ( A ) Two competing models to explain TMEM16A calcium sensitivity have been proposed . It is unclear whether calcium directly binds to TMEM16A-CaCCs ( upper panel ) or whether CaM is required to mediate the calcium sensitivity of the channel ( lower panel ) . ( B ) Representative current traces of wildtype mouse TMEM16A-CaCC ( mTMEM16A ) recorded at +60 mV and −60 mV in response to various intracellular calcium concentrations using an inside-out patch clamp configuration . Table indicates the concentration of calcium used . ( C ) Calcium dose–response of mTMEM16A channel at +60 mV and −60 mV . The smooth curves represent fits to the Hill equation ( ‘Materials and methods’ ) . ( D ) Loss-of-function CaM mutants ( CaM12 , CaM34 , CaM1234 ) did not reduce the apparent calcium sensitivity of the endogenous TMEM16A ( x16A ) channel in Xenopus oocytes . n . s . : non-significant . ( E ) Application of monoclonal anti-CaM antibody CaM85 ( 2 μg/ml ) to the cytosolic face of inside-out patches had no effect on the calcium sensitivity of endogenous Xenopus TMEM16A-CaCC . ( F ) Mutating residues reported by Vocke et al . ( 2013 ) to be in the CaM binding domain of mTMEM16A did not affect apparent TMEM16A-CaCC calcium sensitivity . DOI: http://dx . doi . org/10 . 7554/eLife . 02772 . 00310 . 7554/eLife . 02772 . 004Figure 1—figure supplement 1 . Calmodulin ( CaM ) is not involved in the calcium-dependent activation of TMEM16A-CaCC . ( A ) Acute application of 50 µM W7 , a CaM antagonist , to the cytosolic face of the inside-out patches failed to inhibit endogenous Xenopus TMEM16A-CaCC currents . ( B ) Chronic incubation of Xenopus oocytes with W7 did not reduce the apparent calcium sensitivity of endogenous Xenopus TMEM16A-CaCC . ( C and D ) Barium , a divalent cation that is incapable of binding CaM , can robustly activate mouse TMEM16A-CaCC . ( C ) CaCC currents were recorded with voltage steps in +20 mV increments from −80 mV to +120 mV in isotonic 140 mM NaCl solutions . Both holding and repolarizing potentials were −80 mV . Dotted lines indicate the zero current level . ( D ) Barium dose–response curves of wildtype mTMEM16A channel at +60 mV and −60 mV . Smooth curve represent fits to the Hill equation . DOI: http://dx . doi . org/10 . 7554/eLife . 02772 . 004 In this study , we performed a series of experiments designed to clarify the mechanism by which TMEM16A-CaCCs are gated by calcium . First , with multiple lines of evidence , we confirmed that CaM is not required for TMEM16A-CaCC activation . Then , we performed a comprehensive alanine mutagenesis screen of all highly conserved acidic residues in TMEM16A that are predicted to be accessible to the cytoplasm . Based on our measurements of these mutants' apparent calcium sensitivity in inside-out excised patches , we identify four acidic residues ( including two acidic residues that were recently implicated in the calcium-dependent activation of TMEM16A-CaCCs [Yu et al . , 2012] ) that satisfy several criteria characteristic of calcium-binding sites . Substitution of each of these residues with various amino acids shifts the apparent sensitivity of TMEM16A-CaCC for different divalent cations , and cysteine residues at these sites can be cross-linked by oxidizing intracellular solutions . Our results thus demonstrate that direct binding of calcium to TMEM16A triggers channel activation independently of calmodulin , identify novel interaction sites between calcium ions and TMEM16A , and lay the groundwork for future studies examining the mechanism of calcium-dependent TMEM16 channel activation . We began our study of TMEM16A-CaCC's calcium-dependent activation by testing whether CaM is required for channel activity . Reasoning that apparent calcium sensitivity is the most direct measurement of TMEM16A function , we recorded CaCC currents from GFP-tagged TMEM16A channels at fixed test potentials using an inside-out excised patch configuration . CaCC current amplitudes increased in a calcium-dependent manner until channels reach maximum open probability in our recordings ( Figure 1B ) . By fitting the dose–response curve to the Hill equation ( Figure 1C ) , apparent calcium sensitivity was derived as the half maximal effective concentration ( EC50 ) of calcium required for channel activation . With rapid application of internal solutions , this patch clamp protocol effectively minimized the effect of channel desensitization on our analysis of TMEM16A-CaCC apparent calcium sensitivity ( Figure 1B ) . In addition , deriving the apparent calcium sensitivity from individual inside-out patches allowed us to control for variations in protein expression that complicate the interpretation of data derived from other electrophysiology configurations such as whole-cell patch clamp . Since overexpression of mutant CaM molecules whose EF hand calcium-binding motifs are destroyed has a dominant negative effect on the ability of SK channels to detect changes in intracellular calcium ( Xia et al . , 1998 ) , we tested the effects of these CaM mutations on TMEM16A-CaCCs . We found that co-expression of mutant CaM molecules without functional calcium-binding sites in the N-terminal lobe ( CaM12 ) , C-terminal lobe ( CaM34 ) , or both lobes ( CaM1234 ) did not reduce the apparent calcium sensitivity of the endogenous TMEM16A-CaCC in Xenopus oocytes ( Figure 1D ) , in agreement with a recent study ( Yu et al . , 2014 ) . Because the failure of CaM mutants to reduce TMEM16A calcium sensitivity could conceivably be due to the inability of the mutant CaM molecules to sufficiently displace endogenous CaM from the putative TMEM16A-CaM complexes , we further clarified the role of CaM in TMEM16A-CaCC gating by conducting four additional experiments to manipulate the potential interaction between CaM and TMEM16A . First , we tried to sequester endogenous CaM molecules by directly applying anti-CaM antibodies to the cytoplasmic side of inside-out patches from Xenopus oocytes . In contrast to its potent inhibitory effect on CaM-dependent TRPV1 tachyphylaxis ( Lishko et al . , 2007 ) , the CaM-antibody had no effect on TMEM16A-CaCC currents ( Figure 1E ) . Second , we treated endogenous Xenopus TMEM16A channels with W7 , a potent CaM antagonist that can prevent CaM from binding to its targets . We found that neither acute nor sustained applications of this drug disrupted the calcium sensitivity of TMEM16A channels ( Figure 1—figure supplement 1A , B ) . Third , consistent with previous reports ( Xiao et al . , 2011; Yuan et al . , 2013; Ni et al . , 2014 ) , we found that barium , a divalent cation that cannot bind to CaM ( Chao et al . , 1984 ) , can still potently activate TMEM16A channels in a dose-dependent manner ( Figure 1—figure supplement 1C , D ) . Fourth , mutating three residues ( V310 , Y314 , and L319 ) that were recently reported to be necessary for CaM-TMEM16A interactions ( Vocke et al . , 2013 ) did not reduce the apparent calcium sensitivity of the mouse TMEM16A channel expressed in HEK 293 cells ( Figure 1F ) . Similar to the results reported by Vocke et al . ( 2013 ) , the current amplitudes of these mutant channels were much smaller compared to wildtype channels in our inside-out patch recordings ( data not shown ) . However , the absence of any rightward shift in the dose–response curves suggests that these residues do not contribute toward TMEM16A-CaCC calcium sensitivity . Because we were unable to alter TMEM16A-CaCC activity by inhibiting CaM function , could activate TMEM16A-CaCCs with barium , and could not shift channel EC50 by mutating the putative CaM-TMEM16A binding interface , we conclude that CaM is not responsible for the calcium sensitivity of TMEM16A-CaCCs . After excluding CaM as the calcium sensor for TMEM16A-CaCCs , we hypothesized that calcium might directly bind to TMEM16A itself . A recent study by Yu et al . ( 2012 ) found that mutations altering the charge at E698 and E701 ( numbered E702 and E705 in the splice variant of TMEM16A used in their experiments ) reduced the amplitude of whole-cell currents and altered channel activation and deactivation kinetics , suggesting that a ligand responsible for channel activation may interact with these sites . To test whether the apparent calcium sensitivities of E698Q and E701Q mutant channels are reduced , we measured their steady-state EC50 of calcium-dependent channel activation ( Figure 2A , B ) . We found that their EC50s were shifted to much higher calcium concentrations compared to the EC50 of wildtype TMEM16A channels , suggesting that these two acidic residues might be important for a direct interaction between calcium ions and TMEM16A-CaCC . 10 . 7554/eLife . 02772 . 005Figure 2 . Screen for potential calcium-binding residues in TMEM16A-CaCC . ( A and B ) Quantification of the apparent calcium sensitivity of E698Q and E701Q ( Yu et al . , 2012 ) mutant TMEM16A channels . ( A ) Representative current traces of E698Q and E701Q mutants in response to intracellular solutions with different calcium concentrations recorded at +60 mV . Table indicates the concentration of calcium used . ( B ) Calcium dose–response curves of the mTMEM16A channels at +60 mV . Smooth curves represent fits to the Hill equation . ( C ) Sequence alignment of the calcium-activated TMEM16 channels . h16A , m16A , x16A , m16B , m16F and Fly16 are the human TMEM16A ( Uniprot ID #Q5XXA6 ) , mouse TMEM16A ( Uniprot ID #Q8BHY3-2 ) , Xenopus TMEM16A ( Uniprot ID #B5SVV6 ) , mouse TMEM16B ( Uniprot ID #Q8CFW1 ) , mouse TMEM16F ( Uniprot ID #Q6P9J9 ) and Drosophila TMEM16 channels ( Uniprot ID #Q86P24 ) , respectively . Highly conserved acidic residues that are potentially exposed to the cytoplasm are highlighted in red . Some residues with conserved oxygen-containing side chains in m16F and Fly16 are highlighted in green . Putative transmembrane ( TM ) segments are highlighted in cyan . The controversial TM6 segments are highlighted in gray and labeled as TM6′ and TM6″ , respectively . ‘In’ and ‘Out’ indicate the intracellular and extracellular side of the membrane , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 02772 . 005 In proteins , calcium ions tend to bind to the carboxylate moieties of acidic amino acid residues ( Pidcock and Moore , 2001 ) and typically require six to eight coordinates ( Dudev and Lim , 2003 ) . For a comprehensive survey of putative calcium binding residues in TMEM16A , we identified 38 acidic residues ( including E698 and E701 ) that are conserved in all known calcium-activated ion channels in the TMEM16 family ( including a Drosophila calcium-activated chloride channel [Wong et al . , 2013] ) and that are hypothesized to be solvent accessible from the intracellular compartment ( Figure 2C ) . We then systematically replaced each of these residues in the mouse TMEM16A channel with alanine and measured their apparent calcium sensitivity using inside-out patches excised from HEK 293 cells ( Figure 3 ) . We found that alanine substitutions at E698 and E701 greatly reduced the apparent calcium affinity of TMEM16A-CaCC ( Figure 3 ) , similar to the effects of glutamine substitutions at these residues ( Figure 2A , B ) . Moreover , our screen identified three additional acidic residues ( E650 , E730 , and D734 ) that were critical for the calcium sensitivity of TMEM16A channels ( Figure 3 ) . Similar to E698A and E701A , alanine mutations of these three acidic residues resulted in at least a 10-fold reduction of apparent calcium sensitivity . Interestingly , these five residues identified by our screen are confined to the region between the fifth and seventh putative transmembrane segments ( Figure 2C ) ( Yu et al . , 2012 ) . 10 . 7554/eLife . 02772 . 006Figure 3 . Systematic alanine scan of highly conserved intracellular acidic residues identified five mutations that dramatically reduced the apparent calcium sensitivity of TMEM16A-CaCC . ( A ) Representative current traces of the E650A , E698A , E701A , E730A and D734A mutant channels in response to different intracellular calcium solutions recorded at +60 mV . Table indicates the concentration of calcium used . ( B ) Calcium dose–response curves of these mutant mTMEM16A channels at +60 mV . Smooth curves represent fits to the Hill equation . ( C ) Summary of apparent calcium sensitivity ( EC50s ) of all alanine mutants tested . Dotted line indicates a 10-fold increase in EC50 compared to wildtype channels . DOI: http://dx . doi . org/10 . 7554/eLife . 02772 . 006 Calcium ions interact with different chemical moieties with different affinities depending on the coordination chemistry of the calcium ion ( Sóvágó and Várnagy , 2013 ) . In order to understand the role of E650 , E698 , E701 , E730 , and D734 in TMEM16A-calcium interactions , we replaced the side chains at each of these five potential calcium-binding residues with charge-neutralizing ( Cys , Asn , or Gln ) , charge-reversing ( Arg ) , and charge-conserving ( Asp or Glu ) groups and examined their effects on the apparent calcium sensitivity of TMEM16A channels ( Figure 4 ) . 10 . 7554/eLife . 02772 . 008Figure 4 . The effects of different amino acid side chains on the calcium sensitivity of mutant TMEM16A-CaCC channels indicate that E698 , E701 , E730 and D734 might be directly involved in binding calcium . ( A–E ) Calcium dose–response curves of ( A ) E650 , ( B ) E698 , ( C ) E701 , ( D ) , E730 , and ( E ) D734 mutant mTMEM16A channels at +60 mV . Smooth curves represent fits to the Hill equation . Maximum activity was constrained to 1 for these fittings . ( F ) Summary of apparent calcium sensitivity ( EC50s ) of mTMEM16A mutants . N . C . : no obvious CaCC current recorded . *p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 02772 . 008 Introduction of cysteine substitutions at these sites replaces the glutamate or aspartate side chain with a small , hydrophobic methyl sulfhydryl group . Understandably , these manipulations had effects comparable to those of alanine substitutions ( Figure 3C ) and reduced the apparent calcium sensitivity of the channel ( Figure 4F ) . Similarly , we reasoned that charge-reversing arginine substitutions of residues that coordinate calcium would have dramatic effects due to the introduction of a positively charged guanidinium group . Indeed , E698R , E701R , E730R and D734R were much less sensitive to internal calcium . Compared to wildtype ( WT ) channels ( EC50: 1 . 0 ± 0 . 1 μM ) , the EC50s of E698R and E701R were increased by several thousand-fold to 5 . 0 ± 0 . 3 mM and 3 . 3 ± 0 . 5 mM , respectively , while the effects of E730R were milder , increasing EC50 to 0 . 6 ± 0 . 3 mM . The most substantial changes were caused by the D734R mutation , which elevated EC50 to such a great extent that 10 mM intracellular calcium , the highest concentration of calcium used in this study , was unable to activate the mutant channel . The lack of CaCC current in D734R mutants was not due to defects in protein expression or folding because cadmium ions could still robustly activate the channel ( see below ) . In contrast , the arginine substitution in E650R mutants only produced a small reduction in calcium sensitivity ( EC50: 6 . 5 ± 0 . 8 µM ) , suggesting that the side chain at position 650 may not be directly involved in calcium coordination . Because calcium ions tend to be coordinated by oxygen atoms , we also made several mutations to test the role of carboxylate and carbonyl moieties in TMEM16A-CaCC calcium sensitivity . First , we made charge-preserving mutations ( Asp or Glu ) that alter the length of the amino acid side chain without removing the carboxylate group . If carboxylate moieties are important for TMEM16A-calcium interaction , these mutations may partially preserve TMEM16A-CaCC calcium sensitivity even if the change in side chain length alters the position of the carboxylate group . Second , we made charge-neutralizing mutations ( Asn or Gln ) that remove the charge of the amino acid side chain without changing its size . If carbonyl groups are involved in calcium coordination , these mutations may maintain the orientation of the amino acid residue and preserve the interaction of their carbonyl groups with calcium ions . We found that the shifts in EC50 caused by mutations at E698 , E701 , and D734 were consistent with these acidic residues playing a direct role in calcium coordination ( Figure 4F ) . Charge-preserving mutations in E698D and E701D resulted in only a 3 to 10-fold increase in channel EC50 ( 3 . 3 ± 0 . 9 µM and 10 . 4 ± 3 . 4 µM , respectively ) , suggesting that side chain length at these two positions is not critical for calcium binding , while mutations in D734E shifted the EC50 by about 800-fold to 0 . 8 ± 0 . 3 mM , suggesting that calcium binding may be sensitive to the precise spatial arrangement of the carboxylate group at this position . Although calcium activation was compromised by substituting the acidic residue with a longer side chain , D734E was still more sensitive to calcium than other mutants possessing substitutions at this residue ( Figure 4E ) . In addition , replacing these acidic residues with glutamine or asparagine also increased channel EC50 , suggesting that the negatively-charged carboxylate group is important for calcium-binding . E698Q mutations resulted in a relatively small decrease in calcium sensitivity ( EC50: 0 . 03 ± 0 . 01 mM ) while E701Q and D734N had larger effects ( EC50: 2 . 9 ± 0 . 3 mM and 2 . 3 ± 0 . 3 mM , respectively ) . This data , combined with a dramatic loss of calcium sensitivity in mutants containing arginine , alanine or cysteine mutations at these three sites , supports the idea that side chain carboxylate groups at position E698 , E701 and D734 might be directly involved in binding calcium . Similarly , mutagenesis of E730 suggests that calcium ions may interact with this residue , although not necessarily with its side-chain carboxylate moiety ( Figure 4D , F ) . In this case , charge-preserving aspartate mutations increased EC50 to 0 . 8 ± 0 . 2 mM even though charge-neutralizing glutamine mutations only increased EC50 to 0 . 09 ± 0 . 08 mM . Since channels bearing mutations that preserve the shape but not the charge of this acidic residue are more sensitive to calcium , it appears that TMEM16A-calcium interactions require the proper conformation of residues at this position , possibly to orient the carbonyl oxygen of the peptide backbone . The effects of the charge or size-preserving mutations at E650 further support the notion that this residue may not be directly involved in binding calcium . In contrast to the mild effects of a charge-reversing arginine substitution ( EC50: 6 . 5 ± 0 . 8 μM ) , the charge-preserving aspartate and charge-neutralizing glutamine substitutions increased EC50 to 0 . 8 ± 0 . 2 mM and 4 . 3 ± 0 . 9 mM , respectively , suggesting that neither its carboxylate nor its carbonyl group is involved in calcium coordination . It thus seems likely that E650 might be involved in gating processes downstream of calcium binding while E698 , E701 , E730 , and D734 may interact directly with calcium . This conclusion was reinforced by our study of the effects of E650 mutations on the sensitivity of mutant channels to other divalent cations such as strontium and cadmium ( see below ) . Calcium binding sites often can coordinate other divalent cations that have similar chemical properties or ionic radii similar to those of calcium , such as strontium , barium , and cadmium . Characterizing the selectivity for other divalent cations can provide insight into the composition of a calcium binding site , as shown in previous studies of the cation binding sites involved in the activation of BK channels ( Zeng et al . , 2005; Zhou et al . , 2012 ) . To further test whether E650 , E698 , E701 , E730 , and D734 are involved in divalent cation coordination , we asked whether mutations of these residues also shift the EC50 of TMEM16A channel activation by strontium and cadmium ions . Strontium ions can robustly activate TMEM16A channels ( Yuan et al . , 2013; Ni et al . , 2014 ) even though their ionic radius ( 1 . 13 Å ) is larger than that of calcium ions ( 0 . 99 Å ) . Since both strontium and calcium are ‘hard’ metal ions that prefer coordination with oxygen-containing ligands such as carboxylates and are typically coordinated by six ligands ( Rubin , 1963; Dudev and Lim , 2003 ) , we would expect our various mutations to shift the EC50 of strontium and calcium activation in similar directions . Indeed , we found that the strontium-dependent activity of E698 and E701 mutants were similar to that of their calcium-dependent activity . Compared to the low micromolar strontium sensitivity ( EC50: 7 . 7 ± 0 . 3 μM ) of wildtype channels ( Figure 5A , B ) , alanine mutations at E698 and E701 increased the EC50 of channel activation by strontium to at least the hundred micromolar range ( EC50: 750 ± 70 μM and 7 . 6 ± 1 . 0 mM , respectively ) . The strontium sensitivity of TMEM16A-CaCCs was drastically reduced by charge-reversing arginine mutations ( EC50: >10 mM at both sites ) and was partially preserved by charge-preserving aspartate or glutamate mutations ( EC50: 70 ± 20 μM and 90 ± 10 μM , respectively ) ( Figure 5D–H ) , confirming the importance of side chain charge in divalent cation coordination at these two sites . 10 . 7554/eLife . 02772 . 010Figure 5 . TMEM16A channel sensitivity to strontium ions is disrupted by mutations of the identified calcium-binding sites . ( A ) Representative current trace of wildtype mTMEM16A in response to different intracellular strontium solutions recorded at +60 mV . Table indicates the concentration of strontium used . ( B ) Strontium dose–response curve of wildtype mTMEM16A channels at +60 mV . ( C–G ) Strontium dose–response curves of the ( C ) E650 , ( D ) E698 , ( E ) E701 , ( F ) E730 , and ( G ) D734 mutant mTMEM16A channels at +60 mV . Smooth curves represent fits to the Hill equation . ( H ) Summary of apparent strontium sensitivity ( EC50s ) of mTMEM16A mutants . N . C . : no obvious CaCC current recorded . Upward arrows: estimated strontium EC50 >10 mM and cannot be reported with confidence . p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 02772 . 010 Similarly , side chain charge appears to be important for strontium-dependent activation of D734 mutants . Although we were not able to accurately estimate the true EC50 of any D734 mutants because channel activity was too low to properly normalize current amplitude even at 10 mM strontium , it appears that charge-preserving glutamate mutants are more sensitive to strontium ions than alanine , cysteine , and arginine mutants , which were not activated by strontium at all ( Figure 5H ) . The loss of channel activity in these constructs are not due to improper protein folding or export since these mutants can still be robustly activated by intracellular cadmium ( Figure 6G , H ) . Consistent with our recordings of D734 mutant channels activated by calcium ions ( Figure 4E ) , it appears that although glutamate is able to partially substitute for aspartate at this site , the side chain length as well as the charge of D734 is important for channel activation by hard divalent cations ( Figure 5G10 . 7554/eLife . 02772 . 009Figure 6 . TMEM16A channel sensitivity to cadmium ions is disrupted by mutations at the identified calcium-binding sites . ( A ) Representative current trace of wildtype mTMEM16A in response to different intracellular cadmium solutions recorded at +60 mV . Table indicates the concentration of cadmium used . ( B ) Cadmium dose–response curve of wildtype mTMEM16A channels at +60 mV . ( C–G ) Cadmium dose–response curves of the ( C ) E650 , ( D ) E698 , ( E ) E701 , ( F ) E730 , and ( G ) D734 mutant mTMEM16A channels at +60 mV . Smooth curves represent fits to the Hill equation . ( H ) Summary of apparent cadmium sensitivity ( EC50s ) of mTMEM16A mutants . Upward arrows: estimated cadmium EC50 >10 mM and cannot be reported with confidence . p<0 . 05; p<0 . 01; p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 02772 . 009 ) . Similar to our observations of calcium-dependent channel activation ( Figure 4D ) , TMEM16A’s apparent EC50 for strontium activation seems to depend more on the physical conformation of the side chain than the presence of a carboxylate moiety at residue 730 . Channels containing a charge-preserving aspartate mutation at E730 ( EC50: 0 . 27 ± 0 . 03 mM ) are not much more sensitive to strontium than channels containing a charge-neutralizing alanine mutation ( EC50: 0 . 44 ± 0 . 10 mM ) ( Figure 5F ) . Channels containing the aspartate substitution ( E730D ) had the highest sensitivity for strontium out of the five mutations tested ( Figure 5H ) even though the same mutant channel had the lowest affinity for calcium ( Figure 4F ) . It is conceivable that the smaller aspartate side chain at this position can accommodate the larger strontium ion more easily , supporting its direct interaction with these divalent cations . In contrast to E698 , E701 , E730 , and D734 that display features consistent with their direct interaction with calcium and strontium ions , the effects of the E650 mutations on channel activation by strontium ( Figure 5C ) further support the notion that E650 is not directly involved in metal ion binding , as indicated by their effects on calcium sensitivity ( Figure 4A ) . Among all of the channels with mutations at residue 650 from which we could observe strontium-activated currents , the charge-reversing E650R mutant channel exhibited the highest sensitivity to strontium and was the only substitution at that site that yielded an EC50 for strontium below 10 mM . In contrast , arginine substitutions of E698 and E730 exhibited the lowest strontium sensitivity as compared to other mutations of these residues , and arginine substitutions of E701 and D734 raised EC50 for strontium beyond 10 mM ( Figure 5H ) . This is further evidence that E650 probably plays an indirect role in channel activation by divalent cations and that its side chain is likely involved in an allosteric activation mechanism downstream of metal ion binding . In addition to strontium , we also tested how mutations of these potential calcium binding residues affect the ability of cadmium to activate TMEM16A . Cadmium , with an ionic radius of 0 . 97 Å that is slightly smaller than that of strontium and calcium , is a divalent cation that is able to substitute for calcium and activate wildtype TMEM16A channels ( Figure 6A , B ) . Unlike calcium and strontium ions , however , cadmium is a ‘soft’ ligand typically coordinated by only four residues ( Dudev et al . , 2006 ) , and it prefers coordination by moieties containing nitrogen and sulfur atoms ( Andersen , 1984; Sóvágó and Várnagy , 2013 ) . Reasoning that the TMEM16A calcium-binding site is evolutionarily optimized for ‘hard’ ions like calcium , we hypothesized that the TMEM16A sensitivity for cadmium ions would be low and would not be greatly impacted by mutations that alter the charge of only one acidic residue in the calcium sensor . Indeed , cadmium ( EC50: 940 ± 80 μM ) is much less effective than calcium ( EC50: 1 . 0 ± 0 . 1 μM ) and strontium ( EC50: 7 . 7 ± 0 . 3 μM ) at activating the wildtype TMEM16A channel ( Figure 6A , B ) . TMEM16A channels were poorly activated by 100 μM cadmium even though it reaches its maximum open probability at 100 μM calcium ( Figure 1C ) and 100 μM strontium ( Figure 5B ) . The charge-reversing E650R mutation caused a slight left-shift of the dose response curve of cadmium activation ( Figure 6C ) , in contrast to the right-shift observed with the more conservative size-preserving E650Q mutation , further supporting the notion that this residue does not directly coordinate metal ions but rather works downstream of metal ion binding for channel activation . Although mutations of E698 , E701 , E730 , and D734 shifted the EC50 of cadmium activation , the differences in EC50 between different mutants ( Figure 6D–G ) were much smaller than those observed with calcium and strontium ( Figure 4B–E , Figure 5D–G ) . It is worth noting that the EC50s for cadmium activation of TMEM16A channels containing charge-reversing or charge-neutralizing mutations of these acidic residues were all in the millimolar range , similar to those of wildtype channels and those with charge-preserving mutations ( Figure 6H ) , suggesting that the binding of cadmium ions is not as sensitive as the binding of calcium or strontium ions to side chain perturbations of just one of these putative calcium binding residues . This may reflect the difference in coordination chemistry between the soft cadmium ion and the harder calcium and strontium ions . Since calcium , strontium , and cadmium activation of TMEM16A-CaCC are all affected by mutations of E698 , E701 , E730 , and D734 , it appears that all three divalent cations can interact with the same TMEM16A calcium sensor . Because the effects of each mutation are contingent on the divalent cation species used to activate the channel , it is likely that these sites directly alter the metal ion-binding pocket . In addition , because all of our mutations targeted residues on TMEM16A , our results lend further support to the notion that the TMEM16A polypeptide itself encodes the calcium sensor responsible for channel activation . Our results implicating E698 , E701 , E730 , and D734 as divalent cation coordinating residues in the TMEM16A calcium-sensing domain suggest that these acidic residues are exposed to the cytoplasm . However , due to the high hydrophobicity of the residues in the region between the putative transmembrane segments 5 and 7 , current models of TMEM16A topology ( Model A and Model B in Figure 7C ) do not clearly predict their solvent-accessibility ( Das et al . , 2008; Yang et al . , 2008b; Yu et al . , 201210 . 7554/eLife . 02772 . 007Figure 7 . Cysteine crosslinking suggests that the calcium-binding residues in TMEM16A-CaCC form a metal ion binding pocket that is exposed to the cytoplasm . ( A ) Representative traces of E701C/E730C , E701C , and E730C mTMEM16A mutants recorded under reducing ( DTT ) and oxidizing ( H2O2 ) conditions . ( B ) Comparison of currents recorded in oxidizing conditions of mutants shown in A . When activated for long periods of time , TMEM16A-CaCCs exhibit a persistent decrease in activity , as previously described ( Vocke et al . , 2013 ) . Current amplitudes were measured 60 s after the onset of perfusion and are normalized to currents recorded in reducing conditions . **p<0 . 001 . ( C ) Schematic illustrating the position of the putative calcium binding residues ( E698 , E701 , E730 and D734 ) based on two previous membrane topological models ( Model A and B ) ( Yang et al . , 2008b; Yu et al . , 2012 ) and our experimental data ( Model C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02772 . 007 ) . To confirm that these residues are accessible to internal calcium ions and to demonstrate the spatial proximity of these residues , we tested whether the activity of TMEM16A channels containing double cysteine mutations can be modulated by the redox potential of the internal solution . In support of a TMEM16A topological model shown as Model C in Figure 7C where E698 , E701 , E730 , and D734 are clustered together and facing the cytoplasm , the perfusion of internal solutions containing redox chemicals and 300 μM calcium altered the amplitude of chloride currents through channels containing an E701C/E730C double mutation but not those containing an E701C or an E730C single mutation . Channel activity in oxidizing conditions was significantly smaller than those recorded in reducing conditions , suggesting that formation of a disulfide bridge between E701C and E730C decreases TMEM16A-CaCC activation by calcium ( Figure 7A , B ) . Thus , these residues appear to be spatially clustered at a cytoplasm-accessible site . Calcium is the physiological stimulus that activates calcium-activated chloride channels ( CaCCs ) . Studying how calcium activates the newly discovered TMEM16A-CaCC channel will help us understand its physiological functions . Thus far , the molecular nature of the TMEM16A calcium sensor has been under extensive debate . One theory asserts that a ubiquitous calcium sensing protein , calmodulin ( CaM ) , is responsible for calcium activation of the CaCC ( Tian et al . , 2011; Vocke et al . , 2013 ) , resembling the mechanism of the small-conductance KCa ( SK ) channels . A competing theory asserts that calcium ions can directly bind to the channel itself without the involvement of CaM ( Yu et al . , 2012 , 2014; Terashima et al . , 2013 ) , analogous to the mechanism of the large-conductance KCa ( BK ) channels . To settle this debate and to advance the molecular understanding of how calcium activates TMEM16A , we performed a comprehensive screen of evolutionarily conserved intracellular acidic residues in the TMEM16A protein and mapped its calcium sensor to a region between transmembrane segments 5 and 7 . By manipulating the charge , size , and reactivity of these side chains , we characterized the mutant channels' sensitivity and selectivity for different divalent cations and showed that four critical acidic residues ( E698 , E701 , E730 , and D734 ) in this region are situated at a solvent-accessible site that likely interacts with calcium in the cytoplasm during channel activation . In contrast to previous studies of TMEM16A calcium-dependent activation based on whole-cell recordings , we used the inside-out patch clamp to quantify the apparent calcium sensitivity ( EC50 ) of the channel in this study . Monitoring channel opening in response to calcium applied to the inside-out patch and comparing channel EC50 values is a more direct measurement of calcium-dependent channel activation compared to the methods used in other studies ( Jung et al . , 2013; Terashima et al . , 2013; Vocke et al . , 2013 ) . Although recent electrophysiological studies of TMEM16A-CaCCs using the whole-cell patch clamping configuration provided valuable insights toward understanding channel activation ( Yu et al . , 2012; Jung et al . , 2013; Vocke et al . , 2013 ) , the difficulty in temporally and spatially controlling intracellular calcium concentrations inherent to the whole-cell patch clamp method renders the quantification of calcium sensitivity difficult . By using inside-out patch recordings to accurately assess channel sensitivity to divalent cations and exploiting differences in the coordination chemistry of hard and soft ligands to deduce the functional role of individual acidic residues , our study based on a mutagenesis screen of acidic residues provides a comprehensive survey of the potential high affinity calcium binding sites within the TMEM16A protein . Our study is in agreement with the conclusions of Terashima et al . ( 2013 ) and Yu et al . ( 2014 ) that calmodulin is not involved in the calcium-dependent activation of TMEM16A channels . Similar to these studies , we were unable to detect alterations of TMEM16A channel activity by manipulating calmodulin function . Moreover , mutations of four putative calcium binding residues E698 , E701 , E730 , and D734 differentially shifted the EC50 of channel activation to different calcium concentrations depending on the size and charge of the amino acid side chains , suggesting that these residues are likely involved in binding calcium . Furthermore , the differences in the way these mutations affect channel sensitivity to the hard calcium and strontium ions and the softer cadmium ions suggest that these mutations directly impact channel–ion interactions rather than acting via an allosteric mechanism . It remains possible that CaM may physically interact with TMEM16A channels under certain conditions as evidenced by previous biochemical studies ( Jung et al . , 2013; Vocke et al . , 2013 ) . Nevertheless , this interaction , if present , is not responsible for the calcium-dependent activation of TMEM16A-CaCC channels . It seems likely that calcium directly binds to and activates TMEM16A-CaCC channels in a fashion reminiscent of the calcium activation mechanism of the large conductance , calcium-activated BK potassium channel , one of the most well-characterized calcium-activated ion channels . Both channels contain calcium binding sites that are composed of multiple intracellular acidic residues , and substitution mutagenesis studies of both channels reveal that side chain properties are critical to the binding of calcium and other metal ions such as strontium , barium , and cadmium . Removing carboxylate groups from side chains dramatically reduces or eliminates the ability of both channels to bind calcium , while preserving carboxylate groups with charge-conserving mutations tends to partially rescue calcium sensitivity . Structural and functional characterizations of BK channels over the past three decades have generated a wealth of information that provides valuable inspiration to the study of TMEM16A-CaCC channels . There are three distinct metal binding sites in the pore-forming subunit of BK channels ( Schreiber and Salkoff , 1997; Shi et al . , 2002; Xia et al . , 2002 ) . Two high affinity binding sites ( micromolar ) correspond to the calcium bowl in the RCK2 domain ( Schreiber and Salkoff , 1997 ) and the acidic residue D367 in the RCK1 domain ( Shi et al . , 2002; Xia et al . , 2002; Zhang et al . , 2010 ) , and a low affinity site ( millimolar ) that can nonspecifically bind various divalent cations corresponds to residues from the membrane spanning domain and the cytosolic RCK1 domain ( Shi et al . , 2002; Xia et al . , 2002; Yang et al . , 2008a ) . These three metal ion binding sites act relatively independently to activate BK channels as mutating any one of them has a minimal impact on the other two sites . Extensive mutagenesis along with characterization of divalent cation selectivity has elucidated the properties of the different metal ion binding sites ( Zeng et al . , 2005; Zhou et al . , 2012 ) . Micromolar calcium and strontium ions can activate the channel through both the calcium bowl and the RCK1 high affinity sites but not through the low affinity site . Micromolar cadmium only activates the channel via the RCK1 high affinity site; even 300 μM cadmium cannot act on the calcium bowl site to activate the channel . In contrast , 200 μM barium has no effect on the RCK1 site ( Zhou et al . , 2012 ) . Instead , barium ions activate the channel through the calcium bowl site with five times lower affinity than calcium . On the other hand , millimolar concentrations of all these divalent cations can activate the channel through the low affinity metal ion binding site ( Zeng et al . , 2005 ) . For TMEM16A-CaCC , the apparent calcium ( EC50: 1 . 0 ± 0 . 1 μM ) and strontium ( EC50: 7 . 7 ± 0 . 3 μM ) sensitivities are similar to that of the high affinity cation binding sites in BK channels . However , the TMEM16A channel is much less sensitive to barium ( EC50: 330 ± 30 μM ) and cadmium ( EC50: 940 ± 80 μM ) than BK channels , which responds to 10 μM barium and cadmium via the calcium bowl and the RCK1 high affinity site , respectively . This difference suggests that the high affinity calcium binding site ( s ) in the TMEM16A channel might be less effective in binding barium and cadmium ions due to their less optimal coordination chemistry . Alternatively , barium and/or cadmium might activate the TMEM16A channel via an unknown low affinity cation binding site , which remains latent under physiological conditions . This scenario seems plausible as none of our high affinity site mutations abolishes cadmium activation ( Figure 6H ) . The identity of this putative low affinity , nonselective cation binding site and its potential involvement in TMEM16A channel activation require further study . By following up on our comprehensive alanine mutagenesis with multiple point mutations of the putative calcium binding residues and testing their effects on channel activation by different divalent cations , we were able to take advantage of the well-studied chemical properties of divalent cation coordination to analyze these putative calcium binding sites in detail . After these candidate acidic residues were positively identified , conservative mutations preserving either the negative charge ( glutamate to aspartate or vice versa ) or the approximate side chain volume ( glutamate to glutamine , or aspartate to asparagine ) were expected to have relatively minor effects on calcium coordination in actual binding sites , considering that either the electrostatic or steric side chain properties would be preserved under those conditions . Conversely , introduction of positively charged arginine side chain would have been expected to produce much more drastic effects . Indeed , this pattern was observed for four of the five acidic residues identified in the mutagenesis screen . The calcium sensitivity of E698 , E701 and D734 was drastically reduced by arginine substitutions , in contrast to the milder effects caused by conservative substitutions . Mutations of E730 followed the same pattern consistent with a divalent ion binding site , though with generally less drastic effects . The E650R mutant , on the other hand , retained the highest sensitivity to calcium of all substitutions at that site , whereas conservative substitutions greatly impacted calcium sensitivity . These results suggest that E650 is unlikely to interact directly with a calcium ion , at least not in the same manner as those of the other four , and we have thus excluded it from our proposed model of the primary calcium interaction site ( Figure 7C ) . Given the possibility that E650 may be involved in the transduction of conformational changes associated with calcium binding to trigger the channel's gating machinery , it will be important to further scrutinize its functional role in future studies . Our analyses using different divalent cations to activate various TMEM16A mutants are also consistent with the notion that E698 , E701 , E730 and D734 , but not E650 , are involved in direct interactions with strontium . Strontium is a divalent ion with chemical properties similar to calcium , as it has two electrons in its outer valence shell and is considered a ‘hard’ ligand ( Rubin , 1963; Dudev and Lim , 2003 ) . Although strontium has a slightly larger ionic radius than calcium , it has been shown previously to robustly activate TMEM16A channels ( Yuan et al . , 2013 ) . The effects of side chain substitutions at those four sites on calcium and strontium sensitivity showed similar trends . In contrast , E650 mutants were generally insensitive to strontium activation , and the mutant displaying the highest sensitivity was again E650R , in keeping with our findings regarding calcium activation . In addition to strontium , we also employed cadmium as an alternative TMEM16A activator . Cadmium is among the transition metals and acts as a ‘soft’ ligand , but has an ionic radius nearly identical to that of calcium ( Sóvágó and Várnagy , 2013 ) . Here , we found that while the wildtype channel was far less sensitive to cadmium ions than calcium ions , the apparent cadmium sensitivities of the mutant channels were insensitive to the side chain manipulations and remain in the submillimolar to millimolar range . In our view , there are at least two plausible explanations for this phenomenon . Firstly , as cadmium ions can be adequately coordinated by fewer negatively charged groups , alterations of just one of the interacting acidic residues may have less drastic effects on cadmium coordination . Alternately , as mentioned previously , it is possible that a second , lower affinity and/or less calcium-selective binding site may exist in the channel that can more readily bind cadmium following mutation in the higher affinity site identified in this study . At present , our results do not allow us to distinguish between these two possibilities , particularly as structural and functional studies of TMEM16A gating are still in their infancy . Nevertheless , the identification of residues forming a high affinity binding site , and our finding that cadmium can activate TMEM16A mutants rendered insensitive to calcium , provide key tools for the future study of channel gating processes downstream of TMEM16A’s high affinity binding of divalent cations . Unlike the high affinity sites of BK channels that are located at the large cytosolic C-terminus ( Salkoff et al . , 2006 ) , the high affinity calcium binding residues ( E698 , E701 , E730 , and D734 ) of the TMEM16A channels are between the putative transmembrane ( TM ) segments 5 and 7 ( Figure 7C ) . This is consistent with recent findings ( Yu et al . , 2012; Scudieri et al . , 2013 ) that the third intracellular loop ( between TM6′ and TM7 , Figure 7C , Model B ) is important for the calcium sensitivity of the TMEM16A and TMEM16B CaCC channels . Previous studies ( Yang et al . , 2008b; Yu et al . , 2012 ) have proposed two different models of membrane topology for TMEM16A . The earlier model ( Model A in Figure 7C ) includes a long extracellular or reentrant loop between TM5 and TM6″ ( Yang et al . , 2008b ) , while a subsequent model ( Model B in Figure 7C ) proposes that the peptide between TM5 and TM6″ in Model A actually includes one transmembrane segment ( TM6′ ) ( Yu et al . , 2012 ) . Our current study provides new insight into the membrane topology of this region by showing that residues 701 and 730 are accessible to the intracellular solvent and are in close proximity such that a disulfide bridge can form between E701C and E730C to impact calcium activation of the channel . This functional study implies that the hydrophobic region spanning 28 residues ( including the putative TM6″ ) between E701 and E730 likely forms a short reentrant loop ( Model C in Figure 7C ) , bringing together these four acidic residues at both ends of this loop to coordinate calcium ions . Further structural information is needed to validate this model . In summary , our comprehensive survey of evolutionarily conserved acidic residues has identified several critical residues in the TMEM16A-CaCC that are responsible for its activation by calcium . Our study provides further evidence that the TMEM16A-CaCC channel directly interacts with intracellular calcium without involving CaM . Interestingly , these putative calcium binding acidic residues are highly conserved among proteins in the TMEM16 family , some of which have been shown to be important in cellular processes ranging from mucus secretion ( Huang et al . , 2012b ) to blood coagulation ( Suzuki et al . , 2010; Yang et al . , 2012 ) in mammals and host defense in Drosophila ( Wong et al . , 2013 ) . The identification of these evolutionarily conserved acidic residues that bind calcium in the TMEM16A-CaCC channel will contribute towards a general understanding of the molecular mechanisms of calcium-activated TMEM16 channels as well as their cellular functions in response to calcium signaling in various cell types and organisms . cDNAs ( Open Biosystems cDNA clones number 30547439 , Uniprot identification number Q8BHY3-2 ) derived from mouse TMEM16A ( mTMEM16A ) were subcloned into the pEGFP-N1 vector ( Clontech , Mountain View , CA ) via standard molecular biology techniques . Our clone corresponds to the ‘a’ splice form identified in Caputo et al . ( 2008 ) and lacks alternative exons b , c , and d . Site-directed mutations were generated by PCR with Pfu Turbo DNA polymerase following the Quikchange protocol from Agilent . All mutants were verified by sequencing . HEK 293 cells were cultured in Dulbecco's Modified Eagle Medium ( DMEM ) supplemented with 4 . 5 g/l D-glucose , 110 mg/l sodium pyruvate , 584 mg/l L-glutamine , and 10% fetal bovine serum ( FBS ) and were transfected with Lipofectamine 2000 ( Invitrogen , Carlsbad , CA ) , FuGENE 6 ( Promega , Madison , MI ) , or X-tremeGENE ( Roche , Switzerland ) and cultured for 1–2 days before recording . For some experiments testing the effects of calmodulin on channel function , endogenous CaCC currents were recorded from Xenopus oocytes without exogenously introducing TMEM16A . Female Xenopus laevis were purchased from Nasco ( Fort Atkinson , MI ) . The procedures for harvesting oocytes and housing animals were approved by the UCSF Institutional Animal Care and Use Committee . The loss-of function calmodulin mutants were kindly provided by Dr John Adelman ( Xia et al . , 1998 ) . Defolliculated oocytes were injected with 5–100 ng cRNA and maintained at 17°C in ND96 ( 96 mM NaCl , 10 mM HEPES , 2 mM KCl , 1 mM MgCl2 , pH 7 . 4 ) solution for 2–7 days before recording . The monoclonal anti-CaM antibody ( CaM85 ) and the CaM antagonist W7 ( N- ( 6-Aminohexyl ) -5-chloro-1-naphthalenesulfonamide hydrochloride ) were purchased from Invitrogen and Santa Cruz Biotech ( Santa Cruz , CA ) , respectively . For chronic treatments of W7 , Xenopus oocytes were incubated in ND96 supplemented with 50 μM W7 for 2–6 days before patch recording . 24 hr following transfection , cells were transferred to our standard recording bath solution ( described below ) for inside-out patch clamp recording . Macroscopic currents were recorded from inside-out patches formed with borosilicate pipettes of 1–5 MΩ resistance . Data were acquired using Axopatch 200-B and Axopatch 700-B patch-clamp amplifiers and pClamp10 software ( Molecular Devices , Sunnyvale , CA ) . All experiments were performed at room temperature ( 22–24°C ) . Unless otherwise stated , all solutions used in this study were based on isotonic 140 mM NaCl . Both the basal extracellular solution and the zero calcium intracellular solution contained 140 mM NaCl , 5 mM EGTA , and 10 mM HEPES . Pipette ( extracellular ) solutions were supplemented with 1 mM MgCl2 . Internal solutions with various calcium concentrations ( <100 μM ) were prepared with the pH-metric method ( Tsien and Pozzan , 1989 ) . Briefly , a ‘high calcium’ solution ( 140 mM NaCl , 5 mM Ca-EGTA , 10 mM HEPES ) and a zero calcium intracellular solution were mixed in different ratios to give various calcium concentrations . The basal internal solution ( without calcium buffer ) contained 140 mM NaCl , and 10 mM HEPES . For solutions with [Ca2+]i ≥ 100 μM , CaCl2 was directly added to the basal internal solutions ( without EGTA ) and the free [Ca2+]i was measured with a Ca2+-sensitive electrode ( Thermo Scientific , Waltham , MA ) . The pH of all solutions was titrated with N- methyl-D-glucamine ( NMDG ) or NaOH to 7 . 2 . To test channel activation by Sr2+ or Cd2+ , inside-out patches were exposed to a series of Sr2+ or Cd2+-containing solutions prepared by serial dilution of a base solution containing 140 mM NaCl , 10 mM HEPES , and 10 mM SrCl2 or CdCl2 for a range of divalent concentrations from 10 μM to 10 mM . While most constructs required greater than 10 μM of either divalent to be activated , wildtype TMEM16A required solutions with lower [Sr2+] . As Sr2+ concentration cannot be accurately diluted below 10 μM , Sr2+-EDTA solutions were prepared based on the CaBuf divalent buffering prediction software created by Dr G Droogmans ( Department of Physiology , KU Leuven , Leuven , Belgium ) . For some constructs , 10 mM cation was insufficient to fully activate the channel , and the EC50 could not be calculated with confidence and is not reported . For experiments manipulating the redox potential of the internal solution , 10 mM H2O2 or 10 mM DTT ( dithiothreitol ) from Sigma ( St . Louis , MO ) was freshly added ( ≤1 hr prior to use ) to internal solutions containing either 0 μM or 300 μM free calcium . Data analysis was performed with Clampfit 10 ( Molecular Devices ) and Origin 7 . 5 ( OriginLab , Northampton , MA ) . Concentration dose–response curves were fit to an equation of the form:I/IMAX=Amp1+ ( KD[Ca] ) Hwhere I denotes current , IMAX is the maximum current elicited by the highest concentration of divalent cation , Amp is the maximum value of I/IMAX at a given voltage , KD is the apparent dissociation constant , and H is the Hill coefficient . EC50 values were log-transformed for one-way ANOVA and were compared to wildtype values using Tukey's post-hoc test for significance . Values were considered significantly different if p<0 . 05 . For some mutant channels whose cation sensitivity was greatly reduced , the current did not reach a plateau even as the cation concentration was raised to 10 mM . In these cases , the EC50 values derived from the dose response curves may underestimate the ‘true’ sensitivity of the mutant channels to these metal ions .	Every cell in the body is surrounded by a barrier called the cell membrane . There are , however , a number of ways that molecules can pass through this membrane to either enter or leave the cell . Calcium-activated channels are a group of proteins that are embedded within the cell membrane and that allow different ions to pass through the membrane . These proteins are involved in a number of processes in a variety of tissues , for example in the gut , lungs and nervous system . A family of proteins called TMEM16 includes a number of calcium-activated channels that have been recently identified . However , it is not clear how these TMEM16 channel proteins detect the calcium ions that cause them to open . Two ideas have been suggested: the calcium ions might be detected by a protein called calmodulin , which then forces the channel to open; alternatively , the calcium ions might be detected by the channel protein itself . Tien , Peters et al . have now tested both of these ideas by focusing on a calcium-activated channel protein called TMEM16A , which allows chloride ions to pass through membranes . The possible role of calmodulin was tested in several ways , such as by preventing it from binding to the TMEM16A protein or from binding to calcium . However , none of these changes affected the opening of the channel; so Tien , Peters et al . concluded that calmodulin is not involved in these channels being activated by calcium ions . Next , Tien , Peters et al . tested specific parts of the TMEM16A channel protein itself to see if they were involved in calcium detection instead . Proteins are made from smaller building blocks called amino acids , and it is known that some amino acids are more likely to bind to calcium ions than others . There are 38 of these amino acids in the TMEM16A channel that are also found in other members of the TMEM16 family in both fruit flies and mammals . Tien , Peters et al . found that replacing five of these with other amino acids made the channel less sensitive to calcium . Further experiments suggested that four of these five amino acids are clustered at the site where a calcium ion might bind to the TMEM16A channel protein , which suggests that the protein itself can detect calcium directly . The next challenge will be to understand how calcium ions binding to the site on the TMEM16A channel protein can cause the channel to open to allow the chloride ions to pass through .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2014	A comprehensive search for calcium binding sites critical for TMEM16A calcium-activated chloride channel activity
Interaction between disease-microbiome associations and ageing has not been explored in detail . Here , using age/region-matched sub-sets , we analysed the gut microbiome differences across five major diseases in a multi-cohort dataset constituting more than 2500 individuals from 20 to 89 years old . We show that disease-microbiome associations display specific age-centric trends . Ageing-associated microbiome alterations towards a disease-like configuration occur in colorectal cancer patients , thereby masking disease signatures . We identified a microbiome disease response shared across multiple diseases in elderly subjects that is distinct from that in young/middle-aged individuals , but also a novel set of taxa consistently gained in disease across all age groups . A subset of these taxa was associated with increased frailty in subjects from the ELDERMET cohort . The relevant taxa differentially encode specific functions that are known to have disease associations . Alterations in the gut microbiome have been reported for many diseases ( Duvallet et al . , 2017; Pasolli et al . , 2016 ) . Identifying the specificity of these alterations is necessary for developing microbiome-based diagnostic and therapeutic strategies . In addition to investigating the aspect of causality , three key aspects of microbiome-disease alteration need to be clarified ( Duvallet et al . , 2017; He et al . , 2018 ) . The first aspect involves the disease-associated components ( key taxa and their metabolic functions ) which would help in establishing diagnostics and understanding the mechanisms by which the altered microbiome affects the host . The second aspect concerns the directionality of the associations which is required for designing therapeutic strategies ( e . g . targeted antimicrobials versus microbiome restoration ) . The third aspect is the extent of shared versus disease-specific microbiome alterations that indicate what effect the onset of a given disease may have on the microbiome-related risk of other diseases . Meta-analyses across cases and controls from different geographical locations have identified the components of the microbiome alterations common to multiple diseases , as well as disease-specific alterations ( Duvallet et al . , 2017; He et al . , 2018; Jackson et al . , 2018; Pasolli et al . , 2016 ) . While certain diseases like colorectal cancer ( CRC ) are characterized by an increase ( or gain ) of pathobionts ( such as Fusobacterium , Porphyromonas , Parvimonas ) , the onset of others like Inflammatory Bowel Disease ( IBD ) is associated with the depletion of specific taxa ( e . g . Roseburia , Faecalibacterium ) ( Duvallet et al . , 2017 ) . In contrast , diarrhoeal diseases are accompanied by both an increase of pathobionts ( specifically Enterobacteriaceae ) as well as lower abundance of commensal taxa . Recent meta-analyses have also shown a surprising degree of overlap of microbiome associations ( both at the level of specific taxa as well as specific microbial metabolic pathways ) ( Armour et al . , 2019; Duvallet et al . , 2017; Pasolli et al . , 2016 ) . Despite these overlaps , links between the microbial taxa and the functional pathways that are consistently altered across multiple diseases have not been explored in detail . Microbiome-based diagnostics/therapeutics also need to account for several host factors . Variations in region/ethnicity of subjects are linked to the gut microbiome composition and microbiome-based diagnostics in various diseases ( Deschasaux et al . , 2018; He et al . , 2018 ) . In addition to region/ethnicity-specific variations , multiple studies have also shown age to be a strong covariate of microbiome composition ( Falony et al . , 2016 ) . We , along with others , have previously identified ageing-associated microbiome alterations in the ELDERMET cohort , associated with lower complexity dietary intake and poly-pharmacy ( Claesson et al . , 2012; O'Toole and Jeffery , 2015; Ticinesi et al . , 2017 ) . Besides ageing-associated changes in the microbiome , early and late-onset variants of different diseases are characterized by distinct patho-physiologies ( Duricova et al . , 2014; Yeo et al . , 2017 ) , which could also be linked to age-related microbiome involvement . These observations motivated us to perform an extensive investigation of microbiome-disease alterations across the human age landscape . We aimed to answer the following questions: To what extent does ageing influence microbiome-disease associations for different diseases ? Do different diseases show differential microbiome-signatures ( both in terms of the microbiome components as well as the directionality of the changes ) across age-groups ? If so , to what extent do these changes affect the known list of microbiome-based markers for the different diseases ? Is there a change in the pattern of multiple disease-associated taxa ( as identified by previous meta-analyses ) across age groups ? And lastly , is it possible to identify a link between these disease markers and the microbial metabolic pathways previously associated with these diseases ? Investigating microbiome alterations in multiple diseases across a wide age-range requires a comprehensive dataset of consistently collated microbiome profiles . The curated MetagenomicData repository of the ExperimentHub R library contains taxonomic and functional pathway profiles of more than 6000 shotgun-sequenced human microbiome samples from different human body sites , spanning more than 30 diseases ( and controls ) from more than 25 different studies ( Pasolli et al . , 2017 ) . A further advantage of the curatedMetagenomicData is that these microbiome profiles were created using uniform bioinformatic analysis of the sequenced reads ( using metaphlan2 and humann2 ) ( Franzosa et al . , 2018; Truong et al . , 2015 ) . The repository also provides extensive metadata information for the samples including the study , experimental protocols , region ( or country ) , disease status , age , gender , BMI and antibiotic usage . Although other key factors affecting gut microbiome composition like diet or medication are not provided for all studies , the repository can still facilitate meta-analysis of datasets from multiple studies , to provide initial insights into the effect of other key host associated factors on microbiome disease signatures . Here we analysed the gut microbiome in 15 studies encompassing four diseases , comprising more than 2500 individuals ranging from 20 to 89 years of age , derived from the ExperimentHub repository , the ELDERMET project and recently published studies on IBD ) and colorectal cancer CRC ( Franzosa et al . , 2019; Thomas et al . , 2019; Wirbel et al . , 2019 ) . We first adopted a stepwise methodology to reduce the confounding effects of DNA sequencing/extraction methodologies on the microbiome profiles from the different studies ( See Materials and methods; Figure 1—figure supplement 1 ) , and subsequently retained samples from individuals with age in the range of 20–89 years ( excluding cohorts where the ‘controls’ also consisted of hospitalized patients ) ( Vincent et al . , 2016 ) . We supplemented this data set with 475 shotgun metagenome profiles from recently published studies on IBD ( Franzosa et al . , 2019 ) and CRC ( as ‘Validation’ cohorts ) ( Thomas et al . , 2019; Wirbel et al . , 2019 ) , finally assembling a collated set of microbiome taxonomic profiles from more than 2500 samples ( including the 189 ELDERMET samples used later in the study ) ( see Materials and methods; Supplementary file 1 and Supplementary file 2; Figure 1—figure supplement 2 ) . We next investigated the interaction of metadata with taxonomic profiles . After filtering out redundant and sparse metadata types ( recorded for less than 30% of the samples ) , we performed PERMANOVA linking the effects of each metadata with the gut microbiome accounting for DNA extraction technique as a confounder ( See Materials and methods; DNA extraction technique was still treated as a confounder as it still had a minor effect , R2 = 0 . 019 , on the gut microbiome composition ) . Regional factors namely country and continent had the largest interaction with gut microbiome composition ( Figure 1A ) . Regional factors reflect the ethnicity and other socio-economic properties of the study populations , which has a dominant effect on gut microbiome architecture and microbiome-based disease signatures ( Deschasaux et al . , 2018; He et al . , 2018 ) . Age followed by study-condition ( disease versus control status ) were the second major effect . We thus explored the variation of the apparently ‘healthy’ microbiome across the age landscape using Principal Component Analysis of 1175 gut microbiome profiles from exclusively ‘control’ individuals from the age-groups’ 20–39’ ( categorized as ‘Young’ ) , ’ 40–59’ ( ‘Middle-Age’ ) and’ above 60’ ( ‘Elderly’ ) . PERMANOVA analysis of the gut microbiome composition ( after adjusting for country and the DNA extraction technique ) indicated that the elderly controls had a significantly distinct microbiome composition compared to the young and the middle-aged ( p<0 . 001 ) ( Figure 1B ) , in line with our previous findings ( Claesson et al . , 2012 ) . To further explore this effect , we then investigated the effect of subject age on the gut microbiome composition in five diseases , IBD , type II diabetes ( T2D ) , intestinal polyps , CRC , and liver cirrhosis , for which substantial number of cases were available across at least two age-groups . Samples corresponding to different diseases were available in different cohorts ( Table 1 ) . Each cohort dataset comprised healthy controls and diseased individuals matched with respect to geographical region and environment ( thereby reducing these confounding effects on the gut microbiome ) , so we were able to perform PERMANOVA to quantify the effect of both age group ( ‘Young’ , ‘Middle-Aged’ and ‘Elderly’ ) and disease status ( ‘Control’ and ‘Disease’ ) on the gut microbiome ( separately in each of the different disease-specific cohorts ) . For all five diseases , age had a significant effect . Notably , for three diseases ( T2D , CRC and intestinal polyps ) , this effect was higher than that of the disease itself ( Figure 1C ) . Next , we investigated whether these age-related changes in microbiome composition influenced disease signatures . However , performing this investigation posed a specific challenge . Given that regional factors like country or continent had the highest effect on the gut microbiome composition , any investigation into age-group-specific microbiome signatures for the various diseases required the comparisons to be performed within geographically homogeneous sub-populations ( or study cohorts ) to ensure that any differences in disease signatures were not simply driven by regional variations in gut microbiome composition . However , focusing on specific disease cohorts ( which belong to specific countries ) also imposes limitations including reduction in sample size ( and therefore statistical power ) and bias for certain age-groups . Among the eight disease cohorts ( in the curatedMetagenomicData ) that we analyzed here , we discovered that half of these featured a significant difference in the age of the control and diseased individuals ( i . e . they were not age matched ) ( Figure 1D ) . In addition , there were skews in the representation of specific age-groups for most of the disease cohorts . Adding ‘control’ samples from other cohorts that were collected from the same region ( country or continent ) as the corresponding disease cohorts was expected to circumvent this issue by increasing the number of available samples and thereby the statistical power of comparisons . We addressed this issue by grouping the samples into disease-specific country-/continent-level bins ( See Materials and methods; Figure 1—figure supplement 3 ) . This ensured sufficient representation of samples from the three different age-groups for all diseases . Next , we performed investigations using two different approaches to explore the effect of age on disease-microbiome signatures . In the first analysis , within each disease-specific country-level cohort , we performed PERMANOVA investigating the effect of the interaction between disease signatures and age-group , after adjusting for the effects of country ( given that there were samples from different countries within the different disease cohorts ) and the independent effects of disease and age-group ( Table 2 ) . Interaction between two metadata categories measures the extent to which the microbiome variation with respect one metadata ( in this case , the disease ) is influenced by the variation in the other ( that is the age-group ) . For three of the diseases ( IBD , CRC and Polyps ) , the effect of the interaction of disease with age-group was significant ( all with p<0 . 05; PERMANOVA after adjusting for the confounders as above ) . The effect was also marginally significant for T2D ( p<0 . 08; PERMANOVA ) . This indicated that ( even after considering the regional variations and the individual influences of disease and age ) , for these diseases , the microbiome variation associated with disease was significantly influenced by the variations in the age-group of the individuals . Further investigation of these influences ( measured by the R2 ) also indicated that , while IBD and CRC had the highest influence of age as a covariate of disease signatures , with T2D and Polyps relatively lower but significant influences ( Table 2 ) . In the second analysis , we utilized Random Forest ( RF ) Classifier Models . Random Forest models belong to a category of ensemble-based machine learning methods ( based on decision trees ) , that can be used to predict a given trait ( in this case , disease status ) based on a list of predictor features . Random Forests can not only be used to judge the strength of the association between the predictors and the response ( based on the accuracy or area under the curve ( AUC ) of prediction ) , but also whether associations identified on one set of observations are extrapolatable to another . Consequently , Random Forest models have been routinely used in microbiome based studies to not only quantify and characterize the microbiome associations with various diseases ( Feng et al . , 2015; Karlsson et al . , 2013; Pasolli et al . , 2016; Qin et al . , 2012 ) , but also to study the transferability of the associations predicted in set of individuals on others ( He et al . , 2018; Thomas et al . , 2019 ) . We specifically focused on the disease-specific country-level bins ( to ensure regional homogeneity as much as possible ) . Subsequently , we divided subjects ( in the country level cohorts corresponding to each disease ) into three age bands ( as described above ) and performed 100 iterations , each time training Random Forest classifiers on a subset of samples belonging to an age band , and evaluating the disease classification performance on the samples from the same age band ( excluding the samples used for training ) ( Same Age-group classification ) or the other age bands ( Different Age-group classification ) ( while keeping the training and testing data set sizes constant across age-bands ) ( see Figure 2—figure supplement 1 and Materials and methods for details ) . In summary , in each disease-age-group scenario , we used two different approaches to assess whether disease classification performance was significantly different between Same Age-group classification and Different Age-group classification . In the first approach , for the 100 iterative re-sampled classifier models , we compared the AUCs obtained for the Same age-group classification with those obtained for the Different Age-group classification using paired Wilcoxon Signed Rank tests . In the second approach , using Permutation tests , for the 100 iterative re-sampled classifier models , we compared if the actual difference in AUC for the age-group cohorts ( Same Age-group classification – Different Age-group classification ) is significantly different from what would be expected by random ( null distribution generated by permuting the age-group labels of the subjects ) ( Figure 2—figure supplement 1 and Materials and methods ) . The data revealed large differences across age groups for the various diseases . Specifically , in 10 of the 13 disease-age group scenarios ( covering five diseases ) , classifiers trained and tested on the same age-group had significantly higher disease prediction AUC than when tested on different age-groups ( Figure 2 ) , with the improvement of classification performance significantly higher than would be expected by random chance ( obtained using the permutation test based strategy ) ( Figure 2—figure supplement 2 ) . This confirmed that microbiome-disease associations had age-centric trends . Furthermore , the extent of the differences was pronounced for IBD , CRC and T2D ( the above trend being reflected across all age-groups ) , reflecting the patterns observed in the PERMANOVA analysis ( Table 2 ) . CRC and Polyps were characterized by noticeably similar age-specific trends wherein the elderly age-groups had a noticeable decrease in classification AUCs ( that is lower AUCs for classifiers trained or tested on the elderly age-groups ) , indicating that the microbiome-disease signatures for these diseases in the elderly age-groups are weaker ( Figure 2 ) . For these two diseases , the AUC difference between the Same Age-group and Different Age-group disease classification was also reduced in the elderly ( in the case of Polyps , this difference was not significantly different what would be expected by random chance ) ( Figure 2—figure supplement 2 ) . We next sought to identify differentially associated age-disease markers and how these corresponded with the currently known microbiome-disease associations . Besides quantifying the strength and extrapolatable nature of the microbiome signatures , iterative RF models also provided the list of taxa contributing to the models along with their relative importance scores ( for each disease age-group scenario ) . Therefore , in the first step , based on the RF analysis ( as described above ) for each disease , we computed the age-group specific feature importance scores for each taxa and then shortlisted the taxa having marker scores in the top 85 percentile individually for each of the age-groups ( See Materials and methods; Figure 3—figure supplement 1; Figure 3—source data 1 ) . Comparing the top 85 percentile markers across the different age-groups revealed that the proportion of age-influenced markers ( or markers differentially associated across at least one age-group ) ranged from 30% in CRC to 50% in IBD ( greater than 35% for the other three ) ( Figure 3—source data 1; Figure 3—figure supplement 2 ) . The feature importance scores ( obtained using the iterative RF models ) were then compared across age-groups and taxa having significant differences in the feature importance scores ( across age-group ) ( Benjamini FDR < 0 . 01 Kruskal-Wallis test for IBD , T2D and Cirrhosis and Mann-Whitney Tests for CRC and Polyps ) were then identified ( Figure 3—source data 1 ) . These results indicate that the microbiome associations with disease have both an age-group-specific and an age-group-independent component . However , many of the age-group-specific changes in microbiome-disease association may also be reflections of changes that accompany ageing in general . Age had the second major effect on microbiome composition , and it was important to investigate these age-specific changes in disease-microbiome associations after deconvoluting for the effect of ageing . Secondly , while grouping samples into country-level bins ensured enough regional homogeneity of the compared groups , there were still biases in the representation of certain regions in certain age-groups for the different diseases ( Figure 1—figure supplement 3 ) . Given that RF models cannot intrinsically adjust for these confounding effects , we probed this using a linear regression-based strategy that compared the extent of influence that age-group had on the disease-association pattern of a taxon ( Disease:Age-group ) as compared to the individual influences of disease and age-group ( see Materials and methods; Figure 3—figure supplement 3 ) . We specifically investigated those taxa having significant differences in their feature importance scores ( Figure 3—source data 1 ) and identified those having significantly higher influence of Disease interacting with Age-group than Age-group alone ( Log likelihood test one sided p<0 . 05 ) ( Figure 3—source data 2; Figure 3 ) as the final validated list of ‘strictly age-specific disease markers’ . Notably , the percentage of taxa showing significant differences in their feature importance scores that were also validated in the Linear regression-based approach , showed disease-specific trends reflecting those observed in our PERMANOVA-based ( Table 2 ) and RF-based analysis ( Figure 2 ) . Specifically , for IBD and CRC , more than 63% of features showing significant differences in their marker scores across age-groups were also validated in the linear-regression-based validation , indicating that for these diseases , a majority of age-group-specific disease associations persist even after accounting for the overall effect of ageing . In contrast , for Polyps and Cirrhosis these overlaps showed a progressive decrease ( Figure 3—figure supplement 4A ) . We next compared the age-linked disease markers identified here with the disease markers originally reported . For four of the diseases ( i . e . not CRC ) , we compiled a list of known disease-associated markers from each of the original studies ( Feng et al . , 2015; Franzosa et al . , 2019; Karlsson et al . , 2013; Qin et al . , 2012; Qin et al . , 2014 ) . For CRC , a recently study presented a meta-analysis of all major cohorts ( encompassing those in curatedMetagenomicData ) ( Feng et al . , 2015; Vogtmann et al . , 2016; Zeller et al . , 2014 ) , along with three newly sequenced cohorts ( Thomas et al . , 2019 ) , to produce a refined set of cross-cohort validated CRC markers . We utilized this list to compile the set of known markers for CRC . Comparing the list of these associations with a pre-compiled list of all known marker associations reported in the original studies ( Figure 3—source data 3 ) ( Feng et al . , 2015; Franzosa et al . , 2019; Karlsson et al . , 2013; Qin et al . , 2012; Qin et al . , 2014; Vogtmann et al . , 2016; Zeller et al . , 2014 ) , identified that 40% of taxa identified in the original study for Polyps , more than 63% for CRC and IBD , 58% for T2D and 25% for Cirrhosis , displayed significant differences in their feature importance scores across age-groups , thereby emphasizing that age-linked disease-microbiome associations are prevalent across diseases ( Figure 3—figure supplement 4B ) . Even in the final list of ‘strictly age-specific markers’ that were deconvoluted for the effect of ageing , the overlap varied from 20% for T2D and Polyps to as high as 63% for CRC ( Figure 3; Figure 3—figure supplement 4C ) . The effects were especially pronounced for IBD , where many of the known markers ( species belonging to Lachnospiraceae , Escherichia , Clostridium clostridioforme , Clostridium nexile , Blautia producta ) were not even identified in the top 85 percentile in certain age-groups ( Figure 3 ) . For CRC , some of the well-known markers Fusobacterium nucleatum , Parvimonas micra , Peptostreptococcus stomatis , and Porphyromonas asaccharolytica , although identified in the top 85 percentile across age-groups , the strength of the associations in terms of RF feature importance scores as well as the results of linear modelling showed age-specific trends . The same pattern was also observed for some of the signature taxa for T2D ( Faecalibacterium prausnitzii , Roseburia intestinalis , Haemophilus parainfluenzae , Bacteroides intestinalis and Alistipes indistinctus ) and Polyps ( Bacteroides dorei ) . The variations in the strength and stability of these associations were further evident when we compared the percentage of times each marker is identified with scores greater than 85 percentile in different age-groups ( within the 100 iterations for each age-group ) ( Figure 3—figure supplement 4D ) . Several known markers of CRC ( Parvimonas micra , Eubacterium eligens ) and T2D ( Bacteroides intestinalis , Haemophilus parainfluenzae ) are identified in the top 85 percentile across more than 80% iterations in certain age-groups , but less than 50% in others . This has implications for the efficacy of microbiome-directed diagnostic/therapeutic strategies across age groups . We next tested if the differential trends of disease association observed were affected by cohort-specific differences . For CRC , we observed the strongest effects of age-group influenced disease signatures which affected more than 63% known CRC markers , and that were validated even after considering country-specific representations and deconvoluting the effect of ageing . Furthermore , the three newly sequenced cohorts ( referred to as ‘Cohort1’ , ‘Cohort2’ and ‘WirbelJ_2019’ , as a group ‘Validation Cohort’ ) had sufficient representation of samples from different age-groups ( Supplementary file 1 ) ( Thomas et al . , 2019; Wirbel et al . , 2019 ) , and represented new data that had not been included in our initial identification of differentially associated markers . In the curatedMetagenomicData repository , the CRC disease samples were collected from three cohorts , namely ZellerG_2014 , VogtmannE_2016 and FengQ_2015 ( Feng et al . , 2015; Vogtmann et al . , 2016; Zeller et al . , 2014 ) . To explore the reproducibility of the associations ( irrespective of the general effect of ageing ) as well as to understand the biological basis for these differential associations , we validated the age-aware RF classification models designed on this cohort ( referred to as the Training Set1 ) on the validation cohort ( Figure 4A; upper panel ) ( using a similar strategies as described in Figure 2—figure supplement 1 ) . The age-specific trends of disease classification were similar to those obtained earlier ( Figure 2 ) , that is , classifiers trained on the young/middle-aged individuals had the highest classification performance when tested on the same age-group , with a significant reduction in classification AUCs when tested on the elderly . In contrast , performance decreased significantly for those either trained or tested on elderly ( where the difference between the Same Age-group and Different age-group testing was not observed to be significant ) ( Figure 4A; Figure 4—figure supplement 1 ) . We further re-generated similar iterative age-group specific microbiome disease-prediction models by taking subsets of samples from within the ‘Validation Cohort’ ( referred to as Training Set2 ) and testing on non-training subsets of the same . The classification pattern remained invariant ( Figure 4A; lower panel versus upper panel ) . These results indicated that the variations in age-group specific microbiome-disease signatures remained stable and were not derived from biases in the training datasets . We then examined the overlap of the age-dependent associations of the known CRC markers across the two training models ( Figure 4B ) . From among the refined cross-cohort associated subset of 19 CRC-associated taxa ( Thomas et al . , 2019 ) , 13 ( 67% ) had differential association with either of the age-groups ( with FDR corrected p-value<0 . 1 ) in Training Set 1 . Nine of these ( 77% ) could be replicated with the same directionality in the Training Set 2 ( Figure 4B ) . Interestingly , from among the top CRC-predictive features reported in the recent meta-analysis , six were found to be associated with higher predictive power only in the young/middle age category ( Figure 4B ) . Only one was observed to be associated with the elderly microbiome . We next checked the stability of the feature-associations of CRC across the young/middle and the elderly age-groups across the cohorts . The bootstrapped approach adopted in this study enabled obtaining the microbiome-disease signatures for multiple subsets of diseased and control subjects . In a scenario where the microbiome signature is robust and stable , the microbiome-disease associations ( which in this case are obtained as the feature rank profiles for each of the taxa ) from each iteration of classifier should be similar ( across the cohorts ) . However , in this specific case , while the classifier profiles for the young/middle-aged individuals across the two cohorts were relatively similar , the inter-training-cohort distances obtained for the elderly-specific disease classifiers were significantly higher ( indicating variable disease signature and loss of reproducibility ) ( p<2 . 2e-16; Mann Whitney Test ) ( Figure 4C ) . Although no specific trends were observed when comparing abundances of the known CRC-specific markers , five of the top ten markers ( Fusobacterium nucleatum , Peptostreptococcus stomatis , Gemella morbillorum , Parvimonas micra and Parvimonas spp ) showed significantly higher prevalence rates in the elderly controls ( reproducibly across both datasets ) ( Figure 4D ) . This resulted in a significant reduction in the effect size of their differences ( between control and diseased individuals ) in the elderly age-group . Thus , ageing-associated changes may render elderly individuals susceptible to specific microbiome-related diseases like CRC . We next investigated if the diseases were characterized by distinct patterns of microbial taxon gain or loss , even across the different age-groups . We first focused on the individual study cohorts ( Figure 1C ) . For each disease-age-group scenario , we determined the directionality ( increased versus decreased in disease ) of association of the corresponding top disease-predictors by comparing their abundance trends in ( study-matched ) control and diseased samples ( Mann Whitney Test p<0 . 05 ) . For four of the five diseases ( i . e . except polyps ) , there was a significant change in the directionality of the microbiome alteration , characterized by a significant reduction in the number of gained features with older age ( Fishers’ exact test p<0 . 05; Figure 5—figure supplement 1 ) . To add statistical power and check whether the above trend is retained across a larger cohort , we performed the analysis using ‘continent-matched’ disease and controls ( using reasonably stringent thresholds of Benjamini-Hochberg corrected P value < 0 . 1 obtained from Mann Whitney tests ) . The trend remained similar . For elderly , microbiome alterations ( for all diseases except cirrhosis and polyps ) were characterized by a significant increase in the number of lost taxa ( Figure 5A ) . The list of disease-specific markers with significant directionality of disease association across age-groups are presented in Figure 5—source data 1 . Thus , the microbiome alterations in most of these diseases are characterized by a gradual shift from a state dominated by gained microbiome components to an increasing loss of control-associated taxa in the elderly . The overlap of altered taxa across diseases reported in one study was 51% ( Duvallet et al . , 2017 ) , while another study reported that generic control versus disease classifiers trained by agglomerating disease samples from multiple studies could still distinguish controls from disease with an AUC of greater than 0 . 8 ( Pasolli et al . , 2016 ) . Given that microbiome composition changes with age , we investigated the effect of age on the extent of these shared disease responses . We designed generic disease classifiers ( using Random Forest ) , taking equally sized sub-samples of controls and diseased individuals ( containing equal number of samples from each disease to prevent disease/age-group specific biases in classification performance ) ( See Materials and methods and Figure 1—figure supplement 3 ) . While the performances of the generic disease prediction models in the young/middle-aged was high ( median AUC: 0 . 79 ) and similar to those reported by earlier studies ( Pasolli et al . , 2016 ) , the same models applied to data from elderly subjects had significantly lower performance AUC ( p<1e-7 ) ( Figure 5B ) . Moreover , in these models , while no significant differences were observed with respect to the disease prediction sensitivities ( p<0 . 13 ) , the specificity of prediction ( that is the accuracy of identifying healthy individuals ) was significantly lower for the elderly . This was not an effect of the differential representation of samples from the different diseases , as we had ensured equal representation of all diseases across all age-groups . Thus , in contrast to previous meta-analyses , the shared disease response was significantly lower across elderly subjects , primarily with respect to the discrimination of non-diseased individuals . Furthermore , our clarification of the effect of age on disease-associated taxa ( Figure 3—source data 1; Figure 3—source data 2; Figure 5—source data 1 ) provides a refined set of features for improved microbiome-based diagnostics for these diseases . The above disease-independent changes in the pattern of microbial alterations as well as the inability of the generic disease classifiers to distinguish non-diseased controls were intriguing and could be a consequence of the loss of beneficial bacteria in the gut microbiome with ageing , which in turn could be driven by a multitude of factors like diet and medication . This increasing loss could lead to dysbiotic configurations characterized by higher inter individual variability , increased abundance of pathobionts ( resulting in loss of disease signature ) thereby making the microbiome more susceptible to diseases . Therefore , using intra age-group Spearman distances , we next checked whether the samples from elderly controls had significantly higher variability as compared to young/middle-aged controls , across the different continental regions ( Figure 5—figure supplement 2 ) . In line with our hypothesis , for both Europe and North America , the gut microbiome from elderly individuals was significantly more variable compared to young and middle-aged controls . However , this was not observed in the Asian ( Chinese ) cohort . Interestingly , the cirrhosis group neither shows an association with disease signatures nor the gain versus loss pattern contained only subjects from Asia . Next , we sought to characterize the elements of the shared disease response . Notably , closer inspection of the directionality of the associations indicated specific taxa with consistent trends of association with multiple diseases ( based on the trend shown in Figure 5—source data 1 ) . The overall patterns of taxon gain or loss encompassed several trends observed by earlier studies ( Duvallet et al . , 2017; Pasolli et al . , 2016 ) . For example , Streptococcus anginosus and Fusobacterium nucleatum were detected as gained in multiple diseases . Similarly , species belonging to Roseburia spp . ( R . hominis ) were lost . We identified a total of 61 taxa that showed consistent directionality of association with multiple diseases in either young/middle-aged or the elderly age-groups ( Figure 5C ) . Based on their differential detection profiles in the shared response across age-groups , we assigned these into six different groups , namely G1 ( increased in disease across all age groups ) , G2 ( increased in disease only in the elderly ) , G3 ( increased only in young/middle-aged ) , L1 ( decreased in disease across both ) , L2 ( decreased only in the elderly ) and L3 ( decreased only in the young/middle-aged groups ) ( Figure 5C ) . Many of the species previously reported as associated with shared gain or loss across multiple diseases ( Duvallet et al . , 2017; Pasolli et al . , 2016 ) belonged to the G3 group , that is , they showed similar trends of gain or loss in disease ( as reported earlier ) in the young and the middle-aged groups , but not in the elderly . These included the Streptococci , Fusobacterium nucleatum , and Escherichia coli and Bacteroides fragilis . In contrast , a separate group of species including Ruminococcus torques , Solobacterium moorei and Streptococcus mitis were associated with multiple diseases only in the elderly . Finally , we identified a distinct group of species ( G1 ) that was gained across diseases in both elderly and young/middle aged groups . These included a group of Clostridia ( C . bolteae , C . symbiosum , C . hathewayi , C . citronae , C . asparagiforme ) ( Figure 5C; see Figure 5—source data 1 for individual diseases ) . These taxa have been identified in separate studies of different diseases and/or disease-like states ( T2D , Polyps , CRC and autism ) ( Pequegnat et al . , 2013; Qin et al . , 2012; Sinha et al . , 2019; Yu et al . , 2017 ) , but are shown here , for the first time to be part of a shared gain response across diseases . Based on these findings , we hypothesize that this specific G1 group of taxa constitute a shared disease response associated with a general patho-physiological failure in the affected individual . Frailty in the elderly is characterized by reduced function of multiple systems . We investigated if the taxon associations with frailty could be validated in the ELDERMET cohort ( Supplementary file 2 ) , for whom we had both shotgun metagenome and faecal metabolomic data . Using Random Forest regression ( with five-fold cross validation ) , we could predict the frailty of an individual ( testing both community- and residential care-dwelling subjects ) based on the taxonomic composition of the gut microbiome ( R = 0 . 79; Figure 6A; Figure 6—figure supplement 1 ) . We then calculated the mean feature importance ranks in the frailty prediction model for the six gain/loss generic disease response groups ( from Figure 5C ) . Validating our previous observations , the G1 group of taxa had the highest mean rank of feature importance scores for frailty ( Functional Independence Measure ( FIM ) ) prediction , indicating that this group had the highest predictive power for frailty in the ELDERMET subjects ( Figure 6B ) , followed by the G2 group ( increased across multiple diseases only in the elderly ) . Validating our earlier findings on the elderly-specific markers of shared microbiome response , this was followed by the L1 group . An optimal number of eight taxa had the highest frailty predictive ability ( Figure 5C; Figure 6C; Figure 6—figure supplement 1 ) . All these taxa were more abundant with frailty ( negatively correlated with FIM and positively associated with Frailty ) ( Figure 6—source data 1A–B ) . Five of the eight taxa belonged to the G1 group ( Figure 6C; Figure 5C ) . Thus , these findings independently validated in the ELDERMET cohort our earlier identification of specific elderly-associated generic disease response groups ( Figure 5C ) . If microbiota alterations contribute to disease causation , microbial metabolites may be effector molecules . A distinguishing feature of the current study is that it is based on metagenomic data , as distinct from previous 16S-based meta-analyses of microbiome-disease ( Duvallet et al . , 2017 ) . This allowed us to obtain taxonomic composition at finer taxa level , which in turn we could use to predict metabolite profiles of the microbiomes by exploiting experimentally validated functional profiles of the constituent taxa . For this purpose , we used the Virtual Metabolic Human database ( Noronha et al . , 2018 ) , as well as a recent method to search experimentally-verified functional profiles representing the production and consumption patterns of different metabolites in various gut-associated species ( Sung et al . , 2017 ) . We could thus collate more than 300 metabolic profiles ( consumption/production/degradation ) from 992 species ( Figure 6—source data 2 ) . A total of 82 metabolite profiles were observed to have significant association with FIM scores ( Spearman Rho; FDR of less than 0 . 25 ) ( Figure 6—figure supplement 2 ) . We observed that this association analysis of metabolite profiles reflects frailty-associated changes with respect to bioavailability of specific compounds , many of which have been previously shown to have corresponding associations with health , thereby validating our findings ( Claesson et al . , 2012 ) . Specifically , the onset of frailty is observed to be associated with an increase of SCFA consumption by gut-bacteria ( accompanied by a concomitant decrease of the production of the SCFA butyrate ) , increased consumption of the beneficial amino acid Tryptophan as well as the increased production of the T2D-linked branched chain amino acid Threonine . We first focussed on the group of frailty marker taxa and identified 13 metabolic profiles ( Figure 6D ) that were significantly associated with the eight frailty-marker taxa ( See Materials and methods ) . This subset of 13 metabolite profiles alone could predict frailty with a R value of 0 . 60 ( between the actual and predicted FIM values ) , which is significantly higher than that obtained after removing these features from the metabolite profile ( Figure 6D; Top panel second from left ) . The first set of key metabolic profiles included the degradation of primary bile acids , namely cholic acids ( CA ) and chenodeoxycholic acids ( CDCA ) to produce hydrophobic secondary bile acids ( lithocholic acid: LA; deoxycholic acid: DCA ) . We validated this predicted metabolic functionality against the experimentally measured fecal metabolomic profiles for ELDERMET subjects , where the fecal metabolomes of frail individuals were characterized by significantly higher levels of LCA ( along with its derivatives ) and DCA ( p<0 . 006 ) and significantly lower levels of CA and CDCA ( p<0 . 02 ) , as compared to the non-frail individuals ( Figure 6D ) . The two species associated with this functionality , namely Clostridium scindens and Clostridium leptum , did not belong to either the G1 or G2 groups of species ( contrary to their frailty-association in this study ) . These species were previously implicated in Clostridium difficile resistance and IBD , respectively ( Buffie et al . , 2015; Manichanh et al . , 2006 ) . Contrary to their therapeutic relevance , this specific functionality of these species , namely production of higher levels of hydrophobic bile acids , has also been associated with the onset of CRC and non-alcoholic fatty liver disease ( NAFLD ) ( Tsuei et al . , 2014 ) . One of the recent meta-analyses of CRC gut microbiomes also predicted increased production of secondary bile-acids to be a key CRC-specific functional signature ( Wirbel et al . , 2019 ) . Two other metabolite functionalities associated with the frailty-specific markers that were also validated in the fecal metabolome profile were p-cresol production ( measured p-cresol negatively correlated with FIM: p<0 . 01; R = −0 . 29 with FIM across community and longstay; R = −0 . 47 for only long-stay ) and ethanol production ( significantly higher levels in the fecal metabolome of frail individuals in the long-stay cohort , as compared to the non-frail individuals: p<0 . 02 ) . While p-cresol is a cytotoxic compound ( produced by specific gut bacterial species including C . difficile ) that is associated with microbiome alteration , large bowel cancer and insulin resistance ( Fau et al . , 1976; Khan et al . , 2014 ) , high level ethanol production in the gut has not only been linked to incidence of NAFLD , atherosclerosis and SIBO , but also to an increase in the levels of inflammatory markers ( Elshaghabee et al . , 2016 ) . Similarly , production of acetone , ammonia ( NH3 ) have been associated with increased cytotoxicity , small intestinal bacterial overgrowth , as well as insulin resistance ( Baskaran et al . , 1989; Ghoshal et al . , 2017; Khan et al . , 2014 ) . Although fecal metabolome data was not available for ExperimentHub microbiome samples , the abundance of specific microbial pathways ( obtained using humann2 ) was associated with the production of these metabolites , namely propane-1 , 2-diol to acetone , and allantoin to Ammonia/CO2/Glyoxylate was predicted to be significantly higher in the microbiome of the frail individuals ( Figure 6D; Lower panel ) . Another functionality associated with the frailty markers was choline consumption and Trimethylamine ( TMA ) production . Specific gut microbial species including C . citroniae , C . clostridioforme and C . hathewayi degrade choline to trimethylamine , which is liked with atherosclerosis ( Martínez-del Campo et al . , 2015 ) . Profiling the abundance ( normalized by sequencing depth ) of the specific bacterial CutC enzyme catalysing this conversion revealed significantly higher representation in the gut microbiome of the frail individuals ( Figure 6D; Upper right panel ) . This association with CRC has also been reported in one of the meta-analysis datasets used in the current study ( Thomas et al . , 2019 ) . Thus , there is a prima facie mechanistic case whereby this group of microbiome markers could functionally drive the host to an increasingly pan-disease susceptible state , also accelerating the onset of frailty . Interestingly , identifying similar metabolite associations across the G1-G3 markers indicated overlapping metabolite production signatures of the frailty-associated markers with even the G3 groups of taxa ( specifically the production of Ethanol and NH3 ) , indicating that the different taxa groups enriched in a generalized microbiome-disrupted state , irrespective of differences in their composition , contain certain metabolic signatures that may drive the intestinal eco-system to a state detrimental to the host ( Figure 6—figure supplement 3 ) . Similarly , comparing the metabolic signatures of the taxa in the G1-G3 groups with those of the different groups of lost taxa L1-L3 , also revealed an interesting pattern whereby the ‘gained’ groups were disproportionately associated with the consumption of simple sugars ( lactose , psicose , xylose , sucrose , raffinose ) . The ‘lost’ groups in contrast were enriched for consumption of dietary prebiotics like xylo-oligosaccharides , fructo-oligosaccharides , inulin and production of butyrate and succinate . A recent study of the gut microbiome of Thai individuals migrating to the US observed the loss of metabolic functions linked to degradation of dietary fibers like xylan , arabinoxylan , cellobiose , pullulan , glucomannan and resistant starch , with increasing duration of stay leading to a broader loss of microbiome function ( Vangay et al . , 2018 ) . These patterns , reflective of the metabolic capabilities of specific microbial groups , are especially important for diet-based microbiome restoration strategies in the elderly , where loss of control-associated taxa has a stronger effect on disease , as compared to the young and the middle-aged . By identifying specific age-linked microbiome associations for different diseases , this study can potentially inform the development of microbiome-based diagnostic strategies customized for specific age-groups . Accounting for age in disease-microbiome classifiers has clarified the microbiome alterations common to many diseases and shortened the lists of taxa specifically implicated in common non-communicable diseases . Furthermore , distinct changes observed in the overall directionality of taxon associations ( e . g . the gradual shifting to a loss-of-taxa phenotype in the elderly ) highlight that microbiome restoration strategies may be more crucial for disease amelioration in elderly subjects ( as opposed to antibiotic-based eradication regimens ) . This is especially important since specific changes are observed with ageing that can potentially make the host more susceptible to certain microbiome-related diseases . The above pattern was further observed in our analysis using generic disease classifiers . The specific loss of the ability of the classifiers to distinguish between cases and controls further indicates that with ageing , there is loss of health-associated microbiome signatures . Interestingly , these findings resonate with the results of one of our earlier studies , where with increasing age , a decrease of the core microbiome accompanied by an increase of pathobionts was observed ( O'Toole and Jeffery ) . Finally , reproducible identification of a specific set of species markers associated with multiple diseases and their specific metabolite profile ( sugar consumption and the production of a range of metabolites that are detrimental for the host ) indicates that the acquisition of this subset of disease-associated taxa can shift the metabolic state to a disease-like state . Separating correlation from causation in microbiome-disease studies requires identification of putative taxa correlating robustly with disease , which will be improved by accounting for age in microbiome-disease association studies . Many of the features in the current analysis were not reported in the original publications . There could be several reasons for this , namely variation in the comparative analysis protocol used in the original studies as compared to those in the current study; more power and robust signatures in the current analysis by virtue of combining multiple studies; study-wise biases in the number of samples belonging to different age-groups . In this regard , a key aspect is the relative stability of the age-specific differential associations across regions/ethnicities . In this study , we have used a combination of Random Forest , linear models as well as validations at different levels of regional homogeneity to not only reduce the effects of regional variations on microbiome signatures but deconvolute the effect of general ageing on these age-specific disease signatures . However , biases in number of samples for certain regions in specific disease-age-group scenario still exist . Currently , a majority of datasets in the curatedMetagenomic data repository and the other validation cohorts are from North America , Europe and China . However , future availability of microbiome data from disease cohorts especially from Non-Western populations will further clarify the age-specific trends of disease classification . This is especially important , since in the current study , we identify region-specific changes across age-groups , where in , while the gut microbiomes of elderly controls from North America and Europe have significantly higher variation as compared to region-matched young/middle-aged controls , no such differences are observed within the Asia ( Chinese ) populations . Future availability of new disease-specific shotgun microbiome datasets will further enable cross-cohort validations of disease signatures ( as was performed in the current study for CRC ) on other diseases . Another factor that needs to be considered is the effect of diet . Many of the taxa that are depleted in disease are also taxa whose growth is promoted by complex carbohydrate consumption . Different populations have specific dietary patterns and these associations need to be factored into suitably designed future prospective studies . We could not control for medication intake in subjects/data from the ExperimentHub , because this data was not available ( except for antibiotic treated subjects that were removed ) , and which can be a major confounder ( Forslund et al . , 2015 ) . Previous studies have reported the association of some of the disease markers with medication intake . For example , Streptococcaceae have been shown to be associated with intake of Proton Pump Inhibitors ( Jackson et al . , 2018 ) . Another recent study investigating the effect of metformin on the gut microbiome of two of the T2D cohorts analysed in the current study identified Lactobacillus salivarius to be affected by metformin , accompanied by an increase in Escherichia ( Forslund et al . , 2015 ) . However , none of the other G1 and G2 markers were reported to be affected . The FranzosaEA_2018 dataset ( added along with the curatedMetagenomicData repository ) contained drug usage information for Immuno-suppressants , Steroids and Mesalamine usage ( Franzosa et al . , 2019 ) . For these three drugs , we checked whether the abundance of shared disease markers was affected by medication intake ( Supplementary file 3 ) . We could only identify associations for Peptostreptococcaceae ( decreased in mesalamine treatment ) , Coprococcus comes ( increased in mesalamine treatment ) and Barnesiella intestinihominis ( decreased in steroid treatment ) . Moreover , adjusting for medication did not alter abundance of taxa predictive of frailty in the ELDERMET cohort ( for which the medication usage data was available ) . Each of the top frailty-associated markers were not only observed to have significant associations with the different frailty measures even after taking into account for the effect of the selected sets of medication type , but also identified in the top 15 predictors of frailty , independently in the individuals with high or low medication ( Figure 5—source data 1c ) . However , the lack of associations with this limited list of medications does not completely rule out the possibility of medications being partly responsible as a driver for these differential associations . Consequently , for diseases like T2D , the results obtained here have to be treated with caution . Future cohort studies should record all measurements of diet and medication intake , so that the complex interactions of age , lifestyle , diet , microbiome and health may be further elucidated . Another confounding factor may be age of disease presentation . For example , a recent study identified cancer stage-microbiome interactions ( Yachida et al . , 2019 ) ; however , we also note from the published data that 51% of the younger subjects ( less than 60 years of age ) presented with Stage III/IV cancer , whereas a significantly higher percentage ( 71% ) of older subjects presented with Stage I/II cancer ( p<0 . 01 ) . Variation in disease presentation in different age groups could also influence the disease-microbiome signatures , and therefore needs to be considered in future studies . The last concern is methodological and pertains to the non-independence between the training samples in the iterative RF classifiers generated for each disease-age-group scenario . Classifiers generated in each iteration are likely to share overlap of training ( and/or testing ) samples , thereby potentially resulting in narrower AUC distribution which could inflate significance values . However , the magnitude of the differences as well as consistency of the results ( using multiple methods and study cohorts ) clearly indicate the reliability of the results obtained in this study . Since the focus of investigation was the gut microbiome , we first selected a subset of the curatedMetagenomicData , containing 4195 stool samples ( annotated as ‘body_site’: stool ) ( Pasolli et al . , 2017 ) . We subsequently removed the samples which have not defined age and study-condition , thereby filtering the dataset to 3580 samples . From this set , we removed samples having age less than 20 years of age ( retaining 2564 samples ) . Notwithstanding the uniform bioinformatics analysis strategy applied to this data , two major factors that may contribute an artefactual bias in multi-cohort microbiome datasets ( and which were available in the metadata ) are the read-length ( obtained from the sequencer ) and DNA extraction methodologies ( which are study-specific ) . To test the effect of these factors , we first removed the samples from Peruvian , African and Fijian individuals in order to remove the confounding effects of region/life-style-specific ) , along with those from hospitalized individuals . Subsequently , on the remaining subset , we evaluated the effect of these factors using envfit on the species level profiles by first visually comparing the differences using PCoA and then testing the confidence of these differences using envfit ( https://cran . r-project . org/web/packages/vegan/vegan . pdf ) . For this purpose , we performed 20 boot-strapped envfit iterations , each time taking a subset of samples ( sub-sample size: 200 ) and computing the R2 and the significance ( P-value ) of the differences , and then comparing the distribution obtained with that obtained using a null distribution obtained by taking 200 sub-samples ( after permuting the labels ) . We established that while read lengths had a marginal effect ( R = 9e-3 , p<0 . 09 ) ( Figure 1—figure supplement 1A ) , samples from one of the studies ( SchirmerC_2016 ) ( Schirmer et al . , 2016 ) , using a DNA extraction technique tagged as ‘Illuminakit’ in the metadata , had a distinct taxonomic profile . We removed the 465 samples of this study from all further analyses , thereby reducing the effect of extraction methodology on taxonomic profiles ( p<0 . 06; Figure 1—figure supplement 1B–C ) . To this compiled list , we added the samples from four recently published datasets ( one IBD-specific dataset of 220 samples , referred to as ‘FranzosaEA_2018’ [Franzosa et al . , 2019]; three CRC-Specific datasets referred to as ‘WirbelJ_2019’ , ‘ThomasAJ_Cohort1’ and ThomasAJ_Cohort2’ [Thomas et al . , 2019; Wirbel et al . , 2019] ) . This repository , along with the 189 shotgun sequenced samples from the ELDERMET cohort , resulted in a total of 2564 samples . Schematic representation of the workflow used for preparing a core set of 2564 gut metagenomic samples derived from the publicly available datasets ( curatedMetagenomicData and the four newly added cohorts ) and the ELDERMET repository is provided in Figure 1—figure supplement 2 . The details of the samples belonging to the curatedMetagenomicData and FranzosaEA_2018 dataset included in the current study are provided in Supplementary file 1 . The clinical metadata of the ELDERMET samples are listed in Supplementary file 2 . In this analysis , the taxonomic composition obtained using the metaphlan2 pipeline was obtained at the microbial species level . We have used the term ‘species’ and ‘taxa’ in this manuscript to refer to a taxonomic unit below the level of genus . We specifically investigated the metadata of the samples listed in Supplementary file 1 . Some of the metadata were observed to be redundant and had similar associations ( examples included groups like Country and Dataset Name; Age and Age-category; Study condition and Disease; Antibiotics current use and Antibiotics family ) . In these cases , we retained the former metadata and removed the latter ones . We added another region-specific metadata , namely Continent , for reasons explained in the subsequent section . Subsequently , we filtered out those metadata present in less than 30% of the samples . A total of six metadata remained . We obtained the association of each of these metadata using PERMANOVA ( using the adonis function of the vegan R package ) . Given that DNA extraction methodologies still had a marginal effect on the gut microbiome composition ( R2 = 0 . 019 ) , the PERMANOVA analyses for each of the metadata were performed after adjusting for the DNA extraction method as a confounder . The PERMANOVA analysis ( for each metadata ) was performed using the adonis function of the vegan package using the following pseudocode formula: adonis ( species ~ dna_extraction_method + metadata ) . For investigating the variation of the microbiome with age across the adult-hood landscape , the individuals were binned into three age groups namely young ( 20–39 years of age ) , middle-aged ( 40–59 years ) and elderly ( 60 years and above ) . We removed the antibiotic-treated subjects from all subsequent analyses . Principal component analysis of the microbiome profiles of the samples belonging to the three age-groups was performed and plotted using dudi . pco and s . class function of the ade4 R package . The significance of the association was obtained using the PERMANOVA ( adonis function ) implemented in the ‘vegan’ R package ( with ‘country’ and the DNA extraction method as a confounder ) . Given that regional factors had the highest effect on the microbiome composition , it was important to ensure regional homogeneity for comparative disease-association analysis . However , the majority of disease-specific cohorts either displayed significant differences in age difference in the age of the control and diseased individuals ( i . e . they were not age matched ) or biases for disease patients from specific age-groups ( Figure 2 ) . For each disease , collating samples from the same country or continent-level as the disease cohorts would bypass the issue of limited sample numbers across the age-groups , whilst maintaining regional homogeneity of the cohorts ( Figure 1—figure supplement 3A–B ) . To compare the overall effects of the two regional factors , country and continent , on the microbiome profiles we performed bootstrapped PERMANOVAs ( by taking 20% subsets ) within the control individuals . The results indicated that , although continent was observed to have a marginally lower effect on the microbiome composition compared to nationality ( country ) , performing repeated bootstrapped comparisons indicated the effect of continent to have a higher significance ( calculated as -log of Adonis P-values ) than the country on the microbiome profiles ( Figure 1—figure supplement 3C ) . For each disease , the country/continent specific affiliations of the disease cohorts were first obtained . Subsequently , we performed all the investigations pertaining to each disease by pooling samples belonging to the same country as the corresponding disease cohorts ( referring to them as disease-specific country-level bins ) ( Figure 1—figure supplement 3D ) . This was expected to optimally homogenize the region-specific variations , while ensuring enough representation of various diseases ( and controls ) across age-groups . Wherever applicable , we have adopted a similar disease-specific regional grouping strategy at the level of continents . For each disease , we performed PERMANOVA within the corresponding disease-specific country bins investigating the effect of the interaction of Disease and Age-group ( Disease:Age-group ) after adjusting for the effects of Country , and the independent effects of Disease and Age-group . Briefly , the pseudocode of the formula is provided below: adonis ( species ~ country + disease + age-group + disease:age-group ) . If diseases have age-group specific signatures , then classifiers trained on the same age-group would have significantly better performance when tested on the same age-group as compared to that when tested on different age-groups . To evaluate the performance of disease classifiers trained on one age-group on disease prediction in either the same ( Same Age-group classification ) or different age-groups ( Different Age-group classification ) , we adopted the following strategy ( Figure 2—figure supplement 1 ) . For each disease-age group combination , we performed 100 iterations , such that in each iteration , we trained the classifier on a subset of disease and the same number of control samples ( 50% of the minimum number of diseased samples across any age-group; denoted as ‘training subset’ , the disease-specific training subset sizes as defined in Figure 2—source data 1 ) . The evaluation of each of these disease classifiers for ‘Same Age-group’ and ‘Different Age-group’ classification was then performed using two approaches as mentioned below . In the first approach , we created two ‘re-sampled’ test sets . While one contained Y diseased and Y control individuals from the same age-group ( but not included in the training sub-set ) ( Same Age-group test set ) , the other contained Y diseased and Y control individuals from the other two age-groups ( Different Age-group test set ) ( with Y defined as in Figure 2—source data 1 ) . We tested the classifier on each test set and computed the AUCs . Testing the same classifier on both test sets ensured that the observed variations in disease prediction performance was not due to differences in the subject sub-samples used to create the classification models . Further to ensure that we don’t have biases introduced because of the selection of test sets ( same age-group , different age-group ) , we repeat above steps 20 times ( per classifier in each iteration ) and computed the median AUCs for both the Same Age-group classification and the Different age-group classification . These median AUCs obtained for the Same Age-group and Different age-group classification for each of the 100 iterative sub-sampled classifiers was compared using Wilcoxon Signed Rank tests ( to check if the performance of the classifiers significantly varied when tested on the same or different age-groups ) . In the second approach , using Permutation test framework , we tested whether observed difference in classification AUCs ( i . e AUC for Same Age-group classification – AUC for Different Age-group classification ) was significantly different than what would be expected by random . For this , we needed null distribution of empirical differences of AUC for the same sub-sampled classifier . For this purpose , for each of the 20 iterations corresponding to each of the 100 sub-sample based RF classifier models ( as described in the previous paragraph ) , we first merged the two ‘re-sampled’ test sets ( Same Age-group test set and Different Age-group test set ) , then permuted the age-group labels of the subjects , creating two ‘Permuted’ test tests ( Permuted test set 1 and 2 ) , tested the classifier model for each of the Permuted test sets and finally computed the AUC differences obtained for the same classifier between the two permuted test sets . For each of the RF classifiers ( across the 100 iterations ) , we computed the medians of the differences of AUCs ( for the two permuted test sets ) across the 20 iterations . The median difference of the AUCs obtained for the actual test ( for Same Age-group – Different Age-group ) and the permuted tests ( Permuted Test set 1 – Permuted test set 2 ) obtained for the 100 iterations are then compared with Wilcoxon signed rank-tests . The objective of these permutation tests was to reduce the effects of correlated errors associated with certain ‘aberrant’ samples in influencing the disease prediction performance of RF classifiers for specific age-groups . This permutation test is expected at least ensure that the correlations introduced by sampling and by testing the same classifier on multiple cohorts will also be present in the null distributions . For both the approaches , the p-values of comparison across age-groups were corrected using Holms’ correction . For a given disease , to ensure that the observed changes were not artefactual consequences of differences in sizes of training and testing subsets , we kept the training and testing subset sizes constant across all training age-groups . The classification AUCs , Specificities and Sensitivities were computed using the various modules in the pROC package . For each disease-age group scenario ( as described above ) , we ranked the species-level taxa in decreasing order of their mean importance scores ( mean decrease in GINI ) across all 100 iterations . The iterative sub-sampling based Random Forest analysis avoided overfitting and , by virtue of the uniformity of training and test sizes for each disease across all age-groups , ensured that the differences across age-groups were not a simple consequence of differences in the number of diseased and control individuals . The objective of the iterative approach was also to identify taxa that showed stable association with disease irrespective of the sub-sampling population . This aimed to identify taxa that were always considered for the classification model irrespective of the sampled population of diseased and controls ( Mean Decrease of GINI of at least 0 across at least 95% of iterations ) . So , for each disease/age-group scenario , we aimed to identify that minimum percentile threshold above which the taxa were considered for the classification model across at least 95% of the iterations ( Figure 3—figure supplement 1A ) . We observed that for taxa , having a percentile score of at least 85 , included in training models across at least 95% of the iterations across all the 13 disease age-group scenarios . Further plotting the mean feature scores of taxa arranged in terms of their percentile scores , we observed that the mean feature importance scores remained stable and low till the 80-percentile score and started increasing considerably only after that ( Figure 3—figure supplement 1B ) . Based on these two results , the taxa having importance scores in the top 15 percentile ( that is higher than 85 percentile ) were identified as the top 85 percentile predictors/markers for a given disease in that age-group . In this linear model based validation strategy , within the disease-specific country-level bins , we modelled the ( log transformed to the base 10 ) abundance of each species first as a function of disease-status and country and then as a function of the interaction of disease-status and age-group and country as an independent predictor as described below: Model 1: Log ( Species ) ~Country + Disease + Age-group Model 2: Log ( Species ) ~Country + Disease * Age-group ( which is equivalent to Log ( Species ) ~Country + Disease + Age-group + Disease:Age-group ) The goodness of fit of both models were quantified using adjusted R-squares and AIC values , and the significance of improvement of Model two with respect to Model 1 , was judged using log-likelihood tests . This took care of the country as a confounder as well as provided the extent to which the interaction of disease and age-group had an influence on the abundance of the species as compared to the individual factors . Those species showing significant improvement in performance of Model two with respect to Model 1 ( log likelihood ratio test p<0 . 05 one-sided ) , could be identified as ones for which age-group had significantly high influence on the disease associations . For the case of T2D , however , there were skews in the representation of the diseased samples across countries , where in the diseased samples in the young and middle aged group were only from the Chinese cohort , while those from the Elderly individuals had representation from both the Chinese and Swedish cohort . Consequently , to ensure regional homogeneity , in the above validation for T2D , we included young/middle-aged controls from only the Chinese cohort , while the elderly controls included those from both the Chinese and the European ( Swedish for country-specific comparisons ) cohorts . Taxa having significant differences in their RF feature importance scores ( FDR < 0 . 01 at least one pair of age-groups using Dunns’ test for IBD , T2D and Cirrhosis and Mann-Whitney Tests for CRC and Polyps ) , which were also validated as having log likelihood ratio test p<0 . 05 in the linear model based validation were identified as the final validated list of age-group specific markers . This was done because the validation of each taxa was independent and additional stringent thresholds during validation would result in loss of sensitivity . In order to evaluate the prevalence of these age-linked markers in the known disease-microbiome associations , for four of the five diseases , we obtained the list of associated species ( detected as either significantly different or as a marker with high disease predictive score ) from the original published studies corresponding to each of the disease cohorts ( Feng et al . , 2015; Franzosa et al . , 2019; Karlsson et al . , 2013; Qin et al . , 2012; Qin et al . , 2014; Vogtmann et al . , 2016; Zeller et al . , 2014 ) ( Figure 3—source data 3 ) . For CRC , we utilized the list of markers provided in Thomas et al . ( 2019 ) . We specifically utilized the list from this study , as it had already performed a multi-cohort meta-analyses of CRC-specific taxa adopting a similar metaphlan2-humann2 based classification scheme . The list ( as created above ) was then compared with list of age-linked markers specific to each disease ( Figure 3—source data 3 ) . Iterative age-group specific disease classifiers were devised and tested as described in Figure 2—figure supplement 1 ( by specifically setting the ‘Training’ and ‘Testing’ cohorts ) . For classifiers designed on the validation cohort , the feature score importance scores of the top 85 percentile markers ( obtained during the iterations ) were compared using Mann-Whitney U-tests between each age-group ( in each cohort ) . For testing the reproducibility of the associations , the directionality of the 19 known markers and the significance of their differences in the two training cohorts were then obtained and compared . For investigating the stability of the microbiome signatures ( obtained for the different age-groups ) across training cohorts , we profiled the mutual variations in the feature rank profiles obtained in cohort with respect to those obtained for the other . For a given feature rank profile ( obtained in a given cohort for a specific age-group ) , its mean Spearman distance to all the feature rank profiles obtained for the same age-group in the other cohort was obtained . This was then collated for all feature profiles for the same age-group in both the cohorts . The distribution of the distances would indicate the stability of the microbiome signature across cohorts . Lower cross-cohort distances would indicate stable reproducible signatures and higher cross-cohort distances signify a relative lack of signature . The ageing associated changes in the prevalence of some of the CRC markers and their effect relative variation within disease and controls were profiled in the following manner . In the first step , we compared the prevalence rates of these markers in the elderly controls and the young-middle aged controls using Fishers’ Exact tests separately for the two cohorts . The p-values obtained for each cohort were then merged using the Fisher method . The effect size difference ( disease v/s control ) distribution for the markers computed using age-group specific iterations wherein each iteration , we compared 20 disease samples with 20 controls and computed the effect size using Cohens’ D . Empirically , a Cohens’ D values of less than 0 . 2 indicates low effect and greater than 0 . 2 indicates medium to high effect . To obtain the directionality ( increased or decreased in disease ) of the of the microbiome changes , Mann-Whitney U tests were first performed for all taxa to compare their abundances in ‘study-matched’ control and diseased samples ( for ensuring regional homogeneity ) from the specific age-group ( ref to Figure 1C for the Study cohorts ) . The taxa having significant differences ( Mann Whitney Test p<0 . 05 ) in their abundances between control and diseased samples were identified along with their directionality ( ie . increased in disease or decreased in disease ) . These analyses were finally re-performed at the level of continent ( that is between ‘continent-matched’ disease and control samples ) to ensure statistical power to detect the gain versus loss patterns . For each disease age-group scenario , the top 85 percentile markers having significant change in their abundance with FDR corrected p-values<0 . 01 ( for the continent level comparisons ) were then filtered ( Figure 5—source data 1 ) . The directionality of these markers was assigned based on the trends of their abundance patterns , as either 'Increased' or 'Decreased' in disease . The generic disease prediction classifiers were developed in a manner similar to that shown in Figure 1—figure supplement 3 . The only difference was the agglomeration of equal number ( n = 10 ) samples from each of the diseases ( rather those of a specific disease ) as described below . For each age-group ( young/middle-aged or elderly ) , a generic disease cohort was created by taking equally sized sub-samples from each disease ( to remove biases in the classifiers originating from specific diseases ) . The sub-sample sizes were also kept the same across the age-groups to ensure uniformity in the testing and training sizes of the classifiers across all age-groups . The iteration was subsequently repeated five times using a different ( but equally sized ) subset of diseased and control samples ( as described above ) . The AUC and sensitivities for the five repetitions were then merged and compared across the young/middle-aged and elderly . To remove regional biases in microbiome compositions affecting these results , all analyses were restricted within the disease-specific continent cohorts ( Figure 1—figure supplement 3D ) . We used a random forest model to regress both the Functional Independence Measure ( FIM ) and the Barthel Score ( both an inverse measure of frailty ) of an individual from the microbiome profile . Random forest regression training was performed on 20% of the samples and tested on the remaining 80% . The ranked feature importance scores of the different species were then obtained . Microbiome features ( that is the species ) were ranked in decreasing order of their feature importance scores ( mean decrease in GINI coefficient upon excluding the feature ) . The mean feature ranks for the different groups of species ( G1-L3 ) were then calculated . To identify the most predictive marker set , the regression was repeated by iteratively reducing the number of the top microbiome features , and the mean error was calculated for each iteration . For both FIM and Barthel Score , the number of top microbiome features for which the error was minimum was taken as the set of the top frailty-associated markers . For both FIM and Barthel Score , the minimum number of features were 8 and 6 , respectively . The median FIM values were computed for both the residential care cohort and the overall cohort ( community + residential care ) . For each cohort , individuals having FIM values below and above the corresponding median were classified as ‘Frail’ and ‘Non-Frail’ , respectively . To validate the metabolite signatures , measured levels of actual metabolites , abundances specific microbial pathways and gene-families ( obtained using humann2 ) were then either compared between these groups of ‘Frail’ and ‘Non-Frail’ individuals or correlated with the FIM values . We utilized the literature-curated experimentally annotated species to metabolite ( production/consumption ) associations available as part of the Virtual Metabolic Human database as well as those obtained in a recent meta-analysis by Noronha et al . ( 2018 ) ; Sung et al . ( 2017 ) , to create a species-to-metabolite map of more than 300 metabolite production and consumption profile corresponding to 992 species in a 0 ( absent ) and 1 ( present ) notation ( Figure 6—source data 2 ) . For each microbiome , the metabolite production/consumption capability was then obtained as the matrix inner product of the abundance profile of the species and the species-to-metabolite map thus obtained . Next , we identified the frailty-linked metabolites associated with the eight taxonomic markers of frailty using a two-step strategy . First , we performed a correlation analysis of each metabolite profile ( i . e the cumulated abundance of taxa previously associated in literature with a given metabolic capability as obtained above ) with FIM scores and identified metabolite profiles that showed significant association with FIM scores ( Spearman Rho; FDR of less than 0 . 25 ) . Next , we identified which of these identified metabolite profiles were detected in the taxonomic markers of frailty ( based on previous literature ) at a rate significantly higher than the background detection ( using our Fishers’ exact test approach with FDR corrected p<0 . 25 . The detailed description of the codes is provided in the methods section . The key in-house source codes used in this meta-analysis have been provided as Supplementary file 4 . The shotgun data of the ELDERMET is available for download from the ELDERMET website at http://eldermet . ucc . ie/temp1/eldermet_shotgun_data_filtered_all_sample . tar . The shotgun data for the ELDERMET has also been uploaded at the European Nucleotide Archive ( ENA ) with the project accession number PRJEB37017 .	The human body is an ecosystem made up of both human cells and trillions of microbes , and the largest microbial community is in the gut . This community of gut microbes helps harvest nutrients from our food , modulates our immune system , and even affects our mood . Infectious and chronic diseases appear to cause changes in the make-up of the gut microbiome , while microbiome changes may increase the risk of some non-infectious diseases . Learning more about these disease-linked changes in the gut microbiome may therefore help scientists to develop new tests and treatments . To do this , scientists need to understand which microbes play a role in individual diseases , if risk-related microbes are gained or helpful microbes lost in patients with particular diseases , and if certain changes in gut microbes occur across many diseases . Ageing also changes the gut microbes . This may happen because older individuals eat a less complex diet and are likely to take many medications that may alter the microbes in their gut . Because of this , age may affect changes in gut microbes associated with diseases . This highlights the need for studies that tease apart the importance of ageing-related and disease-related changes in the gut microbiome . Now , Ghosh et al . show that gut microbe changes linked to diseases may vary with a person’s age . The analysis compared the gut microbiomes of more than 2 , 500 individuals aged 20 to 89 . This included individuals with inflammatory bowel disease , colorectal cancer , type 2 diabetes , intestinal polyps and liver cirrhosis . The study revealed that younger people gradually gain disease-associated gut microbes , while older people tend to lose the gut microbes usually found in a healthy gut . Ghosh et al . also identified a set of gut microbes that were gained in many diseases and across age-groups . This set of microbes was also associated with frailty in elderly people . The characteristics of the microbes in this set are all known to have detrimental effects on human health . This analysis shows how important it is to control for age and other factors that may skew the results of microbiome projects . Future studies are needed to understand why these gut microbe changes occur and what the consequences of these changes are for a person’s health and the course of their disease . This may lead to the development of treatment strategies that help promote a healthy gut microbiome and fight disease throughout life .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "microbiology", "and", "infectious", "disease" ]	2020	Adjusting for age improves identification of gut microbiome alterations in multiple diseases
During gastrulation , neural crest cells are specified at the neural plate border , as characterized by Pax7 expression . Using single-cell RNA sequencing coupled with high-resolution in situ hybridization to identify novel transcriptional regulators , we show that chromatin remodeler Hmga1 is highly expressed prior to specification and maintained in migrating chick neural crest cells . Temporally controlled CRISPR-Cas9-mediated knockouts uncovered two distinct functions of Hmga1 in neural crest development . At the neural plate border , Hmga1 regulates Pax7-dependent neural crest lineage specification . At premigratory stages , a second role manifests where Hmga1 loss reduces cranial crest emigration from the dorsal neural tube independent of Pax7 . Interestingly , this is rescued by stabilized ß-catenin , thus implicating Hmga1 as a canonical Wnt activator . Together , our results show that Hmga1 functions in a bimodal manner during neural crest development to regulate specification at the neural plate border , and subsequent emigration from the neural tube via canonical Wnt signaling . The neural crest is a vertebrate-specific stem cell population with the capacity to migrate long distances during embryonic development ( Bronner and LeDouarin , 2012; Le Douarin , 1980; Simões-Costa and Bronner , 2013 ) . Originating at the neural plate border , these cells occupy the leading edges of the closing neural folds during neurulation . Subsequently , premigratory neural crest cells that initially reside within the dorsal aspect of the developing neural tube undergo an epithelial-to-mesenchymal ( EMT ) transition in order to delaminate and migrate extensively . Upon reaching their terminal locations within the embryo , neural crest cells differentiate into a plethora of derivatives , including craniofacial cartilage , pigment cells , smooth muscle , and peripheral neurons and glia ( reviewed in Gandhi and Bronner , 2018 ) . A feed-forward gene regulatory network ( GRN ) underlies the formation of the neural crest , from induction at the neural plate border to final differentiation into a multitude of cell types . This GRN is comprised of transcription factors and signaling pathways , partitioned into developmental modules ( Martik and Bronner , 2017; Simões-Costa et al . , 2015 ) . Recently , new tools like single-cell RNA sequencing ( scRNA-seq ) and Assay for Transposase-Accessible Chromatin using sequencing ( ATAC-seq ) have enabled analysis of the neural crest GRN at a global level , helping to clarify lineage trajectories and elucidate key biological processes therein , ranging from proliferation to differentiation ( Williams et al . , 2019 ) . These approaches have opened the way to extensive functional analysis of important nodes within the GRN , particularly at early stages of neural crest development , which are less well-studied . Neural crest formation begins at the gastrula stage , with establishment of the presumptive neural ectoderm bordering the non-neural ectoderm . Quantitative gene expression analysis of gastrula stage chick embryos has revealed a surprisingly high degree of overlap of multiple transcription factors associated with diverse cell types within single cells in the early neural plate border region , ranging from markers characteristic of the neural crest ( Pax7 ) , to neural ( Sox2 ) and placodal ( Six1 ) cell types ( Roellig et al . , 2017 ) . This is consistent with the possibility that cells in the neural plate border are transcriptionally primed toward multiple cell fates , rather than committed to a particular lineage . What then leads to cell lineage commitment and specification toward neural crest rather than alternative fates , and to their subsequent ability to initiate migration from the neural tube ? One possibility is that previously unidentified transcriptional and epigenetic regulators play a critical role in these processes . In this study , we used scRNA-seq to identify novel transcription factors and chromatin remodelers expressed in neural crest cells of the early chick embryo . We first describe the single-cell transcriptome of early migrating neural crest cells emerging from the hindbrain , with a focus on identifying new transcriptional regulators . One of the genes uncovered in the neural crest cluster was Hmga1 , a non-histone chromatin remodeler that has known roles in tumor metastasis ( Resar et al . , 2018 ) , but has been understudied in development . We first characterized the expression and function of Hmga1 during neural crest development using in situ Hybridization Chain Reaction ( HCR ) and observed Hmga1 transcripts enriched in the neural crest , with the onset of expression preceding neural crest specification in the neural plate border . To test its functional role in neural crest development , we used plasmid- and protein-based CRISPR-Cas9 strategies to knock out Hmga1 in neural crest progenitors with temporal precision . The results demonstrate an early role for Hmga1 in neural crest lineage specification in a Pax7-dependent manner , resulting in the downregulation of neural crest specifier genes such as Snail2 , FoxD3 , and Sox10 . Interestingly , loss of Hmga1 after completion of neural crest specification revealed a distinct set of defects in cranial neural crest emigration and migration . Using in situ hybridization and a fluorescent protein-based reporter , we show that this is a consequence of reduced canonical Wnt activity mediated by Wnt1 , which can be rescued by concomitantly expressing stabilized ß-catenin , thus establishing a separate role for Hmga1 in delaminating neural crest cells as a Wnt pathway activator . Taken together , these results identify a dual role for Hmga1 in neural crest development with an early effect on neural crest specification and a later effect on initiation of migration via the canonical Wnt signaling pathway , mechanisms that may be inappropriately redeployed during tumorigenesis . Many RNA-seq datasets have sought to examine genes that are enriched in cranial neural crest cells compared with other tissues ( Simões-Costa et al . , 2014 ) or axial levels ( Martik et al . , 2019 ) . However , here we aimed to identify highly expressed transcription factors and chromatin remodelers that may have been missed due to overlapping expression between neural crest cells and surrounding tissues . To this end , gastrula stage Hamilton Hamburger ( HH ) four embryos were electroporated with the neural crest enhancer FoxD3-NC2:eGFP and cultured ex ovo until stage HH12 ( Hamburger and Hamilton , 1951 ) . The NC2 enhancer labels early migrating neural crest cells ( Simões-Costa et al . , 2012 ) , thereby facilitating dissection of the region surrounding the rhombomere ( r ) six migratory neural crest stream for dissociation ( Figure 1A–A’ ) . To aid downstream analysis and clustering , we introduced an ‘outgroup’ of dissected primary heart tube cells into the single-cell suspension and generated barcoded Gel Bead-In-Emulsions ( GEMs ) on the 10X Genomics platform . The library was sequenced at a depth of 50 , 000 median reads/cell to profile a total of 1268 cells , out of which 1241 cells passed the quality control filters ( Figure 1—figure supplement 1A–C ) . Following mapping and dimensional reduction , the cells split into distinct cellular subtypes ( Figure 1B ) , including five cell types ( mesoderm , otic , ectoderm , hindbrain , and neural crest ) derived from the dissected tissue , and the spiked-in outgroup ( ‘Heart tube’; Myl2+ , Tnnt2+ ) . Known genetic markers that were enriched in each population served to distinguish the neural crest subcluster ( Tfap2B+ , ItgB3+ ) from the surrounding tissues ( i . e . otic placode ( Cldn3+ , Gbx2+ ) ; hindbrain ( Pax6+ , Zic2+ ) ; ectoderm ( Epcam+ , Crabp2+ ) ; mesoderm ( FoxC2+ , Col1A1+ ) ) ( Figure 1—figure supplement 2A–B ) . Consistent with the anatomical diversity of the mesoderm , the latter was further subdivided into specific cell types like myocardium ( Hand2+ ) and paraxial mesoderm ( Prrx1+ ) ( Figure 1—figure supplement 1D–E ) . We particularly focused our subsequent analysis on the neural crest cluster in the context of the neighboring tissues of hindbrain and ectoderm ( Figure 1C , D , D’ ) . We sought to determine all transcription factors and chromatin regulators that were expressed in the neural crest-specific subcluster , regardless of their expression in other cell types . To this end , we shortlisted all genes associated with the gene ontology terms ‘DNA-binding’ , ‘regulation of transcription’ , and ‘transcription factor’ . This revealed several chromatin remodelers and transcription factors with high levels of expression in neural crest cells ( Figure 1E; Figure 1—figure supplement 2C ) . The identified genes fell into two groups , the first of which was comprised of transcription factors enriched in neural crest cells , with little overlap in surrounding cell types . As expected , many of these genes , including Sox10 , Ets1 , MafB , and Nrip1 , are known for their expression in the neural crest ( Gandhi et al . , 2020; Tani-Matsuhana et al . , 2018 ) . Importantly , the second group was comprised of chromatin remodelers and/or transcriptional regulators previously overlooked in bulk transcriptomic datasets , including Hmga1 , Dact2 , Ssrp1 , and Tbxl1x , due to overlapping expression in other tissues . Indeed , their distribution in low-dimensional space confirmed that a high proportion of cells in the hindbrain and/or ectoderm also expressed these genes ( Figure 1F ) . The expression of a subset of the above genes was validated at HH12 ( Figure 1G–K ) by in situ hybridization chain reaction ( HCR ) , which revealed an abundance of transcripts in both r4 and r6 neural crest streams that emerge from the hindbrain . Taken together , the results show that our single-cell gene expression analysis is sufficient to resolve the underlying heterogeneity of the chick hindbrain . We also identified several genes expressed in migrating neural crest cells not highlighted in previous datasets given their broad expression in other tissues . Of the novel transcriptional regulators that were previously overlooked in bulk transcriptomic datasets , we were particularly intrigued by the chromatin remodeler Hmga1 , due to its extensively studied role in tumorigenesis . A member of the High Motility Group A ( HMGA ) superfamily , Hmga1 encodes a small , nonhistone chromatin remodeling protein that binds to the minor groove of DNA , thereby affecting the chromatin landscape and facilitating the binding of other transcription factors in the opposing major groove ( Grosschedl et al . , 1994 ) . In developing mouse embryos , Hmga1 has been shown to have widespread expression across several tissues , including the brain , where its loss has been correlated with reduced developmental potential of neural precursor cells ( Kishi et al . , 2012 ) . While well-studied in cancer ( Masciullo et al . , 2003; Sarhadi et al . , 2006 ) , little was known about its developmental function in neural crest cells . Given the parallels between the mechanisms that regulate delamination , migration , proliferation , and survival of neural crest cells and tumor cells ( Gallik et al . , 2017 ) , we sought to characterize the expression of Hmga1 during neural crest development . At HH12 , Hmga1 was observed in migrating neural crest streams emanating from rhombomeres 4 and 6 ( Figure 1K ) , and in the cranial mesenchyme , suggesting that its expression is not restricted to the hindbrain neural crest . Therefore , to determine its spatiotemporal pattern at early stages of neural crest development , we performed HCR for Hmga1 at stages ranging from gastrulation ( HH4 ) , when neural crest cells are undergoing induction in the neural plate border ( Figure 2B’’’ ) , to HH10 , when neural crest cells have delaminated from the dorsal neural tube ( Figure 2E’’’ ) and are mid-migration in the cranial region ( Figure 2F’’’ ) . As an early marker for the neural plate border and neural crest ( Basch et al . , 2006 ) , we co-labeled Pax7 transcripts at the aforementioned stages . Hmga1 transcripts were first detected in the neural plate and neural plate border , but not in the non-neural ectoderm at HH4+ , and preceded the expression of Pax7 in the neural plate border . Hmga1 expression remained high at HH5-6 ( Figure 2A–B ) , overlapping in the neural plate border ( Figure 2—figure supplement 1A ) with Pax7 transcripts ( Figure 2B’ ) , as observed in transverse sections ( Figure 2B” , Figure 2—figure supplement 1B–B’ ) . As the neural plate border elevated to form neural folds between HH7 and HH8 ( Figure 2D’’’ ) , expression of Hmga1 was retained in the neural tube ( Figure 2C–D ) and continued to overlap with Pax7 in the dorsal neural folds ( Figure 2D’ , Figure 2—figure supplement 1C–D’ ) . Between stages HH9 and HH10 , when neural crest cells delaminated from the dorsal neural tube ( Figure 2E’’’ ) and started migrating laterally ( Figure 2F’’’ ) , Hmga1 expression was retained in delaminating ( Figure 2E , E’ , E’’ , Figure 2—figure supplement 1E–F’ ) and migrating ( Figure 2F , F’ , F’’ , Figure 2—figure supplement 1G–H’ ) neural crest cells . Interestingly , transverse sections through a representative HH10 embryo revealed that within the migrating neural crest stream , Hmga1 was expressed in both leader and follower cells , as compared to Pax7 , which appeared to be downregulated in the leader cells ( Figure 2F’’ ) . Together , these results show that the onset of Hmga1 in the neural crest occurs in precursors at the neural plate border region prior to their specification and is retained in premigratory and migrating neural crest cells . Given that Hmga1 transcripts were enriched in the cranial neural crest , we sought to interrogate its possible role therein . To this end , we designed guide RNA plasmids ( gRNAs ) targeting the coding sequence of Hmga1 ( Figure 3—figure supplement 1A ) and electroporated them together with constructs encoding Cas9 and nuclear RFP on the right side of HH4 gastrula stage embryos ( Figure 3A ) . The left side of the embryo was electroporated with Cas9 , nuclear RFP , and a control gRNA , chosen for its lack of binding in chick cells ( Gandhi et al . , 2017 ) . Embryos were cultured ex ovo until stage HH9/9+ ( Figure 3B ) , after which they were processed for immunohistochemistry , in situ hybridization , and HCR . We first validated our knockout approach by probing for the expression of Hmga1 itself in knockout embryos using HCR . This revealed a significant reduction in the abundance of Hmga1 transcripts on the knockout side ( Figure 3C ) . We quantified this phenotype in whole-mount embryos and observed a 25% reduction in Hmga1 expression ( Figure 3D , Figure 3—source data 1; p<0 . 05 , Wilcoxon rank test ) . Notably , the loss of Hmga1 transcripts in the neural crest was more dramatic than in the neural tube , due to targeted electroporation of knockout reagents to the presumptive neural plate border region . Next , we investigated the effect of knocking out Hmga1 on Pax7 expression in neural crest cells . We examined Pax7 mRNA expression by HCR in embryos where Hmga1 was knocked out on the right side , relative to the left side which served as an internal control . Consistent with their hierarchical onset of expression , loss of Hmga1 resulted in a notable reduction in Pax7 mRNA levels ( Figure 3E ) . Next , we assessed whether the reduction in Pax7 transcripts would consequently result in a loss of Pax7 protein in the neural crest by immunostaining Hmga1-knockout embryos with a Pax7 antibody . As expected , Pax7 protein levels were dramatically reduced in the migratory cranial neural crest ( Figure 3F , H’ ) , with further analysis revealing a significant decrease in the number of Pax7+ cells on the knockout side compared to the control side ( Figure 3G , Figure 3—source data 2; p<0 . 0001 , student’s t-test ) . Moreover , in the absence of Hmga1 , cranial neural crest cells failed to migrate properly , as depicted by the expression of the migratory neural crest marker HNK1 ( Figure 3—figure supplement 1B ) . A transverse section through the hindbrain ( Figure 3H ) of another representative Hmga1-knockout embryo stained for Pax7 ( Figure 3H’ ) revealed a notable reduction in the expression of Pax7 ( Figure 3I ) as well as the neural crest specifier Snail2 on the knockout side ( Figure 3I’ ) . Furthermore , using in situ hybridization , we found that other neural crest specifier genes including FoxD3 ( Figure 3J , J’ , Figure 3—figure supplement 1D , E’ , E’’ ) , Tfap2b ( Figure 3K ) , Sox10 ( Figure 3L , L’ ) , and c-Myc ( Figure 3—figure supplement 1C ) were also downregulated on the knockout side . In transverse sections through Hmga1-knockout embryos labeled for Sox10 expression ( Figure 3L ) , we also detected fewer Pax7+ cells ( Figure 3N ) and diminished levels of HNK1 expression ( Figure 3M ) . On the other hand , no notable difference in the thickness of the neural tube was observed ( Figure 3H , J’ , M’’ , Figure 3—figure supplement 1E ) , suggesting that the targeted knockout of Hmga1 in the neural plate border affected the neural folds/dorsal neural tube but not the neural plate itself . Taken together , our results indicate that Hmga1 is important for proper specification of neural crest cells . As the expression of Hmga1 precedes that of Pax7 in the neural plate border , and its loss also causes a reduction of Pax7 levels in neural crest cells ( Figure 3F ) , we next asked if this regulation occurs in the neural crest precursors that reside in the neural plate border . In the preceding experiments , we characterized the expression of neural crest markers following CRISPR-plasmid-mediated loss of Hmga1 at stages corresponding to early stages of neural crest emigration and migration . However , one caveat to our plasmid knockout strategy is that it takes time after electroporation of the CRISPR constructs in gastrula stage embryos for Cas9 to transcribe , translate , and properly fold to form a functional Cas9-gRNA complex . As specification of neural crest cells at the neural plate border is ongoing at HH4 , this means that functional Cas9 would not be available until well after initial electroporation . Given that neural crest development occurs in a rostral-to-caudal wave along the body axis ( Gandhi and Bronner , 2018 ) , we speculated that effects of plasmid electroporation would be more penetrant in the hindbrain compared to anterior regions of the embryo , such as the midbrain . To test this possibility , we generated transverse sections through the midbrain ( Figure 4A ) and hindbrain ( Figure 4B ) of Hmga1-knockout embryos and quantified the number of Pax7+ cells as a ratio of cell count on the experimental versus control sides . Accordingly , we noticed that while this ratio was 0 . 948 ± 0 . 036 ( n = 5 ) at the midbrain level , it was significantly reduced in the dorsal hindbrain , with a mean ratio of 0 . 69 ± 0 . 08 ( n = 5; p<0 . 05 , paired Student’s t-test ) ( Figure 4C , Figure 4—source data 1 ) . These results supported our assumption that specification may have already occurred at the midbrain level by the time Cas9 was functionally active . To circumvent this issue and test the earliest effects of knocking out Hmga1 concomitant with its onset of expression , we turned to an alternative CRISPR knockout strategy that enabled the loss of Hmga1 immediately after transfection . To this end , we electroporated recombinant Cas9 protein with in vitro-transcribed Hmga1 or control gRNA as ribonucleoprotein ( RNP ) complexes on the right and left sides of gastrula stage embryos , respectively ( Figure 4D ) , and cultured embryos ex ovo until HH6 . First , to validate the Cas9-protein-mediated knockout strategy , we labeled Hmga1 transcripts in knockout embryos using HCR and observed a very efficient reduction in Hmga1 expression ( Figure 4E ) , especially in the neural plate border , thereby offering precise temporal control over the loss of this gene’s activity . Next , we assayed for the expression of Pax7 in the neural plate border by immunostaining Hmga1-knockout embryos and found that the levels of Pax7 protein in the neural plate border ( Figure 4F , Figure 4—figure supplement 1A ) were severely downregulated . Transverse sections through the experimental compared with control sides revealed that neural plate border cells no longer expressed Pax7 after the loss of Hmga1 ( Figure 4H , I ) , and that this was not a result of premature apoptosis or aberrant cell proliferation ( Figure 4—figure supplement 1B–G’ , Figure 4—figure supplement 1—source data 1 ) . In further support of the latter , the thickness of the neural plate border remained unchanged ( Figure 4H’ , I’ ) . We also quantified the corrected Pax7 total fluorescence intensity ( C . T . C . F ) in the neural plate border and observed a statistically significant difference between the control and knockout sides ( p<0 . 01 , paired student’s t-test ) , with mean Pax7 intensity on the control side being 89 . 853 ± 23 . 388 a . u . ( n = 5 ) as compared to 42 . 763 ± 16 . 079 a . u . ( n = 5 ) on the knockout side ( Figure 4J , Figure 4—figure supplement 1H , Figure 4—source data 2 ) . Taken together , these results suggest that Hmga1 is required for the expression of Pax7 in neural crest precursors that are induced in the neural plate border . The neural plate border was initially thought to contain discrete domains corresponding to neural , neural crest , placodal , and epidermal precursors . However , recent work has demonstrated that cells in this region co-express genes characteristic of different cell fates and exhibit a broad developmental potential , suggesting they are not restricted to individual cell fates until later in development ( Roellig et al . , 2017 ) . In chick embryos , Tfap2a and Msx1 are expressed in the neural plate border , with Tfap2a transcripts spanning both the neural plate border and the non-neural ectoderm ( de Crozé et al . , 2011; Luo et al . , 2003 ) , whereas Msx1 transcripts are expressed within a subset of Pax7+ cells in the neural plate border region ( Khudyakov and Bronner-Fraser , 2009 ) . Given that loss of Hmga1 resulted in a reduction in Pax7 protein levels in the neural plate border , we asked if this was a result of a general neural plate border defect versus a selective effect on Pax7 . To test this , we examined the expression of Tfap2a ( Figure 4K ) and Msx1 ( Figure 4N ) transcripts together with Pax7 using HCR in Hmga1-knockout embryos developed to neurula stages . Consistent with the loss of Pax7 protein in the neural plate border , Hmga1 knockout caused reduced Pax7 mRNA levels on the experimental ( Figure 4M , P ) compared to the control ( Figure 4L , O ) sides . However , the expression of Tfap2a ( Figure 4L’ , M’ ) and Msx1 ( Figure 4O’ , P’ ) was retained in the absence of Hmga1 , together with no noticeable difference in the thickness of the neural plate border ( Figure 4L’’ , M’’ , O’’ , P’’ ) . Since Pax7 and Msx1 are specifically expressed in the neural plate border , we quantified the corrected total cell fluorescence intensity per unit area associated with their transcripts and calculated the ratio between the experimental and control sides . Under control conditions , this ratio would be close to 1 . However , for Pax7 , the mean calculated ratio was 0 . 552 ± 0 . 06 ( n = 6 ) , with a statistically significant difference between the experimental and control sides ( p<0 . 001 , paired Student’s t-test ) . On the other hand , while the mean calculated ratio for Msx1 was 0 . 809 ± 0 . 098 ( n = 3 ) , the fluorescence intensities were not significantly different between the two sides ( Figure 4Q , Figure 4—source data 3 ) . Taken together , these results show that Hmga1 specifically regulates Pax7 expression at the neural plate border . Given that loss of Hmga1 affects neural crest specification , we asked if its overexpression would have the converse effect . To exogenously provide Hmga1 , we designed a plasmid construct containing the coding sequence of Hmga1 under the regulation of a ubiquitous enhancer/promoter combination ( Figure 5A ) . This construct also contained the coding sequence for nuclear RFP downstream of an Internal Ribosome Entry Site ( IRES ) , allowing identification of successfully transfected cells . To test the effect of overexpressing Hmga1 , we electroporated this construct on the right side of gastrula stage embryos and cultured them ex ovo until HH9+ . The left side served as an internal control and was electroporated with an equal concentration of a construct encoding nuclear RFP alone ( Figure 5B ) . The results show that , rather than having the opposite effect to loss of function , overexpression of Hmga1 also resulted in a notable reduction in Pax7 expression on the experimental side ( Figure 5C ) . This suggests that maintaining appropriate levels of Hmga1 is critical for proper neural crest specification . The gold standard to demonstrate specificity for loss-of-function experiments is to perform a rescue . We posited that if modulating the levels of Hmga1 was important for neural crest formation , then exogenous expression of Hmga1 in an Hmga1-knockout background should successfully rescue neural crest cell numbers . Since the Protospacer Adjacent Motifs ( PAMs ) adjacent to both Hmga1 gRNAs are located in the introns ( Figure 3—figure supplement 1A ) , the coding sequence on the plasmid would be guarded against the endonuclease activity of Cas9 . To test our hypothesis , we knocked out Hmga1 using CRISPR plasmids as described above , but co-electroporated the ‘rescue’ construct on the right side . The left side was electroporated with an equal concentration of a plasmid encoding nuclear RFP . Embryos cultured to HH9+ and immunostained for Pax7 revealed that overexpression of Hmga1 concomitant with knocking out the endogenous gene successfully rescued Pax7 levels in cranial neural crest cells ( Figure 5D ) . To quantify the extent of rescue , we calculated the ratio of the number of Pax7+ cells on the experimental versus control sides in both wholemount ‘knockout’ ( from Figure 3G ) and ‘rescue’ embryos ( Figure 5D ) . In unperturbed embryos , this ratio will be close to 1 , reflecting similar numbers of Pax7+ cells on both sides of the embryo . However , in the ‘knockout’ group , we observed a mean ratio of 0 . 767 ± 0 . 041 ( n = 11 ) , which was significantly different from the ratio of 0 . 946 ± 0 . 016 ( n = 8 ) observed in the ‘rescue’ group ( p<0 . 001 , Welch two-sample t-test ) ( Figure 5E , Figure 5—source data 1 ) . Next , we probed for the expression of the neural crest specifier gene Sox10 to ask if rescuing the expression of Hmga1 truly restored the process of neural crest specification . To do this , we processed ‘rescue’ embryos for HCR against Sox10 ( Figure 5F ) together with Pax7 ( Figure 5G ) . Indeed , restoring the levels of Hmga1 was sufficient to rescue the expression of Sox10 ( Figure 5G’ ) and Pax7 ( Figure 5G’’ ) in early migrating crest , and Pax7 in the premigratory crest residing in the dorsal neural tube , as visualized in transverse sections through the embryo . We also confirmed that the expression of the ‘rescue’ construct was restricted to the dorsal neural tube ( Figure 5G’’’ ) , thereby precluding unintended effects on neural tube development . Finally , given that the loss of Hmga1 specifically affected Pax7 expression in neural crest precursors , we asked if exogenous expression of Pax7 would be sufficient to rescue the effects of losing Hmga1 on neural crest specification . We tested this by overexpressing the coding sequence of Pax7 ( Roellig et al . , 2017 ) on the right side of gastrula stage embryos together with CRISPR plasmids targeting Hmga1 ( Figure 5H , H’ ) . Given that the effect of CRISPR-plasmid-mediated loss of Hmga1 was more penetrant posteriorly , we developed ‘Pax7-rescue’ embryos to HH9+ , processed them for HCR against the neural crest specifier genes Tfap2b ( Figure 5I ) and Snai2 ( Figure 5I’ ) , and generated transverse sections through the hindbrain . Tfap2b is expressed in delaminating and migrating neural crest cells ( Simoes-Costa and Bronner , 2016 ) , whereas Snai2 is expressed in premigratory neural crest and is eventually downregulated as the cells begin to migrate ( Taneyhill et al . , 2007 ) . Compared to the Hmga1-knockout embryos in which Tfap2b mRNA ( Figure 3K ) and Snail2 protein ( Figure 3I’ ) levels were notably downregulated , restoring the levels of Pax7 in an Hmga1 knockout background rescued the expression of both Tfap2b ( Figure 5J ) and Snai2 ( Figure 5K ) in the dorsal hindbrain . Together , these results suggest that maintaining the correct levels of Hmga1 is necessary for proper neural crest specification in a Pax7-dependent manner . Neural crest induction , specification , and emigration from the neural tube are intricate processes that require an interplay between Wnt , FGF , and BMP signaling pathways ( Piacentino and Bronner , 2018; Woda et al . , 2003 ) working reiteratively at different stages of development . For example , at the onset of neural crest emigration , Wnt1 is prominently expressed in the dorsal neural tube , where premigratory neural crest cells reside ( Simões-Costa et al . , 2015 ) . As a result , these cells turn on Snai2 , a critical regulator of EMT ( Nieto et al . , 1994 ) known to function downstream of the Wnt signaling pathway . After knocking out Hmga1 using CRISPR plasmids , we noted not only perturbed emigration but also a dramatic reduction in Snail2 levels , even within the subset of Pax7-expressing cells in the dorsal neural folds ( Figure 3I’ ) . These results raised the intriguing possibility that this might be due to an effect on Wnt signaling in already-specified premigratory neural crest cells . Accordingly , we hypothesized that Hmga1 may function as a Wnt activator in these cells . If so , its loss would be predicted to result in reduced Wnt signaling in the dorsal neural tube . To test this possibility , we used a reporter construct expressing nuclear GFP under the regulation of six Tcf/Lef binding sites and a minimal promoter as a readout for canonical Wnt signaling ( Ferrer-Vaquer et al . , 2010; Figure 6A ) . Plasmids encoding Cas9 , gRNAs targeting Hmga1 , nuclear RFP , and Tcf/Lef:H2B-GFP were electroporated on the right side of gastrula stage embryos , while the left control side was electroporated with plasmids encoding Cas9 , control gRNA , nuclear RFP , and Tcf/Lef:H2B-GFP . As described above , this plasmid-based knockout strategy resulted in the loss of Hmga1 after neural crest specification in the midbrain but well before their emigration . Embryos were allowed to develop until HH9 , by which time neural crest cells have started delaminating from the neural tube at the midbrain level ( Figure 6B ) . Consistent with our hypothesis , the results show that Hmga1 knockout caused a notable reduction in canonical Wnt reporter activity on the knockout side compared to the control side ( Figure 6C ) at the midbrain level . Interestingly , these embryos had a neural crest migration defect ( Figure 6—figure supplement 1C–D ) but no notable difference in the number of Pax7+ cells between the two sides ( Figure 6—figure supplement 1E’ ) , as expected if the Hmga1 knockout occurred after specification was complete; this is consistent with previous work showing that perturbation of canonical Wnt signaling following specification does not affect the number of Pax7+ cells at cranial EMT stages ( Hutchins and Bronner , 2018 ) . Quantitation of this effect revealed a significant reduction in reporter activity following the loss of Hmga1 ( Figure 6D , Figure 6—source data 1; p<0 . 01; student’s t-test ) , as measured by comparing the ratio between the knockout and the control sides of transfected cells ( RFP+ ) that turned on Wnt signaling within the Pax7+ domain , therefore expressing GFP . While this ratio was expected to be one for embryos with unperturbed Wnt signaling on both sides , we observed a mean ratio of 0 . 325 ± 0 . 082 ( n = 5 ) , suggesting that Wnt activity was disrupted in the absence of Hmga1 . Next , to investigate the mechanism by which Hmga1 regulates Wnt signaling , we turned to a recently published cranial neural crest chromatin accessibility dataset ( Williams et al . , 2019 ) and looked for open chromatin regions surrounding genes that encode for known Wnt ligands . In particular , Wnt1 expression in the dorsal neural tube is known to be necessary for proper delamination of cranial neural crest cells ( Simões-Costa et al . , 2015 ) . Interestingly , we discovered a putative enhancer downstream of Wnt1 ( Figure 6—figure supplement 1A ) that contained an AT-rich domain consistent with Hmga1-binding motifs ( Figure 6—figure supplement 1B; Reeves and Nissen , 1990 ) . Therefore , we hypothesized that Hmga1 may modulate Wnt signaling by regulating Wnt1 expression . To test this , we knocked out Hmga1 on the right side of gastrula stage embryos using CRISPR plasmids , cultured them ex ovo until HH9 , and examined Wnt1 mRNA expression using in situ hybridization . Indeed , the dorsal neural tube expression of Wnt1 was severely downregulated ( Figure 6E ) in the midbrain . Consistent with the effect of losing Hmga1 after neural crest specification , the number of Pax7+ cells in the dorsal neural tube appeared unchanged ( Figure 6E’ ) . In contrast , no change in Wnt1 expression was observed at the hindbrain level ( Figure 6F ) which , being developmentally ‘younger , ’ instead exhibited a specification defect that resulted in fewer Pax7+ cells in the dorsal neural tube on the experimental side compared to the control side ( Figure 6F’ ) . Interestingly , following Hmga1 knockout , we also observed defects in basement membrane remodeling and laminin channel formation at midbrain levels ( Figure 6G ) , another Wnt-dependent process necessary for neural crest EMT; consistent with Wnt inhibition via Draxin overexpression ( Hutchins and Bronner , 2019 ) , loss of Hmga1 abrogated laminin remodeling and resulted in physical blockage of the channel through which migrating cranial neural crest cells normally transit ( Figure 6H ) . Together , these data indicate that after neural crest specification , Hmga1 is necessary for the expression of Wnt1 and activation of canonical Wnt signaling in the dorsal neural tube , and by extension , Wnt-dependent neural crest delamination/EMT . Finally , given that Hmga1 functions as a canonical Wnt pathway activator , we asked if the migration defects caused by the loss of Hmga1 post-specification can be rescued by restoring canonical Wnt signaling in premigratory neural crest cells . To address this , we expressed GFP-tagged , stabilized ß-catenin ( NC1-∆90 ß-cat ) to upregulate canonical Wnt signaling output specifically in premigratory neural crest cells , thus circumventing the critical process of neural crest induction at the neural plate border ( Hutchins and Bronner , 2018 ) . If loss of Hmga1 in premigratory neural crest cells resulted in migration defects due to reduced canonical Wnt signaling , expression of a stabilized ß-catenin would be predicted to restore those levels , thereby rescuing proper migration . To test this , we knocked out Hmga1 on the right side of gastrula stage embryos using CRISPR plasmids as previously described , but co-electroporated NC1-∆90ß-cat-GFP on the right side . The left side was electroporated with control reagents ( Figure 6I ) . Embryos were cultured ex ovo until HH9+ and processed for immunohistochemistry against Pax7 . Consistent with our hypothesis that Hmga1 functions as a Wnt activator in neural crest cells , expression of stabilized ß-catenin was sufficient to rescue proper cranial neural crest migration from Hmga1 knockdown ( Figure 6J ) . We also calculated the ratio of the area occupied by migrating cranial neural crest cells between the experimental and control sides ( Figure 6K , Figure 6—source data 2 ) . For wildtype embryos , this ratio would be close to 1 , reflecting equal neural crest migration on both sides of the embryo . However , after the loss of Hmga1 , the ratio of areas on experimental versus control sides was 0 . 642 ± 0 . 028 ( n = 5 ) . Importantly , co-expression of stabilized ß-catenin in premigratory neural crest rescued migration , with an average migration ratio of 1 . 015 ± 0 . 035 ( n = 4 ) , which was significantly different from the knockout group ( p<0 . 0001 , ANOVA and post hoc Tukey HSD ) . Similarly , ubiquitous expression of the Hmga1 coding sequence also rescued neural crest migration , with an average migration ratio of 0 . 972 ± 0 . 044 ( n = 5 ) , which was also significantly different from the knockout group ( p<0 . 001 , ANOVA and post hoc Tukey HSD ) . Taken together , our results suggest that Hmga1 mediates the process of EMT by activating canonical Wnt signaling in premigratory neural crest cells , thus enabling them to emigrate from the neural tube . While chromatin modifiers are known to influence gene expression and cell fate decisions at many stages of development ( Cai et al . , 2014; Laugesen and Helin , 2014; Miller and Hendrich , 2018; O'Shaughnessy-Kirwan et al . , 2015 ) , parsing cell-type-specific functions and targets for these proteins is often challenging due to broad expression across multiple tissues and time points . In this study , we have used scRNA-seq to identify a chromatin-remodeling protein , Hmga1 , as highly expressed in neural crest cells . Using high-resolution in situ HCR and temporally controlled knockdowns , we present evidence for a dual role of Hmga1 in the formation and migration of neural crest cells . At early stages , we find that the neural plate border gene Pax7 is a downstream target of Hmga1 , such that loss of Hmga1 blocks neural crest specification in a manner that can be rescued by restoring Pax7 expression . After neural crest specification is complete in the closing neural tube , Hmga1 plays a second role in modulating canonical Wnt signaling via alterations in the levels of Wnt1 in premigratory neural crest cells . This in turn influences neural crest EMT and delamination from the dorsal neural tube ( Figure 7 ) . The canonical Wnt signaling pathway is a major input in a complex GRN that activates transcriptional circuits and controls neural crest specification and cell lineage decisions ( Martik and Bronner , 2017; Simões-Costa and Bronner , 2015; Williams et al . , 2019 ) , influencing multiple aspects of neural crest development from induction at the neural plate border to proliferation , onset of migration via EMT , and differentiation ( Milet and Monsoro-Burq , 2012; Rabadán et al . , 2016; Simões-Costa and Bronner , 2015; Steventon et al . , 2009; Wu et al . , 2003; Yanfeng et al . , 2003 ) . For example , regulation of the levels of canonical Wnt signaling is critical for progressive basement membrane remodeling during neural tube closure and neural crest delamination . Consequently , perturbation of Wnt signaling output at different stages of basement membrane remodeling or delamination causes severe defects in neural crest EMT ( Hutchins and Bronner , 2019; Rabadán et al . , 2016 ) . Interestingly , early inhibition of canonical Wnt signaling in gastrula stage chick embryos has been shown to reduce Pax7 expression ( Basch et al . , 2006 ) , whereas canonical Wnt inhibition after neural crest specification does not alter Pax7 expression but has a marked effect on EMT ( Hutchins and Bronner , 2018 ) . This suggests that there are separable early versus late effects of canonical Wnt signaling during neural crest development . Hmga1 has been shown to upregulate canonical Wnt signaling components and downstream targets in the intestinal stem cell niche , thereby amplifying signaling output ( Xian et al . , 2017 ) presumably through increased promotor accessibility . Our results are consistent with a similar role for Hmga1 in the neural crest , where its loss resulted in decreased output from a canonical Wnt reporter ( Figure 6 ) as well as downregulation of the Wnt1 ligand ( Figure 6 ) and the canonical Wnt target Snail2 ( Figure 3 ) . However , temporally controlled knockdowns revealed that loss of Hmga1 reduced Wnt1 expression following completion of neural crest specification , but not at earlier stages of neural crest induction . Conversely , Hmga1 knockdown affected Pax7 expression during neural crest induction but not after specification is complete . One possible explanation is that once open/remodeled , the chromatin landscape surrounding the Pax7 regulatory regions form topologically associating domains ( TADs ) that are stable and resistant to repression . Alternatively , other Hmga1-independent cis-elements may influence Pax7 expression following neural crest induction . Given that neural crest cells appear to have highly dynamic chromatin accessibility surrounding spatiotemporally regulated enhancer elements ( Williams et al . , 2019 ) , we would predict the latter , although further investigation is needed to distinguish between these possibilities . An intriguing possibility is that Hmga1 , in addition to regulating neural crest EMT , is necessary for maintaining the broad developmental potential of neural crest cells . Neural crest cells exhibit stem-cell-like properties , including multipotency and self-renewal . Thus , the early expression of Hmga1 in neural crest precursors , together with its reported role in maintaining stemness and self-renewal properties in various stem cell and cancer systems ( Battista et al . , 2003; Schuldenfrei et al . , 2011; Shah and Resar , 2012 ) , may indicate a role in maintaining neural crest stemness . In the dorsal neural tube , self-renewal within the premigratory neural crest is driven by the transcription factor c-Myc ( Kerosuo and Bronner , 2016 ) , which together with its binding partner regulates cell cycle progression . Consistent with this possibility , we found that the mRNA expression of c-Myc is downregulated in the dorsal neural tube following Hmga1 knockout . However , given that canonical Wnt signaling also regulates stem cell self-renewal ( Xu et al . , 2016 ) , as well as cell cycle progression in neural crest ( Burstyn-Cohen et al . , 2004 ) , Hmga1 may have additional Wnt-dependent and/or Wnt-independent roles in the maintenance of the neural crest stem cell pool . Hmga1 is a member of the high motility group A ( HMGA ) family of genes that are characterized by their A/T-hook domains and the ability to transform chromatin architecture to regulate transcription of target genes . To date , two members of the HMGA family , Hmga1 and Hmga2 , have been identified in mammals ( Reeves et al . , 2001 ) , each having distinct roles in oncogenesis ( Jun et al . , 2015; Meyer et al . , 2007; Miyazawa et al . , 2004 ) . As both genes share several common targets , they may compensate for each other where they overlap . Indeed , Hmga1/Hmga2 double knockout mice have severe embryonic lethality , compared to less-penetrant effects in individual knockouts ( Federico et al . , 2014 ) . While both Hmga1 and Hmga2 genes are annotated in the chick genome , our scRNA-seq results show that a significant proportion of neural crest cells express Hmga1 but not Hmga2 . Consistent with this , loss of Hmga1 resulted in neural crest-related developmental defects , making it unlikely that there is redundancy and/or compensation by Hmga2 in chick embryos . Interestingly , only the hmga2 paralogue is present in Xenopus laevis embryos ( Macrì et al . , 2016 ) , but morpholino-mediated knockdown of hmga2 did not affect the expression of neural plate border gene pax3/7 . This contrasts with our results in chick embryos , where the loss of Hmga1 affected both Pax7 transcription and protein levels in the neural plate border . Together with the absence of Hmga2+ cells in our single-cell data , this raises the possibility that individual HMGA family members play discrete roles in neural crest development , similar to their distinct roles in tumorigenesis . In addition to Hmga1 , other chromatin remodeling proteins serve similar functions in neural crest cell fate decisions . For example , both the ATP-dependent chromatin remodeler CHD7 and the histone demethylase Jumonji D2A ( KDM4A/JmjD2A ) are necessary for expression of neural crest specifier genes; notably , however , these chromatin modifiers appear to function at later developmental time points than Hmga1 , as neither knockdown of KDM4A in chick embryos nor CHD7 in Xenopus embryos affect Pax3/7 expression levels ( Bajpai et al . , 2010; Strobl-Mazzulla et al . , 2010 ) . Furthermore , KDM4A influences Snai2 and Wnt1 levels , raising the possibility that Hmga1 may act in concert with KDM4A or other chromatin remodelers to restructure the accessibility of neural crest GRN circuits at different cell fate checkpoints . In summary , our data reveal a dual role for the chromatin remodeler Hmga1 during neural crest development . First , during specification at the neural plate border , Hmga1 regulates the completion of neural crest induction as assayed by readout of Pax7 expression at the neural plate border . Later , following neural crest specification , Hmga1 plays a second role in modulating the levels of the canonical Wnt signaling pathway in the closing dorsal neural tube to influence neural crest EMT and delamination at the onset of their migration . Post-embryonically , neural-crest-derived cells are prone to metastasis and give rise to numerous cancers ( Maguire et al . , 2015 ) . Furthermore , neural crest and cancer cells often employ similar mechanisms to drive EMT; in particular , the hallmarks of metastasis often involve disruption of the basement membrane which may also be driven by canonical Wnt signaling ( Gallik et al . , 2017; Powell et al . , 2013 ) . Interestingly , high expression levels of Hmga1 have been associated with premature EMT and prolonged stemness in several cancers of the pancreas ( Abe et al . , 2000 ) , breast ( Flohr et al . , 2003 ) , lung ( Sarhadi et al . , 2006 ) , and ovaries ( Masciullo et al . , 2003 ) . Therefore , it is interesting to note that Hmga1 may play parallel roles in neural crest development and cancer metastasis . Understanding how Hmga1 , and chromatin remodeling in general , alters cell fate decisions and EMT through signaling and transcriptional regulation in neural crest cells will undoubtedly have important and broad implications in human development and disease . Chicken embryos ( Gallus gallus ) were commercially obtained from Sun Valley farms ( CA ) , and developed to the specified Hamburger-Hamilton ( HH ) ( Hamburger and Hamilton , 1951 ) stage in a humidified 37°C incubator . For ex ovo electroporations , embryos were dissected from eggs at HH4 , injected with specified reagents , then electroporated as described previously ( Sauka-Spengler and Barembaum , 2008 ) . Following electroporation , embryos were cultured in fresh albumin/1% penicillin-streptomycin at 37°C and grown to specified HH stages . Once the embryos reached the desired stages , they were screened for transfection efficiency and overall health . Unhealthy and/or poorly transfected embryos were discarded and not included for downstream assays . HH4 embryos electroporated with FoxD3-NC2:eGFP were cultured until HH12 ex ovo at 37°C , following which the hindbrain region spanning rhombomeres 6 , 7 , and 8 was dissected under a fluorescence microscope . For dissociation , several different conditions were tested ( dissociation in a glass dish for 1 hr in Accumax ( EMD Millipore ) , chemical dissociation on a nutator for 15 , 30 , and 45 min , and chemical dissociation with gentle pipetting for 15 , 30 , and 45 min ) . The quality of the single-cell suspension obtained was tested by a trypan blue-based live-dead staining . Accordingly , we pooled dissected tissue washed in chilled 1X DPBS and incubated it in Accumax cell dissociation solution for 15 min at 37°C with gentle mixing every 5 min . Dissociation was terminated using Hanks Buffered Saline Solution ( HBSS ) ( Corning ) supplemented with BSA Fraction V ( Sigma; 0 . 2% w/v ) . The suspension was centrifuged at 300 g for 4 min to collect cells at the bottom , the supernatant was removed , and the pellet was resuspended in 1 mL HBSS-BSA . To remove cell debris and clumps , the 1 mL suspension was passed through a 20 µm filter in a clean hood . This suspension was loaded on a 10X Chromium chip A ( v2 ) to generate GEMs . The library was prepared according to the manufacturer’s protocol and sequenced on the Illumina HiSeq platform using the paired end chemistry . The raw fastq files were aligned to the galgal6 ( GRCg6a ) genome assembly obtained from the ENSEMBL database using the cellranger pipeline downloaded from the 10X Genomics website . For feature counts , a custom galgal6 GTF file , where all annotated 3’ UTRs were extended by 1 kb , was used . This was done to compensate for improper gene annotations in the chick genome . The count matrices were then imported in R for analysis using Seurat ( Butler et al . , 2018 ) . The initial filtering step discarded all cells with fewer than 200 and more than 10 , 000 genes per cell . We also filtered out cells expressing more than 5% mitochondrial or less than 20% ribosomal genes . Next , we removed genes corresponding to small RNAs , micro RNAs , mitochondria , and general housekeeping from the count matrix . Following log normalization and Principal Component Analysis , the cells were clustered using the first 15 dimensions ( calculated from the elbow plot ) . Different resolution parameter values were tested , and a value of 0 . 45 was used to identify subpopulations within the data . Dimensional reductionality was performed using the UMAP ( McInnes et al . , 2018 ) algorithm . All plots were created in R , exported in SVG format , and assembled in Inkscape . HCR v3 was performed using the protocol suggested by Molecular Technologies ( Choi et al . , 2018 ) with minor modifications . Briefly , the embryos were fixed in 4% paraformaldehyde ( PFA ) overnight at 4°C or 2 hr at room temperature , washed in 0 . 1% PBS-Tween , dehydrated in a series of 25% , 50% , 75% , and 100% methanol washes , and incubated overnight at −20°C in 100% methanol . The next day , the embryos were rehydrated , treated with proteinase-K for 2–2 . 5 min , and incubated with 10 pmol of probes dissolved in hybridization buffer overnight at 37°C . The next day , following several washes in ‘probe wash buffer , ’ embryos were incubated in 30 pmol of hairpins H1 and H2 diluted in Amplification buffer at room temperature overnight . The next morning , embryos were washed in 0 . 1% 5x-SSC-Tween and imaged . All probes were designed and ordered through Molecular Technologies . The coding sequence of Hmga1 was obtained from the UCSC genome browser ( Karolchik et al . , 2003 ) . A V5 tag was cloned in-frame at the N-terminus using overlap PCR ( Accuprime ) . The fusion product was cloned downstream of the CAGG promoter , upstream of the IRES-H2B-RFP segment of pCI-H2B-RFP ( Betancur et al . , 2010 ) to clone the final plasmid ( CAGG >V5-HMGA1-IRES-H2B-RFP ) . The Cas9 and gRNA constructs ( Gandhi et al . , 2017 ) , neural crest enhancer FoxD3-NC2:eGFP ( Simões-Costa et al . , 2012 ) , canonical Wnt reporter TCF/Lef:H2B-GFP ( Ferrer-Vaquer et al . , 2010 ) , and neural crest-specific stabilized ß-catenin NC1-∆90ß-cat ( Hutchins and Bronner , 2018 ) have all been previously described and validated . The genomic locus for Hmga1 was obtained from the UCSC genome browser ( Karolchik et al . , 2003 ) . Two gRNAs targeting the coding sequence , the first targeting exon 3 ( 5’-CCAGGAAGAAACCGGAGgta-3’ ) , and the second targeting exon 4 ( 5’-GCCAGCTCCAAAGGCAGGgt-3’ ) , were designed using CHOPCHOP ( Labun et al . , 2019 ) . The protospacers were cloned downstream of the chick U6 . 3 promoter as described in Gandhi et al . , 2017 . For control electroporations , the control gRNA described in Gandhi et al . , 2017 was used . CAGG >nls-Cas9-nls and CAGG >H2 B-RFP were electroporated at a concentration of 2 µg/µl , together with either 0 . 75 µg/µl per Hmga1 gRNA on the right side or 1 . 5 µg/µl of control gRNA on the left side . For Cas9/in vitro-transcribed gRNA RNP experiments , we generated single-guide RNAs ( sgRNAs ) as described previously ( Hutchins and Bronner , 2018 ) , using the following primers: Of the recombinant Cas9 ( M0646; New England Biolabs ) , 2 . 6 µl was mixed with equal volumes of control gRNA or 1 . 3 µl each of the two Hmga1 gRNAs , and heated to 37°C for 15 min . The solution was then incubated at room temperature for 15 min , mixed with 2 µg/µl H2B-RFP and 1 µl of sterilized 2% food dye , and injected in embryos for electroporation . Chromogenic in situ hybridization was performed as described previously for FoxD3 , Sox10 , and Tfap2b , c-Myc , and Wnt1 ( Kerosuo and Bronner , 2016; Simoes-Costa and Bronner , 2016; Simões-Costa et al . , 2015 ) . Immunohistochemistry was performed as described previously ( Gandhi et al . , 2017 ) . Briefly , embryos were fixed for 20 min at room temperature , blocked in 10% goat or donkey serum in 0 . 5% PBS-Triton overnight at 4°C , incubated overnight at 4°C in primary antibodies diluted in blocking solution , washed at room temperature in 0 . 5% PBS-Triton , incubated overnight at 4°C in secondary antibodies diluted in blocking solution , washed at room temperature in 0 . 5% PBS-Triton , and processed for imaging and/or cryosectioning . The following primary antibodies and concentrations were used: Mouse IgM HNK1 ( 1:5; Developmental Studies Hybridoma Bank ( 3H5 ) ) ; Mouse IgG1 Pax7 ( 1:10; Developmental Studies Hybridoma Bank ( RRID:AB_528428 ) ) ; Goat GFP ( 1:500; Rockland Cat# 600-101-215 ) ; Rabbit RFP ( 1:500; MBL Cat# PM005 ) ; Rabbit Snail2 ( 1:200; Cell Signaling Technology ( 9585 ) ) ; Rabbit Laminin ( 1:1000; Sigma-Aldrich ( L9393 ) ) ; Rabbit cleaved-Caspase 3 ( 1:500; R and D Systems Cat# AF835 ) ; Mouse phospho-histone H3 ( 1:500; Abcam Cat# Ab14955 ) . The following species-specific secondary antibodies labeled with Alexa Fluor dyes ( Invitrogen ) were used: Goat/Donkey anti-Mouse Alexa Fluor 647 ( for Pax7 and pH3; 1:250 ) , Goat/Donkey Goat anti-Mouse IgM Alexa Fluor 350/488 ( for HNK1; 1:250 ) , Goat/Donkey anti-Rabbit Alexa Fluor 488 ( for Snail2 , cleaved-Caspase3 , and Laminin; 1:250 ) , Donkey anti-goat Alexa Fluor 488 ( for Citrine; 1:500 ) , and Goat/Donkey anti-rabbit Alexa Fluor 568 ( for RFP; 1:500 ) . Following whole mount imaging , embryos were washed in 5% and 15% sucrose overnight at 4°C . The next day , embryos were transferred to molten gelatin for 3–5 hr at 37°C , embedded in molds at room temperature , frozen in liquid nitrogen , and stored at −80°C overnight . Embedded embryos were sectioned on a micron cryostat to obtain 16 µm sections through immunostained embryos and 20 µm sections through in situ hybridized embryos . The sections were degelatinized at 42°C in 1x PBS for 5 min , washed in 1x PBS , soaked in 1x PBS containing 0 . 1 µg/mL DAPI for 2 min , washed in 1x PBS and distilled water . Fluoromount mounting medium was used to mount coverslips on slides . Whole mount embryos and sections on slides were imaged on a Zeiss Imager M2 with an ApoTome module and/or Zeiss LSM 880 confocal microscope at the Caltech Biological Imaging Facility . Images were post-processed using FIJI imaging software ( Schindelin et al . , 2012 ) . To calculate corrected total cell fluorescence ( CTCF ) , the following formula was used:CTCF=Integrated Density- ( Selected area*Mean background fluorescence ) For cell counts , a median filter was applied to 8-bit images . A Bernsen-based auto local-thresholding method ( Bernsen , 1986 ) followed by watershed segmentation was used to identify cell boundaries . The ‘Analyze particles’ function was used to count the number of cells . All statistical analyses were performed in R . The Wilcoxon rank test was used in datasets that were not normally distributed . In cases where the underlying distribution was normal , a student’s t-test was used to calculate significance . In cases where multiple samples were compared , Analysis of Variance ( ANOVA ) test combined with Tukey HSD correction was used . Post hoc power analysis was used to validate sample size and confirm sufficient statistical power ( >0 . 8 ) .	The neural plate is a structure that serves as the basis for the brain and central nervous system during the development of animals with a backbone . In particular , the tissues at the border of the neural plate become the neural crest , a group of highly mobile cells that can specialize to form nerves and parts of the face . The exact molecular mechanisms that allow the crest to emerge are still unknown . The protein Hmga1 alters how genes are packaged and organized inside cells , which in turn influences how genes are switched on and off . Here , Gandhi et al . studied how Hmga1 helps to shape the neural crest in developing chicken embryos . To do so , they harnessed a genetic tool called CRISPR-Cas9 , and deleted the gene that encodes Hmga1 at specific developmental stages . This manipulation highlighted two periods where Hmga1 is active . First , Hmga1 helped to define neural crest cells at the neural plate border by activating a gene called pax7 . Then , at a later stage , Hmga1 allowed these cells to move to other parts of the body by triggering the Wnt communication system . Failure for the neural crest to develop properly causes birth defects and cancers such as melanoma and childhood neuroblastoma , highlighting the need to better understand how this structure is formed . In addition , a better grasp of the roles of Hmga1 in healthy development could help to appreciate how it participates in a range of adult cancers .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2020	Bimodal function of chromatin remodeler Hmga1 in neural crest induction and Wnt-dependent emigration
A new mutant mouse ( lamb1t ) exhibits intermittent dystonic hindlimb movements and postures when awake , and hyperextension when asleep . Experiments showed co-contraction of opposing muscle groups , and indicated that symptoms depended on the interaction of brain and spinal cord . SNP mapping and exome sequencing identified the dominant causative mutation in the Lamb1 gene . Laminins are extracellular matrix proteins , widely expressed but also known to be important in synapse structure and plasticity . In accordance , awake recording in the cerebellum detected abnormal output from a circuit of two Lamb1-expressing neurons , Purkinje cells and their deep cerebellar nucleus targets , during abnormal postures . We propose that dystonia-like symptoms result from lapses in descending inhibition , exposing excess activity in intrinsic spinal circuits that coordinate muscles . The mouse is a new model for testing how dysfunction in the CNS causes specific abnormal movements and postures . Dystonia , the third-most common human movement disorder , involves 'sustained or intermittent muscle contractions causing abnormal , often repetitive movements , postures or both' ( Albanese et al . , 2013 ) . There is strong evidence that dystonia is a circuit disorder involving various brain regions , including sensory input , premotor and motor cortex , striatum and globus pallidus , subthalamic nucleus and parts of the thalamus , cerebellum , and the tracts connecting them ( Berardelli et al . , 1998; Breakefield et al . , 2008; Lehéricy et al . , 2013; Neychev et al . , 2011; Quartarone and Hallett , 2013; Thompson et al . , 2011 ) . There is also decreased inhibition and a bias toward potentiation in synaptic plasticity ( Hallett , 2011; Quartarone and Pisani , 2011 ) . However , there is little certainty about exactly how circuit and synaptic abnormalities produce the persistent overflow of motor control , often involving only certain muscle groups and the co-contraction of opposing muscles . Until recently , there has been a lack of a phenotypically penetrant genetically defined mouse model , where circuit hypotheses for mechanisms of dystonia can be tested in the context of abnormal movement ( Liang et al . , 2014; Weisheit and Dauer , 2015 ) . The lamb1t mouse introduced here exhibits late postnatal/young adult onset of dystonia-like hindlimb movements and postures , and it has high viability , gene penetrance , and inter-individual consistency . Several aspects of its biology have parallels with dystonia , such as post-developmental onset , an ability to overcome the symptoms , and slow progression . However , the mutant mouse also has symptoms exposed by sleep and anesthesia , and these led to the demonstration that there are circuit abnormalities in the spinal cord . The strategy was to characterize the genetic inheritance and behavior of the mouse; do diagnostic experiments to narrow down the neural substrates; map the gene’s locus and identify the mutation; and check expression of mutant protein . A dominantly-inherited Lamb1 mutation was found . Laminins are present in the extracellular matrix ( ECM ) surrounding neurons where they bind to synaptic proteins , and have been implicated in synaptic and neuromuscular junction structure and plasticity ( Dityatev et al . , 2010; Wlodarczyk et al . , 2011 ) . The mechanistic hypothesis was tested that there is altered synaptic activity in identified laminin β1-positive neurons in the CNS of the mutant mouse . The lamb1t mouse arose spontaneously in a WT C57Bl/6N mouse . It showed dominant inheritance: 140 out of 272 ( 51 . 5% ) mice with one WT parent were symptomatic . Awakening or novel environment typically elicited dystonic movements . The most prominent was hyperextension of one or both hindlimbs that was clearly hyperkinetic . Movement and postural abnormalities also included wide-spread ( extended ) legs during sitting , transiently curvy tail , strong hyperextension response to swimming , and abnormal tail suspension reflexes ( Figure 1 ) . Motor behavior in novel or stressful environments ( unfamiliar tray; elevated rack ) is shown in Video 1 . When unstressed in the home cage , however , the mutant mice could walk normally , climb available structures , rear up while touching the side of the cage , and climb upside down on the food rack . 10 . 7554/eLife . 11102 . 003Figure 1 . The mutant mouse had intermittent dystonic behaviors affecting hindlimbs and tail . ( A ) During ambulation , hyperextension could affect either hindlimb or both , sometimes with inversion of the foot . When hyperextension was unilateral , some mice had a preferred side et al . switched sides . Hyperextension of hindlimbs was seen at the youngest age during locomotion , but by 4 months of age was sometimes seen at rest and sometimes was bilateral . A bilateral , maximally extended posture is within a WT mouse’s normal repertoire because it is shown by nursing dams straddling a large litter . ( B ) Hyperextension often continued when the animals sat . ( C ) Briefly curved tail was sometimes the first symptom in weanlings but was seen in older adults mainly when stressed . The curvature , in the plane of the floor , utilizes lateral muscle groups , and Straub tail was seldom if ever seen . ( D ) While WT mice sometimes have brief periods of rigidity and tilting when dropped in water , the mutant mice adopted an upright posture with extreme hyperextension and spread toes . They soon recovered and swam . ( E ) The normal WT reflex when suspended by the tail . ( F ) Mutants exhibited caudal hyperextensions involving one or both hindlimbs . This is also within the normal repertoire because WT exhibit a hindlimb posture like this when suspended just out of reach of an object and reaching with the forelimbs . ( G ) The mutants also exhibited transient hyperflexions of one or both hindlimbs . This was not a coordinated 'clasped' posture ( limbs held together at the midline ) . Vibration stimulation of the knee joint in awake , hand-held mutant mice sometimes elicited strong dystonic movements when mice were released ( not shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11102 . 00310 . 7554/eLife . 11102 . 004Video 1 . Symptoms . Lamb1t mice displaying dystonic symptoms . 1 ) In a tray ( novel environment ) a mouse walked with kicks , adopted a wide-based sitting stance , then abruptly overcame symptoms to sit and groom . 2 ) A mouse with extreme hindlimb hyperextension on an elevated rack , where it also displayed curvy tail . 3 ) A mouse on an elevated beam traversed it with rigid hindlimbs by pulling itself across with the forelimbs . On a second trial the mouse recovered the ability to perform almost normal walking . In the home cage , mice often resumed normal motor control . Representative of many observations . DOI: http://dx . doi . org/10 . 7554/eLife . 11102 . 004 Dystonia in humans often has intermittent symptoms , and intermittency was a key feature of the impairment in the mice . Video 1 also shows a mouse that straddled a beam with dystonic-looking legs while crossing it , then exhibited almost normal gait in a repeat trial . Unstressed mice ran voluntarily on wheels ( Video 2 ) , making neuropathy or spasticity unlikely , but during forced treadmill running dystonic symptoms emerged . We recorded gait with a DigiGait imaging system , and gait differences between WT and mutant mice are compared in the video . Gait was abnormal and limbs were often propelled to the side ( Figure 2 ) . 10 . 7554/eLife . 11102 . 005Figure 2 . Gait abnormality in lamb1t mice . They were filmed ventrally from below the DigiGait transparent belt . The plane of focus was shallow , so feet that are seen clearly are in contact with the belt , while feet and tail that are more than a few mm above the belt are distorted or not resolved . ( A ) Treadmill-running WT mice placed their hindlimbs immediately behind the forelimbs in the alternating step pattern , and the hindlimb made contact close to the ipsilateral forelimb , as marked with a line . ( B ) In contrast , when mutant mice ran , the pattern was still alternating , but accuracy varied . In this example , the hindlimbs swung wide and did not get close to the forelimb . ( C ) Stance in mid-stride . In the WT the swinging hindlimb stayed close to the body , and mice had only two feet in contact with the belt ( * ) . ( D ) The mutant's hindlimbs were both splayed out , and three feet were in contact with the belt . Utilizing both front feet may have compensated for deficient hindlimb mechanics . WT had three feet in contact only when breaking stride to rest briefly . The mice studied ranged in age from 63 to 132 days old , mean 109 , n = 9 per group . DOI: http://dx . doi . org/10 . 7554/eLife . 11102 . 00510 . 7554/eLife . 11102 . 006Video 2 . Running . Lamb1t mice performed well when relaxed and poorly when stressed . 1 ) Lamb1t mice given a running wheel in the home cage ran voluntarily as soon as the lights turned off , and ran for hours as detected by a meter . This was filmed in the dark with an infrared camera . In the example shown the mouse ran smoothly but a single hindlimb hyperextension terminated the run . A magnet attached to the disk activated the meter; the viewer can use it to count rotations . Representative of n = 4 . 2 ) Forced running on a treadmill moving at fixed speed , in contrast , was stressful and running success varied from trial to trial . 3 ) Slow motion ventral plane videography ( DigiGait ) of WT and lamb1t mice , representative of n = 9 each . WT , wild type . DOI: http://dx . doi . org/10 . 7554/eLife . 11102 . 006 Motor symptoms were first detected between P17 and P28 . Between 2 and 6 months , symptoms became more persistent , and mice often slept with hindlimbs extended . Between 6 and 12 months , there was no obvious qualitative progression separable from age , although secondary morphological changes to muscle and bone similar to those in patients may develop . Brain , spinal cord , and spinal root morphology were indistinguishable from wild type ( WT ) , and weight gain was normal until motor symptoms become persistent . During gentle manual assessment , hyperextended hindlimbs resisted motion at the joint . No cogwheel rigidity ( as in Parkinson’s Disease ) or rate-dependent resistance ( as in spasticity ) was felt . Hindlimb stretch reflexes , elicited by slowly pulling on a foot , were present when the hindlimb was relaxed but suppressed when it was hyperextended . We detected no abnormality in the back , forelimbs , head , or axial posture , and the mice groomed and built nests well . We did not observe the following: slow paroxysmal events that wax and wane , ataxia symptoms like reeling , staggering or imbalance , tremor , loss of righting , circling , camptocormia or kyphosis , hyperekplexia , myoclonus , hopping , or any form of seizure . Social interactions appeared unexceptional , and both sexes bred successfully . The subjective impression was that the mice were alert , curious , and active . Six motor skill tests were applied . The elevated beam was used to assess motor coordination with repeated tests of the same cohort with age . The time to cross the beam was similar when WT and mutant mice were compared , but mutants exhibited many more hindlimb slips ( Figure 3A ) , sometimes pulled with their forelimbs ( Video 1 ) , and when very affected sometimes could not stay on the beam . The rotarod tests ability to match the speed of rotation , requiring limb coordination and strength . Lamb1t mice stayed on the rod 40% as long as WT ( Figure 3B ) . With male WT mice , there was also an effect of weight , illustrated by linear regression of the data ( Figure 3C ) . The Olympic pool ( Figure 3D ) tested two skills combined: the time to orient when dropped in the water and the speed of swimming down a lane to a platform . Mutants were significantly slower ( mutant [6 . 9 ± 0 . 49 s] and WT [2 . 46 ± 0 . 07 s] , p = 4 . 5 x 10−12 by t-test; n = 27 WT and 28 mutant mice ) , but swim speed did not decline with age or weight ( Figure 3E , F ) . 10 . 7554/eLife . 11102 . 007Figure 3 . Quantitative measures of impairment . The mice ( all on the C57Bl/6N background ) were designated affected and unaffected before discovery of the gene . Bar graphs show mean ± SEM and two-way ANOVA was applied . ( A ) Crossing an elevated beam to return to the home cage was tested with one cohort repeatedly at the ages shown ( n = 14 WT , 10 males , 4 females , and n = 7 lamb1t , 4 males , 3 females ) . Symptoms normally appeared between 3 and 4 weeks of age . Young mutant mice ( 1–2 months old ) tended to exhibit only slips , while mature mutant mice ( 3–6 months old ) tended to have a mix of foot slips , hyperextension , and full control , and often could not stay on the beam . Traverse time and foot slip data from trials where a mouse fell were not included in the calculations , and the mouse was given another trial . ( Numbers completing the task: at 4 weeks , WT 13 , mutant 7; 6 and 8 weeks , WT 14 , mutant 7; 12 weeks , WT 13 , mutant 1; 16 weeks , WT 13 , mutant 3; 23 weeks , WT 10 , mutant 4 . ) There was no significant difference between WT and mutant in time to cross at any age ( p ranged from 0 . 28 to 0 . 76 ) . However , the data showed a significant difference in the number of foot slips during beam traverse . At 4 weeks , p = 0 . 0021; 6 weeks , p = 3 . 7 x 10−6; 8 weeks , p = 3 . 1 x 10−6; 12 weeks , n . a . ; 16 weeks , p = 3 . 9x10−11; 23 weeks , p = 0 . 034 for main genotype effect ( * ) , and Bonferroni’s test confirmed significance . At 12 weeks , only one mutant mouse was able to complete the task , and the bar ( † ) was a single data point . ( B ) In the accelerating rotarod task , both male and female mutant mice showed substantially shorter latencies to falling off . After 2 days of training , two trials on the third day were averaged ( WT , n = 17 male and 9 female mice; mutant , 13 males and 14 females; p = 3 . 4 x 10−5 for males and 6 . 5 x 10−7 for females , followed by Tukey’s test ) . In multiple comparisons , all differences were significant except male affected vs . female affected mice . Ages ranged from 60 to 180 days ( averages WT 125 days ± 47 . 5 , SD; mutant 114 ± 44 . 5 ) , and there was no trend with age . ( C ) Weight gain was initially normal in lamb1t mice , but they plateaued at 3–4 months , likely due to the metabolic demands of elevated muscle activity . Weight as a confounding variable is not often considered in rotarod testing . Plotting the rotarod data against weight showed it to be a continuous independent variable in WT males ( linear regression for WT males had a significant slope [R square = 0 . 6299 , F = 24 . 99 , p = 0 . 0002]; slopes in the other groups tested non-significant ) . However , the main effect of genotype dominated the results even though male lamb1t mice weighed less . ( D ) The Olympic pool ( empty ) . Mice swam down a lane to a submerged platform . ( E ) Swimming speed results as a function of age , and ( F ) as a function of weight . Each symbol is the average of two trials for one mouse . There was little overlap between genotypes , and no significant deterioration with age or weight was found by linear regression . SD , standard deviation; SEM , standard error of the mean; WT , wild type . DOI: http://dx . doi . org/10 . 7554/eLife . 11102 . 007 The activity chamber assessed spontaneous ambulation ( 10 min; novel environment ) . There was no difference in either horizontal ( same beam ) or ambulatory activity ( different beams broken successively ) between WT and mutants for either sex; that is , motor impairment did not diminish activity ( two-way ANOVA , p = 0 . 81 males , p = 0 . 40 females . n = 22 WT [9 males and 13 females]; n = 20 lamb1t [10 males and 10 females] ) . The age range was 95–169 days , average 129 days . No difference in forelimb function was detected in the adhesive removal test ( sticker applied to the forehead ) ( Student’s two-tailed t-test , p = 0 . 40 , n = 16 per group , average age 77 days ) . The ability to cling to a food rack from underneath utilizes skills learned in daily voluntary activity . WT mice usually confidently explored the rack . Mutants moved less and fell more ( significant by t-test at all ages , 4 , 6 , 8 , and 12 weeks , p = 0 . 0018 , 0 . 037 , 4 . 9 x 10−5 , and 2 . 4 x 10−12 , respectively , and with Bonferroni correction at 4 , 8 , and 12 weeks; n = 14 WT , 10 males , 4 females; 7 mutants , 4 males and 3 females ) ; however , it was not an ideal quantitative test because of variability due to the mutants adaptively using hindlimbs as hooks and wrapping their tails through the bars . We considered the possibility of peripheral neuropathy ( axonal structure , sensory degeneration , or demyelinating ) that would preferentially affect long axons , since symptoms predominated in the hindlimbs and tail , but no evidence for a peripheral nerve defect was found . Sciatic nerve conduction velocity ( mediated by fast myelinated axons ) in anesthetized mice was measured with EMG electrodes , and it was in the normal range ( 40 to 50 m/s , n = 3 , both legs , for WT and mutant ) . Latency and amplitude were also like WT . Cross sections of fixed sciatic nerve ( mid-thigh ) showed a normal distribution of myelinated axons by toluidine blue or hematoxylin-eosin stain . Both WT and mutant had normal-appearing large and small myelinated fibers and unmyelinated fibers . We assessed the tail withdrawal reflex ( mediated initially by slow unmyelinated nociceptor axons ) with a tail immersion test . Time to tail flick was measured in water at 51oC . Male mice had longer average latencies to tail withdrawal than females; however , WT and mutant were not significantly different ( male WT 1 . 99 s ± 0 . 21; male mutant 2 . 07 s ± 0 . 23; female WT 1 . 44 ± 0 . 19; female mutant 1 . 50 ± 0 . 14; mean ± SEM; male n =15 for WT and 13 for mutant; female n =9 for WT and 14 for mutant , two-way ANOVA , p =0 . 0092 for sex difference , and p = 0 . 75 for genotype effect ) . In the same data set , there was also no increase in latency with age ( assessed by linear regression , range 2 . 4–6 . 6 months , average age WT 141 ± 48 days , mutant 127 ± 45 days ) , as would be expected for progressive neuropathy . In sleep and in isoflurane anesthesia , twitching of lamb1t hindlimbs was often seen when mice were old enough for symptoms to be firmly established ( Video 3 ) . Twitching of hindlimbs during sleep was assessed at hourly intervals through the bottoms of cages with scant litter in 31 mutant mice , average 14 . 5 weeks old . Twitching was seen in 76 of 150 hourly observations where the mouse appeared to be asleep , indicating that it too is intermittent . 10 . 7554/eLife . 11102 . 008Video 3 . Sleep and anesthesia . Sleep and anesthesia disinhibit abnormal spinal activity . 1 ) A WT female ( left ) and lamb1t male ( right ) sleeping in an igloo were filmed from underneath the cage . The mutant slept with both hindlimbs extended to different degrees . 2 ) Spontaneous twitching activity of another lamb1t mouse sleeping in the home cage . 3 ) Lamb1t mice ( hybrids with different unlinked coat color gene combinations ) lying anesthetized in an O2/isoflurane vapor chamber . Representative of many observations . DOI: http://dx . doi . org/10 . 7554/eLife . 11102 . 008 The EMG activity associated with hindlimb twitching was recorded in lamb1t mice under isoflurane anesthesia . Electrode recordings ruled out myotonia ( persistent muscle firing causing prolonged contractions because of defective Cl- channels ) , because characteristic muscle fiber potentials ( positive sharp waves and/or fibrillation potentials ) with waxing and waning frequency and amplitude ( Heller et al . , 1982 ) were not detected , and there were periods of normal electrical silence ( Figure 4A ) . The recordings also ruled out neuromyotonia ( peripheral nerve hyperexcitability ) and myokymia ( quivering of muscles ) because of the absence of either of their EMG signatures , a constant buzz of neuromuscular junction activity ( Isaacs , 1961; Stum et al . , 2008 ) , or regular motor unit action potential activity in multiplet discharges ( Toyka et al . , 1997; Zielasek et al . , 2000 ) . We also did not detect signs of myopathy ( muscle weakness ) , such as low-amplitude , short duration motor unit action potentials ( MUAPs ) ( Hanisch et al . , 2014 ) . Plasma creatine kinase was also normal ( n = 3 WT and 4 lamb1t ) , that is , there was no indication of muscular dystrophy . 10 . 7554/eLife . 11102 . 009Figure 4 . Electromyography ( EMG ) in the lamb1t mouse showed co-contraction . ( A–F ) Electrodes were in opposing hindlimb muscle pairs ( anterior rectus femoris and posterior biceps femoris ) . Simultaneous activity in opposing muscles is a cardinal feature of dystonia . ( A ) Because young mice stopped showing symptoms when they warmed up after awakening , EMG was used to test for myotonia , a muscle channelopathy where a warm-up phenomenon is well-known , but results were negative . ( B , D , F ) Semi-rhythmic MUAPs ( motor unit action potentials ) typical of voluntary movement occurred simultaneously at 10–20 Hz under anesthesia in lamb1t mice . Sometimes recruitment of a second MUAP could be seen ( as occurs with increasing force ) , but there was less recruitment than normal . Vigorous coincident bursts of action potentials occurred either without MUAPs ( C ) , or with MUAPs ( D ) . ( E ) An uncommon complex repetitive discharge at 120 Hz with antagonist muscle group alternation , a spinal discharge pattern usually associated with locomotion . ( F ) Spontaneous large single spikes were random and not seen in the opposing muscle . ( G ) Example of electrodes in the same or opposing muscles in different legs: no co-contraction or synchronization . Images are representative of n = 6 lamb1t mice . Silent recordings of littermate controls ( n = 2 WT ) are not shown . The mice ranged in age from 43 to 113 days , mean 59 . DOI: http://dx . doi . org/10 . 7554/eLife . 11102 . 009 Notably , two-electrode EMG recorded co-contraction driven by motor unit activity . Synchronized MUAPs , typically at 10 Hz , were detected with recordings from opposing ( agonist and antagonist ) muscles ( Figure 4B , D , F ) . Synchronized polyphasic bursts very similar to those seen in sleeping blepharospasm ( eyelid ) and oromandibular ( mouth , jaw , and tongue ) dystonia patients ( Sforza et al . , 1991 ) correlated with twitching movements ( Figure 4C , D , E ) . Most bursts were complex , with wave summation interference that obscures any underlying regularity ( Figure 4C , D ) , but alternating contraction was occasionally seen ( like fictive locomotion in spinalized animals ) ( Figure 4E ) in addition to co-contraction . Some polyphasic bursts occurred in isolation ( Figure 4C ) , others in trains of MUAPs ( Figure 4D ) . Isolated large spikes that occurred randomly were not reflected in the opposing muscle ( Figure 4F ) . A continuous recording from opposing muscles is in Video 4 . To test whether co-contraction was due to a system-wide volley of excessive descending activation , electrodes were moved to the same muscles on contralateral legs or to opposing muscles on contralateral legs . No correlation of firing was observed in either case ( example in Figure 4G ) . 10 . 7554/eLife . 11102 . 010Video 4 . EMG . A continuous reading of motor unit activity recorded by EMG from opposing muscles in a lamb1t mouse . Coincident timing of both MUAPs and polyphasic bursts can be seen . Representative of n = 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 11102 . 010 The mouse’s characteristics made it possible to ask whether the co-contraction EMG activity originated in the brain , or was intrinsic to the spinal cord . The spinal cord has local neuronal circuits ( central pattern generators ) required for locomotor coordination , muscle synergy , and posture . During sleep , pons and medulla normally activate inhibitors in the spinal cord , and volatile anesthetics like isoflurane activate GABAA receptors . The EMG activity exposed by sleep and anesthesia suggests two alternative hypotheses: that in lamb1t , the mutation results in the disinhibition of stimulatory signals descending from the brain , or that there is insufficient inhibition ( descending from the brain or from spinal interneurons ) to control overactive local spinal circuits . Spinal transection under anesthesia was used to determine whether signals descending from the brain elicited or reduced the twitching and its underlying co-contraction . We performed this on six lamb1t mice and four WT . In all mutant mice , the activity observed under anesthesia was undiminished or slightly enhanced by spinal transection ( Video 5 ) . WT mice remained immobile under anesthesia before and after transection . 10 . 7554/eLife . 11102 . 011Video 5 . Spinal transection . Under continuous anesthesia delivered by nose cone , two examples of hindlimb activity before and after spinal transection are shown . In the first case , rigidity and twitching increased . In the second case , gentle stimulation appeared to elicit hyperreflexia . Representative of n = 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 11102 . 011 These observations are instructive for the mechanism of the disorder , because they rule out the possibility that the co-contraction originates in the brain . They also demonstrate aberrant intrinsic spinal activity and suggest that co-contraction is determined by the over-activation of propriospinal circuits designed to execute or oppose reflexes , motion or postural stability . The fact that the relaxed awake lamb1t mouse can suppress abnormalities and that arousal and stress produce dystonia-like symptoms shows that the brain’s ability to suppress or enhance the activity of spinal cord intrinsic circuits may be a fundamental feature of dystonia . Locus mapping was done in hybrids between C57Bl/6N and FVB strains using SNP markers to detect recombination events . In the F1 generation , symptom onset was delayed to 5–7 weeks of age , and symptoms were less obvious . Nonetheless 10 out of 22 offspring were symptomatic . The less robust symptoms in F1 hybrids indicated that there were strain-specific modifier genes , but symptoms did not diminish with further back-crossing to FVB . For 24 individual SNP-mapped symptomatic hybrid mice only one B6 region was shared by all . The defining recombination events restricted the locus to the first 35 gigabases of chromosome 12 ( Figure 5A ) . 10 . 7554/eLife . 11102 . 012Figure 5 . Identification of the Lamb1 mutation . ( A ) SNP locus mapping summary for chromosome 12 in B6/FVB hybrids . The mutation is necessarily in B6 DNA ( yellow ) ; FVB DNA is blue . Mouse centromeres are at the top ( gray ovals ) . There were no informative SNPs between base 0 and 11 , 922 , 132 . ( B ) Exome sequencing result . Nucleotide and protein sequence for Lamb1 ( laminin β1 ) amino acids 1721 to 1741 flanking the mutation . Mutation at a single nucleotide generated a stop codon , TAG , and the sequence in red and beyond ( amino acids 1730 to 1786 ) was truncated . Eight other variants identified by exome sequencing in the locus were in exon-flanking intron sequence or 3'UTR and not predicted to be damaging . ( C ) We validated the mutation by Sanger sequencing . The identified causative Lamb1 mutation was not a reported variant in dbSNP , the Mouse Phenome Database , or the Sanger1 database . To date , the mutation has been verified in 33 symptomatic mice . ( D ) Allele-specific PCR design . Pink blocks are exons 32 and 33 , the red symbol is the mutation , and the yellow square is the normal stop codon . The forward allele-specific primers ( green ) were longer than the reverse allele-specific primers ( violet ) because of high AT content . Each set of otherwise-identical internal primers ended with either T or A . If the mismatch is sufficiently destabilizing , priming will be absent or very low . If the mismatch is not sufficiently destabilizing , another base 5’ of the mutation can be changed to reduce stability and improve selectivity; the reverse WT allele-specific primer also had a substitution of G for C at –2 . Forward and reverse outside primers ( gold and blue ) were predicted in the flanking DNA at convenient distances from the mutation based on melting temperatures matching the allele-specific primers . Diagnostic PCR was done with the gold/violet pair with the mutation , while the gold/blue pair served as a positive control . DOI: http://dx . doi . org/10 . 7554/eLife . 11102 . 012 Exome sequencing was done on DNA from a WT and a mutant on the C57Bl/6NCrl background . There were nine mutant-specific variants in the chromosome 12 locus , but only one was a non-synonymous coding change . The candidate was near the closest recombination site ( Figure 5A ) and was a single base pair transversion in Lamb1 ( T5460A ) that changed a leucine to a stop codon , amino acid p . Leu1730stop ( Figure 5B ) . The mutation was confirmed by Sanger sequencing ( Figure 5C ) and validated by showing complete co-segregation between the mutation and the symptoms in eight littermates on the C57Bl/6 background ( four with symptoms , four without ) , and six littermates on the mixed C57Bl/6-FVB background ( three with symptoms , three without ) , p < 10−8 by two-tailed t-test . Heterozygote x heterozygote matings from the N3 hybrid generation ( N3F1 , N3F2 , and N3F1N1F1 ) were used to determine whether homozygous lamb1t mice were viable . 25% of offspring of het x het crosses of FVB-B6 hybrids should have had two copies of the mutation , but instead average litter size compared to WT x het litters was reduced 28% ( mean ± SEM; n = 255 mutant x WT offspring; n = 178 mutant x mutant offspring . p = 7 . 9 x 10−5 , Student’s two-tailed t-test ) . No homozygotes were found by genotyping 16 dead pups or 4 surviving runts . Intercrossing did produce some hybrids with symptoms as robust as B6 lamb1t , however , and we performed SNP mapping on 16 of those to look for homozygotes: all were heterozygous for the chromosome 12 locus . This is further evidence that homozygosity is lethal and supports the influence of unlinked modifier genes that increase or decrease symptom strength . Allele-specific PCR ( Wu et al . , 2010; You et al . , 2008 ) uses otherwise-identical primers that match or mismatch the mutation at the 3’ end of the primer . Allele-specific nested PCR primers were developed for routine mouse genotyping ( Figure 5D ) . Laminin is a trimer of three different subunits that form a cross when assembled ( Figure 6A ) ( Hohenester and Yurchenco , 2013 ) . The α , β , and γ subunits each have an ECM polymerization domain at the N-terminus , and α has a string of cell receptor-binding ( G ) domains at the C-terminus to bind signaling proteins like integrin and dystroglycan . All three subunits have C-terminal portions that let them associate with each other as a stable trimer of coiled-coil that forms an extended linear rod . The lamb1t premature stop codon deletes the last 57 amino acids from β1’s coiled-coil domain . Cells have a mechanism for degrading mRNA with premature stop codons , but when a stop codon is less than 50 bases from the final exon junction , the mRNA escapes nonsense-mediated decay ( Kervestin and Jacobson , 2012 ) . The lamb1t mutation is in exon 32 , just 36 bases upstream of the next ( and final ) exon junction . In immunoblots of tissue extracts of choroid plexus ( a rich source of Lamb1 mRNA; Figure 7B ) from WT and mutant mice , laminin β1 at ~225 kDa was readily detected . The truncated form should be ~6 kDa smaller , and it was resolved when electrophoresis time was extended ( Figure 6B ) . In agreement with the proposal that the truncated protein is fully expressed , we did not observe a reduction in laminin β1 levels ( Figure 6B ) ; we also saw no reduction in samples from cerebellum and sciatic nerve ( not shown ) . Whether the truncation should disrupt the coiled coil was predicted with MultiCoil . No change was calculated for the majority of coil upstream of the truncation ( Figure 6C ) , implying that mutant β1 should assemble with laminin alpha and gamma subunits . 10 . 7554/eLife . 11102 . 013Figure 6 . Laminin β1 protein structure . ( A ) Diagram of the laminin trimer of α ( gray ) , β ( red ) , and γ ( blue ) subunits . Circles are major globular domains and the largest ones are binding sites for each other and other extracellular matrix components . The G domain repeating globular domains are sites of attachment to cell surfaces through integrin and dystroglycan . The rod-like coiled-coil trimer domain , a quaternary structure , is altered by truncation of the last 57 amino acids in the β1 subunit in lamb1t . Coiled-coils are composed of alpha-helices tightly wound together , and the absence of a portion should destabilize the α and γ segments at that site . ( B ) SDS gel electrophoresis ( NuPage 4–12% polyacrylamide gradient MES gels run 50% longer than normal ) followed by immunoblot with laminin β1-specific antibody . The doublet resolved in the mutant ( * ) is assumed to be the proteins produced from WT and mutant alleles . ( C ) We used MultiCoil software to calculate the propensity of protein sequence to form two-stranded or three-stranded coiled coils ( http://groups . csail . mit . edu/cb/multicoil/cgi-bin/multicoil . cgi ) . Laminin β1 is strongly predicted to form three-stranded coiled coils . Just upstream of the truncation , there was a slight decrease in triple-stranded coil propensity ( red ) and slight increase in double-strand coil propensity ( * , blue ) for 27 amino acids , but no change for the rest . DOI: http://dx . doi . org/10 . 7554/eLife . 11102 . 01310 . 7554/eLife . 11102 . 014Figure 7 . Discrete nervous system expression of Lamb1 . Images are reproduced from the Allen Brain Atlas ( Allen Institute for Brain Science ) ( A , B , D , G ) , or GENSAT ( C , E , F ) . ( A ) In situ hybridization in striatum on lightly counter-stained mouse sections . Gad2 signal marks the abundant medium spiny neurons . Pvalb , Ache , and Sst are markers of different interneuron populations . Cholinergic neurons ( Ache ) are very large . Lamb1-positive cells are apparently in an interneuron population , but unlikely to be cholinergic because of their small size . EGFP expression confirms the in situ hybridization signal for Lamb1 . ( B ) Cerebellum . Lamb1 in situ hybridization is high in choroid plexus ( CP ) and Purkinje cells ( PC ) , and expressed at a lower level in the deep cerebellar nuclei ( DCN ) , including all three , dentate , interpositus and fastigial . Label in dorsal cochlear nucleus ( DCO ) is also present . ( ML ) molecular layer where Purkinje cells arborize . There are strongly stained interneurons scattered in the molecular layer . ( GL ) granular layer , where the granule cells , the abundant excitatory inputs to Purkinje cells , do not express Lamb1 . There are sparse labeled cells , however . ( C ) EGFP expression supports the findings . ( D ) Another major excitatory input to Purkinje cells , the climbing fibers , come from the inferior olivary nucleus ( IO ) , which was not labeled for Lamb1 . ( PT ) pyramidal tract . ( E ) Lumbar spinal cord . EGFP expression was in a diffuse band in lamina 1 of the spinal cord , and in strongly labeled scattered cells in lamina 2 or 3 . There was little or no label in other structures other than blood vessels . ( F ) Higher magnification with dorsal horn ( DH ) , dorsal sensory column ( DC ) , and corticospinal tracts ( CST ) indicated . ( G ) Available in situ hybridization in spinal cord was faint , but a similar subpopulation of cells in lamina 2 or 3 as well as some cells at the surface were labeled . Reproduced with permission . DOI: http://dx . doi . org/10 . 7554/eLife . 11102 . 014 Lamb1 expression data for the mouse are available in the Allen Brain Atlas ( in situ hybridization ) and GENSAT ( EGFP expression marker ) . Their data were largely in concordance where they overlapped , and confirmed published data from Lamb1 promoter-driven β-galactosidase constructs in the mouse brain ( Sharif et al . , 2004 ) . Lamb1 is expressed in several sites implicated in movement disorders , including sites that showed neuropathology in rapid-onset dystonia-parkinsonism ( RDP ) ( Oblak et al . , 2014 ) . These include substantia nigra compacta , cerebellar Purkinje neurons , and the deep cerebellar nuclei ( DCN ) . Lamb1 is also expressed in the striatum , where the low density of labeled cells indicates a discrete population of interneurons ( Figure 7A ) . Only a few neurons of the cortex were labeled , requiring identification . Other cells and nuclei were also stained well , including the hippocampus and thalamus , but overall staining was light and selective . Interestingly , distributions characteristic of either astrocytes or oligodendrocytes were not seen . The positive Purkinje neurons and DCN ( Figure 7B , C ) are the output pathway of the cerebellum . Lamb1 seems to not be expressed in some excitatory inputs to Purkinje cells ( granule cells , or the inferior olivary nucleus , Figure 7D ) but is expressed in inhibitory inputs from molecular layer cells that probably are basket and stellate cells ( Figure 7B , C ) . In the spinal cord , there is diffuse stain in lamina 1 , and strong expression in a sparse population of interneurons in lamina 2 or 3 of the dorsal horn ( Figure 7E–G ) . These are laminae where unmyelinated nociceptive sensory neurons terminate ( Gardiner , 2011 ) . Motor neurons appear to have little or no mRNA for Lamb1 , and the EGFP marker expressed in ventral horn was in blood vessels ( Figure 7E ) . Among the Lamb1-positive neurons , cerebellar Purkinje neurons and deep cerebellar nuclei ( DCN ) neurons have well-characterized synaptic interactions . We hypothesized that firing abnormalities might be detected in those cells if laminin β1 mutation is acting at synapses . DCN provides a major output from the cerebellum to ensure coordinated motor activity , and is one part of the circuits important for dystonia . Single-unit extracellular recordings of spontaneous activity of Purkinje cells and cells in the DCN were performed in awake head-restrained mice . Figure 8 presents the data as box and whisker plots to show differences in median , interquartile range , and skew , while Table 1 has the calculated means and statistical differences . Recordings revealed irregularly firing neurons during abnormal postures in the lamb1t mice compared to WT ( Figure 8A , E ) . To quantify the changes in firing and analyze lamb1t firing patterns during both normal and abnormal postures , several parameters were examined . The average firing rate ( defined as the number of spikes divided by the recorded time ) was decreased in the DCN in the mutant during abnormal postures , whereas there was no significant difference between the WT and the mutant during normal postures ( Figure 8B ) . The predominant firing rate ( defined as the reciprocal of the mode interspike interval ) of the mutant mice was significantly higher during both normal and abnormal postures compared to WT , but highest with abnormal postures ( Figure 8C ) . To measure the irregularity of firing in all conditions , the coefficient of variation of the interspike interval ( CV ISI ) , defined as the standard deviation of the interspike interval divided by its mean , was calculated . There was a significant increase in the CV ISI of DCN cells in the mutant during abnormal postures compared to WT or to mutants during normal postures , confirming that cells in the DCN of mutant animals fire irregularly during abnormal postures . 10 . 7554/eLife . 11102 . 015Figure 8 . Cerebellar neurons in lamb1t mice during abnormal postures exhibited high-frequency bursts . Extracellular recordings were performed in vivoin awake head-restrained wild-type and mutant mice . ( A ) Representative raw traces of spontaneous single-unit recordings in cells of the DCN show abnormal burst firing of cells in the mutant during abnormal postures ( magenta ) when compared to periods where the mouse did not exhibit abnormal postures ( gray ) and also compared to the WT ( black ) . ( B ) Even though there was no significant difference in the average firing rate of cells in the DCN of WT compared to the mutant without postures , during abnormal postures the average firing rate decreased significantly . ( p<0 . 05 , p<0 . 01 , mean ± SEM ) ( C ) The data were binned to construct a histogram of the interspike intervals ( ISI ) from which the peak value ( the mode of the distribution ) was determined for each cell . The predominant firing rate was calculated as the reciprocal of the mode of the ISI . During abnormal postures in the mutant animal , the predominant firing rate of cells in the DCN was more than 3-fold higher compared to the WT and twofold higher compared to conditions when the mutant was not displaying abnormal postures ( *p<0 . 0001 , p<0 . 001 ) ( D ) The coefficient of variation of the ISI ( CV ISI ) was significantly higher in the mutant during abnormal postures ( *p<0 . 0001 ) , whereas during conditions of no postures the CV ISI was similar to the WT ( E ) Raw traces showing irregular firing of Purkinje cells in the mutant when compared to the WT . ( F ) When compared to the WT , the average firing rate was statistically higher in the mutant with no postures ( p<0 . 01 ) but not significant in the mutant during abnormal postures . However , there was no significant difference between mutants with and without abnormal postures . ( G ) Predominant firing rate was significantly higher in the mutant with and without postures ( p<0 . 05 ) when compared to the WT . ( H ) Similar to cells in the DCN , the CV ISI was significantly increased in the mutant Purkinje cells during abnormal postures compared to normal postures and to WT ( ***p<0 . 0001 ) . Scale bars in raw traces: X-axis: 300 ms and Y-axis: 20 µV . WT , wild type . DOI: http://dx . doi . org/10 . 7554/eLife . 11102 . 01510 . 7554/eLife . 11102 . 016Table 1 . Cerebellar firing patterns . DOI: http://dx . doi . org/10 . 7554/eLife . 11102 . 016WTlamb1t normalp , norm . vs . WTlamb1tabnormalp , abnorm . vs . WTp , norm vs . abnorm . Deep cerebellar nuclei neurons # of cells264256average f . r . 37 . 4 ± 3 . 750 . 3 ± 5 . 90 . 349926 . 7 ± 2 . 90 . 01090 . 0029predominant f . r . 61 . 0 ± 6 . 9110 . 1 ± 13 . 80 . 0004221 . 9 ± 20 . 9<0 . 00010 . 0001CV ISI0 . 677 ± 0 . 050 . 751 ± 0 . 050 . 65421 . 272 ± 0 . 07<0 . 0001<0 . 0001Purkinje neurons # of cells617645average f . r . 62 . 9 ± 2 . 873 . 6 ± 2 . 80 . 005262 . 7 ± 5 . 00 . 34000 . 0619predominant f . r . 83 . 2 ± 3 . 498 . 7 ± 3 . 60 . 0141103 . 8 ± 8 . 10 . 01280 . 4462CV ISI0 . 578 ± 0 . 020 . 531 ± 0 . 010 . 30360 . 919 ± 0 . 07<0 . 0001<0 . 0001Means and significance of firing rates and coefficients of variation . Definitions: f . r . , firing rate . CV ISI , coefficient of variation of the interspike interval; that is , variability . The data are means ± SEM , and the p values were determined by the Kolmogorov-Smirnov test for non-parametric distributions , which can be seen in Figure 8 . N = 3 WT and N = 4 lamb1t mice; n for each class of cell recording is given in the table . Since a major source of input to the DCN comes from Purkinje cells , single-unit recordings from Purkinje cells were performed on the same mice to determine whether these cells are a likely contributor to the irregularity of firing in the DCN . Similar to cells in the DCN , Purkinje cells in mutant mice appeared to fire irregularly during abnormal postures when compared to WT mice ( Figure 8E ) . Like in the DCN , average rate was higher in lamb1t during normal postures , but was similar to WT , not lower , when abnormal postures were present ( Figure 8F ) . The predominant firing rate in mutants was higher during both normal and abnormal postures than in WT ( Figure 8G ) . Also like in the DCN , the CV ISI of mutants during normal postures was not significantly different from WT , but during abnormal postures CV ISI was much higher ( Figure 8H ) . In summary , the electrophysiological data demonstrate that Purkinje cells and cells in the DCN of the lamb1t mouse fire irregularly with high-frequency bursts during abnormal postures , and that even during normal postures the mutant exhibits an increased predominant firing rate , but without an increased coefficient of variation . This indicates that the Lamb1 mutation results in a change in the intrinsic properties of both Purkinje cells and DCN neurons or in their synaptic inputs . The data do not exclude the involvement , or the possible primacy , of other circuits in the manifestation of symptoms , and much is yet to be learned from this mutant . The 57 amino acid truncation mutation produces a dominantly inherited phenotype with ~100% penetrance . A phenotype was not exhibited by heterozygous Lamb1 knockout mice ( Yin et al . , 2003 ) . Since the heterozygotes of the knockouts and lamb1t were on the same strain background , the dominant phenotype of lamb1t must be due to assembly of altered protein rather than haploinsufficiency , suggesting a gain of function . Based on the analysis of coiled-coil propensity , the sequence predicts that the small truncation of β1 should assemble with α and γ subunits because the coiled-coil domain required for trimerization is almost intact ( including residues for interchain disulfide bonds ) . Agrin , another important synapse-organizing ECM protein , binds to a central portion of laminin’s coiled-coil domain , and interacts with γ1 ( Kammerer et al . , 1999 ) , and that interaction may be preserved . The β1 truncation is unlikely to impair laminin trimer integration into ECM , because integration utilizes the self-polymerization domains at the N-termini of the three subunits . However , like the Lamb1 knockout ( Miner et al . , 2004 ) , lamb1t mice were embryonic lethal when homozygous , so it is essential to have one good copy in early development . Lamb1t’s mutation is at the receptor-binding end of the laminin complex , and there are some precedents for the kinds of effects it may have . Deletion of just three amino acids from the C-terminus of either γ1 or γ2 abrogated binding of laminin to the receptor integrin , as did substitution of glutamine for glutamate at the -3 position ( Ido et al . , 2007 ) . Chimeras of the last 20 residues of laminin β1 and β2 determined selectivity among different integrins ( Taniguchi et al . , 2009 ) . Synaptic localization of muscle-secreted laminin β2 was determined to be controlled by a 16 amino acid motif 100 residues upstream of the C-terminus , which may still be exposed ( Martin et al . , 1995 ) . Thus , the 57-residue truncation in lamb1t may alter association with cell-surface receptors and affect signaling , without necessarily affecting the structural properties of laminin in the ECM . Alternatively , the truncation may destabilize the coiled-coil domain of the complex enough to make it more sensitive to regulated extracellular proteases , changing the kinetics of protease-mediated synaptic processes ( Dityatev et al . , 2010; Wlodarczyk et al . , 2011 ) . Two LAMB1 mutations in humans show a genotype-structure-phenotype relationship that contrasts with lamb1t . The mutations in two consanguineous families were recessive and produced a severe developmental brain disorder when homozygous , cobblestone lissencephaly ( COB ) , caused by defective basement lamina at the pial surface ( Radmanesh et al . , 2013 ) . This outcome was less severe than the in utero lethality of homozygous lamb1t mutation , but similar brain dysplasia was seen in homozygous mouse knockouts of Lamb2 and Lamc3 ( Radner et al . , 2013 ) . The COB mutations had much larger truncations ( via frameshifts ) than lamb1t’s , which eliminated the coiled-coil domain and would prevent trimerization if expressed . The frameshifts should also trigger nonsense-mediated mRNA decay ( Kervestin and Jacobson , 2012 ) , which in homozygous patients would be equivalent to knockout of LAMB1 . Either problem would explain the fact that COB gene carriers in the families were asymptomatic , like the asymptomatic heterozygous Lamb1 knockout mice . Given the high level of mRNA for Lamb1 in choroid plexus of mice , it is of interest that the LAMB1 homozygous COB patients also had hydrocephalus , likely due to impaired ECM structural integrity in the choroid plexus . We predict that if laminin mutations functionally equivalent to lamb1t are eventually found in patients , they should be limited to residues that interfere with receptor binding , but not with laminin assembly and ECM integrity . ECM is a degradable structure that stabilizes neuronal circuitry , is present in the synapse , and is actively remodeled to facilitate plasticity ( Dityatev et al . , 2010; Wlodarczyk et al . , 2011 ) . Activity-dependent synaptic plasticity , such as long-term potentiation mediated by the expansion of spines , entails remodeling of the ECM via regulated proteases . Tissue plasminogen activator ( tPA ) is secreted; then plasmin cleaves laminin and activates matrix metalloprotease 9 ( MMP-9 ) ; then MMP-9 digests a laminin receptor , dystroglycan ( Wlodarczyk et al . , 2011 ) . If this is facilitated by the lamb1t mutation , it may support evidence that synaptic plasticity is increased in dystonia ( Quartarone and Pisani , 2011 ) Laminin is well-studied for its role in controlling neuromuscular junction formation and ultrastructure at least partly through the receptor-binding domain ( Martin et al . , 1995; Nishimune et al . , 2004; Singhal and Martin , 2011 ) . For example , laminins with β2 ( Lamb2 ) complex with voltage-gated calcium channels to organize active zones , and signal through dystroglycan to modify neuromuscular junction structure and stability . Much is still to be learned about laminin β1 in the CNS , but synaptic vesicle protein 2 ( SV2 ) is a presynaptic ECM receptor , and binds laminins with β1 ( Son et al . , 2000 ) . One lab has produced mechanistic evidence for a role for laminin β1 in synaptic plasticity . Molecular mechanisms of learning and memory were investigated by gene manipulation in vivo in rat hippocampus in combination with water maze testing ( Yang et al . , 2011 ) . Maze learning decreased laminin β1 , and conversely laminin β1 overexpression impaired maze performance . Pertinent to the signaling function of laminins , laminin β1 impaired learning through activation of ERK/MAPK and SGK1 ( Yang et al . , 2011 ) . A plasticity cascade entails JAK/STAT activation ( Nicolas et al . , 2012 ) . Maze learning downregulated STAT1 , and STAT1 overexpression impaired maze learning while strongly upregulating Lamb1 . In a crucial experiment , the effect of STAT1 on maze performance was blocked by siRNA for Lamb1 ( Hsu et al . , 2014 ) . Changes in NMDA receptor subunits were upstream of STAT1 . This predicts that laminin β1 is in the middle or end of a synaptic plasticity cascade . The evidence in this report does not prove that the C-terminal truncation in laminin β1 affects synaptic plasticity; however , the altered firing patterns detected in deep nuclei in the cerebellum provide strong cellular evidence for an effect on synapse function or organization . Very similar irregular firing was seen in both DCN and Purkinje cells in a pharmacological model of ATP1A3 dystonia ( Dyt12 ) ( Fremont et al . , 2014 ) . DCN neurons also had neuropathology in rapid-onset dystonia-parkinsonism specimens ( ATP1A3 ) ( Oblak et al . , 2014 ) , and they are the source of the cerebello-thalamo-cortical tracts that show alterations in TOR1A , THAP1 , and ATP1A3 dystonia patients by diffusion tensor imaging ( Argyelan et al . , 2009; Lehéricy et al . , 2013; Whitlow et al . , 2012 ) . In lamb1t , the irregular firing of DCN neurons could be caused by an increase in the efficacy of inhibitory Purkinje cell-to-DCN synapses ( D'Angelo , 2014 ) . The detected abnormalities of a synapse in lamb1t support a role for cerebellar abnormalities in dystonic symptoms ( Fremont et al . , 2014; Wilson and Hess , 2013 ) . This of course does not rule out effects in other circuits with Lamb1-positive neurons , like striatum and of course spinal cord , that may be required for lamb1t’s aberrant motor control . The lamb1t mouse is a phenotypic model with overt dystonia-like symptoms when awake . Lamb1 gene expression is seen in selected neurons in the basal ganglia , cerebellum , spinal cord , and other locations , but it is not expressed universally in the CNS . Identification of Lamb1-positive neurons and their interactions will be a first step to investigating the underlying mechanisms , as begun here with cerebellar recordings . Do the dystonia-like symptoms of lamb1t represent dystonia as understood clinically ? Four theoretical frameworks coexist in human and animal dystonia research: roles for the basal ganglia , cerebellum , motor areas and the pathways that connect them; alterations in sensorimotor integration; reductions in neuronal inhibition; and increases in the potentiation side of synaptic plasticity ( Berardelli et al . , 1998; Breakefield et al . , 2008; Kreitzer and Malenka , 2008; Lehéricy et al . , 2013; Neychev et al . , 2011; Quartarone and Hallett , 2013; Quartarone and Pisani , 2011; Tanabe et al . , 2009; Thompson et al . , 2011 ) . None of these mechanistic aspects can be ruled out in the lamb1t mouse , and the possibility that the laminin β1 truncation affects synapse organization and plasticity is compatible with all . The post-development onset , slow progression , and plateau of symptoms resembles progression and stabilization in human dystonias . The ability to overcome symptoms resembles the human ability to use sensory tricks , such as touching the face , to temporarily overcome dystonia ( Ramos et al . , 2014 ) . However , in dystonic humans , a rapid alternation of affected muscle groups is unlikely , as seen here when mice switch affected hindlimbs . Furthermore , while electromyographic ( EMG ) activity persists in sleep in a few forms of human dystonia ( Sforza et al . , 1991; Stamelou et al . , 2012 ) , it is not typical . Research on focal dystonia patients suggests altered spinal reflexes and failure of reciprocal inhibition in spinal circuits as the immediate cause of co-contraction ( Berardelli et al . , 1998; Panizza et al . , 1990; Sabbahi et al . , 2003; Tanabe et al . , 2009; Thompson et al . , 2011 ) . Patient research also indicates a role for the brain to supply inhibitory instructions to spinal interneurons ( Berardelli et al . , 1998; Hallett , 2011; Quartarone and Hallett , 2013 ) . These aspects are consistent with traits of the lamb1t mouse , and share with it the convergence of abnormalities in the brain and spinal cord . However , CNS lesions causing human secondary dystonia tend to be in basal ganglia and their connections ( Thompson et al . , 2011 ) . This would be conceptually consistent with the lamb1t mouse’s characteristics if typical dystonia-causing human lesions damage the ability of the brain to control ( inhibit ) circuits in the spinal cord . The hindlimb locus of lamb1t symptoms may be characteristic of the species: hindlimbs have a dominant place in sensorimotor cerebellar integration in mice compared to primates ( Logan and Robertson , 1986; Raike et al . , 2013 ) , and hindlimb symptoms are also prominent in other rodent non-paroxysmal dystonia models ( LeDoux , 2011; Tanabe et al . , 2012; Weisheit and Dauer , 2015 ) . The lack of overt dystonic symptoms in mouse models with engineered dominantly inherited human dystonia mutations may be also influenced by a higher level of automaticity of the control of spinal cord in rodents relative to primates , requiring a stronger spinal physiological abnormality to allow manifestation of symptoms more obvious than hindlimb slips . The presence of Lamb1-expressing neurons in the spinal cord presumably underlies the observed abnormal spinal circuit activity . Neurons that express Lamb1 in dorsal horn may mediate sensory input , or may regulate propriospinal circuits that produce reflexes , coordinate muscles , or stabilize posture . If DRG neurons express Lamb1 in adults , aberrant sensory input could also have a role . Caution should be used before assuming that the causative functional defect is in the spinal cord , however , because one fundamental way that motor output is controlled is gradual enhancement of spinal reflex strength by the brain via descending signals , causing long-term scaling changes in the spinal cord ( Wang et al . , 2012 ) . Spinal circuit abnormalities might accrue slowly secondary to a primary defect in the brain . This would be congruent with slow symptom progression in the mouse , and brain–spinal cord interaction might underlie the slow therapeutic response to deep brain stimulation in patients . Available data ruled out other diagnoses such as myotonia , neuromyotonia , neuropathy , or muscular dystrophy . The features of lamb1t mice are ab initio different from spasticity ( a form of hypertonicity , clinically defined as velocity-dependent resistance to muscle stretch ) . In spinal injury or upper motor neuron disease , spasticity results from spinal reflex changes secondary to missing input from upper motor neurons ( Sheean , 2002 ) . Here , there was never paralysis preceding onset of dystonic symptoms; the mice could run voluntarily; and the symptoms were intermittent . It remains to be seen , however , whether abnormalities in descending inhibition , for example in the output of the pontine inhibitory region , can elicit spasticity-like changes in the spinal cord . Changes in spinal circuits ( such as the serotonin supersensitivity of coordinated muscle activation by central pattern generators in spasticity [Husch et al . , 2012] ) could be endophenotypes that are actually shared by dystonia and spasticity , globally or focally . Clinical spasticity can also be accompanied by spasms of muscle contraction , which are potentially related to the motor unit-driven twitches seen in lamb1t under sleep and anesthesia . At this stage , the relationship of the lamb1t mouse to human dystonia is tentative , but supported by a compatible distribution of gene expression in subcortical motor circuits ( and not in motor neurons ) , and by empirical similarities that are likely to reflect shared cellular mechanisms . There is a common premise that basal ganglia and brainstem select the muscles that co-contract in dystonia , for example by reducing the inhibition of competing pattern generators in pallidum ( Mink , 2003 ) , but this mouse presents evidence for a modified perspective . Although the basal ganglia may control the inhibition of unwanted activity , overactive pattern generators in the spinal cord appear to be the proximal cause of co-contraction in lamb1t . It is likely that some spinal central pattern generator circuits normally generate co-contraction , such as for postural control ( Blood , 2008 ) , just as others generate alternating contraction for locomotion ( Zhang et al . , 2014 ) . Lamb1t’s aberrant spinal activity may be essential for its manifestation of dystonic movements and postures . In awake lamb1t mice , we predict that the lapses in motor control that produce dystonic movements are failures ( due to mutation ) of supraspinal inhibitory control originating in familiar dystonia circuits , and modulated by arousal or stress . All animal research followed the NRC Guide for the Care and Use of Laboratory Animals and the policies of the Massachusetts General Hospital or Albert Einstein College of Medicine: MGH IACUC approved protocol 2011N000108 , and Albert Einstein approved protocol 20130801 . The lamb1t mouse arose in a colony carrying a knockout allele for an unrelated gene , Fxyd2 . The strain ( B6N . 129 ( Cg ) -Fxyd2tm1Kdr ) had been obtained from G . M . Kidder ( Univ . of Western Ontario ) ( Jones et al . , 2005 ) . We had back-crossed the colony to C57BL/6NCrl ( Charles River Laboratories , Wilmington , MA ) for six generations ( currently 12 generations ) and het x het matings produced WT , het , and Fxyd2 KO littermates . The lamb1t proband was WT for Fxyd2 , and was the only symptomatic mouse among 29 siblings . We segregated the dystonic proband and descendants as a separate colony without the Fxyd2 gene modification . Dominant inheritance was readily established by breeding with WT C57Bl/6NCrl mice . Every new generation was produced from colony members mated with WTs obtained from the supplier , and because penetrance remained 100% through six generations , it is evident that the phenotype is monogenic on the C57Bl/6NCrl background . The strain name for lamb1t is proposed to be B6N-Lamb1tr57/Swea . Mice were housed in ventilated racks on a 12-hr light-dark cycle . Ear punches were collected under isoflurane anesthesia at weaning both to mark individuals and as a source of DNA for genotyping . Both sexes bred with a success rate equivalent to other C57Bl/6 colonies housed in the same room , in contrast with the reduced breeding success of the Fxyd2 strain , which has pancreatic islet β cell hyperplasia and elevated insulin ( Arystarkhova et al . , 2013 ) . Longevity was not affected by the mutation: we observed the proband and two male sibs for 2 ½ years , and others for >12 months . All behavioral tests were approved by the IACUC of the Massachusetts General Hospital . We performed all experiments between 2 pm and 7 pm except sleep observations , which were from 8 am to 3 pm . For gait analysis , a DigiGait apparatus ( Mouse Specifics , Framingham , MA ) was used for ventral plane videography of mouse gait kinematics on a moving transparent treadmill belt . Mice were tested in late afternoon close to the beginning of the active period . Each mouse was given 4–6 trials at 15 cm/s , a relatively slow treadmill speed even compared to mice with ethanol or basal ganglia toxin-induced gait impairment ( Amende et al . , 2005; Kale et al . , 2004 ) . The balance beam was a 120 cm rectangular rod 1 . 6 cm in width held 30 cm above the padded surface , resting at one end on the edge of the home cage , which was shaded with a dark box . The accelerating Rotarod test ( Rotamex , Columbus Instruments , Columbus , OH ) entailed accelerating the rod from 4 rpm to 40 rpm over 180 s . Mice were trained for 2 days , 2 trials per day , with a 5-min break between trials . The results of the two trials were recorded on the third day . To test swimming speed , we developed a test that was named Olympic pool . A 86 x 46 cm plastic storage box ( Sterilite 1764; Figure 3D ) was divided into four equal lanes with opaque plastic dividers , and a 5-cm-wide submerged Styrofoam platform was fastened at one end of each lane . The pool was filled with room temperature water 20 cm deep , just over the platform . The starting point was marked so that the total length for mice to swim was 71 cm ( 28” ) . Training consisted of putting the mouse in the water at the opposite end once and allowing it to find the platform . On subsequent trials they would swim to it in seconds , and the measured parameter was swimming speed . For activity analysis , we used an OPTO-Varimex Minor activity meter ( Columbus Instruments ) to monitor activity for 10 min by optical beam-breaks , using an empty rat cage ( 37 x 25 x 19 cm ) as the chamber . Mice were introduced to the activity chamber as a novel environment , and beam-breaks were recorded for the first 10 min . The groups were matched for average age ( WT average 130 ± 27 days; mutant 127 ± 28 ) . WT mice were 11% heavier than lamb1t in males and 5% heavier in females . To detect impairment of forelimb function , the adhesive removal test was performed , in which a sticky square ( 5 mm squares of Post-it ) was applied to the forehead . The time to brush it off was recorded ( with a 60 s cut-off ) , and two trials were averaged . Replicate trials of individuals were combined and averaged . To assess grip strength and skill , a food rack from a rat cage was prepared by covering the edges with a 3-cm-wide masking tape to prevent mice from climbing to the top . Each mouse was placed in the middle and the rack was shaken laterally three times to make them grip it well . It was then inverted 30 cm above a well-padded surface . The time spent clinging to ( or climbing around on ) the rack was recorded , and the test was terminated after 60 s . Statistical analyses were performed with GraphPad Prism software ( La Jolla , CA ) . Genetics data were parametric and analyzed by Student’s two-tailed t-test . Behavioral experiments were not designed a priori for parametric statistical analysis because , besides normal sources of variance , there was additional variance stemming from lamb1t’s responses to stress and their ability to also bring motor behavior under control . The behavioral performance of the mutant was sometimes neither normally distributed nor of equal variance compared to WT , often in the tests measuring obvious abnormalities . All the available mice of appropriate age were tested since there was no ethical reason to limit group size . Statistical significance was estimated using two-way ANOVA followed by software-provided post-hoc multiple comparison tests , or two-tailed independent Student’s t-test , and considered significant at p<0 . 05 . Where error bars are shown , compiled data are reported as means ± SEM . Where variance was large , the data are shown as scatter plots . No outlying data points were excluded except in two cases: when a mouse was unable to stay on the elevated beam at all , and on the rotarod , when a mouse turned around to face the wrong way . All replicates were biological replicates; when duplicate trials of individuals were averaged and used as single data points , it is stated in the legend . Because the lamb1t symptoms in the C57Bl/6N strain background are obvious , it was not possible to perform behavioral tests blinded . Although we used many of the behavioral tests on B6/FVB hybrids to assess phenotype , only C57Bl/6N strain results are presented here , except that hybrids were also included for observations of hindlimb activity during sleep . Cerebellar neuron firing rates were analyzed by the Mann-Whitney test , and data are reported as mean ± SEM . For mapping the locus of the gene , we generated hybrids between mouse strains . FVB/NCrl ( Charles River Laboratories ) was selected because of an adequate number of different SNPs , and FVB's robust breeding ability . N1 hybrids were generated using either female or male B6 mutants paired with FVB mates . Following that , symptomatic hybrids were bred successively to FVB mice , producing N2 to N4 generations . In addition , N3 mice were intercrossed ( N3F1 ) and backcrossed again ( N3F1N1 ) . In the hybrids , symptoms were not as easy to detect , and so they were subjected to a battery of tests from 3 to 8 weeks of age , and scored for the presence and repeatability of symptoms . The criteria were display of symptoms upon awakening; during tail suspension; on the balance beam; during sleep; during isoflurane anesthesia; during a 30 s swim; and after vibration of the knee joint . Vibration was administered with a battery-operated fingernail polisher ( Nail Wizard ) with plastic tubing to cushion the tip . Cumulative scores were used to rank hybrid mice as moderately affected , weakly affected , or no detectable symptoms . We submitted DNA from symptomatic hybrids ( 17 N2 and 7 N3F1N1 individuals ) for SNP mapping to identify the locus . SNP mapping of recombinations was performed on the Illumina mouse medium density linkage panel of 1449 SNPs at the Centre for Applied Genomics at SickKids , University of Toronto . 833 of the SNPs on the panel differed between C57Bl/6 and FVB strains . Through six successive generations of backcross to FVB , 157 out of 302 mice on the hybrid background were Lamb1 mutant as determined by SNP mapping or AS-PCR , indicating no in utero lethality for heterozygotes . Exome sequencing was performed to detect variants ( point mutations and small indels ) that might be causative . Genomic DNA was purified with the Qiagen DNeasy Blood and Tissue Kit . Exome DNA was captured with the Agilent SureSelect Mouse All Exon system , and exome sequencing was performed on an Illumina HiSeq 2500 instrument at the Broad Institute of Harvard and MIT . Variants were called using GATK software ( DePristo et al . , 2011 ) . Post-sequence analysis and variant calling was conducted by ContigExpress ( New York , NY ) . Sequence surrounding the only non-synonymous coding variant in the locus was captured by PCR from each candidate mouse separately with the following primers: F- GCAGACTCTAGATGGCGAACTT , R- TGTAGATGACTGCCTCGGTTT , and purified with the Qiagen QIAquick PCR Purification kit . Sanger sequencing was performed at the DNA Core of the Center for Computational and Integrative Biology at Massachusetts General Hospital . Note that an upstream methionine in the reference sequence NM_008482 may be incorrectly identified as the initiation methionine; we numbered the residues from the second methionine corresponding with other species . Allele-specific PCR primers for Lamb1 mutation at T5460A were designed with the help of BatchPrimer3 V . 1 . 0 ( http://probes . pw . usda . gov/batchprimer3/ ) . Allele-specific primers utilized 26 bases upstream and 20 bases downstream of the mutation ( Figure 5D ) . The genotyping PCR reaction was done with ear punch or tail tip tissue , using the REDExtract-N-Amp tissue PCR kit ( Sigma-Aldrich , St . Louis , MO ) with half the recommended volumes . Thermal cycling was in thin-walled tubes ( Molecular BioProducts , Fisher Scientific ) capped with mineral oil , in an MJ Research PTC-100 thermal cycler . The conditions were 3 min 95oC , 32 cycles of 30 s 95oC denaturation , 30 s 63oC annealing , 30 s 72oC extension , and 2 min 72oC , and the products were resolved on 1 . 2% agarose gels . Relatively high annealing temperature was essential for allele specificity , although it reduced yield . The primer set used for routine genotyping of litters with WT and obligatory heterozygote pups were F-outside , GCCCAAGTACTTTGATATTCCTC; R-outside , TTTCACAAGTTCATCTCCACAGA , and R-mutant , GCTTGCTGTTAGCTTGAGCCT . Reactions of ( F-outside + R-outside ) and ( F-outside + R-mutant ) were run in separate tubes . Choroid plexus was dissected under a microscope from brains submerged in ice-cold Dulbecco’s PBS . It was taken from lateral ventricle using #5 forceps after dorsal incision through the corpus callosum to access the ventricles , and from 4th ventricle by rostrally folding the cerebellum back from the brainstem . Because laminin is an ECM protein , crude homogenates of the choroid plexus , cerebellum , and sciatic nerve were used for analysis of the protein on immunoblots . The buffer was 250 mM sucrose , 20 mM Tris , 1 mM EDTA , pH 7 . 2 containing 1 Roche protease inhibitor tablet per 50 ml , and homogenization was with a small Tenbroeck homogenizer on ice . Protein concentration was determined by BCA assay ( Pierce ) . Gel electrophoresis was with NuPage 4–12% polyacrylamide gradient MES gels ( Life Technologies ) . Twenty-five microgam of protein was loaded per lane from WT or lamb1t . The proteins were transferred to nitrocellulose and stained with laminin β1-specific antibody ( LTE ) from NeoMarkers at a dilution of 1:500 , followed by HRP-conjugated secondary antibody . Development was with Pierce West Dura luminal reagent or WesternBright , Advansta ( Menlo Park , CA ) , and images were collected with a GE Healthcare LAS4000 imaging system . Theoretical analysis of the impact of the mutation on the coiled-coil was done with MultiCoil , which performs computations based on a database of crystal structure data of three- and two-stranded coiled coils ( Wolf et al . , 1997 ) . Lamb1 expression data in the nervous system were found in the Allen Brain Atlas , the Allen Institute for Brain Science ( Lein et al . , 2007 ) ( http://mousespinal . brain-map . org/ ) , and GENSAT ( The Gene Expression Nervous System Atlas [GENSAT] Project , The Rockefeller University [New York , NY] ) ( Schmidt et al . , 2013 ) . ( http://www . gensat . org/daily_showcase . jsp ) . All animal procedures were approved by the IACUCs of the Massachusetts General Hospital or Albert Einstein College of Medicine . Nerve conduction velocity was measured in mice anesthetized using a constant inhaled mixture of oxygen and isoflurane administered by a VetEquip instrument through a nose cone . Animals were placed on a heating pad to maintain their core temperature at 37°C . Hind legs were shaved with a razor and then cleaned using alcohol pads . A pair of monopolar disposable 28G needle electrodes and ring electrodes ( CareFusion ) were lightly coated with electrode gel ( SignaGel ) and used for stimulation and recording , respectively . The active recording ring electrode was placed over the gastrocnemius muscle , with the reference electrode over the tendon . The stimulating cathode was placed 5 mm proximal to the recording electrode in the midline of the posterior thigh . The anode was placed subcutaneously in the midline over the sacrum . A surface electrode ( CareFusion ) was grounded on the mouse’s tail . We performed the studies using a portable electrodiagnostic system ( Cardinal Synergy ) . For the motor nerve conduction studies , the low-pass filter was set at 30 Hz , and the high-pass filter was set at 10 kHz . The nerve was stimulated with single square-wave pulses of 0 . 1 ms duration . Supramaximal responses were gradually generated , and maximal responses were obtained with stimulus currents <20 mA ( most often <10 mA ) . The distance between distal and proximal stimulation sites was measured with a millimeter-graduated tape measure . Data were acquired with a sensitivity of 20 mV/division and sweep speed of 3 ms/division . The distal latency , distal and proximal compound motor action potential ( CMAP ) amplitudes , distal and proximal CMAP durations ( measured from onset of initial negative deflection to initial return to baseline ) , and conduction velocity were determined for each nerve studied . A tail immersion test was used to assess the nociceptive reflex . Naive mice were held above a water bath at 51oC , sitting unrestrained on the top of the hand with the tail pointing down and held in position between two fingers . Then , 1 . 5–2 cm of the tail was immersed , and the latency to flick it out of the water was recorded . Sections of sciatic nerve were fixed in fresh periodate-lysine-paraformaldehye ( McLean and Nakane , 1974 ) , washed with Dulbecco’s PBS and infiltrated with 30% sucrose ( 30 g up to 100 ml ) , cut with a cryostat , then stained with 0 . 2% toluidine blue and examined with a Nikon Diaphot inverted phase-contrast microscope without phase ring . Additional samples were fixed with formalin , sectioned , and stained with hematoxylin and eosin with comparable results . Two-needle EMG was performed in anesthetized mice as described above for nerve conduction velocity . A ground self-adhesive gelled surface electrode was placed over the tail . Potentials were recorded from several sites of the hindlimb muscles with concentric needle electrodes ( 30G ) using a gain of 50 µV/division and a band-pass filter with low and high cut-off frequency settings of 10 or 20 and 10 , 000 Hz , respectively . The entire recording process took 20–30 min per mouse . EMG recordings were done as previously described ( Xia et al . , 2012 ) . Spinal transection was performed under continuous controlled isoflurane/oxygen anesthesia as above . The top of the spine was exposed by dissection , a laminectomy was performed at L1 to L3 with sharp scissors , and the spinal cord was transected with a narrow scalpel . Motor responses under constant anesthesia were observed and filmed for up to 2 min , and euthanasia was then performed by increasing the isoflurane followed by cervical dislocation . We verified completeness of the transection by inspection after death . For in vivo electrophysiology , mice were anesthetized with isoflurane and implanted with a custom-made L-shaped metal bracket fixed onto the skull with three bone screws ( Plastics one Inc . ) and dental cement ( M&S Dental Supply ) . A recording area 2 mm wide was drilled in the skull on top of the cerebellum at AP: −6 . 25 mm; ML: ± 1 . 7 mm . The recording area was surrounded with dental cement and covered with surgifoam and bone wax ( Ethicon ) . Mice were allowed to recover 24 hr before recording sessions . For recordings , the mouse was immobilized by fixing the head bracket with a screw attached to the stereotaxic frame . The bone wax and surgifoam were removed from the recording area . Single-unit neural activity was recorded extracellularly in awake head-restrained mice using a carbon fiber electrode ( Kation Scientific , 0 . 4–1 . 2 MΩ ) . The electrode was advanced into the cerebellum to target the Purkinje cells and neurons of the deep cerebellar nuclei . Cell types were identified based on location and the presence of complex spikes in Purkinje cells . Signals were band-pass filtered ( 200 Hz-20 kHz ) , amplified ( 2000 x ) , and digitized ( 20 kHz ) . Waveforms were sorted offline ( Plexon ) using principal component analysis . During recordings the mouse was closely monitored . Abnormal postures could clearly be seen and these periods were noted . Cells recorded during these episodes were noted as 'abnormal postures' , while all other cells recorded when the mouse did not show abnormal postures were categorized as 'no postures' .	Dystonia is the third most common disorder affecting movement in humans . People with dystonia periodically experience movements that they can’t control . Sometimes the movements are repetitive , for example , abnormal or spasmodic blinking . At other times , two sets of muscles that work against each other become active at the same time , which causes the body to assume a strange position . The symptoms are sometimes painful and they tend to be worse in times of stress . But many individuals with dystonia are able to develop tricks to control their symptoms . For example , some find they can stop the unwanted movements by touching their face or by walking backwards . Researchers believe that various parts of the brain fail to communicate properly in patients with dystonia . Additionally , it is thought that the connections between nerve cells called synapses become hyperactive in these individuals . However , it is not clear exactly how these abnormalities are able to circumvent the systems that usually act to suppress unnecessary movements . Now , Liu et al . have discovered a mutation in mice that causes dystonia-like symptoms . When the mice first wake up , or when they are placed in a new environment , one or both of their hind legs become over extended . The mice walk and climb normally when they are in their usual cage and after they have been awake for longer periods . In the experiments , the mice underwent a series of tests to determine what caused these intermittent symptoms . The experiments suggested that hyperactive synapses in the spinal cord trigger the movements , but that the brain is often able to counteract them . Genetic experiments revealed that the mice have a mutation in the Lamb1 gene , which encodes a protein that forms a structural support in the synapse . Next , Liu et al . examined synapses in some parts of the brain of the mutant mice . During normal movements , the levels of synapse activity in these mice were similar to those observed in normal mice . However , when abnormal movements occurred in the mutant mice , their synapses produced irregular patterns of activity . Further studies of these mice should help researchers to better understand what goes wrong in human patients with dystonia .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	A dystonia-like movement disorder with brain and spinal neuronal defects is caused by mutation of the mouse laminin β1 subunit, Lamb1
The Hippo signaling pathway regulates tissue growth in Drosophila through the transcriptional coactivator Yorkie ( Yki ) . How Yki activates target gene transcription is poorly understood . Here , we identify Nuclear receptor coactivator 6 ( Ncoa6 ) , a subunit of the Trithorax-related ( Trr ) histone H3 lysine 4 ( H3K4 ) methyltransferase complex , as a Yki-binding protein . Like Yki , Ncoa6 and Trr are functionally required for Hippo-mediated growth control and target gene expression . Strikingly , artificial tethering of Ncoa6 to Sd is sufficient to promote tissue growth and Yki target expression even in the absence of Yki , underscoring the importance of Yki-mediated recruitment of Ncoa6 in transcriptional activation . Consistent with the established role for the Trr complex in histone methylation , we show that Yki , Ncoa6 , and Trr are required for normal H3K4 methylation at Hippo target genes . These findings shed light on Yki-mediated transcriptional regulation and uncover a potential link between chromatin modification and tissue growth . The Hippo signaling pathway has recently emerged as a central mechanism in organ size control , tissue regeneration , and stem cell biology ( Harvey and Tapon , 2007; Badouel et al . , 2009; Pan , 2010; Zhao et al . , 2010; Halder and Johnson , 2011; Barry and Camargo , 2013 ) . Initially discovered in Drosophila for its critical role in restricting imaginal disc growth , the Hippo pathway comprises several tumor suppressor proteins acting through a core kinase cascade that ultimately phosphorylates and inactivates the transcriptional coactivator Yorkie ( Yki ) ( Huang et al . , 2005 ) . Consistent with its essential role in normal development and tissue homeostasis , YAP , the mammalian counterpart of Yki , encodes a bona fide oncogene and is overexpressed and/or activated in a wide spectrum of human cancers . Elucidating the molecular mechanism by which Yki functions as a transcriptional coactivator is not only relevant for understanding the fundamental mechanisms of growth control but also has important implications for the development of therapeutic strategies targeting the Hippo pathway in cancer and regenerative medicine . Posttranslational modifications of histones are important features of transcriptional regulation in all eukaryotes . A particularly prevalent modification involved in transcriptional activation is histone H3 methylation . Drosophila contains three COMPASS ( complex of proteins associated with Set1 ) -like histone H3 lysine 4 ( H3K4 ) methyltransferase complexes , each defined by a distinct methyltransferase subunit , namely , Trithorax ( Trx ) , Trithorax-related ( Trr ) , and dSet1 ( Mohan et al . , 2011 ) . Previous genetic analysis has implicated Trx in the maintenance of Hox gene transcription and Trr in ecdysone receptor ( EcR ) -mediated gene transcription ( Sedkov et al . , 2003 ) . Nuclear receptor coactivator 6 ( Ncoa6 ) is a specific subunit of the Trr complex in both Drosophila and mammals ( Mohan et al . , 2011 ) . Although its in vivo function remains undefined in Drosophila , the mammalian Ncoa6 orthologue ( also known as NRC , ASC-2 , TRBP , PRIP , and RAP250 ) is essential for embryonic development ( Kuang et al . , 2002; Antonson et al . , 2003; Zhu et al . , 2003; Mahajan et al . , 2004 ) . The mammalian Ncoa6 has been shown to potentiate the activity of nuclear hormone receptors and other DNA-binding transcription factors , at least in part , by recruiting the H3K4 methyltransferases ( Mahajan and Samuels , 2008 ) . Interestingly , like YAP , the mammalian Ncoa6 is a pro-survival and anti-apoptotic gene ( Mahajan et al . , 2004 ) and is amplified in multiple cancer types such as breast , colon , and lung cancers ( Lee et al . , 1999 ) . In this study , we identify Ncoa6 as a Yki-binding protein that is required for transcriptional regulation by the Hippo signaling pathway . We provide evidence showing that the transcriptional coactivator function of Yki depends on its ability to interact with Ncoa6 and that the Trr methyltransferase complex is functionally required for Hippo-mediated growth and gene expression . We further show that Yki , Ncoa6 , and Trr are required for normal H3K4 methylation at Hippo target genes . Thus , Yki functions as a transcriptional coactivator , at least in part , by recruiting a H3K4 methyltransferase and altering the chromatin state of target genes . We recently reported a genome-wide RNAi screen in Drosophila S2R+ cells using a luciferase reporter driven by a minimal Hippo Responsive Element ( HRE ) from the Hippo target gene diap1 ( Koontz et al . , 2013 ) . Briefly , Drosophila S2R+ cells were transfected with Yki- and Sd-expressing vectors , together with HRE-luciferase reporter and Pol III-Renilla expression vector as an internal control . Transfected cells were then seeded into individual dsRNA-containing 96-well plates . After RNAi depletion , HRE-luciferase reporter activity was measured and normalized to the Renilla control . This RNAi screen allowed us to uncover both positive and negative regulators of the HRE-luciferase reporter . We have previously characterized a negative regulator from the screen , Tgi ( Koontz et al . , 2013 ) . Here , we focus on the positive transcriptional regulators ( Figure 1—figure supplement1 ) . One of the positive regulators identified in our primary screen is Nuclear receptor coactivator 6 ( Ncoa6 ) , which was confirmed by repeating the RNAi experiment in triplicate using re-synthesized dsRNA ( Figure 1—figure supplement1 ) . We were particularly interested in Ncoa6 since it contains three PPxY sequences ( Figure 1A ) , which represent a well-established ligand binding motif for WW domains . Since Yki contains two WW domains , we hypothesized that Ncoa6 may potentiate Yki-mediated transcriptional activation through physical interactions with Yki . Indeed , epitope-tagged Ncoa6 and Yki coimmunoprecipitated with each other in Drosophila S2R+ cells ( Figure 1B ) . This interaction was abolished by mutating the three PPxY motifs in Ncoa6 ( Ncoa63m ) or the two WW domains in Yki ( YkiWM ) ( Figure 1B ) , suggesting that the Ncoa6–Yki interaction was mediated by Ncoa6's PPxY motifs and Yki's WW domains . In agreement with this conclusion , we found that wild-type Ncoa6 , but not Ncoa63m , promoted nuclear accumulation of Yki in S2R+ cells in co-transfection assays ( Figure 1C ) . Of note , a recently published Hippo pathway protein–protein interactome included Ncoa6 as one of 245 proteins that were co-immunopreciated by Yki ( Kwon et al . , 2013 ) . 10 . 7554/eLife . 02564 . 003Figure 1 . Ncoa6 physically interacts with Yki and regulates HRE activity . ( A ) Schematic protein structure of Drosophila Ncoa6 and its human orthologue , which contain three and two PPxY motifs , respectively . ( B ) S2R+ cells expressing the indicated constructs were subjected to immunoprecipitation as indicated . Note the physical interactions between Ncoa6 and Yki , and absence of interactions between Ncoa63m and Yki or between Ncoa6 and YkiWM . ( C ) Drosophila S2R+ cells co-transfected with HA-Yki and FLAG-Ncoa6 or FLAG-Ncoa63m constructs were stained for the indicated epitopes . Cells with or without FLAG expression are marked by arrowheads and arrows , respectively . Both FLAG-Ncoa6 and FLAG-Ncoa63m were localized to the nucleus ( arrowheads ) , while HA-Yki was more concentrated in the cytoplasm ( arrows ) . FLAG-Ncoa6 , but not FLAG-Ncoa63m , induced nuclear accumulation of HA-Yki ( compare arrowheads in the merged channel ) . ( D ) Luciferase activity was measured in triplicates in Drosophila S2R+ cells transfected with the indicated constructs . Ncoa6 , but not Ncoa63m , enhanced Yki/Sd-mediated activation of HRE-luciferase reporter . This enhancement was suppressed by co-expression of Wts . Error bars represent standard deviations . ( E ) Drosophila S2R+ cells expressing FLAG-tagged Ncoa6 were subjected to ChIP analysis using control IgG , antibodies against FLAG or antibodies against endogenous Yki . The enrichment of HRE at the endogenous diap1 locus was measured by real-time PCR . Both Yki and FLAG-Ncoa6 bound to the diap1 HRE . DOI: http://dx . doi . org/10 . 7554/eLife . 02564 . 00310 . 7554/eLife . 02564 . 004Figure 1—figure supplement1 . Identification of Ncoa6 as a positive regulator of the HRE activity from cell-based RNAi screen . ( A ) Identification of Ncoa6 as a positive regulator of HRE activity from the primary RNAi screen . The scatter plot highlights genes whose RNAi resulted in a decrease in Yki/Sd-induced HRE reporter activity , with each gene represented by a single dot . The locations of Ncoa6 ( CG14023 ) , sd , and yki are marked . ( B ) Validation of Ncoa6 as a positive regulator of HRE activity . Luciferase activity was measured in triplicates in Drosophila S2R+ cells transfected with Yki , Sd , HRE-Luciferase , and Pol III-Renilla expression vectors , together with dsRNA of GFP ( control ) or Ncoa6 . Error bars represent standard deviations . ( C ) A list of primary hits from the RNAi screen with Z-scores of less than −2 . 26 . DOI: http://dx . doi . org/10 . 7554/eLife . 02564 . 004 Consistent with our observation that RNAi knockdown of Ncoa6 reduced the HRE reporter activity , overexpression of Ncoa6 , but not the PPxY mutant Ncoa63m , potently enhanced Yki/Sd-mediated HRE reporter activity in Drosophila S2R+ cells ( Figure 1D ) . In addition , the enhancement of HRE reporter activity by Ncoa6 was significantly suppressed by co-expression of the kinase Wts ( Figure 1D ) . These results further support the importance of Ncoa6–Yki interactions in Hippo-responsive transcriptional regulation . Consistent with this notion , chromatin immunoprecipitation ( ChIP ) revealed that Ncoa6 , like Yki , binds to the HRE site in the endogenous diap1 gene locus in S2R+ cells ( Figure 1E ) . Since mutant alleles of Ncoa6 are not available , we used a previously validated transgenic RNAi line ( Herz et al . , 2012 ) to assess the role of Ncoa6 in tissue growth and Hippo target expression in vivo . Expression of UAS-Ncoa6 RNAi by the dpp-Gal4 driver resulted in a significant decrease in the width of the dpp-expression domain in adult wings , which corresponds to the region bordered by veins L3 and L4 ( Figure 2A ) . Examination of third instar larval wing imaginal discs revealed a corresponding decrease in the expression of diap1 and four-jointed ( fj ) , two well-characterized Hippo pathway target genes ( Figure 2B–C , E–F ) . These results suggest that Ncoa6 is required for normal tissue growth and expression of Hippo target genes in vivo . 10 . 7554/eLife . 02564 . 005Figure 2 . Ncoa6 and Trr are required for normal tissue growth and expression of Hippo target genes in Drosophila imaginal discs . ( A ) RNAi knockdown of Ncoa6 and Trr by dpp-Gal4 resulted in decreased area of the dpp expression domain in adult wings . The pictures were taken under the same magnification . The graph shows quantification of the dpp expression domain ( green area in the schematic drawing ) relative to the entire wing area ( mean ± SEM , n = 14 , *p<0 . 001 ) . The complete genotypes are: UAS-Dicer2; dpp-Gal4 UAS-GFP ( control ) , UAS-Dicer2; dpp-Gal4 UAS-GFP/UAS-Ncoa6RNAi ( Ncoa6 RNAi ) , and UAS-Dicer2; dpp-Gal4 UAS-GFP/UAS-trrRNAi ( trr RNAi ) . ( B–G ) RNAi knockdown of Ncoa6 or Trr resulted in decreased expression of Hippo target genes . Wing discs expressing UAS-GFP only ( B and E ) , UAS-GFP plus Ncoa6 RNAi ( C and F ) , or UAS-GFP plus trr RNAi ( D and G ) were stained for Diap1 ( B–D ) or fj-lacZ ( E–G ) . Note the reduction of Diap1 and fj-lacZ levels upon Ncoa6 or Trr RNAi . The complete genotypes are: UAS-Dicer2; dpp-Gal4 UAS-GFP ( B ) , UAS-Dicer2; dpp-Gal4 UAS-GFP/UAS-Ncoa6RNAi ( C ) , UAS-Dicer; dpp-Gal4 UAS-GFP/UAS-trr RNAi ( D ) , UAS-Dicer2; fj-lacZ; dpp-Gal4 UAS-GFP ( E ) , UAS-Dicer2; fj-lacZ; dpp-Gal4 UAS-GFP/UAS-Ncoa6 RNAi ( F ) , and UAS-Dicer2; fj-lacZ; dpp-Gal4 UAS-GFP/UAS-trr RNAi ( G ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02564 . 005 Next , we examined genetic interactions between Ncoa6 and the Hippo pathway . Overexpression of Yki or RNAi of Wts by the GMR-Gal4 driver leads to increased eye size ( Figure 3D , G ) . Both phenotypes were suppressed by RNAi knockdown of Ncoa6 ( Figure 3E , H ) . Conversely , knockdown of Ncoa6 exacerbated the small eye size induced by Sd overexpression ( Figure 3J–K ) . To further investigate the genetic interactions between Ncoa6 and the Hippo pathway , we used Mosaic Analysis with a Repressible Cell Marker ( MARCM ) ( Lee and Luo , 1999 ) to examine the requirement of Ncoa6 in hpo mutant clones . Ncoa6 knockdown suppressed the overgrowth as well as the elevated Diap1 expression in hpo mutant clones ( Figure 4A–D ) . In fact , hpo mutant clones with Ncoa6 knockdown showed a decrease in Diap1 expression , similar to wildtype clones with Ncoa6 knockdown ( Figure 4C–D ) . These findings further implicate Ncoa6 in Hippo-mediated growth control and gene expression . 10 . 7554/eLife . 02564 . 006Figure 3 . Genetic interactions between Ncoa6-Trr and the Hippo pathway . Adult eye images of the indicated genotypes , all taken under the same magnification . ( A ) GMR-Gal4/+ . Wild-type control . ( B ) GMR-Gal4/+; UAS-Ncoa6 RNAi/+ . RNAi knockdown of Ncoa6 resulted in a mild decrease in eye size ( compare B to A ) . ( C ) UAS-Dicer2/+; GMR-Gal4/+; UAS-trr RNAi/+ . RNAi knockdown of Trr resulted in no visible effects on eye size ( compare C to A ) . ( D ) GMR-Gal4 UAS-Yki/+ . Overexpression of Yki resulted in an increase in eye size ( compare D to A ) . ( E ) GMR-Gal4 UAS-Yki/+; UAS-Ncoa6 RNAi/+ . RNAi knockdown of Ncoa6 suppressed eye overgrowth induced by Yki overexpression ( compare E to D ) . ( F ) UAS-Dicer2/+; GMR-Gal4 UAS-Yki/+; UAS-trr RNAi/+ . RNAi knockdown of Trr suppressed eye overgrowth induced by Yki overexpression ( compare F to D ) . ( G ) UAS-Wts RNAi/+; GMR-Gal4/+ . RNAi knockdown of Wts resulted in an increase in eye size ( compare G to A ) . ( H ) UAS-Wts RNAi/+; GMR-Gal4/ UAS-Ncoa6 RNAi . RNAi knockdown of Ncoa6 suppressed eye overgrowth induced by Wts knockdown ( compare H to G ) . ( I ) UAS-Dicer2/+; UAS-Wts RNAi/+; GMR-Gal4/ UAS-trr RNAi . RNAi knockdown of Trr did not obviously suppress eye overgrowth caused by Wts knockdown . ( J ) GMR-Gal4 UAS-Sd/+ . Overexpression of Sd resulted in a decrease in eye size ( compare J to A ) . ( K ) GMR-Gal4 UAS-Sd/+; UAS-Ncoa6 RNAi/+ . RNAi knockdown of Ncoa6 enhanced the small eye phenotype caused by Sd overexpression ( compare K to J ) . ( L ) UAS-Dicer2/+; GMR-Gal4 UAS-Sd/+; UAS-trr RNAi/+ . RNAi knockdown of Trr enhanced the small eye phenotype caused by Sd overexpression ( compare L to J ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02564 . 00610 . 7554/eLife . 02564 . 007Figure 4 . Ncoa6 and Trr are required for Hippo-mediated target gene expression . Wing discs containing GFP-marked MARCM clones were stained for Diap1 ( red ) . For each genotype , the left most panel shows low magnification view of the wing disc ( Hoechst + GFP ) , while the remaining three panels show higher magnification view of the same wing disc ( GFP , Diap1 and GFP + Diap1 ) . ( A–F ) Wing discs containing GFP-marked MARCM clones ( green ) of WT control ( A ) , hpo mutant ( B ) , Ncoa6 RNAi ( C ) , hpo mutant with Ncoa6 RNAi ( D ) , Trr RNAi ( E ) , and hpo mutant with Trr RNAi ( F ) . Note the increased Diap1 levels in hpo mutant clones and the decreased Diap1 levels in Ncoa6 RNAi or Trr RNAi clones . Also note the decreased Diap1 levels in hpo mutant clones with Ncoa6 RNAi or Trr RNAi . DOI: http://dx . doi . org/10 . 7554/eLife . 02564 . 007 The physical interactions between Yki and Ncoa6 , together with the requirement for Ncoa6 in tissue growth and Hippo target gene expression , suggest that Yki may function as a transcriptional coactivator by interacting with Ncoa6 . Since Sd is the primary DNA-binding transcription partner for Yki ( Koontz et al . , 2013 ) , we reasoned that fusing the DNA-binding domain of Sd with Ncoa6 may directly target Ncoa6 to Hippo target genes and therefore stimulate their transcription in a Yki-independent manner ( i . e . , bypassing the genetic requirement for Yki ) . We tested this hypothesis in Drosophila wing discs using the MARCM technique . As reported before ( Huang et al . , 2005 ) , yki mutant clones grew poorly and the rarely recovered clones always showed decreased Diap1 levels ( Figure 5A–B ) . MARCM clones expressing a fusion protein between the DNA-binding domain of Sd and Ncoa6 ( SdDB-Ncoa6 ) resulted in rounded clone morphology and dramatically increased Diap1 levels ( Figure 5C ) . Significantly , the SdDB-Ncoa6 fusion protein , but not wild-type Ncoa6 , rescued the growth defect and the decreased Diap1 levels in yki mutant clone ( Figure 5D–E ) . In fact , yki mutant clones with SdDB-Ncoa6 overexpression were indistinguishable in clone size and Diap1 expression compared to wild-type clones with SdDB-Ncoa6 overexpression ( Figure 5C–D ) . Thus , the SdDB–Ncoa6 fusion protein exhibits gain-of-function activity in a Yki-independent manner . Consistent with this notion , the SdDB-Ncoa6 fusion protein robustly stimulated the HRE-luciferase reporter in S2R+ cells in a manner that was not suppressed by co-expression of Wts ( Figure 5F ) , in contrast to Wts' ability to suppress the HRE-luciferase reporter activity stimulated by the co-transfection of Sd , Yki , and Ncoa6 ( Figure 1D ) . 10 . 7554/eLife . 02564 . 008Figure 5 . Fusion of Ncoa6 with the DNA binding domain of Sd bypasses Yki to stimulate Hippo target gene and tissue growth . ( A–E ) Wing discs containing GFP-marked MARCM clones ( green ) of WT control ( A ) , ykiB5 ( B ) , SdDB-Ncoa6 overexpression ( C ) , ykiB5 with SdDB-Ncoa6 overexpression ( D ) , and ykiB5 with Ncoa6 overexpression ( E ) , were stained for Diap1 ( red ) . For each genotype , the left most panel shows low magnification view of the wing disc ( Hoechst + GFP ) , while the remaining three panels show higher magnification view of the same wing disc ( GFP , Diap1 , and GFP + Diap1 ) . Note the decreased Diap1 expression and undergrowth of ykiB5 clones ( B ) or ykiB5 clones with Ncoa6 overexpression ( E ) . SdDB-Ncoa6 overexpression resulted in elevated Diap1 levels in ykiB5 clones ( D ) . ( F ) Luciferase activity was measured in triplicates in Drosophila S2R+ cells transfected with the indicated constructs . Error bars represent standard deviations . Note the Wts-insensitive stimulation of the HRE-luciferase reporter by SdDB-Ncoa6 . DOI: http://dx . doi . org/10 . 7554/eLife . 02564 . 008 The results presented above suggest that Yki activates gene expression , at least in part , by recruiting Ncoa6 . Since Ncoa6 has been reported to be a specific subunit of the Trr methyltransferase complex , we examined whether Trr , the catalytic subunit of this methyltransferase complex , is also required for Hippo-mediated growth control and gene expression . Similar to Ncoa6 , expression of UAS-trr RNAi by the dpp-Gal4 driver resulted in a significant decrease in the width of the dpp-expression domain in adult wings and a corresponding decrease in the Hippo target genes diap1 and fj ( Figure 2A , D , G ) . Like Ncoa6 , Trr knockdown suppressed eye overgrowth induced by Yki overexpression ( Figure 3D , F ) and aggravated the small eye phenotype caused by Sd overexpression ( Figure 3J , L ) , although it did not visibly suppress eye overgrowth induced by Wts RNAi ( Figure 3I ) . We also used MARCM to examine the requirement of Trr in hpo mutant clones . Similar to Ncoa6 , Trr knockdown suppressed the overgrowth as well as the elevated Diap1 expression in hpo mutant clones ( Figure 4E–F ) . Taken together , these results implicate the Trr methyltransferase complex in Hippo-mediated growth control and target gene expression . The Trr methyltransferase complex in Drosophila mainly affects histone H3K4 monomethylation with subtle effect on H3K4 di- or trimethylation ( Herz et al . , 2012; Kanda et al . , 2013 ) . To determine if Yki , Ncoa6 , and Trr regulate growth in the Hippo pathway by modulating histone H3K4 methylation , we first examined the global levels of histone H3K4 methylation in Drosophila wing imaginal discs . It was reported previously that RNAi knockdown of Trr ( using the en-Gal4 driver ) led to a strong decrease in H3K4me1 in the posterior compartment of the wing imaginal discs , while it had marginal effects on H3K4me2 and H3K4me3 levels ( Mohan et al . , 2011; Herz et al . , 2012 ) . It was also showed that RNAi knockdown of Ncoa6 in the posterior compartment of the wing imaginal disc resulted in a weak reduction in H3K4me1 levels ( Herz et al . , 2012 ) , which we confirmed ( Figure 6—figure supplement1 ) . We further examined H3K4me2 and H3K4me3 levels in these imaginal discs , and observed a very subtle decrease in H3K4me3 levels and no detectable changes in H3K4me2 levels ( Figure 6—figure supplement 1B–C ) . However , when we examined mutant clones of yki in the wing imaginal discs , we could not detect any changes in the global levels of H3K4me1 , H3K4me2 , or H3K4me3 ( Figure 6—figure supplement1 ) . Given the negligible effect of Yki on global levels of H3K4 methylation , we investigated whether Yki modulates local H3K4 methylation on Hippo target genes . It is well established that H3K4 monomethylation marks enhancers and actively transcribed introns , while H3K4 trimethylation is enriched at active promoters and transcription start site ( TSS ) -proximal regions ( Heintzman et al . , 2007; Kharchenko et al . , 2011 ) . Interestingly , a previous genome-wide analysis in Drosophila S2 cells revealed that diap1 and ex , two well-characterized Hippo target genes , display such differential enrichment of H3K4me1 and H3K4me3 at the respective region of each gene ( Herz et al . , 2012 ) ( Figure 6A ) . To examine the contribution of Yki , Ncoa6 , and Trr to H3K4 methylation at these Hippo target genes , we knocked down each protein in S2R+ cells and performed ChIP analysis with antibodies against H3K4me1 and H3K4me3 . RNAi knockdown of Yki , Ncoa6 or Trr resulted in a decrease of H3K4me3 in the TSS-proximal region of diap1 and ex ( Figure 6B ) , which normally showed the strongest enrichment of H3K4me3 marks ( Herz et al . , 2012 ) ( Figure 6A ) . RNAi knockdown of Yki , Ncoa6 or Trr also led to a decrease of H3K4me1 in the intronic HRE of diap1 and an upstream region of diap1 or ex which normally showed the strongest enrichment of H3K4me1 marks ( Herz et al . , 2012 ) ( Figure 6C ) . Collectively , these data are consistent with the view that Yki activates target gene transcription by interacting with the Trr methyltransferase complex and modifying the chromatin state of the target loci . 10 . 7554/eLife . 02564 . 009Figure 6 . Yki modulates local H3K4 methylation at Hippo target genes . ( A ) Schematic view of diap1 and ex genomic loci analyzed by ChIP . Transcriptional start site is labeled as +1 , and p1–p5 are a series of primer sets encompassing following regions of diap1 and ex: diap1: p1: −1951 ∼ −1813 , p2: +228 ∼ +377 , p3: +3993 ∼ +4104; ex: p4: −749 ∼ −608 , p5: +249 ∼ +393 . Note that p3 covers the diap1 HRE . Also shown are the profiles of H3K4me1 ( blue line ) and H3K4me3 ( red line ) binding derived from a previously published ChIP-Seq analysis in S2 cells ( Herz et al . , 2012 ) . ( B and C ) RNAi knockdown of Yki , Ncoa6 or Trr resulted in decreased H3K4me3 ( B ) and H3K4me1 ( C ) modification on Hippo target genes . ChIP analysis of H3K4me1 or H3K4me3 were performed in Drosophila S2R+ cells treated with dsRNA of GFP ( control ) , Yki , Ncoa6 , or Trr . Chromatins were precipitated by control IgG or antibodies against H3K4me1 and H3K4me3 . The enrichment of ChIP products on diap and ex was measured by real-time PCR using the indicated primers . p<0 . 001 , p<0 . 01 , p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 02564 . 00910 . 7554/eLife . 02564 . 010Figure 6—figure supplement1 . Ncoa6 , but not Yki , regulates global levels of H3K4 methylation . ( A–C ) Wing discs with Ncoa6 RNAi in the posterior compartment were stained for mono- , di- , and tri-methylation of H3K4 as indicated . Note the subtle decrease of H3K4me1 ( A ) and H3K4me3 ( C ) , but not H3K4me2 ( B ) , in the GFP-marked posterior compartment . The complete genotype is: UAS-Dicer2; en-Gal4 UAS-GFP; UAS-Ncoa6RNAi . ( D–F ) Wing discs containing ykiB5 mutant clones were stained for H3K4me1 , H3K4me2 , and H3K4me3 as indicated . Note the normal levels of H3K4me1 , H3K4me2 , and H3K4me4 in yki mutant clones ( arrows , GFP-negative ) compared to the wild-type neighbors ( GFP-positive ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02564 . 010 Despite the ever expanding complexity of upstream inputs into the Hippo pathway , all of them converge on the transcriptional coactivator Yki . Thus , understanding the molecular mechanisms by which Yki regulates tissue growth and target gene expression has important implications for developmental and cancer biology . Previous studies have established that Yki functions primarily as a coactivator for Sd and that Yki promotes tissue growth by antagonizing Sd's repressor function ( Wu et al . , 2008; Koontz et al . , 2013 ) . Our current study has extended the previous work by identifying Ncoa6 as a Yki-binding cofactor that is required for the expression of Yki target genes . The ability of the SdDB-Ncoa6 fusion protein to rescue the growth and transcriptional defects in yki mutant clones highlights the importance of Ncoa6 recruitment in the transcriptional output of the Hippo pathway . Our results further suggest that Ncoa6 recruits the Trr methyltransferase complex to Hippo target genes and that Yki regulates target gene transcription by modulating local H3K4 methylation . Consistent with this view , a recent genome-wide chromatin-binding analysis revealed a correlation between Yki-bound chromatin and peaks of H3K4me3 modification in Drosophila wing discs and embryos ( Oh et al . , 2013 ) . We note that , besides the H3K4 methyltransferases , the mammalian Ncoa6 has been reported to potentiate the activity of transcription factors by interacting with histone acetyltransferase CBP/p300 and several RNA binding proteins ( CAPER , CoAA and PIMT ) ( Mahajan and Samuels , 2008 ) . Whether these additional mechanisms also contribute to the function of Ncoa6 in Yki-mediated growth control requires further investigation . Given the biological and clinical significance of the Hippo pathway , further studies into the molecular mechanism of Ncoa6 will advance our understanding of developmental growth control and facilitate the development of novel therapeutic strategies . A full-length Ncoa6 cDNA corresponding to the BDGP annotated RD transcript was generated from mRNA of Drosophila third instar larvae , using SuperScript ( R ) III One-Step RT-PCR System with Platinum Taq High Fidelity ( Life Technologies , Carlsbad , California ) . Mutations of PPxY motifs were generated in Ncoa6 using the QuikChange II XL Site-Directed Mutagenesis Kit ( Agilent , Santa Clara , CA ) , replacing tyrosine ( Y ) with alanine ( A ) . To generate Sd-Ncoa6 , Sd DNA binding domain was inserted to the N-terminal of Ncoa6 . FLAG-tag was inserted to the N-terminal of Ncoa6 , Ncoa63m , and Sd-Ncoa6 and cloned into the attB-UAS vector . Flies with the following genotypes have been described previously: ykiB5 , UAS-Yki ( Huang et al . , 2005 ) , hpo42−48 ( Wu et al . , 2003 ) , fj-lacZ reporter fj9-II ( Villano and Katz , 1995 ) , UAS-Sd ( Halder et al . , 1998 ) , UAS-Wts RNAi ( Stock ID VDRC v106174 ) . The UAS-Ncoa6 RNAi and UAS-trr RNAi lines have been validated previously ( Herz et al . , 2012 ) and were obtained from Bloomington Drosophila Stock Center ( Stock ID 34964 and 29 , 563 ) . attB-UAS-Ncoa6 and attB-UAS-FLAG-SdDB-Ncoa6 transgenes were inserted into the 86Fa attP acceptor site by phiC31-mediated site-specific transformation ( Bischof et al . , 2007 ) . For the MARCM experiments in Figure 4 , the following clones were induced 48–60 hr after egg deposition and heat shocked at 37°C for 15 min:UAS-GFP hs-FLP; FRT42D , Tub-Gal80/FRT42D; Tub-Gal4/+UAS-GFP hs-FLP; FRT42D , Tub-Gal80/FRT42D hpo42−48; Tub-Gal4/+UAS-GFP hs-FLP; FRT42D , Tub-Gal80/FRT42D; Tub-Gal4/UAS-Ncoa6RNAiUAS-GFP hs-FLP; FRT42D , Tub-Gal80/FRT42D hpo42−48; Tub-Gal4/UAS-Ncoa6RNAiUAS-Dicer2/UAS-GFP hs-FLP; FRT42D , Tub-Gal80/FRT42D; Tub-Gal4/UAS-trrRNAiUAS-Dicer2/UAS-GFP hs-FLP; FRT42D , Tub-Gal80/FRT42D hpo42−48; Tub-Gal4/UAS-trrRNAi For the MARCM experiments in Figure 5A–E , the following clones were induced 72–84 hr after egg deposition and heat shocked at 37°C for 10 min:UAS-GFP hs-FLP; FRT42D , Tub-Gal80/FRT42D; Tub-Gal4/+UAS-GFP hs-FLP; FRT42D , Tub-Gal80/FRT42D ykiB5; Tub-Gal4/+UAS-GFP hs-FLP; FRT42D , Tub-Gal80/FRT42D; Tub-Gal4/UAS-SdDB-Ncoa6UAS-GFP hs-FLP; FRT42D , Tub-Gal80/FRT42D ykiB5; Tub-Gal4/UAS-SdDB-Ncoa6UAS-GFP hs-FLP; FRT42D , Tub-Gal80/FRT42D ykiB5; Tub-Gal4/UAS-Ncoa6 Drosophila S2R+ cells were cultured in Schneider's Drosophila Medium ( Life Technologies ) supplemented with 10% fetal bovine serum and antibiotics . HA-Yki and HA-YkiWM have been described previously ( Huang et al . , 2005 ) . Luciferase assay was carried out using Dual Luciferase Assay System ( Promega , Madison , WI ) and a FLUOstar Luminometer ( BMG LabTechnologies , Germany ) . Tansfection , immunopreciptation , and immunofluorescence staining of S2R+ cells were performed using standard protocols as described ( Yin et al . , 2013 ) . ChIP assays were performed according to a previously described protocol ( Wang et al . , 2009 ) . Briefly , ∼5 × 106 ( for ChIP assay with histone methylation antibodies ) or 1 . 5 × 107 ( for CHIP assay with Yki or FLAG antibodies ) S2R+ cells were cross-linked with 1% formaldehyde and sonicated to an average fragment size between 200 bp and 500 bp . Two micrograms of control IgG or specific antibodies , including rabbit ɑ-H3K4me1 ( 8895 , Abcam , England ) and rabbit ɑ-H3K4me3 ( 8580 , Abcam ) , and 50 µl of protein G agarose were used in each ChIP assay . The immunoprecipitated DNA was quantified using real-time PCR . All values were normalized to the input . The primers for analyzing the ChIP NA are provided as follows:p1 Forward: TGTTCTTGTTGGTGCTGCTTp1 Reverse: TTAATGCTGGCATGGTTTCAp2 Forward: TAAAACTGGGGCTCACCTTGp2 Reverse: TCGTGTTCACGGAAAATCAAp3 ( HRE ) Forward: ACGAACACGAAGACCAAAp3 ( HRE ) Reverse: CTCCAAGCCAGTTTGATTp4 Forward: AAAAGAGGGAAGAGGGAGCAp4 Reverse: GAATCGGAATCGGAACTTGAp5 Forward: TCGCACTCGCCTCAATTACp5 Reverse: CAGCACCAACTTTTCGGAGT	Cells need to work together for a multi-celled organism , such as a plant or an animal , to thrive . Many complicated signaling pathways therefore exist that allow cells to communicate with one another and to control their own activity in response to the signals that they receive . One such pathway , called the Hippo signaling pathway , regulates when cells grow and divide , which allows organs and tissues to develop correctly and helps to prevent cancerous tumors from forming . Signaling pathways often control the activity of cells by affecting how particular genes are expressed from DNA . One way of doing this is to activate or inactivate proteins called transcription factors , which bind to sections of DNA to alter the expression of nearby genes . In fruit flies , the Hippo signaling pathway stops cells from dividing by inactivating Yorkie , a protein that binds to and activates certain transcription factors . However , exactly how Yorkie is able to activate these transcription factors was unclear . For transcription factors to function correctly , they must be able to reach the stretch of DNA where they bind . Therefore , another way to alter gene expression is to change how the DNA is packaged in a cell . This can be done by modifying the proteins that the DNA is wrapped around , which are called histones , for example by using enzymes called methyltransferases to add methyl groups to these proteins . Qing , Yin et al . looked at the Hippo signaling pathway in the fruit fly Drosophila , and found that the Yorkie protein only activates transcription factors when another protein called Ncoa6—which is part of a methyltransferase—binds to it . Furthermore , when the Ncoa6 protein was bound directly to the transcription factor , the tissue grew normally even when the Yorkie protein was not present . These findings reveal the importance of histone modifications in controlling tissue growth , and could provide a new direction in the search for cancer treatments .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2014	The Hippo effector Yorkie activates transcription by interacting with a histone methyltransferase complex through Ncoa6
SpoIIIE is a homo-hexameric dsDNA translocase responsible for completing chromosome segregation in Bacillus subtilis . Here , we use a single-molecule approach to monitor SpoIIIE translocation when challenged with neutral-backbone DNA and non-hydrolyzable ATP analogs . We show that SpoIIIE makes multiple essential contacts with phosphates on the 5'→3' strand in the direction of translocation . Using DNA constructs with two neutral-backbone segments separated by a single charged base pair , we deduce that SpoIIIE’s step size is 2 bp . Finally , experiments with non-hydrolyzable ATP analogs suggest that SpoIIIE can operate with non-consecutive inactive subunits . We propose a two-subunit escort translocation mechanism that is strict enough to enable SpoIIIE to track one DNA strand , yet sufficiently compliant to permit the motor to bypass inactive subunits without arrest . We speculate that such a flexible mechanism arose for motors that , like SpoIIIE , constitute functional bottlenecks where the inactivation of even a single motor can be lethal for the cell . The ASCE [Additional Strand Conserved E ( glutamate ) ] division of oligomeric , ring-shaped NTPases encompasses a diverse range of enzymes that function as molecular motors ( Lyubimov et al . , 2011 ) . Within the ASCE division , the FtsK/SpoIIIE family of motors is involved in the fundamental process of DNA segregation prior to cell division . During the Bacillus subtilis sporulation lifecycle , an asymmetric division septum closes around one of the sister chromatids before complete chromosome segregation , trapping about two-thirds of the chromosome in the mother cell compartment ( Burton et al . , 2007; Wu and Errington , 1994 ) . To complete chromosome segregation , SpoIIIE must translocate the DNA from the mother cell into the forespore . SpoIIIE and its closely related Escherichia coli homologue FtsK , contain an N-terminal transmembrane domain that anchors the protein to the division septum , a long unstructured polypeptide linker and a C-terminal-soluble motor domain consisting of subdomains α , β , and γ ( Barre , 2007 ) . Subdomains α and β adopt a RecA-like fold containing ATP binding and hydrolysis motifs ( Massey et al . , 2006 ) , while subdomain γ imparts translocation directionality to the motor ( sequence dependence ) ( Besprozvannaya et al . , 2013; Lee et al . , 2012; Löwe et al . , 2008; Ptacin et al . , 2006; 2008 ) . Crystallography and electron microscopy studies indicate that both FtsK and SpoIIIE form homo-hexameric rings , and that double-stranded DNA ( dsDNA ) is threaded through their central pore ( Cattoni , et al . , 2014; Cattoni et al . , 2013; Massey et al . , 2006 ) . A distinguishing characteristic of SpoIIIE/FtsK is their enormous translocation velocity ( ∼5 kbp/s ) and their ability to work against high forces ( Ptacin et al . , 2008; Saleh et al . , 2004 ) . Previous single-molecule studies of these motors focused primarily on investigating the mechanism of sequence recognition and translocation direction reversal ( Lee et al . , 2012; Pease et al . , 2005; Ptacin et al . , 2006; 2008; Saleh et al . , 2004 ) , studying how they strip off DNA-bound proteins ( Lee et al . , 2014; Marquis et al . , 2008 ) , and determining the amount of supercoils introduced in the DNA during translocation ( Saleh et al . , 2005 ) . However , many fundamental aspects of these motors’ operation remain poorly understood: How does the motor interact with its DNA track during translocation ? What is the motor step size ? How does the motor coordinate the activity of its individual subunits ? How is the subunit coordination mechanism optimized for the motor’s specific biological task ? To answer these questions , we used single-molecule manipulation and measurement techniques . Using modified DNA with a neutral backbone , we show that SpoIIIE makes critical electrostatic contacts with the phosphate backbone on the 5'→3' strand in the direction of translocation . This observation indicates that the individual subunits operate in a well-defined sequential order around the ring . To determine the SpoIIIE step size , we challenged the motor to translocate a DNA molecule containing two neutral segments of variable lengths separated by a charged base pair . This hybrid construct revealed the periodicity of motor–DNA interactions , suggesting that each SpoIIIE subunit takes a 2-bp step per ATP hydrolyzed . Experiments where non-hydrolyzable nucleotides were used to probe the intersubunit coordination within the motor suggest that SpoIIIE can tolerate non-consecutive inactive subunits , implying a degree of flexibility in the sequential operation of the motor . Finally , we propose a two-subunit DNA escort model that can rationalize all these data and that correctly predicts the degree of supercoiling introduced by SpoIIIE during translocation . Modified inserts such as ssDNA , dsDNA with interstrand cross-links , dsDNA with abasic sites , and so on , have been previously used in single-molecule experiments to probe how dsDNA translocases interact with their nucleic-acid track ( Aathavan et al . , 2009; Stanley et al . , 2006 ) . To investigate how SpoIIIE interacts with its substrate , we designed DNA constructs containing a modified insert with a methyl-phosphonate ( MeP ) backbone ( Figure 1b ) , which preserves the base pairing and overall structure of B-form DNA ( Strauss and Maher , 1994 ) . We first monitored SpoIIIE translocation along a substrate containing a 30-bp double-stranded MeP ( dsMeP ) insert , several times longer than the expected step size of the motor . Upon reaching the insert , SpoIIIE undergoes multiple slips followed by translocation recoveries ( Figure 1b inset ) , revealing that it makes several attempts to cross the neutral segment . Despite these attempts , SpoIIIE failed to traverse the 30-bp dsMeP segment ( Figure 1b ) , indicating that motor interactions with the negatively charged phosphates are critical for translocation . Depending on the step size of the motor , and the manner in which it interacts with the DNA , the motor should be able to easily traverse short dsMeP inserts up to some critical length . To determine this length , we designed dsMeP inserts of varying size and recorded the motor’s ability to cross each insert ( Figure 2a ) . We found that SpoIIIE can traverse dsMeP segments of 2 bp , 3 bp , and 4 bp with near-100% probability ( Figure 2a ) , but the traversal probability sharply drops to ∼50% for dsMeP inserts of 5 bp ( Figure 2b ) . Importantly , this sharp reduction in insert crossing probability is accompanied by a dramatic increase in the average traversal time ( Figure 2c ) . Note that further increasing the dsMeP insert length from 5 bp to 7 bp does not significantly increase the mean traversal time ( Figure 2c ) . Taken together , these data show that 5 bp is the minimum length of dsMeP required to disrupt processive translocation by SpoIIIE , and indicate that SpoIIIE makes periodic electrostatic contacts with the DNA every 5 bp or less . 10 . 7554/eLife . 09224 . 005Figure 2 . Monitoring SpoIIIE crossing of contiguous MeP inserts . ( a ) Sample traces of SpoIIIE translocating on DNA with dsMeP inserts of 2 bp ( blue ) , 3 bp ( green ) , 4 bp ( magenta ) , and 5 bp ( black ) . ( b ) Traversal probability for dsMeP inserts of various length . Note the sharp decrease in traversal probability between 4-bp and 5-bp inserts . Error-bars represent the 68% CI estimated via bootstrapping . ( c ) Mean traversal time for dsMeP inserts of various length . Note the large increase in traversal time between 4-bp and 5-bp inserts . Error-bars represent the SEM . P values calculated with a two-tailed Fisher exact test . DOI: http://dx . doi . org/10 . 7554/eLife . 09224 . 005 Note that in rare instances ( 1 out of 32 molecules , Table 1 ) , SpoIIIE managed to traverse the 30-bp dsMeP insert , albeit after several seconds of repeated crossing attempts ( Figure 1b ) . A slightly higher traversal probability ( ∼8% ) was recorded for the 10-bp dsMeP insert ( Figure 2b ) . The dynamics of slipping and re-translocation ( Figure 1b , inset ) at the neutral insert and the lengthy traversal times for dsMeP inserts of 5 bp or longer ( Figure 2c ) suggest that , given a sufficient number of traversal attempts , SpoIIIE can cross even relatively long stretches of dsMeP ( 10–30 bp ) . The drop in traversal probability for longer dsMeP inserts suggests that the motor can hold onto MeP DNA for a short amount of time during which it can either step forward , or lose its grip on the neutral DNA . In other words , forward translocation along the MeP insert is in kinetic competition with backward slipping ( Aathavan et al . , 2009 ) . As a result , to traverse longer MeP inserts , the motor must execute a correspondingly larger number of consecutive productive power-strokes . We hypothesize that other types of motor–DNA interactions ( e . g . steric ) enable the motor to traverse the neutral insert given an arbitrarily large number of crossing attempts . In support of this idea , the φ29 ring ATPase has been shown to require electrostatic contacts with the DNA every 10 phosphates , but relies on steric , non-specific interactions to exert force and translocate the DNA in-between those electrostatic contacts ( Aathavan et al . , 2009 ) . 10 . 7554/eLife . 09224 . 006Table 1 . MeP traversal statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 09224 . 006DNA constructSuccessful crossingsFailed crossingsTotal tracesssMeP modification30 base 3'→5' MeP2312430 base 5'→3' MeP62228dsMeP Modification2 bp dsMeP211223 bp dsMeP160164 bp dsMeP213245 bp dsMeP1313267 bp dsMeP391210 bp dsMeP1192030 bp dsMeP13132Length of MeP Probe Segment4 bp dsMeP probe716233 bp dsMeP probe916252 bp dsMeP probe618241 bp dsMeP probe33437All MeP data was gathered under 5 pN of opposing force and [ATP] = 3 mMAll MeP probe segments carry a 4 bp dsMeP modification upstream of the probe To determine whether SpoIIIE tracks the phosphate backbone of one or both DNA strands , we repeated the above experiment with 30-base inserts where phosphates on either strand were selectively neutralized . 23 out of 24 SpoIIIE molecules traversed the insert containing MeP on the 3'→5' strand in the direction of translocation ( Figure 3a , orange traces ) , compared to only 6 out of 28 SpoIIIE molecules that crossed the insert with a neutral backbone on the 5'→3' strand ( Figure 3a , magenta traces ) . Strikingly , the traversals of the insert with a neutral 3'→5' strand were very fast ( similar to translocation along unmodified DNA ) , whereas the traversals of the insert with a neutralized 5'→3' strand required 5–50 s , likely involving several crossing attempts . These results clearly indicate that SpoIIIE makes essential electrostatic contacts selectively with the charged backbone on the 5'→3' strand in the direction of translocation . 10 . 7554/eLife . 09224 . 007Figure 3 . SpoIIIE favors phosphate interactions on the 5’-3’ strand in the direction of translocation . ( a ) Sample traces of SpoIIIE translocating on DNA with 30 bp of neutral DNA on either the 3’-5’ strand ( orange traces ) or the 5’-3’ strand ( magenta traces ) in the direction of translocation . ( b ) Cartoon illustrating a hypothetical sequential model in which the DNA backbone is handed off between adjacent subunits within a hexameric ring . For clarity only one strand of dsDNA is shown . A highlighted backbone phosphate ( yellow ) is in contact with a motor subunit ( magenta ) . In general , the motor step size in such a model can be anything less than 10 bp . Here we illustrate the model using a 2-bp step size . After one motor subunit ( magenta ) executes its power-stroke , the helical backbone of the dsDNA’s will be shifted , positioning the phosphate backbone in close proximity to the next subunit poised to fire . ( c ) Cartoon illustrating a hypothetical model in which a 10-bp burst enables a hexameric ring ATPase to maintain phosphate contacts on the same DNA strand . Initially , a single subunit ( magenta ) is contacting the phosphate backbone ( yellow ) . After a 10 bp burst , the motor has traversed nearly a full helical turn on dsDNA , bringing the phosphate backbone back in register with the same ATPase subunit . DOI: http://dx . doi . org/10 . 7554/eLife . 09224 . 007 The observation that SpoIIIE tracks only one of the DNA strands imposes geometric constraints on how the DNA is handed off from subunit to subunit during translocation . Strand-tracking is inconsistent with a stochastic coordination mechanism in which the six ATPase subunits execute their power-strokes at random . In such a mechanism , the subunit poised to fire is unlikely to be aligned with the tracked strand and will therefore not engage the DNA substrate . Since the DNA is held under constant tension in our experiments , a stochastically coordinated ATPase ring is expected to slip frequently , contrary to our observations . Two possible scenarios are consistent with strand–tracking: ( i ) Subunits hydrolyze ATP and execute their power-strokes sequentially around the ring , such that after every power-stroke the subunit scheduled to fire next is properly aligned to interact with the phosphate backbone of the tracked strand ( Figure 3b ) . In this scenario , the SpoIIIE subunits would fire consecutively ( subunit 1 fires , followed by subunit 2 , followed by subunit 3 , etc ) . After subunit 1 executes its power-stroke , the helical geometry of the DNA will position the phosphate backbone of the tracked strand in close proximity to subunit 2 , depending on the step size of the motor ( Massey et al . , 2006; Strick and Quessada-Vial , 2006 ) . Thus , motor-DNA contacts would proceed from one subunit to the next in an ordinal fashion . ( ii ) The motor translocates DNA in increments of 10–11 bp , which closely matches the helical periodicity of dsDNA ( 10 . 4–10 . 5 bp/pitch ) . This mechanism would ensure that after each translocation event , the motor contacts the same strand after traversing one full helical turn along the DNA ( Figure 3c ) . Such a mechanism is employed by the φ29 ring ATPase , which translocates DNA in 10-bp bursts and contacts the DNA backbone on the same strand every 10 base pairs ( Aathavan et al . , 2009 ) . The data presented above indicate that SpoIIIE makes periodic contacts with the same DNA strand every five base pairs or less and are consistent only with the sequential ATP hydrolysis mechanism outlined in scenario ( i ) . In order to rationalize how SpoIIIE easily traverses 4 bp of neutral DNA , but has difficulty crossing 5 bp of neutral DNA ( Figure 2b ) , we considered several motor-DNA contact models . To be consistent with the observation that SpoIIIE tracks only one DNA strand ( Figure 3a ) , all potential models require that SpoIIIE subunits fire in a well-defined sequential order around the ring ( Figure 4a ) . Furthermore , since electrostatic interactions with the backbone on one strand are critical for maintaining the motor’s grip on DNA , we reason that at least one electrostatic contact between the motor and the phosphate backbone is needed for SpoIIIE to maintain a stable grip on its DNA substrate . Finally , the pool of potential models can be further narrowed by making an assumption about the state of dsDNA inside the central pore of SpoIIIE during translocation . Although no ring-shaped dsDNA translocase has been co-crystallized with its substrate , it is thought that dsDNA is not significantly distorted within the central pore of ring ATPases ( Massey et al . , 2006; Sun et al . , 2008 ) . For example , the central channel of the FtsK ring can fully accommodate a single , undistorted B-form dsDNA strand ( Massey et al . , 2006 ) . Therefore , we considered models where the helical pitch of dsDNA inside the SpoIIIE channel is similar to that of B-form DNA ( 10 . 5 bp/turn ) . Expanding our models to include a distorted helix similar to A-form DNA ( 11 . 0 bp/turn ) would require only minor corrections and would not affect the overall predictions of these models . 10 . 7554/eLife . 09224 . 008Figure 4 . Possible SpoIIIE translocation models capable of crossing 4 bp of dsMeP . ( a ) Cartoon illustrating the sequential firing of subunits A–F of SpoIIIE as it approaches a 4 bp insert of dsMeP . ( b ) Possible translocation models ( i–v ) depicting the interaction between SpoIIIE subunits ( A–F ) and DNA; in all models subunits fire in a sequential order ( A→B→C→D→E→F→A etc ) . For clarity , we show only the backbone the DNA strand tracked by SpoIIIE . Phosphate and MeP groups are shown in blue and red , respectively . A solid line connecting a SpoIIIE subunit to the backbone represents a stable electrostatic interaction; a dashed line represents a disrupted interaction . DOI: http://dx . doi . org/10 . 7554/eLife . 09224 . 008 Five models ( Figure 4b i–v ) are consistent with the results of the dsMeP experiments ( Figure 2 ) and satisfy the conditions listed above: ( i ) At any time , only one subunit contacts one phosphate on DNA ( Figure 4b i ) ; this model requires a 5-bp step size to cross a 4-bp MeP insert . ( ii ) Only one subunit contacts two adjacent DNA phosphates at any time ( Figure 4b ii ) ; this model requires a 4-bp step size to clear a 4-bp MeP insert . ( iii ) At any time , two neighboring subunits each contact one DNA phosphate; this model requires a 3-bp step size to clear a 4-bp MeP insert ( Figure 4b iii ) . ( iv ) At any time , two neighboring subunits each contact two adjacent DNA phosphates; this model requires a 2-bp step size to cross a 4-bp MeP insert ( Figure 4b iv ) . We disfavor models where SpoIIIE subunit simultaneously contact more than two consecutive phosphate groups on the DNA backbone , and models where three or more consecutive subunits contact the DNA backbone because such models would require large motor/DNA distortions . Finally , we also considered a translocation model similar to the mechanisms proposed for the E1 and Rho helicases ( Enemark and Joshua-Tor , 2006; Thomsen and Berger , 2009 ) , ( v ) at any time five subunits contact five consecutive DNA phosphates; this model requires a 1-bp step size to cross a 4-bp insert ( Figure 4b v ) . Although this E1/Rho-like model requires significant motor/DNA distortions ( 5 SpoIIIE subunits span an arc of 300° while five consecutive phosphates in dsDNA span an arc of ∼170° ) , it is in principle consistent with the data from Figure 2 . While we cannot rule out more complex models that may involve non-consecutive subunits simultaneously contacting the DNA , or stochastic bursts consisting of multiple , rapid consecutive steps ( Cordova et al . , 2014; Sen et al . , 2013 ) , we favor the more parsimonious models presented here . To distinguish among the models proposed above and to determine SpoIIIE’s step size , we challenged the motor with inserts containing a 4-bp MeP segment ( 'register' ) , followed by 1 bp of regular DNA ( 'stepping-stone' ) that is in turn followed by a variable-length MeP segment ( 'probe' ) of 1–4 bp ( Figure 5a ) . Because SpoIIIE easily crosses up to 4 bp of neutral DNA , the 4-bp register is meant to align the stepping phase of SpoIIIE with the stepping-stone segment . Motors that are not in perfect register with the stepping-stone should not establish electrostatic contacts with DNA and will either slip backward or stall . The motors that successfully traverse the entire insert should land on the stepping-stone in a 'front-foot-first' configuration—wherein only the leading motor-DNA contact is securely established—and should be able to cross a probe segment of length equal to the step size minus 1 bp ( Figure 5b ) . We found that SpoIIIE traversed a 1-bp neutral probe with near-100% probability ( Figure 5c orange traces ) , inconsistent with the 5 subunit contact model ( Figure 5b v ) . The crossing probability sharply drops to ∼25% for 2-bp , 3-bp , or 4-bp neutral probes ( Figure 5c , magenta traces and Figure 5d ) . As seen in Figure 5b , only model iv can easily cross a 1-bp probe but not probes of 2 bp or longer . In this model , two consecutive SpoIIIE subunits each contact two adjacent phosphates , and DNA is translocated in 2-bp steps ( Figure 5b iv ) . A 2-bp step size is in good agreement with both enzymatic and structural studies of the related dsDNA translocase FtsK , which translocates 1 . 6–2 . 1 bp per ATP hydrolyzed ( Graham et al . , 2010; Massey et al . , 2006 ) . 10 . 7554/eLife . 09224 . 009Figure 5 . Deducing SpoIIIE’s step size using 'stepping-stone' MeP constructs . ( a ) The design of 'stepping-stone' constructs . Each insert consists of a register segment with 4 bp of neutral DNA , followed by one regular DNA base ( 'stepping-stone' ) , and a probe segment with x bp of neutral DNA . For clarity , only the DNA strand tracked by SpoIIIE is shown . ( b ) Diagrams illustrating the longest probe segment that can be traversed by the models depicted in Figure 4b . Model ( i ) should traverse a probe segment of at most 4 bp , whereas model ( v ) cannot traverse even a probe of 1 bp . ( c ) Sample traces of SpoIIIE translocating on DNA with 'stepping-stone' inserts containing a 1-bp probe ( orange ) or a 2-bp probe ( magenta ) . ( d ) Traversal probability for stepping-stone constructs with various probe lengths . Error-bars show the 68% CI estimated via bootstrapping . p values were calculated using a two-tailed Fisher exact test . ( e ) Diagram illustrating how model ( iv ) can successfully traverse a 'stepping-stone' insert with a probe of 1 bp . The star marks the subunit executing the power-stroke . Once a subunit fires , it cannot fire again ( gray shading ) and must eventually disengage from the DNA ( subunit shown as making no interactions with DNA ) . ( f ) Diagram showing why model ( iv ) fails to cross a 'stepping-stone' insert with a probe of 2 bp . DOI: http://dx . doi . org/10 . 7554/eLife . 09224 . 00910 . 7554/eLife . 09224 . 010Figure 5—figure supplement 1 . Crossing predictions of the E1/Rho-like translocation models . Diagram illustrating two different scenarios of how the E1/Rho-like translocation model would cope with a MeP stepping stone insert . In this model subunits fire in a well-defined sequential order ( firing subunit denoted by the yellow star ) and move by 1 bp on DNA , whereas the remaining subunits escort the DNA during the translocation process . If the ATPase subunits have to fire in a strictly sequential order , a subunit cannot fire multiple consecutive times , or fire out of order , and is grayed out after firing . A motor operating via a strictly sequential E1/Rho-like translocation mechanism is not capable of crossing a MeP stepping stone insert with a 1-bp probe ( frame 8 ) , contrary to our observations ( Figure 5c–d ) . If the same ATPase subunit could fire out of order several times in a row , the motor should traverse a MeP stepping stone insert with a 4-bp probe ( inset ) , again inconsistent with our findings ( Figure 5c–d ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09224 . 010 Figure 5e illustrates how model iv enables SpoIIIE to cross the insert with a 1-bp probe . The first three frames illustrate the sequential firing of subunits A , B , and C , all of which maintain at least one anchoring contact with the phosphate backbone . Because subunit D cannot make electrostatic contacts with the backbone , it fails to propel the DNA upon firing ( Figure 5e , frame 5 ) . As a result , no other subunit can establish new contacts with the DNA backbone . After subunit E translocates the DNA by 2 bp , subunit F can now latch onto a negatively charged phosphate and continue to translocate DNA . Figure 5f illustrates how this model copes with a 2-bp MeP probe: subunit F cannot anchor itself onto the DNA , causing the motor to pause and eventually slip ( as shown in Figure 5c , magenta traces ) . This two-subunit DNA escort model requires one subunit to execute the power stroke while an adjacent subunit maintains its phosphate contacts during this power-stroke , escorting the DNA through the ring . In this model , the ATPase subunits translocate the DNA sequentially around the ring in a highly coordinated fashion that enables SpoIIIE to track one DNA strand . There is increasing structural evidence for this type of motor mechanism involving 'translocating' subunits and 'escorting' subunits; examples include the E . coli’s Rho helicase ( Thomsen and Berger , 2009 ) and the papillomavirus E1 helicase ( Enemark and Joshua-Tor , 2006 ) . However , unlike the E1/Rho mechanisms , which employ four escorting subunits , the SpoIIIE translocation model proposed here requires only one escorting subunit . To quantify the strength of the motor–DNA interaction , we measured the force at which SpoIIIE loses its grip on DNA . In a buffer lacking nucleotides , SpoIIIE could bind to DNA and form tethers between the trapped bead and the micropipette-held bead . Manual pulling experiments revealed that these tethers rupture at ∼3 pN ( Figure 7—figure supplement 1 , apo ) , suggesting that apo-SpoIIIE does not interact strongly with DNA . In buffers containing only ADP or ATPγS , SpoIIIE could bind to DNA , forming tethers that rupture at ∼15 and ∼25 pN , respectively ( Figure 7—figure supplement 1 ) . These results indicate that nucleotides stabilize motor–DNA interactions and suggest that only nucleotide-bound subunits are capable of forming stable electrostatic contacts with the DNA phosphate backbone . To further investigate how individual SpoIIIE subunits coordinate their ATP hydrolysis activity , we monitored DNA translocation in a saturating [ATP] buffer that contained ATPγS – a nucleotide analog that is hydrolyzed very slowly . Structural studies of FtsK indicate that ATPγS binds to the same catalytic pocket as ATP ( Massey et al . , 2006 ) , suggesting that ATPγS should act as a competitive inhibitor to ATP binding for the SpoIIIE/FtsK family of ring ATPases . We observe that ATPγS induces SpoIIIE pausing ( Figure 6a ) and reduces the pause-free translocation velocity ( Figure 6—figure supplement 1a ) . This reduction in pause-free velocity is well described by a simple competitive inhibition mechanism ( Figure 6—figure supplement 1b ) . Similar results were obtained with a non-hydrolyzable ATP analog–AMP-PNP ( Figure 6—figure supplement 1c–d ) . Taken together , these results indicate that nucleotide analogs induce SpoIIIE pausing by binding to a subset of ring subunits and arresting them in a pre-hydrolysis state . 10 . 7554/eLife . 09224 . 011Figure 6 . SpoIIIE pauses when two consecutive subunits are bound to ATPγS . ( a ) Sample SpoIIIE traces acquired at different [ATPγS] and 3mM ATP . Pauses are highlighted in red . ( b ) Mean duration of the ATPγS-induced pauses extracted from fitting the pause duration distribution to an exponential . Error-bars represent the 95% CI of the fit . Insert: histogram of pause durations at 750 μM ATPγS and 3 mM ATP and the exponential fit to this distribution . ( c ) The density of ATPγS-induced pauses at various [ATPγS] and 3mM ATP . Error-bars represent the SEM . ( d ) Pause density versus [ATPγS] corrected to account for missed pauses ( black symbols ) . Error-bars represent the SEM . The shaded regions illustrate the predictions for a sequential hydrolysis model where SpoIIIE pauses when n consecutive subunits are bound to ATPγS . DOI: http://dx . doi . org/10 . 7554/eLife . 09224 . 01110 . 7554/eLife . 09224 . 012Figure 6—figure supplement 1 . SpoIIIE activity inhibition with ATP analogs . ( a ) Pause-free velocity versus opposing force for various [ATPγS] and 3mM ATP . Error-bars represent the SEM . ( b ) Degree of pause-free velocity inhibition i = 1-v ( [ATPγS] ) /vmax versus [ATPγS] at 3 mM [ATP] and 5 pN of opposing force . The dashed curve represents the fit to the competitive inhibition model ( inset ) with Ki = 124 ± 20 µM . ( c ) Pause-free velocity versus opposing force at 3 mM [ATP] in the absence of any nucleotide analogs ( black ) , in the presence 500 µM [ATPγS] ( green ) , or in the presence of 500 µM [AMP-PNP] ( pink ) . Error bars represent the SEM . ( d ) Pause density measured at 3 mM [ATP] and various [ATPγS] ( blue symbols ) . Experiments were also performed with 3 mM [ATP] and 0 . 5 mM [AMP-PNP] ( red symbol ) . Error bars represent the SEM . ( e ) Pause density versus opposing force at 3 mM [ATP] and 1 mM [ATPγS] . ( f ) Mean lifetime of ATPγS-induced pauses calculated from single-exponential fits to the pause duration distribution at 3 mM [ATP] and 1 mM [ATPγS] . Error bars represent the 95% CI of the fits . DOI: http://dx . doi . org/10 . 7554/eLife . 09224 . 012 To determine how many analog-bound subunits are required to arrest DNA translocation , we measured the density and duration of ATPγS-induced pauses as a function of [ATPγS] . The density and duration of analog-induced pauses does not depend on external force ( Figure 6—figure supplement 1e–f ) , consequently pauses detected at all forces were pooled together . Given the spatio-temporal resolution of our experiments , we could reliably detect pauses longer than 30 ms at 125 –250 μM ATPγS , 50 ms at 375–500 μM ATPγS , and 75 ms at 750–1000 μM ATPγS . However , regardless of the shortest detectable pause duration and the ATPγS concentration , the distribution of measured pause durations was well described by a single exponential decay ( Figure 6b , inset ) with the same characteristic life-time of ∼35 ms ( Figure 6b ) . The fact that the characteristic life-time does not depend on ATPγS concentration suggests that all analog-induced pauses are drawn from the same distribution . Furthermore , the single-exponential dependence of the pause duration distribution indicates that the exit from the paused state is governed by a single rate-limiting event—presumably the exchange of an ATPγS molecule with one ATP which is present in saturating concentrations , as proposed for other ring ATPases ( Chistol et al . , 2012; Sen et al . , 2013 ) . Despite the fact that we cannot detect analog-induced pauses shorter than a certain cutoff ( 30–75 ms ) , we can estimate the number of missed pauses—and therefore the true pause density—from the measured pause density ( Figure 6c ) and the characteristic life-time derived from the pause distribution ( Figure 6b ) , assuming that the single-exponential distribution holds for pauses shorter than the cutoff ( Hodges et al . , 2009 ) ( see 'Materials and methods' ) . The MeP experiments support a sequential nucleotide hydrolysis model . Therefore , to explain the density of ATPγS-induced pauses , we considered a sequentially coordinated hexameric ATPase ring that pauses whenever n consecutive subunits are bound to ATPγS ( 1 ≤ n ≤6 ) . The pause density ( PD ) can be analytically expressed in terms of the motor’s pause-free velocity ( vpf ) , the ATPγS dissociation rate ( koff ) , and the probability that the motor is paused ( Ppause ) —which is proportional to the concentrations and dissociation constants of ATP and ATPγS ( Materials and methods ) , as follows: PD = koffvpf·Ppause ( 1- PPause ) Figure 6d shows the expected pause density predicted for different n values . For example , n=1 ( green shaded region ) corresponds to a sequentially coordinated ring that pauses whenever a single subunit is bound to ATPγS; n=2 ( blue shaded region ) corresponds to a sequentially coordinated ring that pauses when two consecutive subunits are both bound to ATPγS , and so on . The ATPγS pause-density data corrected to account for missed pauses ( Figure 6d black symbols ) are best described by the model in which two ( n=2 ) consecutive analog-bound subunits are required to induce a pause in the motor ( Figure 6d , blue-shaded region ) . In other words , analysis of the pause density indicates that the motor can operate processively with non-consecutive inactive subunits but not with two or more consecutive subunits . This conclusion is further supported by the observation that FtsK hexamers readily bypass individual catalytically inactive ATPase subunits ( Crozat et al . , 2010 ) . We have shown that SpoIIIE translocates DNA by making crucial anchoring contacts with the negatively charged phosphate groups on one DNA strand . Although electrostatic interactions with the backbone of the 5’→3’ strand in the direction of translocation are the principal mode of motor-substrate contact , we surmise that other types of interactions ( e . g . steric ) play a secondary role in maintaining SpoIIIE’s grip on DNA . This inference explains why we observe small but non-zero traversal probabilities for neutral inserts of 10–30 bp , much longer than the expected motor step size ( Figure 2b ) . Interestingly , SpoIIIE is more likely to traverse 30-bp inserts with a neutral backbone only on the 5’→3’ strand ( 6 out of 28 molecules ) than 30-bp inserts with a neutral backbone on both strands ( 1 out of 32 molecules ) ( Table 1 ) . To explain this discrepancy , we speculate that during translocation , in addition to the essential electrostatic interactions with the 5’→3’ strand , the SpoIIIE ring may also interact very weakly with charges on the 3'→5' strand . These secondary interactions could be mediated by parts of the motor other than the pore loops which execute the power – stroke . Prior experiments on MeP DNA have established that certain patterns of neutral and charged bases can introduce large DNA distortions . An asymmetric neutralization of the phosphate backbone whereby one 'face' of the dsDNA helix had its charges removed ( e . g . neutralizing bases 1-3 on one strand and bases 3-6 on the other strand ) results in DNA bending toward the direction of the neutralized region ( Strauss and Maher , 1994 ) . However , a symmetric neutralization of the phosphate backbone , whereby the phosphates on both strands were uniformly neutralized , resulted in no significant DNA distortions ( Strauss and Maher , 1994 ) . We do not expect the MeP DNA constructs used in this study to introduce significant distortions on the double helix for two reasons: ( 1 ) Magnesium ions were present in the experimental buffer at a sufficiently high concentration ( 10 mM ) , which had been demonstrated to mitigate the effect of DNA distortions due to asymmetric phosphate neutralization ( Strauss and Maher , 1994 ) . ( 2 ) In designing the MeP inserts used in this study , care was taken to avoid an asymmetric neutralization of the phosphate backbone . The majority of the experiments with MeP inserts tested in this study utilized dsMeP modifications . Thus , the backbone neutralization was symmetrically distributed and therefore not expected to cause significant DNA distortions . We tested only two single-stranded MeP/DNA hybrid constructs , with 30 consecutive neutralized bases on either strand , covering nearly three full helical turns of the DNA . Because the backbone neutralization was distributed evenly in all directions around the DNA helix , the charge neutralization should be symmetrical , and therefore , it is not expected to introduce significant DNA distortions . To rationalize the results of the MeP experiments , we propose a minimal translocation model in which SpoIIIE subunits fire sequentially around the ring , each subunit translocates 2 bp of DNA per power-stroke , at least two consecutive ATPase subunits contact the backbone of one DNA strand , and each subunit contacts two consecutive backbone phosphates . To be consistent with the results of the stepping-stone insert experiments ( Figure 5c–d ) , our two-subunit translocation-escort model requires that subunits fire in a well- defined sequential fashion ( A , B , C , D , E , F , A , B , C etc ) and a subunit cannot fire multiple times in a row or out of order . The structures of homo-hexameric helicases E1 and Rho co-crystallized with their single-stranded nucleic acid substrates strongly support a translocation mechanism where five motor subunits contact five consecutive phosphates on the ssDNA/ssRNA backbone ( Enemark and Joshua-Tor , 2006; Thomsen and Berger , 2009 ) , and the hydrolysis of one ATP is coupled to the translocation of one nucleotide . Although an E1/Rho-like mechanism could potentially rationalize how SpoIIIE easily traverses a dsMeP insert of 4 bp but not 5 bp ( Figure 2 ) , such a model predicts that the motor would fail to traverse a stepping-stone construct with a 1-bp MeP probe ( Figure 5b v ) . Even if the E1/Rho-like model did allow a subunit to fire several times in a row , such a model would still be inconsistent with the stepping stone data ( Figure 5—figure supplement 1 ) . It was previously reported that SpoIIIE supercoils plasmid DNA in vitro , and strand/groove-tracking was proposed to explain this observation ( Bath et al . , 2000 ) . Magnetic tweezers measurements of DNA supercoiling during translocation ruled out a groove-tracking mechanism for FtsK ( Saleh et al . , 2005 ) . Here , we show that SpoIIIE does indeed track one DNA strand by making specific electrostatic contacts with backbone phosphates on the 5'→3' strand in the direction of translocation , a feature reported for other RecA-like NTPases: the φ29 packaging motor , DnaB , and Rho ( Aathavan et al . , 2009; Itsathitphaisarn et al . , 2012; Thomsen and Berger , 2009 ) . Unlike the φ29 ATPase , which translocates dsDNA , DnaB , and Rho are single-stranded nucleic acid translocases , and therefore track one strand by default . Interestingly , both SpoIIIE and the φ29 ATPase encircle dsDNA , positioning multiple subunits in close proximity to the dsDNA helix , while they still possess the ability to discriminate one strand from the other . How these dsDNA translocases achieve strand discrimination remains unclear . Several studies have demonstrated that certain skewed sequences ( i . e . sequences whose presence is biased on the leading vs . the lagging strand ) impart directionality to FtsK/SpoIIIE translocation ( Besprozvannaya et al . , 2013; Cattoni et al . , 2013; Lee et al . , 2012; Levy et al . , 2005; Löwe et al . , 2008 ) . It is tempting to imagine that the strand tracking mechanism presented here could be used by SpoIIIE to read these skewed sequences ( i . e . the SpoIIIE Recognition Sequences , SRS ) . However , the γ domains of SpoIIIE have been shown to be required for reading the SRS ( Ptacin et al . , 2008 ) and co-crystal structures of dsDNA with the domain γ reveals it interacting with both DNA strands at the backbone , the major and minor grooves , and individual bases ( Löwe et al . , 2008 ) . We consider it unlikely therefore , that SpoIIIE enlists the strand-tracking mechanism described here for SRS sequence recognition . The electrostatic interactions between SpoIIIE and the DNA backbone are likely to have two main purposes in vivo: ( i ) They serve a load-bearing function by providing SpoIIIE with the grip necessary to strip-bound proteins off the DNA at the high speeds at which the motor operates ( Marquis et al . , 2008 ) ; and ( ii ) They enable the motor to supercoil the DNA during translocation . The later arises from the symmetry mismatch between the angles spanned by two adjacent SpoIIIE subunits and the angle that separates the phosphates contacted by the motor during a 2-bp step . Figure 7a depicts the SpoIIIE-DNA complex as seen from the N terminus of the motor; in this view , SpoIIIE translocates DNA toward the viewer ( Massey et al . , 2006 ) . In the two-subunit DNA escort model , the DNA is handed over from one subunit to the next and , for each 2-bp step , the motor-DNA contacts rotate clockwise around the ring by 60° . However , since B-form DNA has a helical pitch of 10 . 4 bp/turn ( Wang , 1979 ) , the dsDNA backbone rotates clockwise by 69° for every 2 bp . Therefore , after each motor step , a counter-clockwise 9° rotation of the DNA relative to the motor is required to align the phosphate backbone with the next translocating subunit . As a result , the two-subunit escort model predicts that SpoIIIE should introduce one supercoil for every ∼80 bp of DNA translocated . The magnitude and direction of this prediction is consistent with both in vivo measurements ( Nicholson and Setlow , 1990 ) and in vitro single-molecule experiments . Indeed , magnetic tweezers experiments revealed that SpoIIIE supercoils DNA by one turn per 100 ± 10 bp of DNA translocated ( Jerod Ptacin and Marcelo Nollman , personal communication ) . It was previously reported that strand tracking by the φ29 ATPase , another ring-shaped dsDNA motor , also leads to DNA supercoiling ( Liu et al . , 2014 ) . The results presented here for SpoIIIE suggest that DNA supercoiling may be a general feature of ring-shaped dsDNA motors that track one strand . The amount and direction of supercoiling should depend on the ( i ) the helical pitch of dsDNA , ( ii ) the periodicity of motor-DNA contacts ( step size ) , and ( iii ) the motor’s inter-subunit coordination mechanism . 10 . 7554/eLife . 09224 . 013Figure 7 . Two-subunit DNA escort model . ( a ) Diagram illustrating how SpoIIIE supercoils DNA while tracking the phosphate backbone of one DNA strand . Neighboring subunits in the SpoIIIE hexamer are spaced by 60°; consecutive DNA phosphates are spaced by ∼34°; and the backbone contacts of two adjacent subunits are separated by ∼69° . After a 2-bp translocation step , a 9-degree counter-clockwise rotation of the DNA relative to the motor is needed to align the next translocating subunit and the nearest pair of backbone phosphates . ( b ) Diagram of the SpoIIIE hexamer illustrating how the two-subunit DNA escort model enables the motor to bypass an ATPγS-bound subunit ( red ) . For simplicity , only one DNA strand is shown ( orange ) . DNA is translocated out of the page , and each subunit interacts with two neighboring phosphates . The star marks the subunit slated to execute the power-stroke . DOI: http://dx . doi . org/10 . 7554/eLife . 09224 . 01310 . 7554/eLife . 09224 . 014Figure 7—figure supplement 1 . Assessing the strength of the SpoIIIE–DNA interaction in the presence of different nucleotides . To probe the strength of the motor-DNA interaction in the presence of different nucleotides , we pulled on single SpoIIIE-DNA complexes in a nucleotide-free buffer ( Apo state , green ) , 1mM [ATPγS] ( cyan ) , and 1mM [ADP] ( orange ) and measured the mean pull force , that is , the average force at which the tether ruptured . Error bars represent the SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 09224 . 014 In vivo measurements indicate that , during B . subtilis sporulation , the DNA is negatively supercoiled by 1 turn per 97 ± 6 bp in the forespore and 1 turn per 140 ± 10 bp in the mother cell ( Nicholson and Setlow , 1990 ) . Since SpoIIIE is anchored at the septum during sporulation and the B . subtilis DNA is circular , the two-subunit escort model predicts that SpoIIIE translocation should introduce one negative supercoil in the forespore and one positive supercoil in the mother cell for every ∼80 bp translocated . We surmise that the SpoIIIE mechanism is fine tuned to deliver DNA to the forespore in the appropriate negative supercoiled state , similar to what has been proposed for FtsK ( Saleh et al . , 2005 ) . This mechanism may help to conserve cellular resources by minimizing the amount of maintenance performed by topoisomerases/gyrases , a strategy that may provide a significant advantage in the harsh conditions that prompt sporulation ( i . e . , starvation ) . The results of ATPγS experiments suggest that SpoIIIE can processively translocate DNA with non-consecutive inactive subunits . This conclusion can be rationalized by the two-subunit DNA escort model proposed above: having two adjacent ATPase subunits simultaneously contact the DNA enables the motor to continue translocating if either of the two subunits is inactive , but not when both are disabled . Figure 7b illustrates how the two-subunit DNA escort model can bypass the ATPγS-bound subunit B: ( i ) subunit A fires and translocates DNA while subunit B escorts the DNA , ( ii ) the analog-bound subunit B fails to fire while subunit C was poised to escort the DNA ( iii ) subunit C then fires and translocates the DNA , handing it over to subunit D . During this step , subunit B , which remains bound to ATPγS , escorts the DNA for subunit C . Our conclusions agree with the results of a single-molecule study by Crozat et al . , which found that FtsK hexamers with two diametrically opposed inactive subunits translocated DNA as fast as wildtype hexamers ( Crozat et al . , 2010 ) . The authors of that study proposed a sequential DNA escort mechanism in which at least three motor subunits contact the DNA at any given time . The ATPγS experiments presented here provide additional evidence that SpoIIIE/FtsK motors can bypass individual inactive subunits . Moreover , the strand-tracking behavior reported here for SpoIIIE strongly favors a model in which ring subunits fire sequentially , a key aspect of the DNA escort mechanism proposed here and in the Crozat et al study . This study presents evidence for a type of inter-subunit coordination in ASCE ring NTPases where subunit firing is highly coordinated around the ring , yet the motor possesses sufficient flexibility to bypass non-consecutive inactive subunits . How is this mechanism optimized for the specific biological task of SpoIIIE ? In vivo , SpoIIIE is present in low copy numbers during sporulation , with only two motors responsible for the vital task of chromosome translocation at any given time ( Burton et al . , 2007; Yen Shin et al . , 2015 ) . As a result , each SpoIIIE ring functions as a single-molecule bottleneck for this process , where the failure of either motor is likely to be lethal for the cell . The coordination mechanism proposed here can potentially explain how B . subtilis safeguards against the failure of individual motor subunits during sporulation . We speculate that other ASCE motors that also represent single-points of failure may have evolved into similar flexible operations . Biotinylated SpoIIIE constructs were generated by ligating the biotin tag sequence from plasmid Pinpoint Xa-1 ( Promega , Madison , WI ) to the N terminus of SpoIIIE from plasmid pJB103 ( Bath et al . , 2000 ) . Protein purification was conducted as described previously with the addition of 2 µM biotin in the liquid cultures ( Ptacin et al . , 2008 ) . DNA tethers were generated using a 5' biotinylated primer ( IDT ) to PCR amplify a 21 kb region of lambda phage DNA ( NEB ) and gel extracted . DNA oligos containing MeP inserts ( Gene Link or TriLink BioTechnologies ) were ligated to gel-purified DNA fragments: a 9167-bp fragment with the ACT 3'-overhang ( amplified from λ DNA and digested with AlwNI ) , and a biotinylated 3976-bp fragment with the GAA 3'-overhang ( amplified from φ29 DNA and digested with BglI ) . The final ligation product was gel-purified using the QIAEX II kit ( Qiagen ) . In this study , 2 . 1 µm streptavidin beads ( Spherotech ) were blocked for 30 min in 50 mM Tris–HCl pH 7 . 5 , 10 mM MgCl2 , 4% bovine serum albumin ( BSA ) , w/v and 0 . 1% Tween-20 . Then ∼1 pmol of biotinylated SpoIIIE and ∼1 ng of biotinylated DNA were incubated separately onto streptavidin beads and spatially separated in the fluidics chamber . DNA-bound beads and SpoIIIE-bound beads were brought in close proximity to allow SpoIIIE to engage the DNA . DNA translocation was conducted in a reaction buffer containing 50 mM Tris–HCl pH 7 . 5 , 10 mM MgCl2 , and 3 mM ATP . Pauses were detected using a modified Schwartz Information Criterion ( mSIC ) method ( Chistol et al . , 2012 ) . The number and duration of pauses missed by this algorithm were inferred by fitting the pause durations to a single exponential with a maximum likelihood estimator . After removing the detected pauses , the translocation velocity was computed by fitting the data to a straight line . Force-velocity data ( Figure 6—figure supplement 1a ) was gathered in 'passive-mode' ( Smith et al . , 2001 ) , where the optical trap position is fixed . Single-molecule trajectories were partitioned into segments spanning 2–3 pN , and the velocity was computed for each segment . Tether tension and extension were converted to contour length using the Worm-Like-Chain approximation with persistence length p=30 nm , and stretch modulus S = 1200 pN·nm ( Baumann et al . , 1997 ) . The pause-detection algorithm can detect only pauses longer than tc , but it is possible to account for the duration and number of pauses missed by the algorithm . We observe that pauses are drawn from a single-exponential distribution P ( t ) with a mean pause-duration t . ( 1 ) P ( t ) =A·e−tτ The number of all pauses ( Nall ) , the number of detected pauses ( Ndet ) , and the number of missed pauses ( Nmiss ) are given below . ( 2 ) Nall=∫0∞ P ( t ) · dt=∫0∞A· e−tτ· dt ( 3 ) Ndet=∫tc∞ A· e−tτ· dt ( 4 ) Nmiss=∫0tc A· e−tτ· dt The duration of all pauses ( Tall ) ( 5 ) Tall=∫0∞ t· P ( t ) · dt=∫0∞ t· A· e−tτ· dt The duration of detected pauses is Tdet: ( 6 ) Tdet=∫tc∞t· A· e−tτ· dt The duration of missed pauses: ( 7 ) Tmiss=∫0tct· A· e−tτ· dt Given the number of detected pauses and the mean pause duration calculated by fitting the pause distribution to a single-exponential , it is possible to infer the number of total pauses and the number of missed pauses as follows: ( 8 ) Nall=Ndet· etcτ ( 9 ) Nmiss=Ndet· ( etcτ−1 ) If the pause-detection algorithm identifies and removes only pauses longer than tc , then the measured pause-free velocity ( Vmeas ) is given by the following expression: ( 10 ) Vmeas = DTpf Tmiss Here D is the total distance translocated by the motor , Tpf is the 'pause-free' time ( i . e . only the time spent translocating DNA ) , and Tmiss is the total duration of missed pauses . The true pause-free velocity ( Vpf ) is given by the following expression: ( 11 ) Vpf=DTpf=DDVmeas−∫0tct·A·e−tτ·dt The probability of finding the ring in a pause state Ppause ( i . e . the fraction of the time spent in the paused state ) is: ( 12 ) Ppause=τpτp τx where tp is the total time the motor spends in a pause state and tx is the total time the motor spends translocating DNA . The total pause time tp and the total translocation time tx can be expressed as ( 13 ) τp=<T>·η= ( 1/koff ) ·η ( 14 ) τx = x ( t ) vPF where <T> is the average pause duration , η is the number of pauses , koff is a first-order dissociation rate of ATPγS from the ring , x ( t ) is distance translocated over time and vPF is the pause-free velocity expressed as: ( 15 ) vPF = ∑i = 06pivi where pi is the probability of the ring being bound to i ATPγS molecules and vi is the pause-free velocity of the ring bound to i ATPγS molecules . Substituting Equation 14 into Equation 12 , we can express the pausing probability as: ( 16 ) Ppause=η/x ( t ) η/x ( t ) + koff/vPF=PDPD + koff/vPF We define the pause density ( the number of pauses per distance translocated ) as PD =η / x ( t ) . Thus , the PD can be expressed in terms of the pausing probability ( Ppause ) , the ATPγS dissociation rate ( koff ) , and the pause-free translocation velocity ( vPF ) : ( 17 ) PD=koffvPF·Ppause ( 1−Ppause ) The pausing probability ( Ppause ) can be expressed in terms of the probability that a single subunit is bound to ATPγS ( p ) . p depends on the concentrations of ATP and ATPγS , as well as the dissociation constants ( koff/kon ) of ATP and ATPγS ( KATP and KγS ) for individual ATPase subunits ( Sen et al . , 2013 ) : ( 18 ) p=KATP[ATPγS]KγSKATP KγS[ATP] + KATP[ATPγS] To calculate the expected pause probability , we used the measured Ki of ATPγS ( 124 ± 20 µM ) for KγS ( Figure 6—figure supplement 1b ) . For KATP , we used the measured Km of ATP ( 505 ± 50 µM ) as an upper-bound estimate . For a sequential ordinal ATP binding/hydrolysis inter-subunit coordination model ( i . e . subunit 1 binds ATP , hydrolyzes ATP , releases Pi and ADP , and executes the power-stroke , followed by subunit 2 , then subunit 3 , etc ) , the pausing probability is given by Ppause = pn , were n is the number of ATPγS molecules required to induce a pause .	Bacillus subtilis is a bacterium that lives in the soil . When food is in short supply , B . subtilis stops reproducing and individual bacterial cells transform into spores that lay dormant until conditions improve . While , B subtilis is generally harmless , it forms spores in a similar way to other bacteria that cause diseases such as anthrax . During spore formation , a membrane forms to divide the cell into a large mother cell and a smaller “forespore” cell . Then , a copy of the mother cell’s DNA – which is made of building blocks called bases – moves into the forespore . A group of proteins called SpoIIIE is instrumental in this process as it uses energy from a molecule called ATP to pump the DNA across the membrane at the rapid speed of 5 , 000 base pairs of DNA per second . SpoIIIE contains six individual protein subunits that form a ring-shaped motor structure that spans the membrane . It belongs to a large family of proteins that are found in all living organisms and drive many vital processes . How does SpoIIIE interact with DNA and how do the individual subunits coordinate their behaviour ? Liu , Chistol et al . address these questions by using instruments called optical tweezers , which use a laser beam to hold and manipulate tiny objects . The experiments show that to move a fragment of DNA across a membrane , SpoIIIE only makes contact with one of the two strands that make up the DNA molecule . The experiments suggest that the DNA is handed over from one SpoIIIE subunit to another in a sequential order . This would allow the DNA to remain bound to SpoIIIE at all times as it passes through the membrane . Next , Liu , Chistol et al . measured how SpoIIIE steps along the DNA and found that each subunit takes a small two base pair step when energy is released from a single molecule of ATP . There is an element of flexibility in the system , because SpoIIIE can still move DNA normally even if some subunits cannot use energy from ATP . This provides a fail-safe mechanism that still allows the cells to form spores in the event that one subunit is disabled . Future work will concentrate in understanding how the subunits communicate around the ring to coordinate their sequential use of ATP and their DNA pumping activity .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2015	Two-subunit DNA escort mechanism and inactive subunit bypass in an ultra-fast ring ATPase
A large number of drugs can induce prolongation of cardiac repolarization and life-threatening cardiac arrhythmias . The prediction of this side effect is however challenging as it usually develops in some genetically predisposed individuals with normal cardiac repolarization at baseline . Here , we describe a platform based on a genetically diverse panel of induced pluripotent stem cells ( iPSCs ) that reproduces susceptibility to develop a cardiotoxic drug response . We generated iPSC-derived cardiomyocytes from patients presenting in vivo with extremely low or high changes in cardiac repolarization in response to a pharmacological challenge with sotalol . In vitro , the responses to sotalol were highly variable but strongly correlated to the inter-individual differences observed in vivo . Transcriptomic profiling identified dysregulation of genes ( DLG2 , KCNE4 , PTRF , HTR2C , CAMKV ) involved in downstream regulation of cardiac repolarization machinery as underlying high sensitivity to sotalol . Our findings offer novel insights for the development of iPSC-based screening assays for testing individual drug reactions . A large number of drugs have the undesirable side effect of prolonging cardiac repolarization which can trigger life-threatening cardiac arrhythmias ( Kannankeril et al . , 2010 ) . A typical cause is the inhibition of the inward rectifying potassium channel ( hERG or human ether-a-go-go related gene encoded by KCNH2 ) ( Kannankeril et al . , 2010; Roden , 2008a ) , which results in the prolongation of the repolarization time and induces an increase of the QT interval on the electrocardiogram . The early prediction of drug-induced long QT ( diLQT ) is a major requirement to avoid exposing patients to cardiac arrhythmias and ultimately sudden cardiac death ( Sarganas et al . , 2014 ) . Eventually , diLQT is a major cause of drug withdrawal or of premature termination during development ( Fermini and Fossa , 2003 ) . Regulatory agencies require that new drug candidates be systematically screened for hERG inhibition in animal models or in heterologous over-expression cellular systems ( Sager et al . , 2014 ) . Because of species-dependent differences of cardiac electrophysiology and the failure of heterologous systems to express all cardiac proteins , hERG screening however remains suboptimal and has a poor positive predictive value for subsequent proarrhythmia risk ( Gintant , 2011; Giorgi et al . , 2010; Sager et al . , 2014 ) . The risk of developing diLQT varies markedly between subjects . Previous studies in the general population have shown that a small proportion ( i . e . , less than 10% ) of patients exposed to QT-prolonging drugs actually develop diLQT ( Behr and Roden , 2013; Kannankeril et al . , 2010; Soyka et al . , 1990 ) . Different observations support the existence of a genetic determinism to diLQT including the higher propensity to develop drug-induced repolarization abnormalities in first-degree relatives of patients with diLQT ( Kannankeril et al . , 2005 ) and the similarity to the congenital form of the long QT syndrome ( LQTS ) , a rare inherited cardiac disorder ( Moss and Robinson , 1992 ) . Multiple rare mutations with marked effects in genes encoding ion channels have been reported to underlie LQTS ( Cerrone and Priori , 2011 ) . However , candidate gene screening studies in large cohorts of patients with diLQT have reported a small prevalence ( from 10 to 15% ) of these mutations in the known LQTS genes ( Paulussen et al . , 2004; Yang et al . , 2002 ) . QT prolongation can be induced by mutations in many genes that occupy a substantial region of the human interactome and contain many drug targets ( Berger et al . , 2010 ) . The factors leading to diLQT in some individual subjects remain largely unclear but are probably driven by the additive effect of common genetic variants that reduce the cardiac repolarization reserve ( Behr and Roden , 2013; Kannankeril et al . , 2005; Roden , 2006 , 2008b ) . As a consequence , diLQT is difficult to predict as it usually develops in individuals with a predisposing genetic makeup which favors exaggerated response to a pharmacological challenge with QT-prolonging drugs ( Kannankeril et al . , 2010; Roden , 2008b ) but does not affect cardiac repolarization features at baseline ( Behr and Roden , 2013; Kannankeril et al . , 2010; Sarganas et al . , 2014 ) . Human induced pluripotent stem cells ( iPSCs ) have recently been proposed to model monogenic disorders including congenital LQTS ( Itzhaki et al . , 2011; Moretti et al . , 2010; Sinnecker et al . , 2013 ) . Cardiomyocytes derived from congenital LQTS patient-specific iPSCs lines typically exhibit significant electrophysiological abnormalities due to the deficient function of some key cardiac ion channels ( Itzhaki et al . , 2011; Moretti et al . , 2010 ) . These cells display increased arrhythmogenicity under native conditions and can be used as a test system for customization of anti-arrhythmic drugs in these patients ( Terrenoire et al . , 2013; Wang et al . , 2014 ) . In addition , QT prolonging drugs can trigger changes in the electrical potential developed by iPSC-derived cardiomyocytes ( iPSC-CMs ) ( Liang et al . , 2013; Navarrete et al . , 2013; Nozaki et al . , 2014 ) . To expand upon these observations , our goal was to determine whether iPSC-CMs from a group of subjects with quantitative measure of diLQT in vivo could recapitulate the phenotype in vitro thereby providing a model system for diLQT . In this study , we show that iPSC-CMs from subjects shown to be susceptible to diLQT in vivo but otherwise normal present no abnormalities in their basal characteristics but higher cardiotoxic responses to drug stimulation as compared to control iPSC-CMs . The contrasted responses to sotalol in iPSC-CMs were strongly correlated to the inter-individual differences in response to sotalol as recorded during clinical investigations . Our findings offer novel insights on the use of a genetically diverse panel of induced pluripotent stem cells to model complex pharmacogenetic traits . We first performed a clinical study ( clinicaltrials . gov NCT01338441 ) where we prospectively evaluated cardiac repolarization of healthy subjects in response to a pharmacological challenge with a single 80 mg oral dose of sotalol . Sotalol is a non-selective competitive β-adrenergic receptor blocker with additional Class III antiarrhythmic properties by its inhibition of potassium channels ( Zanetti , 1993 ) . Sotalol is widely recognized as a classical QT-prolonging drug ( Soyka et al . , 1990 ) . A total of ninety-two subjects were enrolled in the study ( Figure 1A ) . All subjects gave their written informed consent to participate the study . Included subjects were aged from 18 to 40 years ( 30 . 1 ± 6 . 2 years ) , had a body mass index ( BMI ) between 19 and 29 kg/m² ( 23 . 9 ± 2 . 3 kg/m² ) , were from Caucasian origin and had no known significant disease or long-term treatment . Forty-four percent ( n = 40 ) were male . Cardiac repolarization parameters were analyzed through standardized measurements of digital high-resolution ECG ( sampling rate 1000 Hz ) . All subjects had a normal sinus rhythm and no significant conduction or repolarization abnormalities . Basal QT interval ranged from 340 to 458 ms ( 393 . 4 ± 26 . 3 ms ) , with a corresponding heart rate ranging from 47 to 88 bpm ( 63 . 3 ± 8 . 7 bpm ) . QT interval was corrected according to Fridericia's formula ( QTcf=QT/[RR]1/3 ) providing corrected QTcf in the physiological range from 350 to 447 ms ( 399 . 0 ± 18 . 6 ms ) . Three hours after sotalol intake , QTcf had increased by 23 . 4 ± 2 . 4 ms . A large inter-individual variability was however noticed with a change in QTcf ranging from −9 ms to +81 ms and a resulting post-sotalol QTcf ranging from 351 to 489 ms ( 412 . 6 ± 43 . 0 ms ) ( Figure 1B ) . 10 . 7554/eLife . 19406 . 003Figure 1 . QTcf changes following Sotalol administration in healthy volunteers . ( A ) Flow chart of the clinical study . ( B ) Distribution of QTcf duration before ( blue ) and 3 hr after sotalol intake ( red ) . ( C ) Distribution of delta change in QTcf showing the wide inter-individual variability in response to the same pharmacological stimulation . Subjects with the most extreme responses were selected as low sensitive ( low-S ) or high- sensitive ( high-S ) as indicated in green and red respectively . ( D ) Average delta change in QTcf in the two groups of selected subjects . **p<0 . 001 . ( E ) Typical ECG recordings before and after sotalol intake in a high-S subject ( upper panels ) and in a low-S subject ( lower panels ) . There is one figure supplement . DOI: http://dx . doi . org/10 . 7554/eLife . 19406 . 00310 . 7554/eLife . 19406 . 004Figure 1—figure supplement 1 . Overview of the clinical and experimental study and role of each partner . This translational study started with a prospective clinical study performed in a clinical investigation center near Paris , France . The objective was to identify healthy subjects with extreme responses to Sotalol 80 mg . Skin punch biopsies were performed in a total of twenty subjects , 10 with low-sensitivity and 10 with high-sensitivity to Sotalol . The fresh skin biopsies were then transferred at the Ectycell company ( Romainville , France ) in close vicinity of the clinical investigation center . The objective was to culture and bank human dermal fibroblasts and to derive and characterize hiPSC . iPSC clones were then anonymized before any further cardiac experiments were performed . This procedure was agreed by all partners to avoid potential biases in interpreting the recording at the cellular level . iPSCs clones were successfully generated for 17 subjects using retroviral infection and were investigated in this study . For the remaining three subjects ( Patients 4 , 14 and 20 ) , iPSCs clones were secondarily generated using episomal plasmids and were thus not investigated . The 17 iPSCs clones were then transferred to the Cardiovascular Research Center at Icahn School of Medicine , New York , USA . The objective was to differentiate iPSCs into cardiac myocytes and to perform MEA recordings and drug testing . All required authorizations were obtained . DOI: http://dx . doi . org/10 . 7554/eLife . 19406 . 004 Twenty subjects with the most extreme responses to sotalol ( i . e . , 10 subjects with high-sensitivity ( high-S ) and 10 subjects with low-sensitivity ( low-S ) ) underwent skin biopsy ( Figure 1A and C ) . These subjects displayed highly contrasted responses to the same pharmacological stimulation ( average change in QTcf: 48 . 5 ± 7 . 1 ms in high-S subjects vs . 0 . 2 ± 4 . 8 ms in low-S subjects; p<0 . 0002; Figure 1C–E ) while having no significant differences in their basal ECG characteristics ( Table 1 ) . Demographic characteristics were well balanced between groups with the exception of gender , subjects with high-S being mostly female ( 90% , n = 9 ) whereas subjects with low-S were mostly males ( 80% , n = 8 ) ( Table 1 ) . Sotalol plasma level was measured 3 hr after sotalol intake and was not significantly different in subjects with high-S as compared to the other participating subjects ( 753 . 1 ± 210 . 9 vs . 620 . 7 ± 195 . 3 μg/ml , p=0 . 10 ) . 10 . 7554/eLife . 19406 . 005Table 1 . Related to Figure 1 . Demographic , clinical and electrocardiographic baseline characteristics of subjects with low- vs . high- sensitivity to Sotalol . DOI: http://dx . doi . org/10 . 7554/eLife . 19406 . 005Low-sensitivity , n = 10 High- sensitivity , n = 10 p-value Age29 . 9 ± 5 . 9 28 . 3 ± 5 . 9 0 . 53Gender , Male ( % ) 8 ( 80 . 0% ) 1 ( 10% ) 0 . 006Body Mass Index ( kg/m² ) 23 . 6 ± 0 . 9 22 . 9 ± 2 . 7 0 . 40SBP ( mmHg ) 116 . 0 ± 8 . 6 113 . 7 ± 8 . 3 0 . 50DBP ( mmHg ) 72 . 2 ± 4 . 8 68 . 0 ± 4 . 2 0 . 07Resting heart Rate ( bpm ) 59 . 6 ± 5 . 9 63 . 2 ± 8 . 4 0 . 37PR interval ( ms ) 166 . 7 ± 24 . 7 143 . 5 ± 18 . 1 0 . 09QRS ( ms ) 88 . 3 ± 5 . 9 85 . 6 ± 7 . 7 0 . 08QTcf ( ms ) 384 . 7 ± 26 . 4 402 . 6 ± 20 . 8 0 . 13 Dermal fibroblasts were collected from the twenty subjects with high or low-sensitivity to sotalol stimulation . Human iPSCs were derived through retroviral infection of dermal fibroblasts with the reprogramming factors OCT4 , SOX2 , c-MYC and KLF4 with a successful generation for 17 out of the 20 subjects ( Figure 1—figure supplement 1 ) . All iPSC clones expressed the characteristic human embryonic stem cell ( ESC ) pluripotency markers TRA-1–60 , NANOG , OCT3/4 and SSEA-4 ( Figure 2—source data 1 ) , had positive alkaline phosphatase staining . Sixteen iPSC clones had a normal karyotype ( Table 2 ) . One clone presented a significant translocation that was observed in originating fibroblasts . All clones showed silencing of the exogenous retroviral transgenes and reactivation of the endogenous pluripotency genes OCT3/4 , NANOG and SOX2 ( Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 19406 . 006Table 2 . Related to Figure 2 . Fibroblasts and iPS quality control parameters . DOI: http://dx . doi . org/10 . 7554/eLife . 19406 . 006ID Mycoplasma HIV* , HBV* , HCV* , HTLV1 and 2* Phosphatase alkaline staining Karyotyping P11007~5924~iPSpolyRoksmANegativeNegativePositive46 , XYP11008~5444~iPSpolyRoksmBNegativeNegativePositive46 , XXP11009~6426~iPSpolyRoksmCNegativeNegativePositive46 , XYP11013~5744~iPSpolyRoksmDNegativeNegativePositive46 , XXP11014-5864-iPSpolyRoksmCNegativeNegativePositive46 , XXP11015~6345~iPSpolyRoksmENegativeNegativePositive46 , XXP11018~5644~iPSpolyRoksmBNegativeNegativePositive46 , XXP11019-6444-iPSpolyRoksmBNegativeNegativePositive46 , XYP11020 ~ 7125 ~ iPSpolyRoksmDNegativeNegativePositive46 , XYP11021 ~ 6544 ~ iPSpolyRoksmCNegativeNegativePositive46 , XXP11023~5525~iPSpolyRoksmANegativeNegativePositive46 , XXP11024~5844~iPSpolyRoksmANegativeNegativePositive46 , XXP11026~6504~iPSpolyRoksmDNegativeNegativePositive46 , XYP11028~6904~iPSpolyRoksmJNegativeNegativePositive46 , XYP11029-6284-iPSpolyRoksmBNegativeNegativePositive46 , XXP11030~5684~iPSpolyRoksmANegativeNegativePositive46 , XX , t ( 1;16 ) †P11031~5204~iPSpolyRoksmCNegativeNegativePositive46 , XYTested in originating fibroblast cell lines . †As found in originating fibroblasts . The generated iPSCs lines were then encoded to ensure that all further biological experiments were performed in a blinded fashion without knowledge of the associated clinical phenotype ( Figure 1—figure supplement 1 ) . The code was kept securely by a blinded third party and was only disclosed once all characterizations and electrophysiological measurements with microelectrode array ( MEA ) mapping system were finalized . All classifications of lines according to the observed response to sotalol were therefore performed before the unblinding process . Subject-specific iPSCs were differentiated using a small molecule-mediated directed differentiation protocol that involves sequential modulation of the canonical Wnt pathway and yields a high proportion of ventricular-like cardiomyocytes as previously described ( Karakikes et al . , 2014 ) . As early as seven days following initiation of cardiac differentiation , spontaneously beating embryoid bodies ( EBs ) appeared . To test for iPSC-derived cardiomyocytes quality and purity we first performed immunocytostaining studies showing positive staining for the sarcomeric proteins troponin T ( cTnT ) , and alpha-actinin and for the gap-junction protein connexin 43 ( Cx43 ) ( Figure 2A ) . Cardiac troponin T staining of iPSC-CMs showed a typical striated pattern ( Figure 2A , upper panels ) . We also determined the cardiomyocytes differentiation efficiency by flow cytometry and found an average of 47 . 8 ± 19 . 9% cTNT positive cells . Importantly , after unblinding the study , we did not observe any differences between iPSC lines from the high-S vs . low-S groups ( 49 . 2 ± 24 . 4% vs . 46 . 7 ± 17 . 3% respectively , p=0 . 42 ) . Finally , in line with previous reports ( Liang et al . , 2013 ) , quantitative PCR demonstrated the presence of major cardiac ion channel genes found in adult left ventricular tissue , including of the relevant KCNH2 gene ( Figure 2B and Figure 2—source data 2 ) . 10 . 7554/eLife . 19406 . 007Figure 2 . Expression of sarcomeric proteins and ion channels in human iPSC-CMs . ( A ) Confocal microscopy imaging of Troponin T ( top ) , alpha-actinin and connexin 43 ( bottom ) in single generated iPSC-CMs ( from line P11015 ) . Nuclei are stained with DAPI ( Blue ) . ( B ) Gene expression of cardiac ion channel KCNH2 ( encoding hERG ) by quantitative PCR; Adult LV tissue is used as a positive control and the level of expression in human ESC-derived cardiomyocytes as a comparator . ( C ) Example of monolayer of iPSC-CMs seeded and attached on a 6-well MEA chip , each well containing nine microelectrodes ( black ) . See also Video 1 . ( D ) Representative field potential duration ( FPD ) recorded before and after application of the hERG blocker E4031 ( from line P11007 ) . There are five figure supplements . DOI: http://dx . doi . org/10 . 7554/eLife . 19406 . 00710 . 7554/eLife . 19406 . 008Figure 2—source data 1 . iPSCs characterization . Representative immunostaining for a panel of pluripotent stem cell markers , including Nanog , OCT4 , SSEA4 and TRA1-60 . Results are reported for each cell line . Typical staining in hESC is reported as a positive control . Scale bar 50 μM . DOI: http://dx . doi . org/10 . 7554/eLife . 19406 . 00810 . 7554/eLife . 19406 . 009Figure 2—source data 2 . Expression of main human cardiac ion channels . Quantitative PCR results for the expression of 10 major cardiac ion channels including SCN5A , CACNA1 , KCNQ1 , KCNE1 , KCNJ2 , KCNJ11 , KChIP2 , HCN1 , HCN2 , HCN4 . KCND3 was not detected . Adult LV tissue is used as a positive control and the level of expression in human embryonic stem cells-derived cardiomyocytes as a comparator . DOI: http://dx . doi . org/10 . 7554/eLife . 19406 . 00910 . 7554/eLife . 19406 . 010Figure 2—figure supplement 1 . Expression of exogenous and endogenous pluripotency genes . Representative gel showing silencing of the four exogenous reprogramming transgenes ( OCT3/4; KLF4; SOX2 ; c-MYC ) and quantitative PCR results for endogenous pluripotent stem cell genes ( OCT3/4; NANOG; SOX2 ) in the generated hiPSCs . DOI: http://dx . doi . org/10 . 7554/eLife . 19406 . 01010 . 7554/eLife . 19406 . 011Figure 2—figure supplement 2 . MEA arrays . ( A ) Picture of a 6-well MEA array as used in the study . Each well contains nine micro-electrodes and a limited volume capacity allowing easier contact between cells and micro-electrodes after re-seeding and rapid response to drugs after application in a limited volume of medium . ( B ) Typical micro-electrodes map of the six well array . ( C ) Dose-curve responses to the hERG blocker E4031 . Increasing concentrations of E4031 were applied as described in Materials and methods section . Individual data were plotted and a hill equation was fitted allowing estimates of EC50 for each cell lines . Results were obtained in 15 cell lines as two cell lines display no changes in FPD in response to E4031 , suggesting expression of a non-functional hERG in these iPSC-CMs . For graphical purposes , cell lines from the low-S group are shown in green and from the high-S group in red . DOI: http://dx . doi . org/10 . 7554/eLife . 19406 . 01110 . 7554/eLife . 19406 . 012Figure 2—figure supplement 3 . Quality of FPD adjustment . ( A ) Plot of adjusted FPD using the Bazett’s formula against the individual inter-beat interval . This plot shows the miss-correction achieved using the Bazett’s formula with a significant dependence of FPD to beating rate even after adjustment . ( B ) Plot of adjusted FPD against the individual inter-beat interval after re-estimation of the correction factor to the data set . The plot shows perfect correction with linear regression indicating the lack of influence of the beating rate on the adjusted FPD . DOI: http://dx . doi . org/10 . 7554/eLife . 19406 . 012 We then screened for common single nucleotide polymorphisms ( SNPs , total of 4130 ) in the 15 genes associated with congenital long QT ( AKAP9 , ANK2 , CACNA1C , CALM1 , CALM2 , CAV3 , KCNH2 , KCNE1 , KCNE2 , KCNJ2 , KCNJ5 , KCNQ1 , SCN4B , SCN5A , SNTA1 ) using a high-density chip ( Human Omni 2 . 5 genotyping array ) . We found 25 SNPs in ANK2 , SCN5A , KCNQ1 , CACNA1C , CALM1 , KCNE2 , KCNH2 and KCNJ5 as being significantly imbalanced between low-S vs . high-S groups ( Table 3 ) . This suggests the presence of an allelic series comprising multiple variants of unknown significance that can create a particular predisposing genetic background . Of note , the CALM1 c . 1952C ( rs3814843 ) mutated allele was recently associated with increased risk of sudden cardiac death in patients with heart failure ( Liu et al . , 2015 ) . The allele is rare in Europeans ( anticipated allelic frequency of 1 . 8% ) but displayed a 22 . 2% allelic frequency in the high-S group . In addition , the rare C allele of KCNH2:c . 307 + 1932G>C polymorphism ( rs3778873 ) was significantly more frequent in high-S as compared to low-S ( minor allelic frequency: 44 . 4% vs . 6 . 2% respectively , p=0 . 02 ) . KCNH2 encodes for the hERG potassium channel that is targeted by sotalol and this variant was identified in a large genome-wide association study on the physiological regulation of QT interval ( Pfeufer et al . , 2009 ) . 10 . 7554/eLife . 19406 . 013Table 3 . Related to Figure 2 . Single nucleotide polymorphisms in ANK2 , SCN5A , KCNQ1 , CACNA1C , CALM1 , KCNE2 , KCNH2 and KCNJ5 as being significantly imbalanced ( p<0 . 05 ) between low-S vs . high-S groups . Anticipated minor allelic frequency ( MAF ) was defined using the HapMap-CEU European data . The MAF in cells from the high-S vs . low-S groups was determined once the study was unblinded . Fisher’s exact test was used to compare observed MAFs in high-S vs . low-S groups . DOI: http://dx . doi . org/10 . 7554/eLife . 19406 . 013Gene Rs number Anticipated MAF Observed MAF Low-S Observed MAF High-S p value ANK2 rs170459356 . 8% 0% 22 . 2% 0 . 03rs6231490140% 56 . 25% 27 . 8% 0 . 037rs1767625610 . 2% 12 . 5% 38 . 9% 0 . 03rs96709938 . 1% 25% 44 . 4% 0 . 027rs483432146 . 9% 37 . 5% 55 . 6% 0 . 02rs3530837048 . 3% 25% 66 . 7% 0 . 025rs93183842 . 5% 31 . 25% 66 . 7% 0 . 048SCN5A rs737512319 . 2% 43 . 75% 11 . 2% 0 . 007rs124919877 . 6% 0% 22 . 2% 0 . 03rs987138534 . 1% 6 . 25% 33 . 3% 0 . 02rs981814819 . 2% 56 . 25% 16 . 7% 0 . 049KCNQ1 rs425552015 . 5% 18 . 75% 0% 0 . 043rs15128832 . 7% 0% 22 . 2% 0 . 031rs718579rs1102299639 . 8% 41 . 2% 12 . 5% 27 . 8% 0 . 034rs15121240 . 8% 50% 33 . 3% 0 . 05CACNA1C rs379429912 . 4% 18 . 75% 0% 0 . 043rs4765661rs223801815 . 8% 17 . 3% 31 . 25% 5 . 6% 0 . 027CALM1 rs38148431 . 8% 0% 22 . 2% 0 . 031rs23005027 . 5% 6 . 25% 33 . 3% 0 . 023KCNE2 rs2840936830 . 8% 18 . 75% 38 . 9% 0 . 0239KCNH2 rs207241137 . 5% 18 . 75% 44 . 4% 0 . 027rs377887315 . 8% 6 . 2% 44 . 4% 0 . 02KCNJ5 rs792441624 . 6% 0% 27 . 8% 0 . 013 To evaluate the electrophysiological properties of iPSC-CMs , we used a microelectrode array ( MEA ) mapping system ( Figure 2—figure supplement 2 ) . iPSC-CMs were seeded onto a 6-well MEA chamber in order to form a monolayer in contact with the electrodes . Few days after seeding iPSC-CMs regain spontaneous beating activity thus allowing recordings of field potential duration ( FPD ) ( Figure 2C–D and Video 1 ) . FPD is analogous to the QT interval in the electrocardiogram ( ECG ) and was previously shown to correlate with action potential duration ( APD ) and can be used to test drug effects on repolarization ( Navarrete et al . , 2013 ) . IPSC-CMs were first tested for their responses to E4031 , a specific and potent experimental hERG blocker . E4031 resulted in a dose-dependent increase in FPD with typical flattening of the waveform in all but two cell lines ( Figure 2D and Figure 2—figure supplements 2 and 3 ) . There was moderate inter-line variability , except for line P11019 which demonstrated higher sensitivity to E-4031 and was found to belong to the high-S group after unblinding . There was however no significant differences in sensitivity to E4031 between groups as assessed by computations of half maximal effective concentration EC50 ( 2 . 4 ± 0 . 7 × 10−8 vs 6 . 6 ± 3 . 0 × 10−8 M in high-S and low-S groups respectively , p=0 . 59 ) or of maximal effect Emax ( 49 ± 20% vs 30 ± 2% in high-S and low-S groups respectively , p=0 . 76 ) . Two lines displayed no responses to E4031 and were found as having the lowest expression of hERG ( P11021 and P11023 , Figure 2B ) . Because of the lack of response to a pure hERG blocker and of appropriate recordings to detect effects of QT prolonging drugs with the MEA system , sotalol response was not further tested on these two lines . 10 . 7554/eLife . 19406 . 014Video 1 . Spontaneous beating activity of monolayer iPSC-CMs seeded onto one well of a 6-well MEA chamber . DOI: http://dx . doi . org/10 . 7554/eLife . 19406 . 014 The response to the clinically relevant drug sotalol was then tested in a blinded fashion in the remaining lines . For quality purposes , results from the iPSC line presenting a significant translocation were excluded from further analyses . However , the inclusion of the results obtained with this iPSC line did not change the statistical significance of the results . Final analyses were thus performed on a total of 14 lines ( 7 vs . 7 lines in low-S and high-S groups respectively ) . After unblinding , we found that the different recordings showed a significantly higher response to sotalol in the CMs generated from iPSCs derived from high-S subjects as compared to low-S subjects ( Figure 3A–D ) . Increasing sotalol concentrations was associated with a significant prolongation in FPDs in all iPSC-CMs , an effect that was however significantly enhanced in lines from the high-S group ( Figure 3A ) . There was a dose-dependent sotalol-induced FPD prolongation that typically appears for sotalol concentrations above 30 μmol/L ( Figure 3A ) , as seen in other reports ( Navarrete et al . , 2013 ) . Of note , the application of low sotalol concentration ( 10 μmol/L ) , a concentration this is closer to typical sotalol plasma levels , was already associated with a significant increase in FPD in lines from the high-S group as compared to low-S ( p=0 . 03 ) . The maximal FPD prolongation observed in response to sotalol was also significantly higher in the high-S as compared to low-S iPSC-CMs ( 52 ± 6% vs . 27 ± 6% , p=0 . 02 , Figure 3B ) . Sotalol is known to induce arrhythmic events including ectopic beats and short-long-short rhythm . We found that 5 of the seven lines derived from the high-S group displayed arrhythmias in response to sotalol stimulation as they showed development of irregular spontaneous beating rate . On the other hand , arrhythmias only occurred in one of the eight lines from the low-S group ( Figure 3C ) . We plotted the relative change in FPD in response to sotalol 30 μM against the observed change in QTcf in donors ( Figure 3D ) . With the exception of one cell line in each group , the remaining six low-S and six high-S cell lines were correctly discriminated according to the clinical phenotype . These data indicate that the contrasted responses to sotalol in iPSC-derived CMs are strongly related to the inter-individual differences in response to sotalol as recorded during clinical investigations . 10 . 7554/eLife . 19406 . 015Figure 3 . Differences in iPSC-CMs responses to Sotalol stimulation according to clinical sensitivity to Sotalol . ( A ) Adjusted FPD ( aFPD ) measured in iPSC-CMs derived from subjects with low-sensitivity ( green ) vs . high-sensitivity ( red ) in response to increasing concentrations of sotalol . aFPD are normalized to baseline values to account for inter-lines variability in aFPD values . Two-way analysis of variance demonstrates a significant influence of sotalol concentrations ( p<0 . 0005 ) and of the sensitivity group ( p<0 . 02 ) . p<0 . 05 for post-hoc comparison between groups; high-S vs . low-S . N = 2–5 recordings per cell lines per concentrations . ( B ) Maximal change in aFPD observed during sotalol stimulation . ( C ) Proportion of observed arrhythmias after sotalol application . ( D ) Data plot graph showing the correlation between aFPD observed in iPSC-CMs and the DeltaQTcf observed in donors . The aFPD data are reported for sotalol 30 μM concentration . Data points are clustered in two distinct groups . Except for two lines ( one line in each group ) , a threshold a 25% in aFPD change ( dashed line ) correctly discriminates cells from both groups . DOI: http://dx . doi . org/10 . 7554/eLife . 19406 . 015 We also evaluated the effect of sotalol on action potential duration ( APD ) measured in a representative iPSC-CMs line from each group ( P11009 as low-S and P11029 as high-S ) . Experimenter for APD measurement was however kept blinded of the observed drug sensitivity of selected iPS lines . Sotalol ( 100 and 300 µM ) significantly prolonged the action potential , the duration of 90% repolarization ( APD90 ) of high-S ( P11029 ) iPSC-CMs ( 289 ± 22 in drug-free vs . 492 ± 66 and 435 ± 92 after 100 and 300 µM sotalol , respectively n = 5 ) but not that of the low-S line ( P11009 ) even at high dose ( 339 ± 63 vs . 450 ± 77 ms after 300 µM sotalol , n = 5 ) ( Figure 4A–B and Table 4 ) . 10 . 7554/eLife . 19406 . 016Figure 4 . Patch-clamp analysis of action potential ( AP ) in representative iPSC-CMs . ( A ) Representative AP tracings of the iPSC-CMs generated from the low-S ( P11009 ) and high-S ( P11029 ) lines in control and sotalol-treated conditions . ( B ) bar chart summarizing the APD90 in control and sotalol-treated conditions for the hiPSC-CMs generated from both low-S ( P11009 ) and high-S ( P11029 ) hiPSC cell lines ( n = 5 for each condition ) . p<0 . 05; **p<0 . 01 , ANOVA , followed by Tukey's , sotalol-treated versus respective control without sotalol application . DOI: http://dx . doi . org/10 . 7554/eLife . 19406 . 01610 . 7554/eLife . 19406 . 017Table 4 . Related to Figure 4 . AP parameters of low-S and high-S iPSC-derived cardiomyocytes at baseline control condition . AP data are mean ± SE . APD50/APD90 , AP duration measured at 50% or 90% repolarization; MDP , maximum diastolic potential . None of the baseline AP parameters was significantly different between the two cell lines . DOI: http://dx . doi . org/10 . 7554/eLife . 19406 . 017Low-S ( 11009 ) ( n=7 ) High-S ( 11029 ) ( n=5 ) Firing Frequency ( mV ) 1 . 26 ± 0 . 29 1 . 03 ± 0 . 28 Amplitude ( mV ) 88 . 3 ± 3 . 6 85 . 1 ± 4 . 7 Upstroke velocity ( mV/ms ) 11 . 7 ± 1 . 0 14 . 0 ± 2 . 0 Decay velocity ( mV/ms ) −7 . 2 ± 1 . 4 −11 . 1 ± 3 . 5 APD50 ( ms ) 306 . 4 ± 57 . 9 299 . 4 ± 22 . 0 APD90 ( ms ) 341 . 8 ± 58 . 6 338 . 4 ± 21 . 4 MDP ( mV ) −69 . 4 ± 3 . 2 −68 . 0 ± 1 . 3 To gain insight into molecular mechanisms leading to differential susceptibility to developing diLQT , we performed a transcriptomic profiling of iPSC-CMs from high-S vs . low-S groups using RNA-sequencing . We looked for differentially expressed genes ( DEGs ) between the two groups ( all cell lines included ) and used a prior knowledge based approach to identify potential risk markers that could be mechanistically connected to QT prolongation development . By analyzing the LQTS neighborhood in the human interactome we recently demonstrated that graph theoretical models allow the prediction of new gene variants and drug targets that are involved in disease pathogenesis ( Berger et al . , 2010 ) . Here , we used a similar approach and searched for differentially expressed direct neighbors of the known congenital LQTS genes in the human interactome ( Figure 5A ) . Of note , expression of these LQTS genes was similar between high-S and low-S groups . We identified four up-regulated ( DLG2 , KCNE4 , PTRF , and HTR2C ) and one down-regulated direct neighbors ( CAMKV ) ( Figure 5A–B ) . Among these newly identified candidates , DLG2 ( a member of the family of anchoring proteins called MAGUK ) , KCNE4 ( a regulatory sub-unit of potassium channel ) , CAMKV ( a kinase-like protein that , in the presence of calcium , interacts with calmodulin ( CALM1 ) , and PTRF ( Polymerase I and transcript release factor , or cavin-1 ) were relevant candidates as downstream regulators of cardiac ion channels . 10 . 7554/eLife . 19406 . 018Figure 5 . Identification of dysregulated genes as direct neighbors of QT-associated network in high-S iPSC-CMs . ( A ) Known LQTS genes were used as seed nodes ( green squares ) in the human interactome and five differentially expressed direct neighbors were identified ( circles ) ( path length 1 ) . Up-regulated genes are colored orange , down-regulated genes are in blue . ( B ) Relative expression of identified genes in each group . Males are represented in blue and females in red . Individual samples are represented by the same symbols in all diagrams . ( C ) Comparison of the log2-fold changes between the high-S and the low-S groups with the normalized counts of how often a gene ( blue or orange dots ) was found to be up- or down-regulated between the male and female groups . As there are more females in the high-S group and more males in the low-S groups , report of genes in the lower left or upper right quadrants indicate a gender-specific effect while the lower right and upper left quadrants argue for the lack of gender-specific effect . Except for HTR2C , dysregulation of all other candidate genes was suspected to occur independently of gender . There is one figure supplement . DOI: http://dx . doi . org/10 . 7554/eLife . 19406 . 01810 . 7554/eLife . 19406 . 019Figure 5—figure supplement 1 . Prediction of sex hormones-related transcription factors . Predicted regulatory transcription factors of the identified DEGs ( DLG2 , KCNE4 , CAMKV , PTRF , HTR2C ) based on two transcription factor target databases ( Chea background and Transfac ) and ranked by significance . The sex-hormone related transcription factors ( in orange ) were ranked 16 ( AR ) , 20 ( ESR1 ) and 24 ( ESR2 ) or 27 ( PGR ) and 180 ( ESR1 ) while the top 10 candidates were independent of sex hormones influence ( in blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19406 . 019 The expression levels of the genes between the low-S and the high-S groups show that the risk specific expression ( i . e . low expression for CAMKV and high expression for the other direct neighbors ) does not accumulate on a few high-S subjects but is more or less equally distributed over all high-S subjects ( Figure 5B ) . Each subject has a risk-associated expression of at least one direct neighbor , arguing for multiple ways to disturb the LQTS genes and therefore confer different susceptibility to develop diLQT . Since most of the high-sensitivity subjects were females , while most of the low sensitivity subjects were males , we then estimated whether gender could impact our results . To neutralize gender-specific effects on gene expression , we generated new groups consisting of one high-S and one low-S subject of the same gender ( 16 female groups , six male groups , Table 5 ) . DEGs were identified between all possible combinations of one female versus one male group ( 16 × 6 = 96 sets of DEGs ) . We counted for each gene how many sets of DEGs identified the gene as significantly up-regulated or down-regulated . Down-regulated counts were subtracted from up-regulated counts and resulting netto counts were normalized to the number of total comparisons ( i . e . divided by 96 ) . Using this approach , a gene that is up-regulated in the high-S group because of gender specific expression , should be identified as significantly up-regulated in most of the female versus male groups sets of DEGs . The same accounts for down-regulated genes . As shown in Figure 5B and C , this trend was not observed for most of the genes , arguing against gender differences being the reason for differential gene expression between the high-S and low-S group . 10 . 7554/eLife . 19406 . 020Table 5 . Related to Figure 5 . Generation of new groups associating one high-S and one low-S line of the same gender ( six groups for males and 16 groups for females ) in order to neutralize gender-specific effects on gene expression . These groups were then used to determine DEGs between the high-S and low-S groups independent of gender . DOI: http://dx . doi . org/10 . 7554/eLife . 19406 . 020Male groups Female groups P11019 ( high ) + P11028 ( low ) P11008 ( high ) + P11030 ( low ) P11019 ( high ) + P11026 ( low ) P11015 ( high ) + P11030 ( low ) P11019 ( high ) + P11020 ( low ) P11013 ( high ) + P11030 ( low ) P11019 ( high ) + P11009 ( low ) P11029 ( high ) + P11030 ( low ) P11019 ( high ) + P11031 ( low ) P11018 ( high ) + P11030 ( low ) P11019 ( high ) + P11007 ( low ) P11024 ( high ) + P11030 ( low ) P11023 ( high ) + P11030 ( low ) P11021 ( high ) + P11030 ( low ) P11008 ( high ) + P11014 ( low ) P11015 ( high ) + P11014 ( low ) P11013 ( high ) + P11014 ( low ) P11029 ( high ) + P11014 ( low ) P11018 ( high ) + P11014 ( low ) P11024 ( high ) + P11014 ( low ) P11023 ( high ) + P11014 ( low ) P11021 ( high ) + P11014 ( low ) Finally , to further investigate if the DEGs are the consequence of sex specific gene expression , we predicted regulatory transcription factors of the DEGs based on two transcription factor target databases and ranked them by significance ( Figure 5—figure supplement 1 ) . The sex-hormone related transcription factors were ranked 16 ( AR ) , 20 ( ESR1 ) and 24 ( ESR2 ) or 27 ( PGR ) and 180 ( ESR1 ) , and no sex hormone related transcription factor was among the top 10 candidates . These data suggest that the observed differences in expression patterns between high-S and low-S iPS-CMs are mostly indicative of a signature associated with drug sensitivity rather than a gender effect . Our data illustrate the potential of using a genetically diverse panel of subject-specific iPSCs to model complex and acquired phenotypes . So far , iPSCs technology has been successful in recapitulating monogenic diseases with clear causative mutations including congenital long QT syndrome ( Itzhaki et al . , 2011; Moretti et al . , 2010; Sinnecker et al . , 2014 ) . In contrast , diLQT develop in clinically normal individuals who have a genetic predisposition requiring an additional stressor to become manifest ( Roden , 2006 , 2008b ) . The exact nature of this predisposing genetic background remains uncertain ( Petropoulou et al . , 2014 ) but is likely made of an allelic series of common variants with modest independent effect on cardiac repolarization but potential for a stronger impact in a polygenic model , as exemplified by our genetic profiling . Here , we provide evidence that susceptibility to develop diLQT can be best reproduced in vitro using a targeted library of iPSCs derived from patients presenting with an extreme pharmacodynamic response to drug stimulation in vivo . Extreme phenotype selection is a well-defined methodology that helps enrich the frequency of alleles that contribute to a trait and this methodology has been successfully applied in genetic studies aimed at identifying variants associated with complex traits ( Cirulli and Goldstein , 2010 ) . The use of this methodology is particularly suitable when investigating variability in drug responses by segregating individuals with opposite predisposing genetic makeup , a characteristic that can be transmitted to iPSCs as shown by our data . Importantly , in order to avoid experimental biases , in vitro assessment of iPSC-CMs sensitivity to sotalol was performed blinded to the clinical phenotype . In addition , as we aimed at reproducing a more complex phenotype , we designed our study with a larger sample size than typical studies using iPSCs for disease modeling of monogenic disorders . As a consequence , we were only able to explore one iPSC clone per patient and we did not evaluate potential inter-line variability . The likelihood that inter-line variability would have systematically biased our results to reproduce the anticipated effect as seen in vivo and thus explain our results is however unlikely on a panel of 17 different cell lines as appreciated by low p-values for comparison between groups ( Figures 3 and 4 ) . Under our hypothesis , it is likely that the discrepant results obtained in two cell lines reflect that potential false results associated with inter-line variability from a same patient . However , external replication would further strengthen the potential of subject-specific iPSCs to model complex phenotypes as shown in our seminal study . This library also represents a unique opportunity to explore the molecular mechanisms underlying pro-arrhythmic adverse drug reactions . The transcriptomic comparison of the generated iPSC-CMs from both groups combined with bioinformatics network analysis revealed significant changes in expression of genes related to the regulation of ion channels homeostasis ( DLG2 , KCNE4 , CAMKV and PTRF ) . Of note , while none of these genes have been previously associated with the occurrence of diLQT , CAMKV is coding for a kinase-like protein that , in the presence of calcium , interacts with calmodulin ( CALM1 ) , a critical regulator of ion channels involved in congenital long QT syndrome 14 ( Crotti et al . , 2013 ) and in loperamide-induced long QT ( Berger et al . , 2010 ) . Additionally , mutations in PTRF ( Polymerase I and transcript release factor , or cavin-1 ) cause a particular form of congenital generalized lipodystrophy ( type 4 , CGL4 ) that is associated with features of long QT syndrome and high rate of sudden cardiac death ( Rajab et al . , 2010 ) . PTRF is involved in the formation of caveolae ( Nabi , 2009 ) , that are critical in cardiac ion channels trafficking to the plasma membrane ( Maguy et al . , 2006 ) . In addition to the involvement of DLG2 and KCNE4 genes in the direct regulation of scaffolding , trafficking and gating kinetics of some cardiac potassium channels ( Leonoudakis et al . , 2004; Levy et al . , 2010 ) , our study suggest for the first time that diLQT might be related to changes in downstream regulation of the cardiac repolarization machinery . Further experiments will be required to understand the exact contribution of each of these candidates and whether they might act in combination to confer susceptibility to develop diLQT . Similarly , future studies using single-cell RNA sequencing on cardiomyocytes derived from our library of subject-specific iPSCs would be needed to better identify and characterize differentially expressed genes and pathways . Lastly , we identified an up-regulation in HTR2C ( 5-hydroxytryptamine receptor 2C ) , a G-protein coupled receptor for serotonin that is expressed in the heart . Upon activation HTR2C activates down-stream signaling cascades to promote the release of Ca2+ from internal stores . Interestingly , serotonin receptor agonists have been associated with higher risk for diLQT ( Keller and Di Girolamo , 2010; Sarganas et al . , 2014 ) . Intriguingly , we found non-significant trends for differences between groups in response to E-4031 , a selective hERG channel blocker . Sotalol primarily prolongs cardiac repolarization through hERG inhibition but also have a non-selective competitive β-adrenergic receptor blocker activity and can thus affect other cardiac repolarization components that are associated with higher susceptibility to develop diLQT . Whether sotalol has a direct or indirect effect on the candidates identified in this study would deserve further investigations . Many studies have reported a higher susceptibility to diLQT in premenopausal women as compared to men including in response to sotalol ( Darpo et al . , 2014; Makkar et al . , 1993 ) , an observation that is in line with the results of our prospective clinical investigation . It has been suggested that sex hormones can influence cardiac repolarization by modulating expression and function of cardiac ion channels but their effect is complex to appreciate as levels fluctuate with time and age and the different sex hormones can have counterbalancing effects ( Hulot et al . , 2003; Odening and Koren , 2014 ) . Here , we were able to reproduce susceptibility to develop diLQT in vitro in a hormone-free environment . We performed specific analyses that did not support an important role of gender to explain our results , including the lack of a gender-specific gene expression signature and the lack of significant sex hormone-related transcription factors . Importantly , our data suggest that the identified genes should work as predisposing genes in both genders thus arguing for the existence of an intrinsic predisposing background that is carried by iPS cells but do not depend on gender . Of note , it has recently been shown that women have in mean a greater intrinsic sensitivity to sotalol as compared to men , an effect that however remains widely variable at the individual level ( Darpo et al . , 2014 ) . The development of a new generation of human-based screening assays for testing individual drug reactions ( Inoue and Yamanaka , 2011; Mann , 2015 ) could help in better predicting adverse drug effect that usually develop in a relatively low proportion of susceptible individuals in the general patient population ( Budnitz et al . , 2006 ) . Interestingly , we were able to detect significant differences in response to the lowest sotalol concentrations ( i . e . , 10 µM ) in iPSC-derived cardiomyocytes from subjects clinically presenting with high sensitivity to sotalol . This suggests that the iPSC-derived platform could efficiently detect individual differences in the range of concentrations that is near the typical sotalol plasma levels . Finally , further developments in differentiation protocols ( Burridge et al . , 2014 ) , in functional maturation of iPSC-derived cells ( Lundy et al . , 2013 ) or in iPSC-based bioengineering ( Schaaf et al . , 2011; Turnbull et al . , 2014 ) might also improve the overall predictability of the method . Our data illustrate the potential of iPSCs technology for the prediction of individual risk and for its use in precision medicine . We found a limited number of lines with discrepant results in vitro as compared to the clinical phenotype , suggesting a good accuracy of the approach . However , as we only investigated patients with extreme responses to sotalol , our study was not designed to accurately assess the predictive value of the test . This should be adequately addressed in larger prospective studies . In conclusion , this study underscores the power of developing a panel of iPSCs to model complex traits such as susceptibility to develop cardiotoxic drug response . To participate to the study , male or female volunteers had to fulfill the following inclusion criteria: aged 18 to 40 years , body mass index from 19 to 29 kg/m2 , no previous history of cardiac disease , no major disorders , no medications known to affect cardiac repolarization ( as defined by https://www . crediblemeds . org/everyone/composite-list-all-qtdrugs ) . All volunteers had normal laboratory evaluations and a normal clinical examination with a resting heart rate ≥50 beats min−1 and a systolic blood pressure ≥100 mmHg . Cardiac conduction and repolarization was assessed on a 12-lead resting electrocardiogram and participants were excluded in case of atrio-ventricular conduction disorder ( PR interval >200 ms ) , QRS >100 ms and in case of prolonged QTcF >450 ms . Other exclusion criteria were familial or personal history of sudden death or unexplained syncope , concomitant use of illicit drugs , asthma . The protocol was approved by the local institutional ethics committee and all subjects gave their written informed consent for participation to the study . All investigations were performed prospectively at a single clinical center ( Biotrial Paris , Rueil Malmaison , France ) . In a first phase , eligible subjects were admitted at 7:30am at the clinical investigation center after an overnight fast . Subjects were placed in a quiet room and rested for 30 min in the supine position . Cardiac rhythm and non-invasive blood pressure monitoring were started . The quality of ECG recordings was assessed and baseline recordings were performed during the 5 min preceding sotalol oral administration ( 80 mg tablet ) . ECG recordings were then performed 3 hr after sotalol intake , i . e . around peak plasma concentration . Blood samples for the determination of sotalol concentration were obtained at baseline and 3 hr after sotalol administration . Volunteers were allowed to leave the clinical investigation center 5 hr after sotalol intake after cardiac repolarization had normalized . In a second phase ( 7 to 35 days after phase 1 ) , 20 subjects with the most extreme responses to sotalol ( 10 with smallest QTcF changes and 10 with the largest QTcF changes from baseline in response to sotalol ) were asked to come to the clinical investigation center and had a skin biopsy under local anesthesia using a sterile 3 mm skin punch . Before and 3 hr following sotalol administration , 30 s digital 12-lead electrocardiograms were recorded using a Cardioplug device ( Cardionics Inc , Brussels , Belgium ) connected to a personal computer . All electrocardiogram recordings were then read by the same investigator , blindly of drug administration . The same chest lead with the largest T-wave amplitude was selected for QT interval measurements in a given subject . QT interval was measured manually directly on the computer screen by changing position of cursors indicating the start and the end of the cardiac interval: RR ( interval between two successive R waves ) and QT ( tangent method ) . Baseline QTc was assessed as the mean of three electrocardiographic recordings obtained within 5 min before drug administration . QT was corrected by the Fridericia cubic root formula ( QTcf ) , which minimizes the errors due to the square root Bazett formula . The clinical phenotype was defined by modifications in QTcf duration ( ms ) 3 hr after sotalol intake . Fibroblast derivation was performed immediately after biopsy arrival at Ectycell / Cellectis Stem Cell Company ( Evry , France ) . Skin biopsy of a 3 mm diameter punch was cut into about 20 small explants with scalpels and placed in 6-well plates and cultured with fibroblast growth medium containing DMEM medium ( Gibco ) , 10% of FBS ( PAA Laboratories ) , 10 ng/ml FGF-2 ( Invitrogen , Carlsbad , CA ) and 1% penicillin-streptomycin ( Gibco ) . Cells were fed every 2–3 days . In this culture , fibroblasts appeared and became confluent on average 33 days after plating . The fibroblasts were then subcultured using 0 . 25% trypsin ( Invitrogen ) then re-suspended and cultured into T75 flasks using fibroblast growth medium without antibiotics . An average of 62 million fibroblasts per biopsy was obtained in 4–7 weeks at passage 1 . The fibroblasts were tested for mycoplasma contamination and no contamination was detected ( MycoAlert TM Mycoplasma Detection kit , Lonza France ) . Reprogramming of fibroblasts derived from patient’s skin biopsies into iPSCs was carried out using polycistronic retroviral vectors . The fibroblasts were seeded in 6-well plates at a defined cell density ( 100 , 000 cells/well ) and cultured with fibroblast growth medium ( DMEM medium containing 10% of FBS and 10 ng/ml FGF-2 ) . For transduction , cells were incubated with polycistronic retroviral vectors ( provided by Vectalys , Toulouse , France ) carrying human Oct4 , Sox2 , Klf4 , and c-Myc ( OSKM ) expression factors in fibroblast medium supplemented with 6 µg/ml polybrene ( Sigma , St Louis , MO ) overnight . The cells were transduced at a multiplicity of infection of 5 . Four days post-infection , the cells were split by using 0 . 25% trypsin and plated at 100 , 000 cell/well in 0 . 1% gelatin-coated 6-well plates in fibroblast growth medium . After 24 hr , the medium was switched to the human pluripotent stem cell medium ( DMEM/F12 , 20% knockout serum replacement , 1× non-essential amino acid , 2 mM L-glutamine , 0 . 1 mM 2-mercaptoethanol , 20 ng/ml FGF-2 and 1% penicillin-streptomycin ) with 0 . 5 mM Valproic acid ( VPA ) . The media were changed every day . Around 20–25 days post-infection , the human ESC-like colonies appeared . The iPSCs colonies were picked and expanded in mTeSR1 ( Stemcell technologies , Canada ) on matrigel matrix ( BD Biosciences France ) coated plates . The iPSCs were passaged with Accutase ( 10 ml pre 75 cm2 surface area , Millipore , Billerica , MA ) and the culture medium was changed daily . Colonies fulfilling established ‘stemness’ criteria were selected and sent to the Cardiovascular Research Center at Mount Sinai School of Medicine , New York , USA for differentiation toward the cardiomyocyte lineage and pharmacological characterization . The iPSCs were differentiated into cardiomyocytes using a directed differentiation method . Cardiomyocytes differentiation was initiated in suspension cultures on ultra-low attachment dishes ( Corning , France ) in mTESR1 medium supplemented with BMP4 10 ng ml-1 ) and Blebbistatin ( 5 μM ) for 24 hr . The medium was then replaced with the basal differentiation medium ( StemPro34 , 50 μg ml-1 ascorbic acid , 2 mM GlutaMAX-I ) supplemented with BMP4 ( 10 ng ml-1 ) and Activin-A ( 25 ng ml-1 ) for 48 hr ( days 1–3 ) and then switched to basal differentiation medium for another 36 hr ( days 3–4 . 5 ) . Finally , the cells were differentiated in basal differentiation medium supplemented with IWR-1 ( 2 . 5 μM ) for 96 hr ( Day 4 . 5–8 . 5 ) . The differentiated cardiomyocytes were maintained in basal differentiation media for up to four weeks . All cytokines were purchased from R and D . The small molecules were purchased from Sigma . All differentiation cultures were maintained in 5% CO2/air environment . iPSCs were cultured on matrigel-coated coverslips , fixed in paraformaldehyde and permeabilized in blocking/permeabilization buffer ( 2% BSA/2% FBS/0 . 05% NP-40 in PBS ) for 45 min and incubated with primary antibodies overnight at 4°C . Then the cells were washed in PBS and incubated with Alexa-conjugated secondary antibodies ( Invitrogen ) diluted in blocking/permeabilization buffer ( 1:750 ) . Finally , after washing in PBS the cells were counterstained with DAPI . Immunofluorescence images were acquired using an Olympus X41 microscope . The following antibodies were used: mouse monoclonal anti-OCT4 ( Santa Cruz biotechnology , Germany ) , goat polyclonal anti-NANOG ( R and D systems ) , mouse monoclonal anti-SOX2 ( R and D systems ) , mouse monoclonal anti-SSEA-4 ( R and D systems , Minneapolis , MN ) , mouse monoclonal anti-TRA-1–60 ( R and D systems ) . Similarly , iPSC-derived cardiomyocytes were dissociated and cultured on matrigel-coated coverslips for 4–5 days , fixed in paraformaldehyde and permeabilized in blocking/permeabilization buffer for 45 min . The following primary antibodies were used: mouse monoclonal anti-cardiac troponin T ( Thermo Fisher Scientific , France ) , mouse monoclonal anti-connexin 43 , and mouse monoclonal anti-α-actinin . Confocal imaging was performed using a Leica SP5 confocal system . Single-cell suspensions were obtained by dissociating EBs with 0 . 025% trypsin for 15 min at 37°C . The cells were then fixed with 4% paraformaldehyde for 15 min and washed twice with phosphate buffered saline ( PBS ) . The fixed cells were first permeabilized in permeabilization buffer ( 0 . 2% Triton X-100 in PBS ) for 30 min and then blocked with 10% goat serum for 25 min . Cells were then incubated with the primary antibody ( anti-cardiac troponin T; Thermo Fisher Scientific ) . After 1 hr , the cells were washed in PBS , incubated with Goat anti-mouse IgG1 - Alexa 488 secondary antibody for 45 min , and finally washed twice with PBS . All procedures were performed at 4°C . Fluorescence-activated cell sorting analysis was carried out using a BD LSR analyzer ( BD Biosciences ) . Relative gene expression was determined using a two-steps quantitative real-time PCR method . Total RNA was isolated with the RNeasy Isolation kit with on-column DNase I treatment ( Qiagen , Germany ) and reverse-transcribed using the cDNA Synthesis Kit ( Biorad , Hercules , CA ) . Quantitative RT-PCR was performed with the Quanta SYBR Green Supermix ( Quanta biosciences , Beverly , MA ) on the ABI Prism 7500 Real Time PCR System ( Applied Biosystems , Foster City , CA ) . Fold changes in gene expression were determined using the comparative CT method ( △△Ct ) with normalization to the housekeeping gene B2M . Single nucleotide polymorphism ( SNP ) genotyping analysis was performed using the Illumina HumanOmni2 . 5–8 beadchip genotyping array , which comprise a comprehensive set of around 2 . 6 million SNPs ( with MAF >2 . 5% ) across the genome . The list of mapped SNPs can be found at http://support . illumina . com/downloads/infinium-omni2-5-8-v1-3-support-files . html . All genomic DNA was isolated from iPSC clones using Quick-gDNA Mini Prep kit ( Zymo Research , Irvine , CA ) . Input genomic DNA ( 200 ng ) was processed , hybridized to the array and scanned on an Illumina HiScan at the Mount Sinai Genomic Core Facility . Based on chromosome coordinates , we extracted SNPs in AKAP9 , ANK2 , CACNA1C , CALM1 , CALM2 , CAV3 , KCNH2 , KCNE1 , KCNE2 , KCNJ2 , KCNJ5 , KCNQ1 , SCN4B , SCN5A , SNTA1 using Genome Studio ( Illumina , San Diego , CA ) . Genotypes were estimated and compared between low-S and high-S groups . We used HapMap-CEU data to estimate the anticipated minor allelic frequency in European population . Total RNA was extracted using Zymo columns and 2 µg were used to generate a RNA-seq sequencing library . Poly-A selection and mRNA-SEQ library preparation were performed at the Mount Sinai Genomics Core Facility . Sequencing ( 50 bases , paired ends ) was performed using an Illumina HiSeq2500 . Annotated reads were obtained using STAR and HTSeq and normalized to full library size . Sequencing reads were aligned to the human reference genome 'hg19' using Tophat 2 . 0 . 8 ( Trapnell et al . , 2009 ) , samtools-0 . 1 . 7 and bowtie 2 . 1 . 0 ( Langmead and Salzberg , 2012 ) and differentially expressed genes were identified with cufflinks 1 . 3 . 0 ( Trapnell et al . , 2010 ) . The total number of sequenced reads in a sample influences the likelihood to detect a lowly to moderately expressed gene , especially in case of lower read counts ( McIntyre et al . , 2011 ) . In consequence , it could happen that fewer genes are detected in a sample with a lower read count than in a sample with a higher read count . Such an experimental artifact might distort normalization including total reads as well as upper quartile normalization , the two normalization options that are offered by cufflinks . Both normalization approaches only change the number of reads that are associated with a gene , but not the number of identified genes . Consequently , the same number of normalized reads might be distributed over a different number of genes in two different samples , causing the detection of an equally expressed gene in both samples as differentially expressed . To prevent such an experimental artifact we randomly removed reads from the sequenced samples until every sample had the same number of read counts . Reads were aligned to the human reference genome hg19 with Tophat using the ensemble GTF file as a gene annotation reference and the option 'no-novel-juncs' . Output BAM files were directly subjected to Cufflinks to identify differentially expressed genes , using the options 'multi-read-correct' , 'upper-quartile-norm' and 'frag-bias-correct' against the hg19 genome . Differentially expressed genes were identified based on a FDR of 5% and a minimum fold change ( log2 ( ( FPKMcondition1+1 ) / ( FPKMcondition2 + 1 ) ) >= ± log2 ( 1 . 3 ) ) . We determined differentially expressed genes between the high-S ( P11008 , P11013 , P11015 , P11018 , P11019 , P11021 , P11023 , P11024 , P11029 ) and the low-S ( P11007 , P11009 , P11014 , P11020 , P11026 , P11028 , P11030 , P11031 ) cell lines . To compare gene expression values of the identified candidates across individual samples , we subjected all samples as individual conditions to the cufflinks analysis pipeline in the same way as described above . We used the FPKM values of the cufflinks results file as a measure for normalized gene expression . To analyze , if differentially expressed genes between the high-S and low-S group are caused by gender differences between the two groups , we generated all possible combinations of one high-S female and one low-S female ( 16 groups ) and one high-S male and one low-S male ( six groups ) . We determined differentially expressed genes between all possible combinations of one female group versus one male group ( 16 × 6 sets of DEGs ) . For each gene we counted in how many sets of DEGs it was detected as significantly up-regulated or down-regulated , subtracted the counts for down-regulated detections from the counts for up-regulated detections and divided the resulting number by the counts for sets of DEGs ( i . e . by 96 ) . To search for differentially expressed direct neighbors of LQTS disease genes , we generate a human interactome by merging all protein-protein interaction databases of the Expression2 kinases suite ( Chen et al . , 2012 ) , except the 'Predicted PPI' database , and a recently published protein-protein interaction network ( Rolland et al . , 2014 ) . Small letter symbols of the merged interactome were replaced by their human homologues , using the Mouse Genome Informatics mouse-human orthology database and the NCBI mouse-human homologene database . Finally , we removed all network nodes that were not official human gene symbols as reported in the NCBI geneInfo database . Differentially expressed genes in the direct neighborhood of LQTS disease genes ( 'AKAP9' , 'ANK2' , 'CACNA1C' , 'CALM1' , 'CALM2' , 'CAV3' , 'KCNE1' , 'KCNE2' , 'KCNH2' , 'KCNJ2' , 'KCNJ5' , 'KCNQ1' , 'SCN4B' , 'SCN5A' , 'SNTA1' ) were identified ( path length = 1 ) . All differentially expressed genes were subjected to transcription factor target enrichment analysis using the Chea-background and Transfac database ( downloaded from EnrichR [Chen et al . , 2013] ) , as described previously ( Karakikes et al . , 2014 ) . For MEA recordings , we used 6well-MEA arrays , which contain six independent culture chambers , separated by a macrolon ring ( 60–6 well MEA 200/30 iR-Ti-rcr , Multichannel Systems , Germany ) . Inside each well , there is a field of nine electrodes with an internal reference electrode ( Figure 2—figure supplement 2 ) . The 6well-MEA arrays were prepared by pipetting 5 μl fibronectin solution ( 100 μg/mL , BD Biosciences ) and incubated at 37°C for at least 1 hr . iPSC-CMs were dissociated using 0 . 025% trypsin for 5 min at 37°C and seeded onto prepared MEA plates using 5 μl of cell suspension in StemPro34 medium and then incubated at 37°C/5% CO2 . The day after the cells were covered with 200 µL of StemPro34 media . MEA recordings were performed once the cells started to beat again ( 5–7 days ) . Field potentials of spontaneously beating cardiomyocytes were recorded using a high-resolution Micro Electrode Array ( MEA ) recording system ( MEA60 system , Multi Channel Systems , Reutlingen , Germany , http://www . multichannelsystems . com ) at 37°C . The baseline steady state was achieved following an equilibration period of about 15 min in vehicle medium ( StemPro34 media ) . The tested drugs ( E4031 and Sotalol , Sigma ) were then added directly to each well . Dose-response experiments were performed with the following sequence: drugs were diluted in 200 μL of StemPro34 media; increasing concentrations of drug were added to a well of the MEA array in a cumulative manner; the FPD recordings were started 3 min after application of a given concentration of a drug , a timing that was found optimal to achieve steady-state changes; FPD recordings were then performed for two additional minutes . Each drug was tested in at least three different wells . Raw MEA data were acquired using QT-Screen Lite and MC-Rack softwares ( Multi Channel Systems , Reutlingen , Germany ) . The data were then exported on QT-Analyzer Software to analyze the field potential duration and the inter-beat interval ( Multi Channel Systems , Reutlingen , Germany ) . Parameters were averaged on the 2 min recordings . Inter-beat interval ( in ms ) was used to calculate the instantaneous beat rate . To account for the dependency between repolarization rate and the beating rate , FPD was firstly adjusted using the popular Bazett’s formula ( adjustedFPD = rawFPD / RR1/2 ) . We however found that , as expected , this formula was associated with an important over-correction thus supporting the need for an alternative adjustment formula according to Funck-Brentano et al . ( Funck-Brentano and Jaillon , 1993 ) ( Figure 2—figure supplement 3 , panel A ) . We thus plotted the FPD / inter-beat-interval pairs recorded at various beating rates under baseline conditions . The following equation with the same dimension as a Bazett’s formula: adjustedFPD = rawFPD / RRα , where alpha is a regression parameter , was applied to raw data in order to have a new estimate of the correction factor and results in appropriate corrections of FPD ( Figure 2—figure supplement 3 , panel B ) . Cells were considered as developing arrhythmias in response to sotalol as they presented ectopic beats or spontaneous beating rates became irregular . hiPSC-derived embryoid bodies were enzymatically dissociated into single cardiomyocytes and plated on matrigel-coated glass coverslips . The action potential of cardiomyocytes was assessed in the current-clamp mode using whole-cell patch-clamp technique with a HEKA EPC10 amplifier and Pulse/PulseFit software ( HEKA , Germany ) . Pipette solution consisted of: 110 mM potassium aspartate , 20 mM KCl , 10 mM HEPES , 1 mM MgCl2 , 0 . 1 mM ATP ( disodium salt ) , 5 mM ATP ( magnesium salt ) , 5 mM phosphocreatine ( disodium salt ) and 1 mM EGTA ( pH 7 . 2 ) . The external bath solution contained 140 mM NaCl , 5 mM KCl , 1 mM MgCl2 , 10 mM D-glucose , 1 mM CaCl2 and 10 mM HEPES ( pH 7 . 4 ) . hiPSC-CMs were perfused with either control or sotalol-containing Tyrode's solution for 5 min . The whole-cell condition was then reached . A hyperpolarization current ( −30 pA ) was applied to silence the spontaneous automaticity . The AP recording was then performed with a train of current pulse injection ( 1000 pA , 5 ms ) at 0 . 5 Hz steady pacing . The assay was performed at 37°C and finished within 30 min upon the drug perfusion . All analyses were performed using Prism 6 . 0 ( GraphPad , La Jolla , CA ) . Continuous data are presented as Mean ± SEM . A p-value≤0 . 05 was considered significant . Continuous variables were compared using non-parametric Mann-Whitney test and binary variables were tested using Chi-square or exact Fisher’s test as appropriate . For the in vivo investigations , we planned to identify 20 extreme responders to sotalol ( i . e . , 10 with high response defined by a deltaQTcf >35 ms and 10 with low response defined by a deltaQTcf <5 ms , therefore defining a minimal difference of 30 ms between groups ) . We then anticipated that this in vivo difference will be reproduced in vitro with a reduction by 33% . We thus determined that this sample size will give 80% power to demonstrate a difference of 20 ms between groups , with a standard error of 15 ms and an alpha-error risk of 0 . 05 ( nQuery Advisor version 4 . 0 ) . Responses to drugs , as measured in vitro , were normalized to baseline values and expressed as changes to baseline to account for inter-line variability . All recorded data were analyzed . A two-way analysis of variance with repeated measures was performed to analyze response to sotalol . Sotalol concentrations define the first factor and the sensitivity group the second factor . Individual comparisons were performed when the overall analysis was significant . Dose response curves were built to estimate EC50 using the Hill equation .	Common medications can disturb the electrical signals that cause the heart to beat , potentially resulting in sudden death . Many of the drugs that have these “cardiotoxic” effects were not designed to affect the heart , and include anti-allergenics and anti-vomiting drugs . In general , only a small proportion of individuals treated with these drugs will be at risk of fatal side effects; this risk variation is thought to be due to genetic differences . If these people could be reliably identified , the drugs could be used to treat others who will not develop cardiotoxic reactions , but it is difficult to predict the effect a drug will have on the beating of the heart . Stillitano , Hansen et al . have now investigated whether skin cells can be used to predict an individual’s likelihood of developing cardiotoxic side effects . Skin cells can be reprogrammed to form pluripotent stem cells , which have the ability to develop into any of the cell types in the adult body – including heart muscle cells . The effects of drugs could then be tested on these artificially created heart cells , yet it is not clear whether these effects would be the same as those seen in actual heart cells Stillitano , Hansen et al . created heart cells from skin samples collected from many different people and treated the cells with a drug that affects the rhythm of the heart . Some of the cells came from people whose heart rhythm is strongly affected by the drug , and others came from people whose heart rhythm is barely altered . The response of the lab-grown cells was closely related to whether the cells came from a person who was susceptible to the effects of the drug . Further investigation revealed that the genes that are important for maintaining a regular heartbeat differ in people who experience strong cardiotoxic side effects from those that do not . Overall , the results presented by Stillitano , Hansen et al . support the idea that induced pluripotent stem cells could be used to predict an individual’s risk of developing cardiotoxic reactions . Further work is now needed to develop this approach .	[ "Abstract", "Introduction", "Results", "Discussion", "Material", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine", "medicine" ]	2017	Modeling susceptibility to drug-induced long QT with a panel of subject-specific induced pluripotent stem cells
Evolution is at the core of the impending antibiotic crisis . Sustainable therapy must thus account for the adaptive potential of pathogens . One option is to exploit evolutionary trade-offs , like collateral sensitivity , where evolved resistance to one antibiotic causes hypersensitivity to another one . To date , the evolutionary stability and thus clinical utility of this trade-off is unclear . We performed a critical experimental test on this key requirement , using evolution experiments with Pseudomonas aeruginosa , and identified three main outcomes: ( i ) bacteria commonly failed to counter hypersensitivity and went extinct; ( ii ) hypersensitivity sometimes converted into multidrug resistance; and ( iii ) resistance gains frequently caused re-sensitization to the previous drug , thereby maintaining the trade-off . Drug order affected the evolutionary outcome , most likely due to variation in the effect size of collateral sensitivity , epistasis among adaptive mutations , and fitness costs . Our finding of robust genetic trade-offs and drug-order effects can guide design of evolution-informed antibiotic therapy . Treatment of infectious diseases and cancer often fail because of the rapid evolution of drug resistance ( Bloemberg et al . , 2015; Davies and Davies , 2010; Gottesman , 2002; Zaretsky et al . , 2016 ) . Optimal therapy should thus anticipate how resistance to treatment evolves and exploit this knowledge to improve therapy ( Gatenby et al . , 2009; Imamovic and Sommer , 2013 ) . One promising strategy is based on evolved collateral sensitivity: the evolution of resistance against one drug A concomitantly causes hypersensitivity ( i . e . , collateral sensitivity ) to a second drug B ( Szybalski and Bryson , 1952 ) . If evolved collateral sensitivity is reciprocal , it can – in theory – trap the bacteria in a double bind , thereby preventing the emergence of multidrug resistance during treatment ( Baym et al . , 2016; Pál et al . , 2015; Roemhild and Schulenburg , 2019 ) . Recent large-scale studies have demonstrated that evolved collateral sensitivity is pervasive in laboratory strains and clinical isolates of distinct bacterial species ( Barbosa et al . , 2017; Imamovic et al . , 2018; Imamovic and Sommer , 2013; Jansen et al . , 2016; Jiao et al . , 2016; Lázár et al . , 2014; Lázár et al . , 2013; Oz et al . , 2014; Podnecky et al . , 2018 ) as well as cancer cells ( Dhawan et al . , 2017; Pluchino et al . , 2012; Shaw et al . , 2016; Zhao et al . , 2016 ) . More importantly , evolved collateral sensitivity can slow down resistance evolution during combination ( Barbosa et al . , 2018; Rodriguez de Evgrafov et al . , 2015; Munck et al . , 2014 ) and sequential therapy ( Kim et al . , 2014; Roemhild et al . , 2015 ) , and also limit the spread of plasmid-borne resistance genes ( Rosenkilde et al . , 2019 ) . Although collateral sensitivity appears to be pervasive , its utility for medical application is still dependent on several additional factors . Firstly , the evolution of collateral sensitivity should ideally be repeatable for a given set of conditions ( Nichol et al . , 2019 ) . This means that independent populations selected with the same drug should produce identical collateral effects when exposed to a second one . Such high repeatability is not always observed . Recent work even identified evolution of contrasting collateral effects ( i . e . , some populations with evolved collateral sensitivity and others with cross-resistance ) for different bacteria , including Pseudomonas aeruginosa ( Barbosa et al . , 2017 ) , Escherichia coli ( Nichol et al . , 2019; Oz et al . , 2014 ) , Enterococcus faecalis ( Maltas and Wood , 2019 ) , and a BCR-ABL leukemeia cell line ( Zhao et al . , 2016 ) . These patterns are likely due to the stochastic nature of mutations combined with alternative evolutionary paths to resistance against the first selective drug , subsequently causing distinct collateral effects against other drugs ( Barbosa et al . , 2017; Nichol et al . , 2019 ) . Secondly , the evolution of collateral sensitivity should ideally be repeatable across conditions , for example different population sizes . This is not always the case . For example , an antibiotic pair , which consistently produced collateral sensitivity in small Staphylococcus aureus populations ( e . g . , 106 ) , instead produced complete cross-resistance in large populations ( e . g . , 109 ) and thus an escape from the evolutionary constraint , most likely due to the higher likelihood of advantageous rare mutations under these conditions ( Jiao et al . , 2016 ) . A third and largely unexplored factor is that evolved collateral sensitivity and , hence , the resistance trade-off should be stable across time . This implies that bacteria either cannot evolve to overcome collateral sensitivity and thus die out , or , if they achieve resistance to the new drug B , they should concomitantly be re-sensitized to the original drug A . Two recent studies , both with different main research objectives , yielded some insight into this topic . One example was focused on historical contingency during antibiotic resistance evolution of P . aeruginosa ( Yen and Papin , 2017 ) . As part of the results , the authors identified lineages with evolved resistance against ciprofloxacin that simultaneously showed increased sensitivity to piperacillin and tobramycin . The reverse pattern ( i . e . , evolved high resistance to either piperacillin or tobramycin and increased sensitivity to ciprofloxacin ) was not observed and , thus , this case represents an example of uni-directional collateral sensitivity . The subsequent exposure of the ciprofloxacin-resistance lineages to either piperacillin or tobramycin led to the evolution of resistance against these two antibiotics and substantial ( yet not complete ) re-sensitization to ciprofloxacin . The second study focused on evolving E . coli populations in a morbidostat , in which the bacteria were exposed to repeated switches between two drugs ( Yoshida et al . , 2017 ) . The evolution of multidrug resistance was only prevented in the two treatments with polymyxin B that were also characterized by evolved collateral sensitivity , although again only in one direction ( Yoshida et al . , 2017 ) . To date , the general relevance of this third factor is still unclear , especially for conditions when collateral sensitivity is reciprocal and when the evolving populations are also allowed to go extinct ( i . e . , they cannot overcome the evolutionary trade-off ) . Here , we specifically tested the potential of the model pathogen P . aeruginosa to escape reciprocal collateral sensitivity through de-novo evolution . We focused on the first switch between two drugs , because the evolutionary dynamics after this first switch will reveal the ability of the bacteria to adapt to the second drug , against which they evolved sensitivity , and , if so , whether this causes re-sensitization to the first drug . These two aspects are key criteria for applicability of a treatment strategy that exploits evolved collateral sensitivities . Our analysis is based on a two-step evolution experiment . Bacteria first evolved resistance against a first drug A and concomitant sensitivity against a second drug B . Thereafter , bacteria were subjected to a second evolution experiment , during which they were allowed to adapt to the second drug B , either alone or additionally in the presence of the first drug A . Phenotypic characterization of the evolved bacteria was combined with genomic and functional genetic analyses , in order to determine the exact targets of selection under these conditions . We finally validated the findings made by performing further , independent sets of similar two-step evolution experiments . The imposed antibiotic selection frequently caused population extinction ( Figure 2 , Figure 2—source data 1 ) , even though sublethal drug concentrations were used . Extinction events occurred significantly more often when selection for the original resistance was maintained by the presence of both drugs ( extinction in constrained treatments vs . only one drug , χ2 = 12 . 9 , df = 1 , p<0 . 0001; Figure 2 ) . In treatments with only the second drug B , extinction occurred significantly more often under strong , but not mild concentration increases ( strong vs . mild increases in unconstrained environments , χ2 = 5 . 5 , df = 1 , p=0 . 019 ) . Drug switches with the antibiotic pair STR/PIT was particularly successful , with 33 extinction events ( ~51% , Figure 2 ) . The results differed for the CAR/GEN pair , which produced only seven extinctions ( ~10% ) , all restricted to one drug order , GEN>CAR , suggesting asymmetry in the ability to counter collateral sensitivity . The observed differences in the extinction levels of the pairs considered here are unlikely to be the result of differences in combined synergy , since both combinations ( PIT/STR and CAR/GEN ) are synergistic against P . aeruginosa ( Barbosa et al . , 2018 ) . From this , we conclude that strong genetic constraints against an evolutionary response to collateral sensitivity caused frequent population extinctions for STR/PIT switches , whereas evolutionary rescue was possible for the GEN/CAR pair , although influenced by drug identity and order . We subsequently focused our analysis of the evolutionary dynamics on the surviving populations of the CAR/GEN pair and identified rapid adaptive responses , especially when not constrained by the presence of the two drugs ( Figure 3 ) . We measured bacterial adaptation using relative biomass ( see Materials and methods and Roemhild et al . , 2018 ) and found it to have increased in all surviving populations ( Figure 3a and b , Supplementary file 1–Figure 3–supplementary table 1 , Figure 3—source data 1 ) . For both drug orders , the increase was significantly slower in the constrained treatments , and , to a lesser extent , for the strong concentration increases ( Figure 3a and b ) . Consistent with the asymmetry in extinction , the CAR>GEN switch ( with no extinction ) maintained a high relative biomass across time , while the reverse direction GEN>CAR ( with extinction ) produced lower relative biomass levels . These results indicate that P . aeruginosa can evolve resistance against a drug , to which it had previously shown hypersensitivity , and that such evolutionary rescue is favored for the suboptimal switch . For the STR/PIT pair , we generally obtained similar results ( Figure 3—figure supplement 1a and b , Figure 3—source data 1 ) ; yet because of the few surviving populations and high variation among these rare survivors , the results remained inconclusive . We thus continued to focus on the evolved populations for the CAR/GEN pair and asked how the new adaptation influenced the original drug resistances . Adaptation in the surviving populations of the CAR/GEN pair caused multidrug resistance in the suboptimal switch , but re-sensitization to similar levels to those of the PA14 ancestor ( Figure 3—figure supplement 2 , Figure 3—figure supplement 2—source datas 1 and 2 ) in the alternative switch ( Figure 3c and d , Figure 3—source data 2 ) . In detail , all surviving populations significantly increased resistance against the second drug ( IC90-fold change between two and >64 times that of the starting populations; Figure 3c and d , Supplementary file 1–Figure 3–supplementary tables 2 and 3 , Figure 3—source data 2 ) – in agreement with the recorded biomass dynamics . In the suboptimal switch , CAR>GEN , all populations maintained their original resistance , thereby yielding bacteria with multidrug resistance . This was different for the alternative direction GEN>CAR , where the original resistance was only maintained when both drugs were present in combination ( constrained environments ) . Only under unconstrained evolution , we observed cases of significant re-sensitization to the first drug . We conclude that drug order can determine treatment efficacy , enhance or minimize multidrug resistance , and , in specific cases , lead to a re-sensitization towards the first drug in the surviving populations , as required for applicability of collateral sensitivity cycling ( Imamovic and Sommer , 2013 ) . We hypothesized that the contrasting evolutionary outcomes in constrained versus unconstrained treatments of the GEN>CAR switch were caused by an additional trade-off , in this case between drug resistance and growth rate . We obtained a proxy for such a general trade-off , which is comparable across the distinct antibiotic treatments , by measuring maximum exponential growth rate under drug-free conditions and standardizing it against the corresponding growth rate for the ancestral PA14 strain . Even though measured in drug-free environments , a possible reduction in growth rates may still indicate a general growth constraint or adaptation trade-off , which is also relevant under other conditions ( i . e . , under antibiotic exposure ) . The starting clones for the second evolution experiment indeed showed significantly impaired growth rates under drug-free conditions , with up to 50% reductions relative to the ancestor ( Barbosa et al . , 2017 ) . As a consequence , selection may have favored variants for which both trade-offs ( i . e . , the collateral sensitivity and also the general adaptation trade-offs ) were ameliorated during evolution . For the GEN>CAR switch , we indeed found a significant increase in growth rate relative to the wild type PA14 in the unconstrained treatments ( Figure 3f , Figure 3—source data 3 ) . Constrained populations for this particular switch still showed a significantly reduced growth rate relative to PA14 , with however significantly improved values relative to the starting population ( Figure 3f , Figure 3—source data 3 ) . The alternative switch did not show similar variations , mainly due to the fact that the costs of the initial population were not as high ( Figure 3e , Figure 3—source data 3 ) . Altogether , this data suggests that selection under the GEN>CAR unconstrained treatments provided the dual advantage of reversing two previously acquired evolutionary trade-offs , namely hypersensitivity to a second drug and increases in growth rate . Thus , re-sensitization could have been favored over multidrug resistance because of the associated adaptation trade-off that can ultimately influence treatment outcome upon collateral sensitivity switches . We used population genomic analysis to characterize specific functional changes that were likely targeted by antibiotic selection and allowed populations to survive the second evolution experiment for the CAR/GEN pair ( Figure 4 , Figure 4—source data 1 ) . In particular , we sequenced whole genomes of the resistant starting clones from the beginning and 21 surviving populations from the end of the second evolution experiment . Our results reveal that the evolution of multidrug resistance in the suboptimal switch CAR>GEN can be explained by the sequential fixation of mutations including , under unconstrained conditions , those in ptsP ( Figure 4a , Figure 4—source data 1 ) , a main component of the global regulatory system of ABC transporters and other virulence factors ( Feinbaum et al . , 2012 ) . Similarly , under constrained conditions , we found mutations in the NADH-dehydrogenase genes nuoD or nuoG ( Figure 4a , Figure 4—source data 1 ) , which are known to influence proton motive force and resistance against aminoglycosides upon inactivation ( El'Garch et al . , 2007 ) . For the more effective switch GEN>CAR , multidrug resistance in the constrained treatments coincided with mutations in mexR , phoQ , and cpxS , an independent regulator of MexAB-OprM ( Li et al . , 2016 ) and two-component regulators involved in aminoglycoside resistance ( Gooderham and Hancock , 2009 ) and envelope stress response ( Roemhild et al . , 2018 ) , respectively . The re-sensitization towards the first drug in the unconstrained GEN>CAR treatments was associated with two main types of mutational changes at high frequencies across several replicates , including ( i ) mutations in nalC and nalD that upregulate the expression of the multidrug-efflux system MexAB-OprM in P . aeruginosa ( Li et al . , 2016 ) ; and ( ii ) large deletions in pmrB , which is part of a two-component regulatory system ( Figure 4b , Figure 4—source data 1 ) . Mutations in nalC were previously shown to mediate both resistance to CAR and hypersensitivity to GEN ( Barbosa et al . , 2017 ) . Thus , re-sensitization to GEN may be caused by antagonistic pleiotropy of nalC mutations that override the resistance of the original pmrB mutation ( Figure 4 , Figure 4—source data 1 ) . In addition , there may be epistasis between the two functional modules . A complementary mechanism for re-sensitization against GEN is the re-mutation of pmrB ( Figure 4 , Figure 4—source data 1 ) . In three cases nalC mutations coincided with mutations in pmrB , including two deletions of 17 and 225 base pairs . Whilst the original SNP in pmrB altered gene function , the latter deletions may have suppressed the expression of the original SNP by pseudogenizing the gene . We conclude that mutations in the nalC or nalD regulators of the MexAB-OprM pump , sometimes in combination with follow up mutations in pmrB are likely to account for the re-sensitization towards the first drug GEN . We next investigated whether epistasis between the two functional modules of efflux regulation ( MexAB-OprM regulation by nalC or nalD ) and surface charge modification ( pmrB ) may have contributed to re-sensitization using functional genetic analysis . The respective single and double mutations were re-constructed in the common ancestral background of PA14 ( see Materials and methods for the specific mutations ) and changes in resistance against CAR and GEN were measured using fold-change of minimal inhibitory concentrations ( MIC , Figure 4c and d , Figure 4—source data 2 ) . On CAR , the pmrB mutant had half of the MIC of PA14 ( confirming collateral sensitivity ) , whilst nalC and nalD mutants had increased resistance to CAR . The double mutants had lower MIC on CAR than the nalC and nalD single mutants ( Figure 4c , Figure 4—source data 2 ) . The extent of MIC changes in the double mutants corresponded to the product of the individual effects in the respective single mutants , thus indicating an additive interaction among mutations on CAR . On GEN , however , the double mutants had substantially lower MICs than expected from the single mutants ( Figure 4d , Figure 4—source data 2 ) , strongly suggesting negative epistasis . In detail , GEN-resistance relative to PA14 was 0 . 4x for nalC ( collateral sensitivity ) , 1x for nalD , and 57x for pmrB ( Figure 4d ) . The pmrB , nalD double mutant had 3x lower MIC to GEN than expected from the individual effects . The pmrB , nalC double mutant had >30 x lower MIC to GEN than expected from the individual effects , resulting in greater sensitivity than PA14 ( Figure 4d , Figure 4—source data 2 ) . Altogether , we conclude that re-sensitization to GEN is mediated by antagonistic pleiotropy and negative epistasis . To validate and generalize our findings , we repeated evolution experiments for 14 cases of collateral sensitivity , using a total of 38 distinct resistant populations ( Figure 5a; Figure 5—source data 1 ) , obtained from independent biological replicates of our previous study ( Barbosa et al . , 2017 ) . We initiated experiments with the available populations ( Barbosa et al . , 2017 ) , rather than clones , and applied a less severe bottleneck of 107 cells . Therefore , any adaptative changes may not exclusively rely on de-novo mutations but could also result from available genetic diversity , thus reflecting a more general scenario for evolutionary adaptation to collateral sensitivity . The 14 cases of collateral sensitivity included the same four examples tested above and 10 additional cases , encompassing both reciprocal and also uni-directional collateral sensitivities . The effect size of collateral sensitivity , measured as fold-IC75 , differed substantially among replicates ( Figure 5a , Figure 5—figure supplement 1; Figure 5—source data 1 ) . In the current study , we now subjected the total of 38 resistant populations to either strong or strong+constrained increases of the second drug ( Figure 5a ) and assessed treatment outcome by measuring population extinction rates and changes in resistance profiles of the surviving populations ( Supplementary file 1–Figure 5-supplementary tables 1-2 , Figure 5—figure supplement 1—source data 1 ) . Our validation experiment now demonstrated that extinction events were frequent and occurred significantly more often when selection for the original resistance was maintained by the presence of both drugs ( extinction in constrained vs . strong environments , χ2 = 16 . 204 , df = 1 , p<0 . 0001; Figure 5b; Figure 5—source data 2 ) . Extinction rates showed pronounced variation by drug identity and order . For example , CIP-resistant populations all survived when challenged with either of the aminoglycosides ( STR , GEN ) , whereas more than 50% of these populations went extinct when exposed to β-lactams ( CAR , PIT , Figure 5b ) . We specifically assessed the importance of drug target , collateral sensitivity effect size ( fold-IC75 ) , and relative growth rate under drug-free conditions , using a generalized linear model ( GLM; Figure 6—source data 1 ) . We extracted data on the relative exponential growth rate of the resistant populations in drug-free media from our previous publication ( Barbosa et al . , 2017 ) . Our analysis revealed that variation in extinction was significantly associated with only the molecular target of the second antibiotic ( GLM , combined extinction , n = 14 drug switches , F = 11 . 016 , p=0 . 011; Supplementary file 1–Figure 6–supplementary table 1 , Figure 6—source data 1 ) , but not with growth rate , extent of collateral sensitivity , or target of the first antibiotic . Extinction frequencies were on average twice as high in the six treatments that switched to β-lactam antibiotics that target the cell wall , as compared to the eight treatments that switched to aminoglycoside antibiotics that target the ribosome ( Figure 6a ) . It thus appears that extinctions in our experiments are stochastic events that are mainly influenced by the target of the current antibiotic , but not the preceding evolutionary conditions . Taken together , we conclude that switching from one to two drugs generally increases treatment potency and that this is influenced by drug order and identity . The preferred drug order is to use a β-lactam as the second antibiotic , which is consistent with the detailed dynamics on the clonal level . Resistance gains to drug B were frequently associated with re-sensitization to the original drug A in the unconstrained treatments , while resistance to the new drug B increased in all 14 tested cases ( leading to changes of 2 up to >64 times the IC90 of the starting populations; Supplementary file 1–Figure 5–supplementary tables 1 and 2 ) . In detail , we observed significant antibiotic re-sensitization for 5 of the 14 cases ( 36% , Figure 5b , Figure 5—source data 3 , Supplementary file 1–Figure 5–supplementary table 1 , Figure 5—source data 1; Note that the single available replicate for STR>PIT showed a similar trend but could not be evaluated statistically ) . Within reciprocal collateral sensitivity , re-sensitization was not always observed in both directions , as illustrated by CAR/STR and STR/PIT . Re-sensitization did not occur in the constrained treatments . We noticed some differences between evolutionary outcomes of drug switches in clonal and population experiments . The resistance changes showed an overall categorical agreement of 75% ( 9/12 comparisons; no comparisons for STR>PIT , as this switch was tested with a single replicate ) . Yet , clonal evolution produced re-sensitization for only one sequential treatment direction for CAR/GEN , whereas population evolution showed re-sensitization for both directions . That re-sensitization occurred in both directions at the population level is expected if there was a residual frequency of cells lacking the collateral-sensitivity mutation in the inoculum . The population data further highlighted a lack of re-sensitization for PIT>STR , where we previously observed re-sensitization on the clonal level , albeit not at both concentration regimes . In addition , there is a mild significant increase of resistance to GEN in the strong constrained treatments in GEN>CAR , compared to no change . Overall our results indicate that re-sensitization is frequent for populations with collateral sensitivity , but not imperative ( Figure 5b ) . What are the drivers of the observed resistance gains and re-sensitization ? Our statistical analysis revealed that the extent of re-sensitization in the unconstrained treatments was significantly associated to equal degrees with the target of drug A ( GLM , resistance against drug A , n = 14 , F = 15 . 27 , p=0 . 0019; Supplementary file 1–Figure 6–supplementary table 2 , Figure 6—source data 2 ) , the effect size of collateral sensitivity ( F = 15 . 91 , p=0 . 004 ) , and growth rate in drug-free media ( F = 18 . 38 , p=0 . 0027 ) , while target of drug B did not have a significant influence . Re-sensitization was stronger in drug switches that started with an aminoglycoside , as compared to β-lactam , and fluoroquinolone ( Figure 6b ) . This result is consistent with the drug-dependent negative epistasis ( Figure 4 ) , and validates the relative stability of collateral sensitivity switches from aminoglycosides to β -lactams . Growth rate in drug-free conditions and collateral sensitivity effect size showed small negative effects , indicating that they decelerate re-sensitization . In contrast , the extent of resistance gains in unconstrained treatments was not associated with drug targets of either drug A or B , but instead significantly associated with both the initial effect size of collateral sensitivity ( GLM , resistance against drug B , n = 14 , F = 6 . 71 , p=0 . 0321; Supplementary file 1-Figure 6–supplementary table 2 , Figure 6—source data 2 ) , and the initial growth rate of the resistant population in drug-free media ( GLM , resistance against drug B , n = 14 , F = 7 . 90 , p=0 . 0228 ) . Collateral sensitivity effect size showed a negative association with the resistance gains to drug B ( Figure 6c ) . High sensitivity thus appears to be prone to more rapid resistance gains , possibly because it opens a fitness space for rapid adaptation . The population resistance changes against drug B were generally larger than the initial collateral sensitivity ( Figure 5b; Figure 5—source data 3 ) ; Supplementary file 1-Figure 5-supplementary tables 1-2; evolved lineages showed higher resistance than the PA14 wild type ) , indicating that they were not solely due to competition of pre-existing sub-populations , but involved new genetic adaptations . As observed above , the presence of a general adaptation trade-off ( measured as reduction in growth rate relative to ancestral PA14 in drug-free environments ) inhibited rapid adaptation ( Figure 6d ) . Taken together , our results at the population level suggest that the stability of collateral sensitivity switching is shaped predominantly by drug targets , but additionally influenced by the effect size of collateral sensitivity and a general adaptation trade-off . The most stable collateral sensitivity was identified for the switch from aminoglycoside to β-lactam . Within a particular drug switch , the highest stability was achieved for low collateral sensitivity effect sizes . Large adaptation trade-offs tended to inhibit evolutionary change , that is both desirable re-sensitization , and new resistance . The conclusions are consistent with the data obtained from the clonal evolution experiments . Collateral sensitivity is a pervasive feature of resistance evolution , but its potential for medical application is currently debated ( Nichol et al . , 2019; Podnecky et al . , 2018; Roemhild and Schulenburg , 2019 ) . Its promise as a treatment focus is dependent on the premise that the exploited trade-off is evolutionarily stable and cannot be easily overcome . As a consequence , it should either drive bacterial populations to extinction or minimize the emergence of multidrug resistance by re-sensitization to one of the antibiotics . We here tested the validity of these key predictions with the help of evolution experiments and the model pathogen P . aeruginosa . We found that the effective exploitation of evolved collateral sensitivity in sequential therapy is contingent on drug order and combination , collateral sensitivity effect size , general adaptation trade-offs , and also epistatic genetic interactions . Evolved reciprocal collateral sensitivity generally limited bacterial adaptation . Adaptation was particularly constrained in treatments that switched to a β-lactam , as reflected by the elevated extinction rates . The effect was strongest when the first antibiotic was maintained and a second was added . This finding may point to a promising , yet currently unexplored treatment strategy , namely single-drug therapy followed by combination therapy ( for example the addition of a β-lactam ) , that can maximize exploitation of the evolutionary trade-off . Yet , extinction rates were still high under unconstrained conditions , when drugs were replaced , and in spite of a relatively mild selection intensity . In the detailed analysis with the CAR/GEN drug pair , we observed higher extinction and slower growth improvements in strong , compared to mild dose increases . This finding is generally consistent with previous studies , performed in different context , in which narrowed mutation space upon fast environmental deterioration increased extinction frequencies ( Bell and Gonzalez , 2011; Lindsey et al . , 2013 ) . Interestingly , extinction rates are often not reported as an evolutionary outcome in related studies , possibly because of a different main focus of the study ( Yen and Papin , 2017 ) , or because extinction could not be recorded due to the particular experimental set-up ( i . e . , usage of a morbidostat; Yoshida et al . , 2017 ) . Considering that antimicrobial therapy usually aims at elimination of bacterial pathogens and extinction frequencies are known from previous evolution experiments to vary among treatment types ( Barbosa et al . , 2018; Hansen et al . , 2017; Roemhild et al . , 2018; Torella et al . , 2010 ) , their consideration should help us to refine our understanding of treatment efficacy . In the treatments that replaced drugs , evolutionary stability of the resistance trade-off was determined by drug order . In both datasets ( clonal and population-level evolution ) , we observed high relative stability for collateral sensitivity treatments switching from an aminoglycoside to a β-lactam . Our detailed analysis for the CAR/GEN pair at the clonal level demonstrates that this stability may be caused by slower adaptation ( Figure 3a ) , and efficient re-sensitization ( Figure 3c ) . Also , the slow adaptation rates may be the result of selection favoring the reversal of two evolutionary trade-offs ( Figures 3e and 6d ) together with a relative paucity of accessible resistance mutations . The latter is highlighted by several instances of re-mutation of pmrB , where the effect of the ancestral collateral-sensitivity SNP was countered by additional larger deletions , potentially leading to a loss of function of the same gene ( Figure 4b ) . The efficient re-sensitization that was observed during switches from an aminoglycoside to a β-lactam may be explained by pleiotropy and drug-specific negative epistasis ( Figure 4c and d ) . On the other hand , collateral sensitivity treatments that switch to aminoglycosides , tended to show lower stability , as reflected by the lower levels of extinction ( Figure 6a ) and the lack of re-sensitization ( Figure 6b ) at the population level . For instance , the collateral sensitivity treatments CAR>STR and CIP>STR were frequently countered by the evolving bacteria . These treatments hardly resulted in extinction but allowed for rapid loss of susceptibility to both drugs ( Figure 5b ) . While our data cannot fully explain the high instability , it appears that small values for a general adaptation trade-off ( i . e . , indicated by reduced growth rates in drug-free environments ) , which are also frequently observed for CIP resistance mutations ( Huseby et al . , 2017 ) , accelerated adaptation ( Figure 6d ) . Overall , our finding of strong variation in the stability of collateral sensitivity treatments in pathogenic P . aeruginosa indicate the importance for a careful evaluation of new treatment options . Our detailed analyses for GEN>CAR emphasize the importance of epistasis for the stability of collateral sensitivity . A recent publication confirmed that the expression of particular collateral sensitivity mutations strongly depended on the genetic background and could even cause opposite effects in the closely related species of E . coli and Salmonella enterica due to epistasis ( Apjok et al . , 2019 ) . Our work on the GEN>CAR switch now shows , that epistasis is of high importance also for the temporal stability of antibiotic sensitivity . Here , drug re-sensitization in the unconstrained treatments was likely dependent on negative epistasis among pleiotropic resistance mutations . Mutations in pmrB and the efflux regulators nalC and nalD interacted negatively with each other and caused a complete re-sensitization of bacteria that were previously resistant against GEN . While re-sensitization reliably occurred for the GEN>CAR treatment , it did not occur in the reverse case . Similar examples of antibiotic re-sensitization were previously reported for E . coli and P . aeruginosa , but these relied on different mechanisms . For E . coli , repeated alternation between two antibiotics led to re-sensitization as a consequence of clonal interference between variants in two genes , secD and/or basB ( Yoshida et al . , 2017 ) . The change between drugs prevented fixation of the competing variants , thus maintaining pleiotropic alleles and thereby the allele causing resistance to one drug and hypersensitivity to the other ( Yoshida et al . , 2017 ) . In the previous example for P . aeruginosa , hypersensitivity to a β-lactam depended on an expression imbalance of the MexAB-OprM and the MexEF-OprN efflux systems after exposure to a fluoroquinolone ( Maseda et al . , 2004; Sobel et al . , 2005; Yen and Papin , 2017 ) . Interestingly , partial re-sensitization against the aminoglycoside tobramycin was dependent on inducible resistance , a phenomenon mediated by the MexXY-OprM efflux pump , whereby expression , and consequently resistance , is induced by the presence of the drug , but then reverted after its removal ( Hocquet et al . , 2003; Yen and Papin , 2017 ) . We conclude that our finding of negative epistasis between pleiotropic resistance mutations is a previously unknown mechanism underlying re-sensitization . Whilst positive epistasis can substantially amplify resistance gains ( Wistrand-Yuen et al . , 2018 ) , negative epistasis can limit evolutionary trajectories ( Weinreich et al . , 2006 ) , thus possibly contributing to efficacy of treatment in our case . The mutations observed in this study are commonly associated with variants observed in clinical isolates , particularly in those obtained from cystic fibrosis patients ( Hancock and Speert , 2000; Jansen et al . , 2016; Marvig et al . , 2015; Tueffers et al . , 2019 ) . Both efflux regulators ( including nalC and nalD ) and two-component regulatory systems ( mainly pmrAB and phoQF ) were repeatedly reported to be associated with intermediate and highly resistant isolates of P . aeruginosa , E . coli , Acinetobacter baumannii and other pathogenic species against colistin and aminoglycosides ( Cao et al . , 2004; Gerson et al . , 2019; Sato et al . , 2018 ) . This overlap suggests that the negative epistasis between the genes involved in resistance against β-lactam and aminoglycosides observed here could also be encountered and exploited in clinical settings . We anticipate that the findings of our study could help to guide the design of sustainable antibiotic therapy that controls the infection , whilst reducing the emergence of multidrug resistance . In principle , the refined exploitation of collateral sensitivity could represent a promising addition to new evolution-informed treatment strategies , including as alternatives specific combination treatments ( Barbosa et al . , 2018; Chait et al . , 2007; Rodriguez de Evgrafov et al . , 2015; Gonzales et al . , 2015; Munck et al . , 2014 ) , fast sequential therapy ( Nichol et al . , 2015; Yoshida et al . , 2017 ) , or treatments utilizing negative hysteresis ( Roemhild et al . , 2018 ) . The success of this treatment strategy depends on several key factors . One critical prerequisite is that collateral sensitivity does evolve in response to treatment , which may vary depending on alternative evolutionary paths to resistance against the initially used drug A ( Barbosa et al . , 2017; Nichol et al . , 2019 ) . Our new data additionally suggest that treatment can be further optimized by switching from an aminoglycoside to a β-lactam and/or by focusing on drugs , for which resistance is associated with large general adaptation trade-offs , and/or by using drug combinations with small collateral sensitivity effect sizes . These new insights clearly warrant further evaluation , for example by using clinical isolates of P . aeruginosa and/or patient-related test conditions . All experiments were performed with P . aeruginosa UCBPP-PA14 ( Rahme et al . , 1995 ) and clones obtained from four antibiotic-resistant populations: CAR-10 , GEN-4 , PIT-1 and STR-2 ( Barbosa et al . , 2017 ) . The resistant populations were previously selected for high levels of resistance against protein synthesis inhibitors from the aminoglycoside family , gentamicin ( GEN; Carl Roth , Germany; Ref . HN09 . 1 ) and streptomycin ( STR; Sigma-Aldrich , USA; Ref . S6501-5G ) , or alternatively cell-wall synthesis inhibitors from the β-lactam family , carbenicillin ( CAR; Carl Roth , Germany; Ref . 6344 . 2 ) and piperacillin/tazobactam ( PIT; Sigma-Aldrich , USA; Refs . P8396-1G and T2820-10MG ) . Resistant clones were isolated by streaking the resistant populations on LB agar plates supplemented with antibiotics and picking single colonies after an overnight growth at 37°C . Antibiotic stocks were prepared according to manufacturer instructions and frozen in aliquots for single use . Evolution experiments and resistance measurements were performed in liquid M9 minimal media supplemented with glucose ( 2 g/l ) , citrate ( 0 . 5 g/l ) and casamino acids ( 1 g/l ) . The previously reported collateral sensitivity trade-off ( Barbosa et al . , 2017 ) was confirmed for this study , by measuring sensitivity of the resistant populations CAR-10 to GEN , GEN-4 to CAR , PIT-1 to STR , and STR-2 to PIT , in comparison to PA14 . Populations were grown to exponential phase , standardized by optical density at 600 nm ( OD600 = 0 . 08 ) , and inoculated into 96-well plates ( 100 µl volumes , 5 × 106 CFU/ml ) containing linear concentrations of antibiotics ( 10 concentrations , eight replicates each ) . Antibiotic concentrations were spatially randomized . Plates were incubated at 37 °C for 12 hr , after which growth was measured by OD600 with a BioTek plate reader . Antibiotic susceptibility was quantified from dose-response curves using the lowest concentration required to inhibit growth by a defined value compared to wild type growth in drug free medium . IC95 ( inhibitory concentration 95 ) refers to the smallest concentration required to inhibit growth by 95% . The metrics IC90 and IC75 refer to the concentrations that inhibit growth by 90% or 75% , respectively . Inhibitory concentrations were determined from the dose-response data using linear interpolation between the two closest OD600 values , as inferred with the approx function in the statistical environment R . In our case , the IC75 or IC90 metrics show higher accuracy and precision than the commonly used metric of the minimal inhibitory concentration ( MIC , equivalent to IC100 ) , because we inferred growth characteristics from OD600 measurements , which are subject to unfavorable signal-to-noise ratios close to OD values of zero and thus close to the IC100 condition . Please note that analysis of IC75 and IC90 values produced consistent results ( e . g . , Supplementary file 1-Figure 6-supplementary tables 1 and 2 ) . To test the evolutionary stability of reciprocal collateral sensitivity , we challenged clones from previously evolved resistant populations with increasing concentrations of new antibiotics against which the resistant populations showed hypersensitivity ( so called collateral sensitivity ) : CAR-10 with GEN , GEN-4 with CAR , PIT-1 with STR , and STR-2 with PIT . Stability was assessed with 12 day evolution experiments using a serial transfer protocol ( 100 µl batch cultures , 2% serial transfers every 12 hr; the starting population size for the different populations was approx . 106 CFU/ml ) , as previously described ( Roemhild et al . , 2018 ) . Each population was evaluated with eight replicate populations ( 4 clones x two technical replicates distributed in two plates: plate A and plate B ) for each of 5 treatment groups: ( i ) untreated controls; linearly increasing concentration of hypersensitive antibiotic to a low level ( ii ) or high level ( iii ) , without maintaining selection for previous resistance ( unconstrained evolution ) ; or linearly increasing concentration of hypersensitive antibiotic to a low level ( iv ) or high level ( v ) , with simultaneous selection for previous resistance ( constrained evolution ) . Concentration increases were started with defined initial inhibition levels of 50% ( IC50 ) and concluded when concentrations were above the IC95 of the hypersensitive strain ( mild increases ) or IC95 of the wild type PA14 strain ( strong increases ) , as specified in Supplementary file 1-Supplementary Table 1 . Antibiotic selection was applied in 96-well plates and population growth was monitored throughout treatment by continuous measurements of OD600 in 15 min intervals ( BioTek Instruments , USA; Ref . EON; 37 °C , 180 rpm double-orbital shaking ) . Extinction frequencies were determined at the end of the experiment by counting cases in which no growth was observed after an additional transfer to antibiotic-free media and 24 hr of incubation . Surviving evolved populations were frozen at −80 °C in 10% ( v/v ) DMSO , at the end of the experiment . The continuous measurements of optical density during treatment provided a detailed growth trajectory that accurately describes the dynamics of resistance emergence . Relative biomass was defined as total optical growth relative to untreated control treatments , and was calculated by the ratio of the areas under the time-OD curves of treated compared to untreated controls that are passaged in parallel , as previously described ( Roemhild et al . , 2018 ) . Resistance of evolved populations was measured for the respective antibiotic pairs ( GEN/CAR or STR/PIT ) , as described above , but using two-fold concentrations ( 1/4 to 8x the MIC of the starting clone ) . The respective starting clones of each evolved population served as controls and were measured in parallel . Resistance changes were quantified by subtracting the area under the dose-response curve of the evolved populations from that of the ancestral clones . Positive values indicate that the evolved lineages are more resistant than their ancestor , values close to zero indicate equivalent resistance levels , and negative values denote a loss of resistance . The cases of re-sensitization against GEN were validated by repeating the measurements , whereby the PA14 ancestor was included as an additional control ( Figure 3—figure supplement 2 ) . Maximum exponential growth rates of evolved and ancestral populations were calculated from growth curves in drug-free media , using a sliding window approach . For measurements , sample cultures were diluted 50x from early stationary phase into 96-well plates ( 100 µl total volume ) and growth was measured by OD600 every 15 min for 12 hr . Growth rate were calculated from log-transformed OD data for sliding windows of 1 hr , yielding two-peaked curves indicating initial growth on glucose and citrate . The reported values the maximum values of the first , larger peak . The values reported in Figure 3 are the changes of growth rate in evolved populations relative to their resistant ancestors . These values are taken as a proxy for a general adaptation trade-off , which is distinct to the collateral sensitivity trade-off . We re-sequenced whole genomes of 5 starting clones ( CAR-10 clone 2 , GEN-2 clones 1–4 ) , and 21 evolved populations ( all descendants of these clones from plate B , including five untreated evolved control populations and 16 populations adapted to different treatment conditions ) using samples from the end of the evolution experiments . Frozen material was thawed and grown in 10 ml of M9 minimal medium for 16–20 hr at 37 °C with constant shaking . Genomic DNA was extracted using a modified CTAB buffer protocol ( von der Schulenburg et al . , 2001 ) and sequenced at the Institute for Clinical Microbiology , Kiel University Hospital , using Illumina HiSeq paired-end technology with an insert size of 150 bp and 300x coverage . For the genomic analysis , we followed an established pipeline ( Jansen et al . , 2015 ) . Briefly , reads were trimmed with Trimmomatic ( Bolger et al . , 2014 ) , and mapped to the UCBPP-PA14 reference genome ( available at http://pseudomonas . com/strain/download ) using bwa and samtools ( Li and Durbin , 2010; Li et al . , 2009 ) . We used MarkDuplicates in Picardtools to remove duplicated regions for single nucleotide polymorphisms ( SNPs ) and structural variants ( SVs ) . To call SNPs and small SV we employed both heuristic and frequentist methods , only for variants above a threshold frequency of 0 . 1 and base quality above 20 , using respectively VarScan and SNVer ( Wei et al . , 2011 ) . Larger SVs were detected by Pindel and CNVnator ( Abyzov et al . , 2011; Ye et al . , 2009; Ye et al . , 2009 ) . Variants were annotated using snpEFF ( Cingolani et al . , 2012 ) , DAVID , and the Pseudomonas database ( http://pseudomonas . com ) . Variants detected in the untreated evolved populations were removed from all other populations and analyses as these likely reflect adaptation to the lab media and not treatment . The fasta files of all sequenced populations here are available from NCBI under the BioProject number: PRJNA524114 . To understand re-sensitization , we analyzed candidate mutations from the GEN > CAR switch . The nalD mutation 1551588G > T ( resulting in amino acid change p . T11N , as observed in replicate populations b24_G8 , b24_D9 , and b24_A9 ) was introduced into the PA14 genetic background using a scar-free recombination method ( Trebosc et al . , 2016 ) . The same techniques were previously used to construct the mutants nalC ( deletion 1391016–1391574 ) and pmrB ( 5637090T > A , resulting in amino acid change p . V136E ) in the PA14 ancestor background ( Barbosa et al . , 2017 ) . Based on these mutants and with the same techniques , we constructed the double mutants pmrB , nalD ( pmrB p . V136E + nalD p . T11N ) , and pmrB , nalC ( pmrB p . V136E + nalC deletion c . 49–249 , as observed in population b24_F7 ) . Genetic manipulation and confirmation by sequencing was performed by BioVersys AG , Hochbergerstrasse 60 c , CH-4057 Basel , Switzerland . Resistance of constructed mutant strains was measured in direct comparison to wildtype PA14 , as described above . Relative fold-changes in MIC were calculated from dose-response curves . The expected relative resistance of the double mutants was calculated by multiplication of the mutation’s individual effects , as previously described ( Wong , 2017 ) . For example , if mutation A conferred a 2-fold increase in resistance and mutation B conferred a 4-fold increase of resistance , the expected resistance of the double mutant AB would be 2 × 4 = 8 . A deviation from this null model indicates epistasis , which can be either positive ( greater resistance than expected ) or negative ( lesser resistance than expected ) . The evolutionary stability of collateral sensitivity in genetically diverse populations was investigated by using the same general procedure as described above ( section ‘Experimental evolution initiated with resistant clones’ ) , but using an inoculum of roughly 107 cells instead of a single clone . This experiment was reduced to the treatments groups ‘strong’ and ‘strong+constrained’ , and performed for a total of 38 ( 6 × 6 + 2 ) different resistant starting populations from our previous publication ( Barbosa et al . , 2017 ) . Six replicate populations each from previous evolution for resistance to CAR , GEN , STR , PIT , ciprofloxacin ( CIP ) and cefsulodin ( CEF ) , were challenged with increasing concentrations of antibiotic against which they showed collateral sensitivity . The population were each evolved against two new antibiotics . Due to variation in collateral sensitivity profiles among replicate populations , a seventh population had to be selected for CAR and PIT to assemble a set of 6 collaterally-sensitive populations for each of the switching directions . Only one STR-resistant population showed collateral sensitivity to PIT so that this test was conducted with a single replicate . CIP-resistant populations showed general collateral sensitivity and were tested against four new antibiotics . In total , we thus arrived at 79 ( 14 × 6 - 5 ) evolutionary switches to collateral sensitivity . To test for association of predictive factors with evolutionary stability , we used a generalized linear model ( GLM ) analysis , because it allows us to combine an evaluation of both categorical and continuous predictors and to assess the influence of each factor in consideration of the contributions of the other factors ( which is not possible when using for example correlation analysis ) . For our analysis , we used the functions lm and anova in the statistical environment R and the main effects model: response ~target drug A + target drug B + collateral sensitivity effect size + drug-free relative growth rate .	Over time bacteria can undergo a number of genetic mutations that allow them to evolve in response to changes in their surrounding environment . This process of ‘bacterial evolution’ is one of the major causes of antibiotics resistance , whereby disease-causing microorganisms become resistant to multiple drugs and can no longer be destroyed using antibiotic treatment . However , when bacteria become resistant to a drug this can result in an evolutionary trade-off known as ‘collateral sensitivity’ – when evolving resistance to one drug causes bacteria to gain increased sensitivity to another . Now , Barbosa , Roemhild et al . have investigated whether this evolutionary trade-off could be exploited to tackle the antibiotic crisis and prevent bacteria adapting to different treatments . If this evolutionary trade-off is to be used medically , it must be stable long enough for the bacteria population to either become extinct , or less able to evolve multi-drug resistance . To test how stable collateral sensitivity is over time , Barbosa , Roemhild et al . studied the bacterium Pseudomonas aeruginosa which is known to evolve collateral sensitivity to certain drug treatments . P . aeruginosa were subjected to two rounds of evolution: first , bacteria were evolved to resist ‘Drug A’ and at the same time became more sensitive to another drug , ‘Drug B’ . The bacteria were then allowed to adapt to Drug B either alone or in the presence of Drug A . These evolutionary experiments revealed that the following factors affected the stability of the trade-off: the molecular structure of the antibiotic bacteria evolved sensitivity to , the strength of the original evolutionary trade-off ( i . e . how sensitive bacteria became ) , the order drugs were administrated , and whether resistance came at a large fitness cost ( i . e . when the genetic mutations promoting resistance affect bacteria’s ability to replicate and survive in normal conditions ) . According to the World Health Organization , P . aeruginosa is the second most problematic multi-drug resistant bacteria . The data collected in this study could therefore be used to develop a new antibiotic therapeutic strategy for fighting this bacterium , as well as other microbes which are resistant to multiple drugs .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "microbiology", "and", "infectious", "disease" ]	2019	Evolutionary stability of collateral sensitivity to antibiotics in the model pathogen Pseudomonas aeruginosa
RNA-binding proteins contribute to the formation of ribonucleoprotein ( RNP ) granules by phase transition , but regulatory mechanisms are not fully understood . Conserved fission yeast NDR ( Nuclear Dbf2-Related ) kinase Orb6 governs cell morphogenesis in part by spatially controlling Cdc42 GTPase . Here we describe a novel , independent function for Orb6 kinase in negatively regulating the recruitment of RNA-binding protein Sts5 into RNPs to promote polarized cell growth . We find that Orb6 kinase inhibits Sts5 recruitment into granules , its association with processing ( P ) bodies , and degradation of Sts5-bound mRNAs by promoting Sts5 interaction with 14-3-3 protein Rad24 . Many Sts5-bound mRNAs encode essential factors for polarized cell growth , and Orb6 kinase spatially and temporally controls the extent of Sts5 granule formation . Disruption of this control system affects cell morphology and alters the pattern of polarized cell growth , revealing a role for Orb6 kinase in the spatial control of translational repression that enables normal cell morphogenesis . Many cellular processes , such as cell morphogenesis , migration , and asymmetric cell division , require eukaryotic cells to alter polarity and growth patterns ( Lalli , 2014; Tahirovic and Bradke , 2009; Woodham and Machesky , 2014; Knoblich , 2008 ) . Understanding the conserved mechanisms by which cells tune polarized cell growth has implications for studies of neuronal cell morphogenesis , neurodegenerative diseases , stem cell differentiation , and cancer ( Yoshimura et al . , 2006; Tahirovic and Bradke , 2009; Yamashita et al . , 2010; Tanos and Rodriguez-Boulan , 2008 ) . The fission yeast Schizosaccharomyces pombe is an excellent model system to study cell morphogenesis and growth because cells have a defined cylindrical shape that enables straightforward evaluation of changes in growth and polarity . Under exponential growth conditions , fission yeast cells display a paradigmatic pattern of cell growth , growing in a monopolar fashion during early interphase and activating bipolar growth at the new cell tip once a minimal cell length has been achieved ( Mitchison and Nurse , 1985 ) . Further , S . pombe displays a distinct morphological response to nutrient deprivation , which causes cells to divide at a shorter cell length and grow in a monopolar fashion ( Su et al . , 1996; Yanagida , 2009; Yanagida et al . , 2011 ) . The NDR ( Nuclear Dbf2-Related ) kinase family with roles in cell morphogenesis , cell growth and proliferation , mitosis , and development , is highly conserved in cells ranging from yeast to mammalian neurons ( Verde et al . , 1995; Verde et al . , 1998; Zinn , 2004; Hergovich et al . , 2006 ) . In humans , this subset of the AGC kinase group comprises NDR1 and NDR2 and the closely related kinases LATS1 ( large tumor suppressor 1 ) and LATS2 ( Hergovich et al . , 2006 ) , which function downstream of the MST/Hippo kinases ( Meng et al . , 2016 ) . While LATS1 and LATS2 kinases are central to the Hippo pathway that plays a role in organ size and tumor suppression , dysregulation of NDR kinases has been implicated in cancers such as progressive ductal cell carcinoma , melanoma , non–small-cell lung cancer , and T-cell lymphoma ( Adeyinka et al . , 2002; Millward et al . , 1998; Hauschild et al . , 1999; Ross et al . , 2000; Cornils et al . , 2010 ) . In addition to their link to cancer , NDR kinases function also in neuronal growth and differentiation , dendritic branching , and dendritic tiling , and have been implicated in memory and fear conditioning ( Emoto et al . , 2004; Zallen et al . , 2000; Koike-Kumagai et al . , 2009; Stork et al . , 2004 ) . Recent work has shown that mammalian NDR1 and NDR2 promote polarity in neurons upstream of the polarity protein Par3 ( Yang et al . , 2014 ) . However , the mechanisms by which NDR kinases control cell growth and polarity are not fully understood . The fission yeast NDR kinase Orb6 is a central component of the conserved morphogenesis ( MOR ) regulatory network ( Hergovich et al . , 2006 ) . We previously showed that NDR kinase Orb6 has a role in the establishment of cell polarity and the control of polarized cell growth ( Verde et al . , 1995; Verde et al . , 1998 ) . Orb6 kinase regulates cell polarity , in part , by spatially controlling conserved GTPase Cdc42 ( Das et al . , 2009 ) , via inhibitory phosphorylation of Cdc42 guanine exchange factor ( GEF ) Gef1 ( Das et al . , 2015 ) . Here , we describe a novel role for Orb6 kinase , genetically separable from its control of the Cdc42 pathway , in promoting polarized cell growth by inhibiting translational repression . Translational repression , carried out in part by the assembly of cytoplasmic granules of ribonucleoprotein particles ( RNPs ) , is a quick and reversible cellular strategy for inhibiting cell growth in response to stress , such as nutritional deprivation , oxidative stress , or osmotic stress ( Coller and Parker , 2005; Decker and Parker , 2012; Kedersha et al . , 2005; Jud et al . , 2008 ) . P-bodies , stress granules , and other RNPs such as neuronal transport granules and germ granules play important roles in mRNA regulation with implications for human diseases such as ALS , frontotemporal lobar degeneration , and viral infection ( Ramaswami et al . , 2013; Chahar et al . , 2013 ) . P-bodies in particular contain mRNA decay machinery and serve as sites of storage or degradation for mRNAs during times of cellular stress ( Decker and Parker , 2012 ) . In this work , we describe a novel mechanism whereby NDR kinase Orb6 negatively regulates the recruitment of mRNA-binding protein Sts5 into RNP particles and Sts5 localization to P-bodies at least in part by promoting Sts5 interaction with 14-3-3 protein Rad24 . This mechanism of control prevents the degradation of mRNAs encoding proteins important for polarized cell growth and cell morphogenesis during exponential cell growth , and promotes morphological adaptation during nutritional stress . We observed that loss of Orb6 kinase activity by chemical inhibition of analog-sensitive Orb6-as2 kinase by the ATP analogue 1-NA-PP1 leads to cell separation defects ( Figure 1A , c; B ) and slow growth , in addition to polarity defects ( Das et al . , 2009; Das et al . , 2015 ) . By complementation screening of the orb6-as2 allele with mutants of other orb genes ( Snell and Nurse , 1994; Verde et al . , 1995 ) , we found that sts5 mutants ( allelic to orb4; see Figure 1—figure supplement 1A ) suppress the cell-separation defect associated with chemical inhibition of Orb6-as2 kinase ( Figure 1A , d; Figure 1B; Verde et al . , 1995 ) as compared to control cells ( Figure 1A , c; Figure 1B ) . sts5 encodes an mRNA-binding protein with significant sequence homology to Ribonuclease II ( RNB ) –domain and Ribonuclease R–domain proteins ( Toda et al . , 1996; Jansen et al . , 2009 ) . Closest homologues of Sts5 include S . cerevisiae Ssd1 ( Jansen et al . , 2009 ) , S . pombe Dis3L2 , and the human exonuclease Dis3L2 , which has been associated with diseases such as Perlman syndrome and Wilm’s tumor , as well as Rrp44/Dis3 ( Figure 1C ) ( Malecki et al . , 2013; Robinson et al . , 2015; Lv et al . , 2015; Astuti et al . , 2012 ) . Sts5 and Dis3L2 contain conserved domains ( cold shock domains CSD1 and CSD2 and the S1 domain ) that mediate interaction with the single-stranded RNA substrate ( Faehnle et al . , 2014 ) . However , both Sts5 and Dis3L2 lack the PIN domain and CR3 signature amino acids that are implicated in the association of Rrp44/Dis3 to the exosome ( Indicated by • in Figure 1C ) ( Malecki et al . , 2013; Schaeffer et al . , 2012; Makino et al . , 2013; Robinson et al . , 2015 ) . Furthermore , Sts5 lacks conserved amino acids involved in RNA hydrolysis ( marked by ▲ in Figure 1C ) , indicating that it is unlikely to have exonuclease activity ( Uesono et al . , 1997; Jansen et al . , 2009 ) . 10 . 7554/eLife . 14216 . 003Figure 1 . Loss of RNA-binding protein Sts5 suppresses the cell viability defects of orb6 mutants . ( A ) Deletion of sts5 suppresses the cell separation phenotype of analog-sensitive orb6-as2 mutants . ( a ) wild-type , ( b ) sts5∆ , ( c ) orb6-as2 and ( d ) orb6-as2 sts5∆ mutants treated with 50 μM 1-NA-PP1 inhibitor for 2 hr at 32°C . Bar = 5 μm . ( B ) Septation index quantification of cells in experiment shown in A based on 3 independent experiments ( N>295 per strain ) . Orb6-as2 cells exhibit a significantly higher septation index as compared to control cells ( P = 0 . 0004 ) and as compared to orb6-as2 sts5∆ double mutants ( P = 0 . 0011 ) . P values were determined using analysis of variance ( ANOVA ) with SPSS statistics package 22 . 0 , followed by Tukey’s HSD test . Error bars indicate SD . ( C ) Sts5 protein sequence includes the RNB domain ( a . a . 531–975 ) , with homology to the catalytic domain of E . coli ribonuclease II , three conserved OB-fold domains that promote interaction with RNA , the CSD1 ( a . a . 293–353 ) , CSD2 ( a . a . 430–524 ) , and S1 ( a . a . 981–1031 ) domains . Sts5 is related to the exoribonuclease Dis3L2 that is conserved from S . pombe to humans . ▲ indicates 3 catalytic residues in RNB domain . • indicates CR3 motif residues for exosome targeting . ( D ) Interactions between MOR network proteins . Mor2 serves as a scaffold that enables activation of Mob2-bound Orb6 by the Nak1-Pmo25 complex ( o indicates 2-hybrid interaction; • indicates IP interaction ) . ( E ) sts5-276 mutation suppresses the temperature-sensitive growth of MOR mutants . The indicated cells were spotted on YPD solid medium ( approximately 5 × 104 cells in the left spots for each plate and then diluted 4-fold in each subsequent spot ) and incubated at 25°C , 34°C , and 36°C for 3 days . DOI: http://dx . doi . org/10 . 7554/eLife . 14216 . 00310 . 7554/eLife . 14216 . 004Figure 1—figure supplement 1 . Loss of RNA-binding protein Sts5 does not suppresses the polarity defects observed upon Orb6-as2 kinase inhibition . ( A ) Description of orb4-A9 and sts5-276 early stop codon mutations in the sts5 gene . No significant differences were observed in the phenotype of these mutations as compared to the sts5∆ deletion . ( B ) Loss of sts5 ( sts5∆ ) does not suppress the cell polarity phenotype of orb6-as2 cells following Orb6 kinase inhibition for 5 hr . Bar = 5 μm . ( C ) Loss of sts5 ( orb4-A9 allele ) does not suppress CRIB-GFP mislocalization in orb6-as2 mutants . Cells were grown at 32°C and treated with 1-NA-PP1 inhibitor for 30 mins . Bar = 10 μm . Note that only the subset of cells that do not contain a septum ( non-septating cells ) were analyzed . ( D ) Quantification of CRIB-3xGFP localization as shown in ( C ) ( N≥70 cells per condition ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14216 . 00410 . 7554/eLife . 14216 . 005Figure 1—figure supplement 2 . Deletion of gef1 or dis3L2 does not suppress the growth defect observed upon Orb6-as2 kinase inhibition . ( A ) gef1Δ cells do not suppress the growth defect observed upon Orb6-as2 kinase inhibition . Growth properties of the gef1Δ orb6-as2 double mutant compared with wild-type , orb6-as2 , and gef1Δ cells were assayed by spotting the indicated cells on minimal solid medium in the presence of DMSO or 10 μM 1-NA-PP1 inhibitor and incubated the plates at 32°C for 3 days ( approximately 5 × 105 cells in the left spots for each plate and then diluted 10-fold in each subsequent spot ) . ( B ) Loss of the sts5 homologue dis3L2 does not suppress the growth defect observed upon Orb6-as2 kinase inhibition . Growth properties of the dis3L2Δ orb6-as2 double mutant compared with wild-type , orb6-as2 , and dis3L2Δ cells were assayed by spotting the indicated cells on minimal solid agar as described in A . DOI: http://dx . doi . org/10 . 7554/eLife . 14216 . 005 Next , we investigated whether sts5Δ suppresses the loss of viability observed with temperature-sensitive orb6 mutants and mutants of other components of the Orb6 pathway ( Verde et al . , 1995; Verde et al . , 1998 ) . Orb6 kinase belongs to the morphogenesis ( MOR ) network , which includes Nak1 kinase and its binding partner Pmo25 , the scaffolding protein Mor2 , and the Orb6 binding partner Mob2 ( Figure 1D ) ( Kanai et al . , 2005; Hou et al . , 2003 ) . As with orb6 mutants , temperature-sensitive MOR mutants exhibit a loss of viability at the restrictive temperature . In a spot growth assay , we found that the sts5-276 mutation ( which truncates the sts5 gene to a short 53-bp fragment; see Figure 1—figure supplement 1A ) suppresses the temperature-sensitive growth defect associated with orb6 mutation , as well as the growth defects of other MOR network mutants ( Figure 1E ) . Finally , we tested the idea that the function of sts5 deletion in suppressing the loss of viability of orb6 mutant cells is independent of Cdc42 GTPase . We found that sts5 deletion does not suppress the cell rounding induced by prolonged Orb6 kinase inhibition ( Figure 1—figure supplement 1B ) . Further , mislocalization of the Cdc42 reporter CRIB-GFP in orb6 mutants ( Tatebe et al . , 2008; Hoffman and Cerione , 2000 ) , a hallmark of Orb6 kinase inhibition ( Das et al . , 2009 ) , is not suppressed by loss of sts5 ( Figure 1—figure supplement 1C and 1D ) . Conversely , deletion of the Cdc42 GEF gef1 does not suppress the loss of viability of orb6 mutants ( Figure 1—figure supplement 2A ) , while it suppresses the polarity defect associated with Orb6 kinase inhibition ( Das et al . , 2009 ) . Together , these results suggest that Sts5 mediates the cell separation defects and loss of viability observed in orb6 mutants . Further , this novel role of Orb6 kinase in growth control is genetically separable from its previously established function in the spatial control of Cdc42 GTPase . We used fluorescence microscopy to study the localization of Sts5-3xGFP . We found that during exponential growth Sts5-3xGFP localization is mostly diffuse in the cytoplasm during interphase ( I ) ( Figure 2A ) . Sts5 coalesces into cytoplasmic puncta during mitosis as previously reported ( Vaggi et al . , 2012 ) . 10 . 7554/eLife . 14216 . 006Figure 2 . Sts5 proteins assemble into puncta during mitosis and during nutritional starvation . ( A ) Sts5-3xGFP proteins coalesce into cytoplasmic particles in cells undergoing mitosis ( M ) but appear mostly diffuse in the cytoplasm of growing interphase ( I ) cells ( a , c ) . ( b ) P-body formation , as visualized by P-body marker Dcp1-mCherry is not induced in mitotic cells . Bar = 5 μm . ( B ) Sts5-3xGFP proteins are recruited and colocalize with the P-body marker Dcp1-mCherry upon growth for 1 hr in minimal medium minus glucose ( b , e , h ) . Sts5-3xGFP recruitment and colocalization with Dcp1-mCherry in P-bodies also occurs upon 1 hr of growth in minimal medium minus nitrogen ( c , f , i ) . Sts5-3xGFP recruitment was observed as early as 15 min after transfer to glucose- or nitrogen-depleted medium . Images are deconvolved projections from 12 Z-stacks separated by a step size of 0 . 3 μm . Experiment was performed using prototrophic strain FV2267 . Bar = 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14216 . 00610 . 7554/eLife . 14216 . 007Figure 2—figure supplement 1 . Sts5 protein contains an intrinsically disordered domain . Sts5 protein contains an intrinsically disordered domain as predicted by DisProt software ( using VL3 , VSL2B , and VLXT algorithms ) ( DisProt - Database of Protein Disorder , RRID:SCR_007097 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14216 . 007 We found that upon nutrient starvation Sts5-3xGFP proteins rapidly coalesce into distinct , larger cytoplasmic puncta . Many of these puncta colocalize with P-body ( Processing body ) marker Dcp1-mCherry , a component of the mRNA decapping complex ( Wang et al . , 2013 ) . Sts5 localization to the P-bodies is particularly strong during glucose deprivation ( Figure 2B , b , h ) and occurs also during nitrogen starvation , although puncta appear smaller and co-localization of Sts5 with Dcp1 is partial ( Figure 2B , c , i ) . Conversely , Sts5 recruitment into puncta during mitosis occurs in the absence of substantial P-body formation , as visualized with P-body marker Dcp1-mCherry ( Figure 2A ) . Consistent with these findings , Sts5 contains a region predicted to be intrinsically disordered in the first 301 amino acids of the Sts5 protein , a feature shared by many proteins that undergo assembly into RNP particles ( Lin et al . , 2015; Patel et al . , 2015; Elbaum-Garfinkle et al . , 2015; Wang et al . , 2014; Kato et al . , 2012; Han et al . , 2012 ) ( Figure 2—figure supplement 1 ) . These results indicate that Sts5 proteins can organize in cytoplasmic puncta during mitosis and in response to nutritional stress . Furthermore , these puncta co-localize with P-bodies following glucose or nitrogen starvation . It is possible that Sts5 modulates cell growth by controlling the levels of the mRNAs it binds . To investigate the functions of Sts5-regulated transcripts , we first used microarray analysis to identify mRNAs that are elevated ( at least 1 . 9-fold ) in sts5Δ mutants as compared to wild-type cells , under exponential growth conditions ( See Figure 3—source data 1 ) . This analysis identified 140 mRNAs , and showed significant overrepresentation of genes with roles in polarized cell growth , adhesion and cell wall biogenesis by gene ontology enrichment analysis ( Figure 3 , * ) ( See Figure 3—source data 1 ) . Remarkably , we identified ssp1 , cmk2 , tea5/ppk2 , ksg1 and lkh1 , which encode protein kinases with a role in the activation of bipolar cell growth ( Figure 3; Figure 3—source data 1 ) ( Koyano et al . , 2010 ) . Intriguingly , these mRNAs encoding polarity regulators all contain a consensus sequence in their 5’UTR which functions as a potential recognition sequence for targets of the S . cerevisiae homolog of Sts5 , Ssd1 ( Figure 3 , † ) ( Hogan et al . , 2008 ) ( Wanless et al . , 2014 ) . The fully categorized list of mRNAs identified in the microarray analysis includes mRNAs involved in cell wall biogenesis and secretion , cytoskeletal organization , nutrient transport , and meiosis ( See Figure 3—source data 1 ) . 10 . 7554/eLife . 14216 . 008Figure 3 . mRNAs detected at higher levels in sts5∆ cells by microarray analysis . Total mRNA was extracted from sts5∆ and control cells for microarray analysis . A complete list of mRNAs increased in sts5∆ cells ( ≥1 . 9 fold ) is shown in Figure 3—source data 1 . Several of these mRNAs have established functions in bipolar growth activation and contain putative Sts5-binding sites in their 5’ UTRs ( Hogan et al . , 2008; Wanless et al . , 2014 ) . Gene ontology enrichment analysis of terms that are significantly enriched among the set of mRNAs with sts5Δ/WT ratio ≥1 . 90 in the microarray results . Fold enrichment plotted per gene ontology category among all significant terms ( P<0 . 05 , modified Fisher Exact P-value with the Benjamini P-value correction ) for Cellular Compartment ( CC ) , Biological Process ( BP ) and Molecular Function ( MF ) Gene Ontology terms . †Sts5 binding site: HNNYAHTCHWW ( where H = A , T , C / N = A , T , C , G / Y = C , U / W = T , A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14216 . 00810 . 7554/eLife . 14216 . 009Figure 3—source data 1 . Microarray analysis results . A complete list of mRNAs detected at higher levels ( ≥1 . 9 fold ) in sts5∆ cells by microarray analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 14216 . 009 The mRNAs encoding proteins with known roles in polarized cell growth were selected for further analysis by qPCR , which confirmed that sts5Δ cells exhibit increased levels of ssp1 , cmk2 , tea5/ppk2 , lkh1 , efc25 , and psu1 mRNA , genes with diverse functions in cell morphogenesis ( Figure 4A ) ( Figure 3; Figure 3—source data 1 ) . Further , we determined that Sts5-3xGFP protein co-purifies with efc25 , ssp1 , and psu1 mRNAs ( Figure 4B ) . Consistent with a functional interaction with sts5 and orb6 , ssp1 was previously identified as an extragenic suppressor of sts5 mutants ( Matsusaka et al . , 1995 ) , while psu1 functions as a multicopy suppressor of orb6 mutants ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 14216 . 010Figure 4 . Loss of Sts5 leads to increased levels of mRNAs involved in growth control and bipolar growth activation . ( A ) qPCR analysis confirmation that several of these transcripts are more abundant in the sts5∆ strain as compared to control cells based on 3 independent experiments . Tea4 is shown as an example of a transcript that is not altered . Housekeeping genes were nda3 , act1 , cdc2 , and cdc22 . Error bars indicate SD . ( B ) Interaction of ssp1 , efc25 , and psu1 mRNAs with Sts5-3xGFP as established by co-immunoprecipitation with Sts5-3xGFP followed by qPCR as described in the Materials and Methods . cdc2 is shown as an example of a transcript that does not interact with Sts5-3xGFP . Error bars indicate SD . Three independent experiments were performed . ( C ) Western blotting against Ssp1-HA performed as described in Materials and Methods in WT and sts5Δ cells cultured in YE medium at 25°C and 36°C . Tubulin levels were determined as a loading control . ( D ) Quantification of Ssp1-HA/Tubulin ratio normalized to WT levels was based on 3 independent experiments . Change in Ssp1-HA level is significantly greater in sts5∆ cells as compared to controls at 36°C ( P = 0 . 034 , Student’s t-test ) . Error bar=SD . ( E ) Western blotting against Ssp1-HA performed as described in Materials and Methods in WT and edc3Δ and pdc1Δ cells cultured in supplemented minimal medium at 30°C . Tubulin levels were determined as a loading control . ( F ) Quantification of Ssp1-HA/Tubulin ratio normalized to WT levels at 25°C based on 3 independent experiments . Change in Ssp1-HA level is significantly greater in edc3Δ ( P = 0 . 018 , Student’s t-test ) and pdc1Δ ( P = 0 . 0154 , Student’s t-test ) cells as compared to controls . Error bar = SD . ( G ) RNA FISH visualization of ssp1 mRNA in fixed cells cultured for 20 min in supplemented minimal medium containing 0% glucose . Hybridization used 20-mer DNA oligos ( Stellaris ) labeled with Quasar 705 fluorochromes . Bar = 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14216 . 01010 . 7554/eLife . 14216 . 011Figure 4—figure supplement 1 . Overexpression of Psu1 suppresses the temperature-sensitive growth defect of orb6-25 mutant cells . Psu1 was expressed in orb6-25 and control cells from the pRep3X plasmid under the control of the nmt1 promoter , at restrictive temperature ( 36°C ) , in the presence ( a ) or absence ( b ) of Thiamine . DOI: http://dx . doi . org/10 . 7554/eLife . 14216 . 01110 . 7554/eLife . 14216 . 012Figure 4—figure supplement 2 . Deletion of sts5 alters cell shape and Myc-Efc25 protein levels . ( A ) Calcofluor staining of WT ( a and c ) and sts5∆ ( b and d ) cultured in YE medium at 25°C and 36°C . Bar = 5 μm . ( B ) Western blotting against Myc-Efc25 performed as described in Materials and Methods in WT and sts5Δ cells cultured in minimal medium at 25°C . Tubulin levels were determined as a loading control . ( C ) Quantification of Myc-Efc25/Tubulin ratio normalized to WT levels at 25°C based on 3 independent experiments . Change in Myc-Efc25 level is significantly greater in sts5Δ cells as compared to controls ( P = 0 . 0087 , Student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14216 . 01210 . 7554/eLife . 14216 . 013Figure 4—figure supplement 3 . Extent of colocalization between ssp1 mRNA , Sts5-3xGFP , and Dcp1-mCherry in fixed cells cultured in the presence and absence of glucose . ( A ) RNA FISH visualization of ssp1 mRNA in fixed cells cultured for 20 min in supplemented minimal medium containing 2% glucose or 0% glucose . Hybridization used 20-mer DNA oligos ( Stellaris ) labeled with Quasar 705 fluorochromes . Bar = 5 μm . ( B ) Object-based quantification of the proportion of ssp1 puncta contained within Sts5-3xGFP or Dcp1-mCherry puncta in the presence and absence of glucose . For ssp1 dots contained within Sts5 granules in 0% glucose vs 2% glucose , P = 0 . 0293 ( Student’s t-test ) . For ssp1 dots contained within Dcp1 granules in 0% glucose vs 2% glucose , P = 0 . 0011 ( Student’s t-test ) . Object-based quantification based on the distance between centers of mass performed using ImageJ plugin JACoP ( Bolte and Cordelières , 2006 ) . Error bars denote SD among the means of 3 independent experiments ( N≥25 cells ) . For details , see Materials and methods . ( C ) Object-based quantification of the proportion of Sts5-3xGFP or Dcp1-mCherry puncta containing ssp1 puncta in the presence and absence of glucose . For Sts5 granules containing ssp1 dots in 0% glucose vs 2% glucose , P = 0 . 0042 ( Student’s t-test ) . For Dcp1 granules containing ssp1 dots in 0% glucose vs 2% glucose , P = 0 . 0053 ( Student’s t-test ) . Object-based quantification based on the distance between centers of mass performed using ImageJ plugin JACoP ( Bolte and Cordelières , 2006 ) . Error bars denote SD among the means of 3 independent experiments ( N≥25 cells ) . For details , see Materials and Methods . DOI: http://dx . doi . org/10 . 7554/eLife . 14216 . 013 We chose to measure protein levels of HA-tagged Ssp1 and Myc-tagged Efc25 to gauge how Sts5 regulation of ssp1 and efc25 mRNAs affected the levels of Ssp1 and Efc25 proteins . We determined that Ssp1 protein levels significantly increase in sts5∆ mutants , as compared to controls , when cells are exposed to higher temperatures ( 36°C; Figure 4C , D ) . Interestingly , sts5∆ cells display a temperature-sensitive morphological phenotype , growing over a wider area of the cell surface and developing a rounded cell shape at 36°C ( Figure 4—figure supplement 2A , d ) . Consistent with increased Ssp1 protein levels playing a role in promoting abnormal morphogenesis , the aberrant morphological phenotype of sts5 mutants is partially suppressed by loss of ssp1 ( Matsusaka et al . , 1995; Toda et al . , 1996 ) . In addition , we determined that loss of sts5 leads also to increased levels of Myc-Efc25 proteins as compared to the control sts5+ cells ( Figure 4—figure supplement 2B , C ) . Taken together , these findings indicate a role for Sts5 in regulating the cellular abundance of specific mRNAs , affecting cell morphology in particular during cell stress . Next , we tested if loss of P-body components affects the protein levels of Ssp1 . We assayed the levels of Ssp1 protein in pdc1∆ and edc3∆ mutant cells . Both Pdc1 and Edc3 are P-body components and bind to the mRNA decapping complex catalytic subunit Dcp2 ( Fromm et al . , 2012; Wang et al . , 2013 ) . pdc1 encodes an mRNA decapping scaffolding protein and pdc1∆ mutants display reduced levels of P bodies and reduced mRNA decapping ( Wang et al . , 2013 ) . edc3 encodes an enhancer of mRNA decapping and edc3∆ mutants display decreased decapping of nuclear-transcribed mRNA ( Fromm et al . , 2012; Wang et al . , 2013 ) . We found that levels of Ssp1 protein are increased , as compared to tubulin control , in both the pdc1∆ and edc3∆ mutant backgrounds ( Figure 4E , F ) . This effect is seen even in the absence of starvation , during exponential cell growth , consistent with the idea that P-body components modulate mRNA abundance even in the absence of large P-body formation ( Decker et al . , 2007; Eulalio et al . , 2007 ) . Finally , we tested if ssp1 mRNA localizes to the P-bodies during glucose starvation , using RNA FISH methodology . We found that ssp1 mRNA readily co-localizes to Sts5- and Dcp1-containing granules in cells re-diluted in minimal medium ( EMM ) lacking glucose for 20 min ( Figure 4G; Figure 4—figure supplement 3A–C ) . Conversely , no co-localization was observed in control cells re-diluted in growth medium containing 2% glucose ( Figure 4—figure supplement 3A-C ) . Our data reveal a role for Sts5 and P-body components in regulating the levels of ssp1 mRNA and Ssp1 protein . To test the role of Orb6 kinase in the control of Sts5 , we used the analog-sensitive orb6-as2 mutant to determine whether loss of Orb6 kinase function alters the localization of Sts5 . In orb6-as2 cells treated with the 1-NA-PP1 inhibitor , Sts5-3xGFP rapidly coalesces into cytoplasmic puncta that colocalize with the P-body marker Dcp1-mCherry ( Figure 5A , d , h , and l; see quantification in Figure 5B ) , while DMSO-treated cells exhibit diffuse cytoplasmic localization of Sts5-3xGFP and Dcp1-mCherry ( Figure 5A , c , g , and k and Figure 5B ) , supporting the idea that Orb6 kinase negatively regulates Sts5-3xGFP recruitment into RNP granules and Sts5 co-localization with P-bodies . 10 . 7554/eLife . 14216 . 014Figure 5 . Orb6 kinase inhibits Sts5 recruitment and localization to P-bodies . ( A ) Loss of Orb6-as2 kinase activity leads to Sts5-3xGFP recruitment into puncta that colocalize with the P-body marker Dcp1-mCherry . Cells were treated with inhibitor or DMSO for 1 hr ( shown ) and 5 hr . Bar = 5 μm . ( B ) Quantification of three sets of experiments as shown in A . DOI: http://dx . doi . org/10 . 7554/eLife . 14216 . 014 Recent work has shown that the formation of RNP granules often occurs as a result of liquid-liquid phase transition controlled by the concentration of RNP component proteins ( Kroschwald et al . , 2015; Lin et al . , 2015; Elbaum-Garfinkle et al . , 2015; Patel et al . , 2015; Kato et al . , 2012; Hyman et al . , 2014; Brangwynne et al . , 2009; Lee et al . , 2013; Brangwynne , 2013; Becker and Gitler , 2015 ) . To establish if Sts5 has a role in promoting P-body formation , we inhibited Orb6-as2 kinase and measured the number of Dcp1-mCherry containing granules in the presence or absence of Sts5 . Interestingly , we found that the number of Dcp1-mCherry granules was significantly reduced in the sts5∆ background , indicating that Sts5 recruitment has a role in promoting P-body formation ( Figure 6A–C ) . We also observed residual formation of Dcp1-mCherry dots , suggesting that Orb6 kinase can induce P-body formation via Sts5-dependent as well as Sts5-independent mechanisms . Conversely , Dcp1-mCherry granules were not induced by DMSO or 1-NA-PP1 in orb6+ control and sts5∆ cells ( Figure 6—figure supplement 1A ) . 10 . 7554/eLife . 14216 . 015Figure 6 . Orb6 kinase inhibits Sts5-dependent P-body formation and translational repression . ( A ) Dcp1-mCherry localization in orb6-as2 ( a , c ) compared with sts5Δ orb6-as2 ( b , d ) cells grown in supplemented minimal medium in the presence of 50 μM 1-NA-PP1 ( c and d ) or DMSO ( a and b ) for 1 hr . Loss of Sts5 in the sts5∆ orb6-as2 strain decreases the number of P-bodies induced by Orb6 kinase inhibition . Images are deconvolved projections from 12 Z-stacks separated by a step size of 0 . 3 μm . Bar = 5 μm . ( B ) Quantification of the experiment shown in A based on 3 independent experiments ( n > 24 cells per sample in each experiment ) . The number of P-bodies per cell was significantly lower in sts5∆ orb6-as2 cells as compared to orb6-as2 cells upon Orb6 kinase inhibition relative to DMSO-treated orb6-as2 cells ( P = 0 . 0186 , Student’s t-test ) . No significance difference was observed when comparing orb6-as2 vs sts5∆ orb6-as2 cells treated with DMSO , P = 0 . 2458 ( Student’s t-test ) . Error bars indicate SD . ( C ) Quantification of the experiment shown in A based on 3 independent experiments ( n > 24 cells per sample in each experiment ) . The total P-body fluorescence intensity per cell was significantly lower in sts5∆ orb6-as2 cells as compared to orb6-as2 cells upon Orb6 kinase inhibition relative to DMSO-treated orb6-as2 cells ( P = 0 . 0013 , Student’s t-test ) . No significance difference was observed when comparing orb6-as2 vs sts5∆ orb6-as2 cells treated with DMSO ( P = 0 . 1837 , Student’s t-test ) . Error bars indicate SD . ( D ) mRNA levels of Sts5-regulated transcripts decrease upon Orb6-as2 kinase inhibition as compared to control , as established by qPCR analysis based on 3 independent experiments . act1 is shown as an example of a transcript that is not altered . Housekeeping genes were nda3 , cdc2 , and cdc22 . Error bars indicate SD . ( E ) Ssp1-HA protein levels in control , orb6-as2 , sts5Δ , and sts5Δ orb6-as2 cells cultured in the presence of 50 μM 1-NA-PP1 inhibitor in supplemented minimal medium at 25°C . Tubulin levels were determined as a loading control . ( F ) Quantification of Ssp1-HA/Tubulin in 4 independent experiments , as shown in E , normalized to wild-type levels . Ssp1-HA levels are significantly reduced upon Orb6-as2 kinase inhibition ( P = 0 . 031 ) , and are restored to wild-type levels in the sts5∆ orb6-as2 strain ( P = 0 . 023 ) . P values were determined using analysis of variance ( ANOVA ) with SPSS statistics package 22 . 0 , followed by Games-Howell test . Error bars indicate SD . DOI: http://dx . doi . org/10 . 7554/eLife . 14216 . 01510 . 7554/eLife . 14216 . 016Figure 6—figure supplement 1 . Orb6 kinase inhibits Sts5-dependent translational repression . ( A ) Treatment with DMSO and 1-NA-PP1 does not induce P-body formation in control and sts5Δ cells . Dcp1-mCherry localization in control ( a , c ) compared with sts5Δ ( b , d ) cells grown in supplemented minimal medium in the presence of 50 μM 1-NA-PP1 ( c and d ) or DMSO ( a and b ) for 1 hr . Images are deconvolved projections from 12 Z-stacks separated by a step size of 0 . 3 μm . Bar = 5 μm . ( B ) Temperature-sensitive inactivation of Orb6-25 kinase promotes Sts5-dependent mRNA degradation . Decreased mRNA levels in orb6-25 cells growth at 36°C as compared with wild type as confirmed by qPCR for a selected group of mRNAs , which were identified by microarray analysis ( Figure 3—source data 1 ) . ( C ) Western blots showing levels of Ssp1-HA protein in orb6-25 and control cells at 25°C and at 36°C . Tubulin levels were determined as a loading control . ( D ) Quantification of Ssp1-HA/Tubulin ratio in the experiment shown in C , normalized to wild-type levels , based on 3 independent experiments . At the restrictive temperature ( 36°C ) , the change in Ssp1-HA level is significantly reduced in orb6-25 as compared to control cells ( P = 0 . 044 ) . Ssp1-HA levels are significantly higher in sts5Δ ( P<0 . 0001 ) and sts5Δ orb6-25 ( P<0 . 0001 ) cells as compared to orb6-25 cells at 36°C . P values were determined using analysis of variance ( ANOVA ) with SPSS statistics package 22 . 0 , followed by Dunnet’s t test . Error bars denote SD . ( E ) Western blot showing levels of Myc-Efc25 protein in WT , sts5Δ , orb6-as2 , and sts5Δ orb6-as2 strains cultured in the presence of 50 μM 1-NA-PP1 inhibitor in supplemented minimal medium at 25°C . Tubulin levels were determined as a loading control . Right panel: Quantification of Myc-Efc25/Tubulin ratio , normalized to wild-type levels , based on 3 independent experiments . Myc-Efc25 level is significantly reduced in orb6-as2 as compared to control cells ( P = 0 . 0048 ) . When comparing orb6-as2 with sts5Δ orb6-as2 cells , P<0 . 0001 . Myc-Efc25 levels are significantly higher in sts5Δ ( P = 0 . 0010 ) and sts5Δ orb6-as2 ( P<0 . 0140 ) cells compared to wild-type cells . P values were determined using analysis of variance ( ANOVA ) with SPSS statistics package 22 . 0 , followed by Tukey’s HSD post-hoc test . Error bars = SD . DOI: http://dx . doi . org/10 . 7554/eLife . 14216 . 01610 . 7554/eLife . 14216 . 017Figure 6—figure supplement 2 . Overexpression of Orb6 inhibits Sts5 granule assembly , and Sts5 plays a role in P-body formation . ( A ) Overexpression of Orb6 reduces the intensity of Sts5-3xGFP ( P = 0 . 0227 , Student’s t-test ) and Dcp1-mCherry ( P = 0 . 0352 , Student’s t-test ) puncta in cells cultured in supplemented minimal medium minus glucose for 1 hr . Quantification based on 3 independent experiments ( N>22 cells per condition ) . Bar = 5 μm . ( B ) Dcp1-mCherry localization in control ( a ) compared with sts5Δ ( b ) cells grown in supplemented minimal medium in the presence of 0 . 01% glucose or 2% glucose for 1 hr . Loss of Sts5 decreases the total intensity of P-bodies induced by glucose deprivation . Bar = 5 μm . ( C ) Quantification of the number of P-bodies in the experiment shown in B based on 3 independent experiments relative to control cells cultured in 2% glucose . Error bars indicate SD . ( N > 29 cells per condition ) . ( D ) Quantification of the total P-body intensity in the experiment shown in B based on 3 independent experiments relative to control cells cultured in 2% glucose . Error bars indicate SD . The total P-body intensity per cell was significantly lower in sts5∆ cells as compared to control cells upon 1 hr of growth in supplemented minimal medium containing 0 . 01% glucose relative to control cells ( P = 0 . 0069 , Student’s t-test; N > 74 cells per strain ) . The total P-body intensity was not significantly different among the two strains in 2% glucose ( P = 0 . 8475 , Student’s t-test; N>29 cells per strain ) . Error bars indicate SD . DOI: http://dx . doi . org/10 . 7554/eLife . 14216 . 017 To test the role of Orb6 kinase in preventing the degradation of Sts5-regulated mRNAs , qPCR analysis was performed to probe the levels of specific transcripts following Orb6-as2 kinase inhibition with 1-NA-PP1 . We found that mRNA levels of ssp1 , efc25 and psu1 declined upon Orb6 kinase inhibition ( Figure 6D ) . Consistent with the idea that Orb6 kinase prevents degradation and translational repression of Sts5-regulated mRNAs , immunoblotting analysis showed that Ssp1-HA protein levels decrease in orb6-as2 cells upon inhibition with 1-NA-PP1 ( Figure 6E , F ) . Normal Ssp1-HA protein levels were maintained in sts5∆ orb6-as2 cells upon inhibition of Orb6-as2 kinase , in accordance with the findings that sts5 suppresses the viability phenotype of orb6 mutants and sts5Δ cells accumulate ssp1 mRNA . These observations held true also for Ssp1-HA protein levels in temperature-sensitive orb6-25 cells cultured at the non-permissive temperature ( 36°C ) in the presence and absence of Sts5 ( Figure 6—figure supplement 1B–D ) . Similarly to Ssp1-HA , Myc-Efc25 protein levels also declined in orb6-as2 mutants grown in the presence of 1-NA-PP1 ( Figure 6—figure supplement 1E ) , and sts5Δ abolished the reduction of Efc25 levels in orb6-as2 mutants ( Figure 6—figure supplement 1E ) . Finally , we tested if Orb6 kinase over-expression alters Sts5 recruitment and P-body formation following glucose deprivation . Indeed , we found that cells over-expressing Orb6 kinase display significantly smaller Sts5- and Dcp1-containing particles than control cells , following growth for 1 hr in minimal medium without glucose ( Figure 6—figure supplement 2A ) . Further , consistent with Sts5 having a role in promoting , at least in part , P-body formation during nutritional stress we found that sts5∆ cells form significantly smaller and dimmer Dcp1-containing P bodies , as compared to control cells , following 1 hr growth in glucose deprivation conditions ( shift from 2% to 0 . 01% glucose ) ( Figure 6—figure supplement 2B–D ) . Together , these findings support the idea that Orb6 kinase prevents Sts5 recruitment to Dcp1-containing granules and attenuates P-body formation , in a manner that is at least in part Sts5-dependent . Additionally , the function of Orb6 kinase activity has the effect of decreasing the degradation of specific mRNAs . We performed an in-vitro kinase assay by purification of Orb6 kinase regulatory subunit Mob2 , as previously reported ( Wiley et al . , 2003; Das et al . , 2009 , 2015 ) , using bacterially expressed Sts5 . We found that the immunoprecipitate readily phosphorylates Sts5 ( Figure 7A ) , suggesting that Orb6 kinase phosphorylates Sts5 . Sts5 contains several putative NDR kinase consensus sequences ( Hao et al . , 2008; Mazanka et al . , 2008; Gógl et al . , 2015 ) that are consistent with 14-3-3 binding sites ( RxxS ) when phosphorylated ( Yaffe et al . , 1997 ) . We previously showed that 14-3-3 protein Rad24 has a role in negatively regulating another Orb6 substrate , Cdc42 GEF Gef1 ( Das et al . , 2015 ) . In order to establish whether Sts5 may be subject to regulation by Rad24 , we performed a pull-down assay to test whether Sts5 binds Rad24 . This assay confirmed that Sts5-HA physically associates with Rad24-GST and not with GST alone ( Figure 7B ) . Consistent with Rad24 negatively regulating Sts5 recruitment , we found that Sts5-3xGFP forms cytoplasmic puncta in rad24∆ mutants even when cultured in rich medium ( YE ) in the presence of glucose ( Figure 7C , D ) . This effect occurs in growth conditions where cells are not starved and P-body formation is not strongly induced in either rad24∆ or control cells ( Figure 7C ) . Accordingly , we found that ssp1 mRNA levels do not significantly change in rad24∆ mutants as compared to control cells ( Figure 7E ) . 10 . 7554/eLife . 14216 . 018Figure 7 . 14-3-3 protein Rad24 negatively regulates Sts5 recruitment into puncta . ( A ) Orb6 kinase phosphorylates Sts5 in vitro . Mob2-associated Orb6 kinase was immunoprecipitated for a kinase assay as described in the Materials and Methods and incubated with bacterially expressed Sts5 in the presence of [γ32P]ATP . ( B ) Endogenously expressed Sts5-HA co-purifies with bacterially expressed GST-Rad24 but not with GST alone in a pull-down assay . Three independent experiments were performed . ( C ) Sts5-3xGFP and Dcp1-mCherry aggregation in 2% glucose YE in WT vs rad24Δ cells . Images are deconvolved projections from 12 Z-stacks separated by a step size of 0 . 3 μm . Bar = 5 μm . ( D ) Quantification of the experiment shown in C based on 3 independent experiments ( n > 27 cells per strain in each experiment ) . The number of Sts5 particles is significantly higher in rad24∆ relative to wild-type control cells ( P = 0 . 0005 , Student’s t-test ) . Error bars indicate SD . ( E ) qPCR analysis showing ssp1 mRNA levels are unchanged in rad24∆ cells compared with WT ( P = 0 . 160 ) and increased in sts5∆ cells compared with WT ( P = 0 . 044 ) . When comparing sts5∆ with rad24∆ cells , P=0 . 006 . P values were determined using analysis of variance ( ANOVA ) with SPSS statistics package 22 . 0 , followed by Games-Howell post-hoc test . Housekeeping genes were nda3 , act1 , and cdc2 . Error bars indicate SD . Three independent experiments were performed . ( F ) Physical association between endogenously expressed Sts5-HA and bacterially expressed GST-Rad24 is lower upon inhibition of Orb6-as2 with 50 μM 1-NA-PP1 compared with DMSO treatment ( lanes 7 and 8 ) . Sts5-HA association with GST-Rad24 remains unchanged in wild-type cells in the presence or absence of the inhibitor ( lanes 5 and 6 ) . GST-only control is shown in lanes 1–4 . DOI: http://dx . doi . org/10 . 7554/eLife . 14216 . 01810 . 7554/eLife . 14216 . 019Figure 7—figure supplement 1 . Quantification of the physical association between endogenously expressed Sts5-HA and bacterially expressed GST-Rad24 . Quantification of pull-down experiment depicted in Figure 5D showing a significant reduction ( P = 0 . 0025; Student’s t test ) in the physical interaction between endogenously expressed Sts5-HA and bacterially expressed GST-Rad24 upon inhibition of Orb6-as2 kinase with 50 μM 1-NA-PP1 . Error bars denote SD . Three independent experiments were performed . DOI: http://dx . doi . org/10 . 7554/eLife . 14216 . 019 Finally , we tested the effects of Orb6 kinase activity inhibition on the association of Rad24 to Sts5 . We found that that Sts5-3xGFP association with GST-Rad24 is abrogated by inhibition of Orb6-as2 kinase activity following exposure of orb6-as2 cells to 1-NA-PP1 , and not in orb6-as2 cells exposed to DMSO or in control orb6+ cells ( Figure 7F; see quantification in Figure 7—figure supplement 1 ) . Collectively , our findings indicate that Sts5 protein associates with 14-3-3 protein Rad24 in a manner that is dependent on Orb6 kinase activity , and that this association prevents Sts5 coalescence into cytoplasmic puncta . Orb6 kinase localization is enriched at the growing cell tips during interphase , in a manner that depends on the pattern of growth of the cell ( Figure 8A; Figure 8—figure supplement 1A , a , b , c and B ) ( Verde et al . , 1998 ) . In smaller cells , which grow in a monopolar manner ( M ) , Orb6 kinase localization is higher at the old growing end and lower at the non-growing new end ( Figure 8A , Figure 8—figure supplement 1A , a ) . To establish whether Orb6 kinase has a role in spatially controlling Sts5 in interphase cells during exponential cell growth , we tested the extent of Sts5 recruitment into small granules in smaller , monopolar cells ( 9 . 1 μm on average ) , which grow from the old end only . As shown earlier ( See Figure 2A ) , during growth in rich medium , Sts5 localization is generally diffuse . However , a closer inspection indicated that most cells contain a few small Sts5 puncta ( Figure 8B ) . We consistently found an increase in the number and intensity of Sts5 puncta at the non-growing end of smaller cells ( Figure 8C; Figure 8—figure supplement 1A , d , e and f; Figure 8—figure supplement 1B ) , indicating an inverse correlation between Orb6 kinase localization at the growing tip and Sts5 aggregation ( Figure 8A–C ) . 10 . 7554/eLife . 14216 . 020Figure 8 . Role of Orb6 kinase in Sts5 granule assembly during the cell cycle . ( A–F ) Active Orb6 kinase localization spatially anti-correlates with Sts5 recruitment into puncta in interphase cells . A . Orb6-GFP localizes to the growing cell tip in small monopolar wild-type cells . Orb6-GFP is enriched at the growing old cell end as compared to the non-growing new cell end . Bar = 5 μm . ( B ) Sts5-3xGFP aggregation increases towards the new cell end in monopolar wild-type cells . Images are deconvolved projections from 12 Z-stacks separated by a step size of 0 . 3 μm . Bar = 5 μm . ( C ) The average number of Sts5-3xGFP puncta per cell at the non-growing new end is significantly higher as compared to the growing old end ( P<0 . 0001 , Student’s t-test ) . Error bars denote SD . Three independent experiments were performed ( N = 31 cells ) . ( D ) Orb6-GFP localizes to the growing cell tip in tea1∆ cells . Bar = 5 μm . ( E ) Sts5-3xGFP recruitment onto puncta increases towards the non-growing tip in tea1∆ cells . Images are deconvolved projections from 12 Z-stacks separated by a step size of 0 . 3 μm . Bar = 5 μm . ( F ) The average number of Sts5-3xGFP puncta per cell at the non-growing end in tea1∆ cells is significantly higher as compared to the growing old end ( P<0 . 0009 , Student’s t-test ) . Error bars denote SD . Three independent experiments were performed ( N = 24 cells ) . We used calcofluor staining to identify growing tips and measured monopolar tea1∆ cells that were growing from the previous old end , which facilitated definitive identification of the nongrowing cell end . ( G–H ) Orb6 kinase activity temporally anti-correlates with Sts5 assembly into puncta during mitosis . ( G ) ( a , b and c ) Localization of Sts5-3xGFP in cells undergoing cell division; ( d , e and f ) visualization of Rlc1-Tomato; ( g , h and i ) calcofluor staining of cell wall and septum . Bar = 5 μm . ( H ) Quantification of the number of Sts5 puncta in dividing cells during cytokinetic ring formation , ring constriction , and septation . Ring formation vs septation , P<0 . 0001; ring constriction vs septation , P<0 . 0001; ring formation vs ring constriction P = 0 . 588 ( N>20 cells per condition ) . P values were determined using analysis of variance ( ANOVA ) with SPSS statistics package 22 . 0 , followed by Tukey’s HSD post-hoc test . Three independent experiments were performed . DOI: http://dx . doi . org/10 . 7554/eLife . 14216 . 02010 . 7554/eLife . 14216 . 021Figure 8—figure supplement 1 . Additional images of Orb6-GFP and Sts5-3xGFP localization in monopolar WT cells and quantification of total Sts5-3xGFP granule intensity at growing and nongrowing tips . ( A ) ( a , b , c ) Localization of Orb6-GFP in wild-type monopolar ( M ) ( a ) , bipolar ( B ) ( b ) , and septated ( S ) ( c ) cells cultured in supplemented minimal medium . ( d , e , f ) Localization of Sts5-3xGFP puncta ( see arrows ) in wild-type monopolar ( M ) cells ( d , e , f ) . indicates growing tip . Bar = 5 μm . ( B ) The average total intensity per cell of Sts5-3xGFP puncta at the non-growing new end is significantly higher as compared to the growing old end ( P = 0 . 0059 , Student’s t-test ) . Error bars denote SD . Three independent experiments were performed ( N = 31 cells ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14216 . 02110 . 7554/eLife . 14216 . 022Figure 8—figure supplement 2 . Additional images of Orb6-GFP and Sts5-3xGFP localization in monopolar tea1Δ cells and quantification of total Sts5-3xGFP granule intensity at growing and nongrowing tips . ( A ) ( a ) Localization of Orb6-GFP in tea1Δ monopolar cells , and ( b–f ) localization of Sts5-3xGFP puncta ( see arrows ) in tea1Δ monopolar cells cultured in supplemented minimal medium . * indicates growing tip . Bar = 5 μm . ( B ) The average total intensity per cell of Sts5-3xGFP puncta at the non-growing end in tea1∆ cells is significantly higher as compared to the growing old end ( P = 0 . 0435 , Student’s t-test ) . Error bars denote SD . Three independent experiments were performed ( N = 24 cells ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14216 . 02210 . 7554/eLife . 14216 . 023Figure 8—figure supplement 3 . Orb6 kinase inhibition prevents the dissolution of Sts5-3xGFP puncta after completion of mitosis . ( A ) ( a , b , c ) Localization of Sts5-3xGFP in cells undergoing cell division and ( d , e , f ) expressing Rlc1-Tomato during cytokinetic ring formation , ring constriction , and septation in orb6-as2 cells treated with DMSO . Bar = 5 μm . Right panel: Quantification of Sts5-3xGFP puncta in DMSO-treated orb6-as2 cells . Ring formation vs septation , P<0 . 0001; ring constriction vs septation , P<0 . 0001; ring formation vs ring constriction P = 0 . 345 ( N = 18 cells per condition ) . P values were determined using analysis of variance ( ANOVA ) with SPSS statistics package 22 . 0 , followed by Tukey’s HSD post-hoc test . Three independent experiments were performed . ( B ) ( a , b , c ) Localization of Sts5-3xGFP in cells undergoing cell division and ( d , e , f ) expressing Rlc1-Tomato during cytokinetic ring formation , ring constriction , and septation in orb6-as2 cells treated with 50 mM 1-NA-PP1 for 1 hr . Bar = 5 μm . Right panel: Quantification of Sts5-3xGFP puncta in 1-NA-PP1-treated orb6-as2 cells . Ring formation vs septation , P = 0 . 764; ring constriction vs septation , P<0 . 773; ring formation vs ring constriction P = 0 . 392 ( N = 18 cells per condition ) . P values were determined using analysis of variance ( ANOVA ) with SPSS statistics package 22 . 0 , followed by Tukey’s HSD post-hoc test . Three independent experiments were performed . DOI: http://dx . doi . org/10 . 7554/eLife . 14216 . 02310 . 7554/eLife . 14216 . 024Figure 8—figure supplement 4 . sts5∆ cells display increased cell lysis during cell separation . ( A ) sts5∆ cells display increased cell lysis during cell separation ( d ) at the restrictive temperature ( 35 . 5°C ) . Arrows indicate sister cells that have lysed . ( B ) Quantification of the percentage of septated cells that lyse in WT and sts5∆ cells at 25°C and 35 . 5°C . Quantification based on 3 independent experiments . Percentage of septated cells that lyse is significantly greater in sts5Δ cells as compared to controls at 35 . 5°C ( P = 0 . 0 . 0022 , Student’s t-test , N>176 cells per condition ) . When comparing WT and sts5∆ cells 25°C , P = 0 . 1583 ( Student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14216 . 024 To further investigate this effect , we visualized Sts5-3xGFP in longer tea1∆ cells that grow from one end only ( Snell and Nurse , 1994; Verde et al . , 1995 ) . tea1Δ cells display monopolar Orb6-GFP localization at the only growing cell tip ( Figure 8D; Figure 8—figure supplement 2A , a ) . In these cells , Sts5-3xGFP recruitment into small puncta was clearly seen increasing towards the non-growing cell tip ( Figure 8E-F; Figure 8—figure supplement 2A , b–f; Figure 8—figure supplement 2B: asterisk marks the growing tip ) . Collectively , our findings indicate that Orb6 kinase activity negatively regulates Sts5 recruitment into cytoplasmic puncta , via interaction with 14-3-3 protein Rad24 , in a manner that is spatially significant: an asymmetry of Orb6 distribution between growing and non-growing tips correlates with an asymmetry in Sts5 recruitment . Orb6 kinase activity is repressed during mitosis by the septation-initiation network ( SIN ) , which triggers cytokinesis ( Kanai et al . , 2005; Gupta et al . , 2013 , Gupta et al . , 2014 ) . SIN signaling remains active until completion of cytokinesis , which is marked by closure of the contractile ring and a fully formed cell septum ( García-Cortés and McCollum , 2009; Alcaide-Gavilán et al . , 2014 ) . After cytokinesis , the primary septum must be degraded for cell separation to occur . The derepression of MOR signaling , and of Orb6 activity , that results from inactivation of the SIN pathway promotes cell separation upon completion of cytokinesis ( Gupta et al . , 2014 ) . Consistent with Orb6 kinase re-activation , Sts5-3xGFP puncta dissipate when the septum is fully closed and the actomyosin contractile ring protein Rlc1 , encoding myosin II regulatory light chain ( Le Goff et al . , 2000; Wu et al . , 2006 ) , disappears from the plane of cell division ( Figure 8G , H; Wei et al . , 2016 ) . Orb6-GFP is still physically present at the site of cell division during septum formation and cytokinetic ring constriction ( Figure 8—figure supplement 1A , c ) while it is enzymatically repressed by SIN signaling ( Kanai et al . , 2005; Gupta et al . , 2013 , Gupta et al . , 2014; García-Cortés and McCollum , 2009; Alcaide-Gavilán et al . , 2014 ) . Consistent with a role for Orb6 kinase reactivation in mediating Sts5-3xGFP granules dissipation , Orb6 kinase inhibition maintains Sts5-3xGFP granules even following actin ring closure and Rlc1 disappearance ( Figure 8—figure supplement 3A , B ) . Supporting a role for Orb6 kinase and its substrate target Sts5 in cell separation , microarray analysis found that Sts5 negatively regulates transcripts that encode cell wall proteins with potential functions in cell separation . Transcripts for a predicted β-1 , 3 glucanase ( encoded by SPBP23A10 . 11C ) and predicted β-glucosidase Psu2 are more abundant in sts5Δ cells ( See Figure 3—source data 1 ) , consistent with a role for β-1 , 3 glucan degradation in the primary septum during the process of cell separation ( Martín-Cuadrado et al . , 2003 ) . In addition , sts5Δ cells accumulate transcripts of the transcription factor Mbx1 ( See Figure 3—source data 1 ) that cooperates with the transcription factor Ace2 to promote expression of endo-glucanase Agn1 , a hydrolytic enzyme involved in septum degradation ( Suárez et al . , 2015 ) . Thus , it is possible that Sts5 recruitment into puncta during mitosis , mediated by SIN pathway-dependent inhibition of Orb6 kinase , functions to translationally repress mRNAs encoding cell wall hydrolytic enzymes that would interfere with the deposition of the primary septum . Consistent with this idea , we found that sts5∆ cells are prone to rupture at the site of cell separation ( Figure 8—figure supplement 4A , d and B ) similarly to cells that express ectopically active Orb6 during mitosis ( Gupta et al . , 2013 , Gupta et al . , 2014 ) . When cultured at 25°C , sts5Δ cells appear normal with a cylindrical shape ( Toda et al . , 1996 ) . However , we found that sts5Δ cell cultures display an increased percentage of cells growing in a bipolar fashion , as compared to similarly sized control cells , under exponential growth conditions ( optical density at 595 nm <0 . 4 ) in both rich ( Figure 9A , B ) as well as in minimal medium ( Figure 9—figure supplement 1A ) . These findings suggest that Sts5 has a function in partially constraining growth at the new end during the exponential growth phase ( OD<0 . 4 ) , without affecting overall growth rates ( Figure 9—figure supplement 1B ) or cell length at division , which are the same as control cells ( Figure 9C ) . 10 . 7554/eLife . 14216 . 025Figure 9 . Sts5 modulates bipolar growth activation during exponential cell proliferation and during nutritional stress . ( A ) sts5∆ cells display a delayed morphological response to nutritional stress induced by high cell density as compared with wild type cells . Cells were stained with calcofluor . Bar = 5 μm . ( B ) Quantification of the percentage of bipolar cells in control versus sts5∆ cells in the experiment depicted in A . Percentage bipolar cells was significantly higher in sts5∆ cells versus control cells during exponential growth ( OD600 <0 . 4 ) ( P = 0 . 0013 , Student’s t test ) and at OD600 = 1 . 4 ( P<0 . 0001 , Student’s t test ) , OD600 = 3 ( P<0 . 0001 , Student’s t test ) , and OD600 = 4 . 5 ( P = 0 . 0003 , Students’ t test ) . Error bars indicate SD . At least 3 independent experiments were performed ( N>64 for each strain per cell density condition ) . Cells undergoing cell division were not included . ( C ) Quantification of cell size ( defined as cell length at division ) in control versus sts5∆ cells in the experiment depicted in A . Cell size was significantly longer in sts5∆ cells versus control at OD600 = 1 . 4 ( P<0 . 0001 , Student’s t test ) , OD600 = 3 ( P<0 . 0001 , Student’s t test ) , and OD600 = 4 . 5 ( P<0 . 0001 , Student’s t test ) . Error bars indicate SD . At least 3 independent experiments were performed ( N>16 for each strain per cell density condition ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14216 . 02510 . 7554/eLife . 14216 . 026Figure 9—figure supplement 1 . Increased bipolarity of sts5Δ vs wild-type cells is not due to changes in cell size or overall cell growth . ( A ) Quantification of the percentage of bipolar cells in wild type versus sts5∆ cells comparing small cells ( <10 mm in length; P = 0 . 0050 , Student’s t-test ) , larger cells ( 10–11 . 99 mm in length; P = 0 . 0 . 0012 Student’s t-test ) and cell with a length longer than 12 µm cells; P = 0 . 0256 ) . Quantifications based on 4 independent experiments ( N>239 cells per strain ) . ( B ) Growth curves ( OD 595 nm ) of wild-type versus sts5∆ cells as measured by the TECAN system . Cell were grown in supplemented minimal medium . DOI: http://dx . doi . org/10 . 7554/eLife . 14216 . 026 Since the pattern of cell growth is altered by nutritional stress , inhibiting bipolar growth activation and increasing the percentage of monopolar cells , ( Su et al . , 1996; Yanagida , 2009; Yanagida et al . , 2011 ) we hypothesized that Sts5 recruitment into puncta , a form of RNP granules , might have an adaptive role to modulate the morphological response during nutritional starvation . As cell density increases , S . pombe cells respond to limiting nutrient availability by entering mitosis at a shorter cell size ( Costello et al . , 1986; Su et al . , 1996; Yanagida , 2009; Yanagida et al . , 2011 ) . We found that , whereas wild-type cells divide at a shorter length upon starvation induced by high cell concentration , as determined by optical absorbance at 595 nm ( Figure 9 ) , sts5Δ cells maintain a longer length at cell division as cell concentration increases ( Figure 9A , C ) . Similarly , a higher proportion of sts5Δ mutants continue to activate bipolar growth , as compared to wild-type cells at the same concentration ( Figure 9B ) . These observations suggest that Sts5 has a role in partially constraining bipolar cell growth , a function that is important for cellular adaptation to nutrient limitation . Consistent with this idea , we find that sts5∆ cells display decreased viability after prolonged starvation ( I . N . and F . V . , unpublished observation ) . Collectively , our results indicate that NDR kinase Orb6 inhibits the recruitment of mRNA-binding protein Sts5 into cytoplasmic puncta by promoting its interaction with 14-3-3 protein Rad24 . Further , Orb6 kinase has a role in negatively controlling P-body formation , in a manner that is at least in part Sts5-dependent . This mechanism controls the levels of mRNAs encoding proteins important for polarized cell growth and cell separation . During interphase , Orb6 inhibits Sts5 recruitment in a manner that is biased towards the old end in small cells , thus promoting normal cell morphogenesis and partially constraining extensile growth at the second , newer cell tip . Extensive Sts5 recruitment into smaller puncta during mitosis and into larger RNP granules during nutritional stress may allow proper septum deposition and modulates morphological adaptation to limiting nutrient availability . In this article , we define a novel mechanism that spatially regulates polarized cell growth and cell morphology in fission yeast during exponential cell proliferation and in response to environmental stressors , such as increased temperature or cell density . Under exponential growth conditions , fission yeast grow in a monopolar fashion during early interphase and activate growth at the new cell tip once a minimal cell length has been achieved . Different control mechanisms cooperate in the activation of the second tip , a process known as NETO ( New End Take Off ) , including the microtubule-dependent Tea1 complex ( Martin and Arkowitz , 2014; Sawin and Nurse , 1998; Martin et al . , 2005; Tatebe et al . , 2005 ) , the availability of Cdc42 regulators ( Coll et al . , 2003; Tatebe et al . , 2008; Das et al . , 2012 ) , cell transcription ( Vjestica et al . , 2013 ) , and a diverse array of signaling kinases ( Koyano et al . , 2010; Matsusaka et al . , 1995; Rupes et al . , 1999; Koyano et al . , 2015; Grallert et al . , 2013 ) . We have previously shown that NDR kinase Orb6 promotes cell polarity and regulates bipolar growth by spatially restricting the activation of Cdc42 GTPase , a key morphology control factor ( Das et al . , 2009 ) . We recently showed that Orb6 negatively regulates Cdc42 activation by promoting the association of Cdc42 Guanine Exchange Factor ( GEF ) Gef1 with 14-3-3 protein Rad24 , and thus limiting Gef1 activity at the membrane ( Das et al . , 2015 ) . This function has the effect of spatially regulating Cdc42 activation , thus promoting the emergence of cell polarity . In this article , we describe a genetically separable role for Orb6 kinase in the control of polarized cell growth and cell separation . We report that Orb6 kinase regulates the association of mRNA-binding protein Sts5 , another Orb6 substrate target , with 14-3-3 protein Rad24 . This association prevents the recruitment of Sts5 into cytoplasmic puncta ( see hypothetical model in Figure 10A ) . Orb6 localization varies during the cell cycle , increasing at the cell tips during interphase and at the cell septum during cell division ( Verde et al . , 1998; Wiley et al . , 2003 ) . In small cells that have not yet undergone NETO , or in the monopolar tea1Δ mutant cells , Orb6 is enriched at the one growing cell tip . Consistent with a role for Orb6 in inhibiting Sts5 assembly into puncta , we have observed that Sts5-3xGFP puncta are spatially anti-correlated with Orb6 kinase activity and are preferentially localized near non-growing cell tips , which are depleted of Orb6 kinase ( Figure 10B ) . Because Orb6 kinase is enriched at growing cell tips , this method of Sts5 regulation might promote the availability of Sts5-targeted mRNAs for translation near sites of polarized growth , while partially constraining growth at the non-growing , newer cell tip . Consistent with this idea , an increased percentage of sts5Δ cells exhibit a bipolar pattern of growth , as compared to control cells of similar length . 10 . 7554/eLife . 14216 . 027Figure 10 . A model of spatial control of translational repression and polarized growth by Orb6 kinase and mRNA binding protein Sts5 . ( A ) Orb6 kinase prevents Sts5 recruitment into larger RNP granules by promoting the association between Sts5 and the 14-3-3 protein Rad24 . Upon Orb6 kinase inhibition , Sts5 proteins are recruited into larger RNP granules and co-localize with P-bodies , leading to reduced mRNA levels and translational repression . ( B ) In small monopolar cells Orb6 kinase is localized at the growing old end . Sts5 recruitment in larger granules is observed at the new end , lacking Orb6 kinase activity . In larger bipolar cells Orb6 kinase is localized at both cell tips , and Sts5 recruitment is reduced at both cell ends and throughout the cell . During mitosis , Orb6 kinase is inactivated by the SIN pathway , which allows Sts5 recruitment into larger RNP granules and translational repression . Once cell separation is complete , Orb6 kinase activity resumes , promoting Sts5 disassembly , translational derepression , and cell separation . DOI: http://dx . doi . org/10 . 7554/eLife . 14216 . 027 By microarray and qPCR analysis we found that several transcripts , previously implicated in bipolar growth activation , accumulate in sts5Δ cells . These transcripts encode the putative CAMK kinase Cmk2 , CAMKK kinase Ssp1 , LAMMER kinase Lkh1 , pseudokinase Tea5/Ppk2 , and PDK1 kinase Ksg1 , which promote bipolar growth activation in S . pombe ( Koyano et al . , 2010 ) , as well as the Ras1 GEF Efc25 , which affects Cdc42 activity ( Papadaki et al . , 2002 ) . In higher eukaryotes , the homologues of these Sts5-regulated transcripts are also implicated in cell growth and morphology . CAMK signaling regulates cytoskeletal organization , plays a role in neuronal development and dendritic spine morphology , and has been shown to be increased in mouse models of cardiac hypertrophy ( Penzes et al . , 2008; Passier et al . , 2000 ) . The Drosophila LAMMER kinase Doa inhibits proliferation of germ cells ( Zhao et al . , 2013 ) . Also , Ras signaling has conserved roles in cytoskeletal organization with implications for cancer development ( Shields et al . , 2000 ) . Activation of PI3K signaling via PDK1 kinase has been implicated in cancer , and PDK1 has been found to regulate cell growth , proliferation , and migration ( Li et al . , 2010; Mora et al . , 2004 ) . Consistent with Orb6 kinase inhibiting the extent of Sts5 recruitment during exponential growth , Sts5-containing cytoplasmic puncta assemble during the later stages of mitosis when Orb6 kinase activity is blocked by the SIN pathway ( Figure 10B ) ( Vaggi et al . , 2012; Gupta et al . , 2013 , Gupta et al . , 2014 ) . Sts5 recruitment during cytokinesis may function to prevent inappropriate translation of proteins involved in septum degradation while the septum is still forming . Indeed , sts5∆ cells lyse at the cell septum , in particular upon stress . Once cytokinesis is complete and Orb6 kinase is once again active , Sts5 localization is again diffuse in the cytoplasm , perhaps to allow expression of hydrolases required for cell separation . Consistent with this idea , orb6 mutants delay cell separation . We did not find obvious induction of P-body formation during mitosis or interphase , likely preventing the degradation of Sts5-regulated transcripts that will eventually be needed for cell separation and polarized cell growth . This result suggests that P-body independent , Sts5-containing RNPs are formed during mitosis and that additional stress signals are required for Sts5 to seed P-body formation under nutritional limitation conditions . Sts5 bears closest homology to RNA exonucleases such as fission yeast Dis3L2 ( Malecki et al . , 2013 ) . In humans , hDis3L2 is of particular interest because it has been implicated in the congenital Perlman syndrome , which confers fetal overgrowth and susceptibility to Wilms tumor ( Astuti et al . , 2012 ) . Whereas Sts5 lacks crucial amino acids required for exonuclease activity ( Figure 1C ) ( Malecki et al . , 2013; Uesono et al . , 1997 ) , our work shows that Sts5 promotes mRNA degradation likely by promoting the interaction of Sts5-associated transcripts with P-body components . Similar to Sts5 , Dis3L2 also localizes to P-bodies ( Malecki et al . , 2013 ) , as well as the S . cerevisiae homologue of Sts5 , Ssd1 ( Jansen et al . , 2009; Kurischko et al . , 2011 ) . These findings suggest that , differently from the related exonuclease Dis3 ( Robinson et al . , 2015 ) , this group of RNA-binding proteins employs a mechanism of mRNA degradation that does not involve interaction with the exosome . However , it is likely that Sts5 and Dis3L2 have different roles in P-body assembly and/or recruitment of mRNAs to P-bodies . Indeed , deletion of the sts5 homologue dis3L2 does not suppress the growth defect observed upon Orb6-as2 kinase inhibition ( Figure 1—figure supplement 2B ) . In S . cerevisiae and C . albicans , Sts5 homologue Ssd1 , with roles in cell wall hydrolysis ( Jansen et al . , 2009; Wanless et al . , 2014 ) , fertility ( Bourens et al . , 2009 ) , and transcription ( Lee et al . , 2015 ) , has been proposed to promote localized mRNA translation because of its enrichment at the bud site ( Kurischko et al . , 2011; Lee et al . , 2015 ) . We have not observed enrichment of Sts5 at the sites of polarized growth in S . pombe cells , which may employ a different strategy for ensuring adequate mRNA localization in daughter cells . Our work indicates that Sts5 may function as a seed for P-body formation under stress and upon Orb6 kinase inhibition . P-bodies and similar RNPs have been shown to have liquid properties and their formation has been described as condensation by a phase-separation mechanism once P-body components reach critical concentrations ( Kroschwald et al . , 2015; Lin et al . , 2015; Elbaum-Garfinkle et al . , 2015; Hyman et al . , 2014; Brangwynne et al . , 2009; Lee et al . , 2013; Brangwynne , 2013; Becker and Gitler , 2015 ) . It has been proposed that mRNA-binding proteins may play a role in the aggregation of mRNPs into larger structures , especially proteins containing intrinsically disordered domains , which have the potential to promote phase separation by forming multiple weak protein-protein interactions ( Lin et al . , 2015; Patel et al . , 2015; Elbaum-Garfinkle et al . , 2015; Wang et al . , 2014; Kato et al . , 2012; Han et al . , 2012; Malinovska et al . , 2013; Toretsky and Wright , 2014; Kroschwald et al . , 2015 ) . Thus , Sts5 may have the ability to function in mRNP granule formation through interactions with other P-body proteins , and one function of Orb6 phosphorylation may be to limit Sts5 recuitment to prevent inappropriate seeding of P-bodies . Consistent with this idea , Sts5 displays a predicted disordered domain at the N-terminus , which is phosphorylated in vivo ( Kettenbach et al . , 2015; Carpy et al . , 2014; Koch et al . , 2011; Wilson-Grady et al . , 2008 ) . Future experiments will address how phosphorylation of this domain modulates the function of Sts5 and affects its properties in vitro . We found that Sts5 has a role in the morphological response to nutritional stress . Wild type S . pombe cells mount characteristic morphological responses to changing environmental conditions: increased temperature ( Mitchison and Nurse , 1985 ) , decreased nutrient availability ( Costello et al . , 1986; Su et al . , 1996; Yanagida , 2009; Yanagida et al . , 2011 ) , and hyper-osmotic stress decrease the incidence of bipolar growth activation and alter overall cell dimensions ( Rupes et al . , 1999; Robertson and Hagan , 2008 ) . The mechanisms that modulate cell morphogenesis and polarized cell growth in response to varying growth and environmental conditions are still poorly understood . We find that sts5Δ mutants delay the adaptation to starvation conditions , maintaining bipolar growth and a longer cell length as cell density increases and the medium becomes depleted of nutrients . This effect appears to be further exacerbated at higher temperatures ( 36°C ) , where sts5Δ cells become enlarged and bloated , with increased protein levels of the CAMKK Ssp1 . Our findings indicate that sts5Δ mutants have defects in adapting to nutrient deprivation or temperature increase , and fail to manifest the appropriate morphological response to varying extracellular conditions . Thus Sts5 may integrate diverse nutritional and environmental signals to coordinate changes to the pattern of cell growth . In summary , our results support a role for NDR kinases in the spatial control of polarized cell growth , during cell proliferation and in response to the nutritional environment by mediating the translational availability of specific mRNAs . Future research will seek to identify nutrient-sensitive signaling pathways upstream of Orb6 and define the specific roles of Orb6 kinase and Sts5 in the control of P-body assembly . Due to the conservation of these factors , this work has the potential to open new avenues of research linking nutrient-sensitive signaling and P-body regulation with implications for studies of cancer and neurodegenerative diseases . S . pombe strains used in this study are listed in the supplement in Supplementary file 1 . All strains used in this study are isogenic to the original strain 972 . Cells were cultured in yeast extract ( YE ) medium or minimal medium ( EMM ) plus required supplements . Prototrophic strain FV2267 was cultured in unsupplemented EMM . For glucose and nitrogen starvation experiments , cells were washed in glucose-free or nitrogen-free EMM before transfer to EMM either lacking or containing 2% glucose or 0 . 5% nitrogen , respectively . Exponential growth was maintained for at least eight generations before experimental analysis , and genetic manipulations and analysis were carried out using standard techniques ( Moreno et al . , 1991 ) . Approximately 5 × 107 cells of the nak1-125 ( KP1-6D ) , orb6-25 ( DH433-12C ) , or mor2-276 mutant ( DH107-4C ) were spread per one YPD plate ( 1% yeast extract , 2% polypeptone , 2% dextrose , and 2% agar ) containing 10 mg/ml Phloxine B ( Sigma-Aldrich , P2759 ) ( called YPDP plate ) , and the plates were incubated at 35 . 5°C for 4 days . Spontaneously developed Ts+ colonies at 35 . 5°C were picked up on YPDP plate and incubated at 35 . 5°C for 3 days . To investigate the cold sensitivity and cell morphology of the mutants , the Ts+ colonies were replica plated on 2 YPDP plates and incubated at 18°C and 35 . 5°C . In this screening , we selected Ts+ and cold sick ( red colony ) at 18°C , and isolated 1 , 2 , or 1 sts5 mutant alleles from mor2 , nak1 , or orb6 mutants , respectively . Genetic linkage ( allelism ) between the suppressors and sts5 was confirmed by tetrad analysis . Cells expressing fluorescently tagged proteins were photographed using an Olympus fluorescence BX61 microscope ( Melville , NY ) equipped with Nomarski differential interference contrast ( DIC ) optics , a 100X objective ( NA 1 . 35 ) , a Roper Cool-SNAP HQ camera ( Tucson , AZ ) , Sutter Lambda 10 + 2 automated excitation and emission filter wheels ( Novato , CA ) and a 175 W Xenon remote source lamp with liquid light guide . Images were acquired and processed using the Intelligent Imaging Innovations ( Denver , CO ) SlideBook image analysis software and prepared with Adobe Photoshop CC ( San Jose , CA ) and ImageJ64 ( U . S . National Institutes of Health ) ( ImageJ , RRID:SCR_003070 ) . For measurements of Sts5-3xGFP and Dcp1-Cherry puncta , we subtracted the contribution of the cytoplasmic background for each cell as previously described ( Das et al . , 2012 ) . This process was performed using an ImageJ plugin that sets a subtraction threshold to 3 standard deviations from cytoplasmic-region mean . Pilot studies were used to obtain means and standard deviations to be used for sample size estimation before determining how many cells to measure in each independent experiment of Sts5-3xGFP or Dcp1-mCh aggregation . The following formulas were used for sample size estimation , assuming an alpha of 0 . 05 , beta of 0 . 2 , and power of 0 . 8: k= ( n2/n1 ) = 1 n1 = [ ( σ12 + σ22/K ) ( z1 – α/2 + z1 – β ) 2] / Δ2 Δ = \|μ2-μ1\| = absolute difference between two means σ1 , σ2 = mean variances n1 = group 1 sample size n2 = group 2 sample size α = probability of type I error ( set to 0 . 05 ) β = probability of type II error ( set to 0 . 2 ) z = critical Z value for a given α or β Cells were grown under normal conditions ( eight generations of exponential growth ) prior to the start of the experiment . Cells were then treated in accordance with the particular experiment . The RNA was extracted from the yeast using the ZR Fungal/Bacterial RNA MiniPrep kit ( Zymo Research ) . After elution of the RNA , the remaining genomic DNA was digested with TURBO DNA-free ( Ambion ) . The digestion of genomic DNA was confirmed by PCR amplification of the housekeeper genes . RNA was quantified via NanoDrop , and cDNA was prepared using the iScript cDNA Synthesis Kit ( Bio-Rad ) . The qPCR reaction was done with SsoFast Evagreen Supermix ( Bio-Rad ) using primers design with Beacon in a Bio-Rad CFX96 Real-Time PCR system . Data was analyzed with Bio-Rad CFX Manager 2 . 0 software using a regression Cq determination mode . Our housekeeper genes were nda3 , act1 , cdc2 , and cdc22 ( depending on the experiment ) . Each condition was run at least in triplicate and 3 independent experiments were performed . RNA was provided to the Oncogenomics Facility ( http://sylvester . org/research/shared-resources/laboratory-resources/oncogenomics-core-facility ) for the Bioanalyzer to assess RNA quality and amount , followed by microarray hybridization and scanning using the Affymetrix GeneChip Yeast Genome 2 . 0 Array . Data was then analyzed with MEV ( http://www . tm4 . org/mev ) ( TM4 Microarray Software Suite: TIGR MultiExperiment Viewer , RRID:SCR_001915 ) after conversion to RMA via RMAExpress ( http://rmaexpress . bmbolstad . com/ ) ( RMA Express , RRID:SCR_008549 ) . Gene ontology enrichment analysis was performed using Database for Annotation , Visualization , and Integrated Discovery ( DAVID ) Bioinformatics Resource 6 . 7 ( DAVID , RRID:SCR_001881 ) . Cultures were grown of wild-type cells and sts5-3xGFP cells for harvesting . Cell pellets were broken in breaking buffer ( 20 mM Tris-HCl ( pH 8 . 0 ) , 140 mM KCl , 1 . 8 mM MgCl2 , 0 . 1% NP-40 , 0 . 2 mg/ ml heparin , 0 . 5 mM DTT , protease inhibitors ( complete EDTA-free protease inhibitor cocktail tablets ( Roche Applied Science ) ) , 100 U/ml Rnasin Plus ( Promega ) ) with a Savant FastPrep FP120 bead beater . The Sts5 protein was then immunoprecipitated with anti-GFP ( Roche; RRID:AB_390913 ) and protein G magnetic resin ( Invitrogen ) . After extensive washing of the resin with wash buffer ( 20 mM Tris-HCl ( pH 8 . 0 ) , 140 mM KCl , 1 . 8 mM MgCl2 , 10% glycerol , 0 . 5 mM DTT , 0 . 01% NP-40 , 10 U/ml Rnasin Plus , and protease inhibitors in the beginning washes ) , the RNA was eluted from the resin by treating the resin with proteinase K . The RNA was then purified with a spin column kit ( ZR Fungal / Bacteria RNA MiniPrep Kit , Zymo Reseach ) . After elution of the RNA , the remaining genomic DNA was digested with TURBO DNAse ( Ambion ) and the digestion was confirmed by PCR . qPCR was then used to determine the relative levels of target mRNA in WT ( null IP ) versus the Sts5-3xGFP IP . Sts5 ORF ( a . a . 1–1066 ) was tagged with N-terminal His6 by cloning into pET15b expression vector . The construct was transformed in BL21 cells , and His6-Sts5 expression was induced by incubation with 1mM IPTG for 1 hr . Native His6-Sts5 was purified using Ni-NTA spin columns ( Qiagen ) following the manufacturers instructions . Western blot using anti-His6 antibody ( Covance; AB_10063707 ) was performed to confirm the purification of His6-Sts5 . In vitro kinase assay for phosphorylation of Sts5 was performed as described in Wiley et al . ( 2003 ) . Briefly , Myc-tagged Mob2 and untagged Mob2 were expressed in S . pombe cells grown to mid-log phase at 32°C . Cells lysis was performed using Savant FastPrep FP120 bead beater in HB buffer ( 25 mm MOPS , pH 7 . 2 , 60 mM β-glycerophosphate , 15 mM p-nitrophenyl phosphate , 15 mM MgCl2 15 mM EGTA , 1 mM dithiothreitol , 0 . 1 M sodium vanadate , 1% Triton X-100 , 1 mM phenylmethylsulfonyl fluoride , and protease inhibitors ( complete EDTA-free protease inhibitor cocktail tablets ( Roche Applied Science ) ) ) . Extracts from cells expressing Myc-tagged Mob2 and from wild-type cells were incubated with Protein A agarose ( Sigma-Aldrich ) beads bound to rabbit anti-Myc antibodies ( Santa Cruz Biotechnology; RRID:AB_631274 ) for 1 hr , washed twice with HB buffer , and then washed once with kinase buffer ( 50 mM Tris-HCl , pH 7 . 5 , 100 mM NaCl , 10 mM MgCl2 , 1 mM MnCl2 ) . The resin was resuspended in 25 μl of kinase buffer containing 10 μCi of [γ-32P]ATP ( 6000 Ci/mmol ) and 20 μM ATP and combined with 5 μl bacterially expressed Sts5 . The kinase reaction was stopped after 20 min at 30°C . Proteins were separated on an SDS polyacrylamide gel . The protein extraction was performed as previously described ( Matsuo et al . , 2006 ) . 10-ml cultures of exponentially growing cells were harvested by centrifuging at 5000 rpm for 5 min . The cell pellet was first washed in 1 mL of distilled water and then resuspended in 300 μL of distilled water . Then , 300 μL of 0 . 6 M NaOH was added , and cells were incubated at room temperature for 10 min and collected by centrifugation . After removing the supernatant , cells were resuspended in modified SDS sample buffer ( 60 mM Tris HCl pH 6 . 8 , 4% β-mercaptoethanol , 4% SDS , 0 . 01% bromophenol blue , and 5% glycerol ) and boiled for 3 min . The samples were then loaded on 4–15% Mini-PROTEAN TGX gels ( Biorad ) for routine western analysis . The primary antibodies used were mouse monoclonal anti-HA ( Covance; RRID:AB_2314672 ) , rabbit polyclonal purified antibody c-Myc ( A-14 ) ( Santa Cruz Biotechnology , Inc . ; RRID:AB_631274 ) rat monoclonal anti-α-tubulin [YL1/2] ( Novus Biologicals; RRID:AB_305328 ) , mouse monoclonal anti-α-tubulin clone B-5-1-2 ( Sigma-Aldrich; AB_477579 ) and rabbit polyclonal anti-GST ( Z-5 ) ( Santa Cruz; AB_631586 ) . The secondary antibodies used were IRDye 800 conjugated anti-mouse antibody ( Rockland Immunochemicals , Inc; RRID: RRID:AB_10703265 ) , IRDye 800 conjugated anti-rabbit antibody ( Rockland Immunochemicals , Inc; RRID:AB_220152 ) , and IRDye700 conjugated anti-rat antibody ( Rockland Immunochemicals Inc . ; RRID: AB_220171 ) . The blots were analyzed using the Odyssey Infrared Imaging system ( LI-COR Biosciences ) . Design and construction of the orb6-as2 analog-sensitive mutant was previously described ( Das et al . , 2009 ) . Inhibition of Orb6-as2 kinase was carried out using the ATP-analog 1-NA-PP1 ( 4-Amino-1-tert-butyl-3- ( 1’-naphtyl ) pyrazolo [3 , 4-d]pyrimidine; Toronto Research Chemicals ) diluted in DMSO . In liquid media , a final concentration of 50 μM 1-NA-PP1 was used to achieve Orb6-as2 kinase inhibition . In solid media , the final concentration of 1-NA-PP1 used was 10 μM . Bacterially expressed GST and GST-Rad24 were bound to Glutathione linked sepharose beads or magnetic beads ( Pierce ) . The beads were then mixed with fission yeast protein extract from wild type and Sts5-HA tagged strains incubated for overnight at 4°C . The beads were then washed with TRIS lysis buffer ( 50 mM TrisCl , PH 7 . 7; 150 mM NaCl; 5mM EDTA; 5% Glycerol; 1% Triton X; 1 mM PMSF; complete EDTA-free protease inhibitor cocktail tablets ( Roche Applied Science ) ) and separated by SDS polyacrylamide gel and analyzed by western blot using mouse monoclonal Anti-HA antibodies ( Covance; RRID:AB_2314672 ) . To inhibit Orb6 kinase , cells were incubated with either DMSO or 50 μM 1-NA-PP1 for 15 min at 32°C . The subcellular localization of ssp1 mRNA in cells expressing Sts5-3xGFP and Dcp1-mCherry was visualized using FISH , and our method was adapted from previously described protocols ( Heinrich et al . , 2013; Nilsson and Sunnerhagen , 2011; Brengues and Parker , 2007 ) with the following modifications . Custom Stellaris DNA probes targeted against ssp1 mRNA were coupled to Quasar 705 ( BioSearch Technologies ) . Cells were fixed with 4% paraformaldehyde for 20 min at room temperature and washed with buffer B ( 1 . 2 M sorbitol , 100 mM KHPO4 , pH 7 . 5 ) Cell walls were digested for 30 min in spheroplast buffer ( 1 . 2 M sorbitol , 100 mM KHPO4 at pH 7 . 5 , 20mM vanadyl ribonucleoside complex , 20 μM β-mercaptoethanol ) containing 5% Zymolyase 20T at room temperature . Cells were pelleted ( taking care to spin cells at ≤500 rpm for 3–5 min between washes in steps after the Zymolyase digestion ) then washed in buffer B . Cells were then incubated in 1 mL of -80°C methanol , stored overnight at -20°C , incubated in 1 mL of acetone for 1 min , and then washed twice in 1 mL of 2X SSC ( 0 . 3 M NaCl , 30 mM sodium citrate ) . Cells were preincubated at 37°C in 50 μl of hybridization buffer , consisting of a 1:1 ratio of Buffer F ( 20% formamide , 10 mM NaHPO4 at pH 7 . 0 ) and Buffer H ( 4X SSC , 4 mg/ml , 1 purified BSA and 20mM vanadyl ribonuclease complex ) and 2 μl of 10-mg/ml salmon-sperm DNA ( which was boiled for 3 min at 95°C ) . After 1 hr of prehybridization , 0 . 5 μl of 12 . 5 μM Quasar 705-conjugated ssp1 probe was added , and the cells were incubated at 37°C for 5 hr . Cells were washed two times with 2X SSC and resuspended in 2X SSC buffer . Object-based colocalization analysis ( based on the distance between centers of mass ) was performed using the ImageJ plugin JACoP ( Just Another Colocalization Plugin ) ( Bolte and Cordelières , 2006 ) . For threshold selection , we adapted a method similar to one previously described for image threshold selection of RNA FISH images ( Raj et al . , 2008 ) . Specifically , we applied a Laplacian of Gaussian filter to reduce noise and highlight areas of rapid change in each cell and chose the threshold where the histogram reached a plateau , indicating a region where above-background pixels can be clearly detected . pREP3X-Orb6- and pREP3X-carrying cells expressing Sts5-3xGFP and Dcp1p-mCherry were grown in absence of thiamine for 18 hr at 32°C . Cells were then washed once in minimal medium minus glucose and resuspended in minimal medium containing 2% or 0% glucose and the required supplements . Cultures were incubated at 32°C for 1 hr before visualizing the localization of Sts5-3xGFP and Dcp1-mCherry using fluorescence microscopy .	Living cells can grow to adopt a range of different shapes . For example , fission yeast cells maintain their rod-like shape by growing from their ends and then splitting in the middle to produce two new cells of an equal size . Like many other cells , fission yeast often responds to shortages of nutrients or other environmental stressors by growing more slowly or stopping its growth altogether . One way that this stress response is achieved is by preventing certain growth-promoting proteins from being made by storing or degrading the RNA molecules that are needed to make these proteins . Fission yeast uses an enzyme called Orb6 to control its growth and its overall shape . This enzyme is a kinase , meaning that it adds phosphate groups on to other proteins . Orb6 controls cell growth in part by defining the scope of action of an important growth control factor , Cdc42 . This is done by preventing the localization of a molecule that activates Cdc42 at the wrong places of the cell membrane , such as , for example , the cell sides . Now , Nuñez et al . show that the Orb6 enzyme also controls growth via a completely separate mechanism . Orb6 prevents an RNA-binding protein called Sts5 from being recruited into clusters of RNA molecules and proteins called granules , and directs Sts5 to interact with another protein called Rad24 instead . Since the RNA molecules in the granules tend to end up being degraded , preventing Sts5 from being recruited to the granules protects the RNA molecules that bind to Sts5 . Many Sts5-bound RNAs encode proteins required for cell growth , and in this manner Orb6 promotes the production of these Sts5-controlled proteins to encourage normal cell growth . In the future , Nuñez et al . would like to determine how Orb6 recognizes and responds to environmental signals to control cell growth . Further studies could also explore how Sts5 and Orb6 kinase affect the assembly of RNA-protein particles that have been implicated in diseases in humans .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2016	Spatial control of translation repression and polarized growth by conserved NDR kinase Orb6 and RNA-binding protein Sts5
Cell-penetrating peptides ( CPPs ) allow intracellular delivery of bioactive cargo molecules . The mechanisms allowing CPPs to enter cells are ill-defined . Using a CRISPR/Cas9-based screening , we discovered that KCNQ5 , KCNN4 , and KCNK5 potassium channels positively modulate cationic CPP direct translocation into cells by decreasing the transmembrane potential ( Vm ) . These findings provide the first unbiased genetic validation of the role of Vm in CPP translocation in cells . In silico modeling and live cell experiments indicate that CPPs , by bringing positive charges on the outer surface of the plasma membrane , decrease the Vm to very low values ( –150 mV or less ) , a situation we have coined megapolarization that then triggers formation of water pores used by CPPs to enter cells . Megapolarization lowers the free energy barrier associated with CPP membrane translocation . Using dyes of varying dimensions in CPP co-entry experiments , the diameter of the water pores in living cells was estimated to be 2 ( –5 ) nm , in accordance with the structural characteristics of the pores predicted by in silico modeling . Pharmacological manipulation to lower transmembrane potential boosted CPP cellular internalization in zebrafish and mouse models . Besides identifying the first proteins that regulate CPP translocation , this work characterized key mechanistic steps used by CPPs to cross cellular membranes . This opens the ground for strategies aimed at improving the ability of cells to capture CPP-linked cargos in vitro and in vivo . Cell-penetrating peptides ( CPPs ) are short non-toxic sequences of 5–30 amino acids present in proteins able to cross membranes such as homeoproteins and some viral components . CPPs can also be used to deliver bioactive cargos ( siRNAs , DNA , polypeptides , liposomes , nanoparticles , and others ) in cells for therapeutic or experimental purposes ( Bechara and Sagan , 2013; Futaki et al . , 2013; Guidotti et al . , 2017; Illien et al . , 2016; Jones and Sayers , 2012; Koren and Torchilin , 2012; Madani et al . , 2011; Mueller et al . , 2008; Ruseska and Zimmer , 2020; Trabulo et al . , 2010; Vasconcelos et al . , 2013 ) . Even though they differ in their origin ( Frankel and Pabo , 1988; Green and Loewenstein , 1988; Joliot et al . , 1991; Oehlke et al . , 1998 ) and physico-chemical properties , the majority of CPPs carry positive charges in their sequence ( Bechara and Sagan , 2013; Guidotti et al . , 2017; Jones and Sayers , 2012; Madani et al . , 2011 ) . Polyarginine ( e . g . R9 ) , HIV-1 TAT47-57 , and Penetratin ( Antennapedia43-58 ) are among the most used and studied CPPs . The mode of CPP cellular entry is still debated and no proteins have been identified that regulate this process . CPP entry starts after the initial electrostatic interactions between the positively charged CPP and the negatively charged components of the cell membrane ( Bechara and Sagan , 2013; Futaki et al . , 2013; Guidotti et al . , 2017; Jones and Sayers , 2012; Koren and Torchilin , 2012; Madani et al . , 2011; Ruseska and Zimmer , 2020; Trabulo et al . , 2010; Vasconcelos et al . , 2013 ) . Interaction with acid sphingomyelinase ( Verdurmen et al . , 2010 ) and glycosaminoglycans ( Amand et al . , 2012; Bechara et al . , 2013; Butterfield et al . , 2010; Fuchs and Raines , 2004; Futaki and Nakase , 2017; Ghibaudi et al . , 2005; Gonçalves et al . , 2005; Hakansson and Caffrey , 2003; Rullo et al . , 2011; Rusnati et al . , 1999; Ziegler , 2008; Ziegler and Seelig , 2004; Ziegler and Seelig , 2011 ) , local membrane deformation ( Hirose et al . , 2012 ) , as well as calcium fluxes ( Melikov et al . , 2015 ) have been suggested to play a role in CPP internalization . CPPs enter cells through a combination of two non-mutually exclusive mechanisms ( Illien et al . , 2016; Bechara et al . , 2013 ) : endocytosis and direct translocation ( Bechara and Sagan , 2013; Futaki et al . , 2013; Guidotti et al . , 2017; Jones and Sayers , 2012; Koren and Torchilin , 2012; Madani et al . , 2011; Ruseska and Zimmer , 2020; Trabulo et al . , 2010; Vasconcelos et al . , 2013 ) . The nature of these entry mechanisms is debated and not fully understood at the molecular level . The vesicular internalization of CPPs has been suggested to occur through clathrin-dependent endocytosis , macropinocytosis , and caveolin-1-mediated endocytosis ( Bechara and Sagan , 2013; Futaki et al . , 2013; Guidotti et al . , 2017; Jones and Sayers , 2012; Koren and Torchilin , 2012; Madani et al . , 2011; Trabulo et al . , 2010 ) . However , recent data indicate that CPP endocytosis proceeds via a newly discovered pathway that is Rab14-dependent but Rab5- and Rab7-independent ( Trofimenko et al . , 2021 ) . When CPPs are endocytosed , access to the cytosol requires that the CPPs break out of endosomes through a poorly understood process called endosomal escape . Direct translocation allows the CPPs to access the cytosol through their ability to cross the plasma membrane . There is currently no unifying model to explain mechanistically how direct translocation proceeds and no genes have yet been identified to modulate the manner by which CPPs cross cellular membranes . Direct translocation across the plasma membrane often seemed to originate from specific areas of the cells , suggesting discrete structures on the plasma membrane involved in CPP entry ( Allolio et al . , 2018; Duchardt et al . , 2007; Hirose et al . , 2012; Wallbrecher et al . , 2017; Ziegler et al . , 2005 ) . There is a general consensus though that an adequate plasma membrane potential ( Vm ) is required for direct translocation to occur based on live cell experiments ( Rothbard et al . , 2004; Wallbrecher et al . , 2017; Zhang et al . , 2009 ) , as well as in silico studies ( Gao et al . , 2019; Lin and Alexander-Katz , 2013; Moghal et al . , 2020; Via et al . , 2018 ) . Electrophysiological and pharmacological Vm modulations have revealed that depolarization blocks CPP internalization ( Rothbard et al . , 2004; Zhang et al . , 2009 ) and hyperpolarization improves the internalization of cationic CPPs ( Chaloin et al . , 1998; Henriques et al . , 2005; Moghal et al . , 2020; Rothbard et al . , 2004; Wallbrecher et al . , 2017 ) . By itself , a sufficiently low Vm ( i . e . hyperpolarization ) appears to trigger CPP direct translocation in live cells ( Rothbard et al . , 2004; Wallbrecher et al . , 2017; Zhang et al . , 2009 ) . In silico modeling has provided evidence that membrane hyperpolarization leads to the formation of transient water pores , allowing CPP translocation into cells ( Gao et al . , 2019; Herce and Garcia , 2007; Herce et al . , 2009; Lin and Alexander-Katz , 2013; Via et al . , 2018 ) , but the free energy landscape governing CPP translocation has not been determined . Moreover , the nature and the structural characteristics of the pores used by CPPs to cross the plasma membrane have not been investigated in live cells . Here , we provide the first genetic evidence that validates the importance of Vm for CPP direct translocation and we characterize the diameter of the water pores used by CPPs to enter live cells . We also determined the role of the Vm in modulating the free energy barrier associated with membrane translocation and the impact of the Vm on CPP translocation kinetics . In the present work , we have used TAT-RasGAP317-326 as a model compound to investigate the molecular basis of CPP cellular internalization . This peptide is made up of the TAT48-57 CPP and a 10 amino acid sequence derived from the SH3 domain of p120 RasGAP ( Michod et al . , 2004 ) . TAT-RasGAP317-326 sensitizes cancer cells to chemo- , radio- , and photodynamic therapies ( Chevalier et al . , 2015; Michod et al . , 2009; Pittet et al . , 2007; Tsoutsou et al . , 2017 ) and prevents cell migration and invasion ( Barras et al . , 2014 ) . This peptide also exhibits antimicrobial activity ( Heulot et al . , 2017; Georgieva et al . , 2021; Heinonen et al . , 2021 ) . Some cancer cell lines , such as Raji ( Burkitt’s lymphoma ) , SKW6 . 4 ( transformed B-lymphocytes ) , and HeLa ( cervix carcinoma ) , are directly killed by this peptide ( Heulot et al . , 2016 ) . The manner by which TAT-RasGAP317-326 kills cells has recently been uncovered ( Serulla et al . , 2020 ) . The peptide first accesses the cell’s cytosol by direct translocation through the plasma membrane . It then binds to specific phospholipids , such as phosphatidylserine and phosphatidylinositol-bisphosphate that are enriched in the inner leaflet of the plasma membrane . This binding allows the peptide to disrupt the cell’s membrane causing its death by necrosis . Most CPPs can enter cells by direct translocation and by endocytosis ( Bechara and Sagan , 2013; Futaki et al . , 2013; Guidotti et al . , 2017; Illien et al . , 2016; Jones and Sayers , 2012; Koren and Torchilin , 2012; Madani et al . , 2011; Mueller et al . , 2008; Ruseska and Zimmer , 2020; Trabulo et al . , 2010; Vasconcelos et al . , 2013 ) . This is also the case for TAT-RasGAP317-326 ( Figure 1A–B and Videos 1–3 ) . Two types of staining were observed in cells incubated with this peptide: ( i ) vesicular only and ( ii ) vesicular and cytosolic ( Figure 1A and Figure 1—figure supplement 1A ) . When the peptide cytosolic signal was strong , it masked the vesicular staining ( Figure 1A ) . In our experimental settings , the cytosolic acquisition of TAT-RasGAP317-326 occurred only through direct translocation and not through endosomal escape ( Figure 1—figure supplement 2 , Video 4 ) and was not due to phototoxicity ( Figure 1—figure supplement 3 ) as can occur in some settings ( Dixit and Cyr , 2003; Ha and Tinnefeld , 2012; Levitus and Ranjit , 2011; Zheng et al . , 2014 ) . As TAT-RasGAP317-326 needs to translocate through the plasma membrane to reach the cytosol , a prerequisite for the peptide to kill cells ( Serulla et al . , 2020 ) , we used the killing ability of the peptide in a CRISPR/Cas9 screen to identify genes involved in CPP direct translocation in two different cell lines ( Raji and SKW6 . 4 cells ) ( Figure 2—figure supplement 1A ) . The top candidate genes identified through this approach were specific potassium channels or genes coding for proteins known to regulate such channels indirectly ( e . g . PIP5K1A; Suh and Hille , 2008; Figure 2A and Figure 2—figure supplement 1B ) . KCNQ5 , identified in Raji cells , is a voltage-dependent potassium channel . KCNN4 and KCNK5 , identified in SKW6 . 4 cells , are calcium-activated channels and belong to the two-pore ( voltage-independent ) potassium channel family ( Shieh et al . , 2000 ) , respectively . These potassium channels were pharmacologically or genetically inactivated ( Figure 2B and Figure 2—figure supplement 2A-C ) to validate their involvement in the direct translocation of TAT-RasGAP317-326 through the plasma membrane and the resulting death induction . The KCNQ family inhibitor , XE-991 ( Schroeder et al . , 2000 ) , as well as KCNQ5 genetic invalidation ( Figure 2—figure supplement 2A ) , fully blocked peptide internalization in Raji cells and protected them from the killing activity of the peptide ( Figure 2B and Figure 2—figure supplement 2D ) . SKW6 . 4 cells individually lacking KCNN4 or KCNK5 ( Figure 2—figure supplement 2B ) , or SKW6 . 4 cells treated with TRAM-34 , a KCNN4 inhibitor ( Wulff et al . , 2001; Wulff et al . , 2000 ) , were impaired in their ability to take up the peptide and were partially protected against its cytotoxic activity ( Figure 2B and Figure 2—figure supplement 2D ) . Inhibition of KCNN4 activity with TRAM-34 in KCNK5 knock-out cells did not further protect the cells against TAT-RasGAP317-326-induced death . In HeLa cells , TRAM-34 , but not XE-991 , inhibited TAT-RasGAP317-326 internalization and subsequent death ( Figure 2B ) . Thus , in HeLa cells , KCNN4 channels regulate the membrane translocation of the peptide . This was confirmed by knocking out KCNN4 in these cells ( Figure 2B and Figure 2—figure supplement 2C ) . Resistance to TAT-RasGAP317-326-induced death in KCNQ5 knock-out Raji cells and KCNN4 knock-out SKW6 . 4 or HeLa cells was restored through ectopic expression of the corresponding FLAG- or V5-tagged channels ( Figure 2—figure supplement 2E-F ) , ruling out off-target effects . We next determined whether vesicular internalization or direct translocation were affected in cells with impaired potassium channel activities . Compared to their respective wild-type controls , the percentage of cells with diffuse cytosolic location of FITC-TAT-RasGAP317-326 was drastically diminished in cells lacking one of the CRISPR/Cas9 screen-identified potassium channels in the respective cell lines ( Figure 2C and Figure 2—figure supplement 3A-B ) . This was mirrored by an increase in the percentage of knock-out cells with vesicular staining . The invalidation of potassium channels did not affect transferrin or dextran internalization into cells ( Figure 2—figure supplement 3C ) or the infectivity of vesicular stomatitis virus ( Torriani et al . , 2019 ) , substantiating the non-involvement of these channels in endocytic pathways . One possibility to explain the above-mentioned results is that the absence of potassium channels reduces peptide binding to cells , thereby hampering subsequent peptide cellular uptake . At a 20 μM concentration , TAT-RasGAP317-326 is readily taken up by wild-type Raji cells but not by KCNQ5 knock-out cells . At this concentration , peptide binding was slightly lower in knock-out than in wild-type cells ( Figure 2—figure supplement 3D , upper graph ) . However , augmenting the peptide concentrations in the extracellular medium of KCNQ5 knock-out cells to reach surface binding signals equivalent or higher than what was obtained in wild-type cells still did not result in peptide cellular internalization unless ≥80 μM of the peptides were used and even in this case , the uptake remained inefficient ( Figure 2—figure supplement 3D ) . Difference in peptide binding is therefore not the cause of the inability of potassium channel knock-out cells to take up TAT-RasGAP317-326 . We then assessed whether the role of potassium channels in cellular internalization also applied to TAT cargos other than RasGAP317-326 . TAT-PNA is an oligonucleotide covalently bound to TAT , which can correct a splicing mutation within the luciferase-coding sequence ( Abes et al . , 2007; Kang et al . , 1998 ) . This can only occur if TAT-PNA reaches the cytosol . The luciferase activity triggered by TAT-PNA was diminished in the presence of potassium channel inhibitors and in potassium channel knock-out cell lines ( Figure 2—figure supplement 4A ) . Cytosolic access of TAT-Cre , which can recombine a loxP-RFP-STOP-loxP-GFP ( D’Astolfo et al . , 2015; Wadia et al . , 2004 ) gene construct , was then assessed . Switch from red to green fluorescence occurs only when TAT-Cre reaches the nucleus . This took place in wild-type Raji cells but not in the KCNQ5 knock-out cells ( Figure 2—figure supplement 4B ) . We finally tested a clinical phase III therapeutic D-JNKI1 compound ( Guidotti et al . , 2017; Vasconcelos et al . , 2013 ) used in the context of hearing loss and intraocular inflammation . The internalization of this peptide was completely blocked in Raji cells lacking KCNQ5 ( Figure 2—figure supplement 4C , left ) . D-JNKI1 internalization was also diminished in SKW6 . 4 cells lacking KCNN4 and KCNK5 channels , as well as in HeLa cells lacking the KCNN4 potassium channel ( Figure 2—figure supplement 4C , middle and right panels ) . These data demonstrate that the absence of specific potassium channels diminishes or even blocks the entry of various TAT-bound cargos . Potassium is the main ion involved in setting the plasma membrane potential ( Vm ) . The potassium channels identified in the CRISPR/Cas9 screen may therefore participate in the establishment of an adequate Vm permissive for CPP direct translocation ( Chaloin et al . , 1998; Henriques et al . , 2005; Moghal et al . , 2020; Rothbard et al . , 2004; Wallbrecher et al . , 2017; Zhang et al . , 2009 ) . Figure 3A ( left graph ) shows that genetic disruption or pharmacological inhibition of KCNQ5 in Raji cells led to an increase in their Vm ( from –26 to –15 mV , validated with electrophysiological recordings; see Figure 3—figure supplement 1A ) . Surprisingly , such minimal increase in Vm in the KCNQ5 knock-out Raji cells practically abolished CPP internalization ( Figure 3B , left graph ) , indicating that above a certain threshold , the Vm is no longer permissive for CPP direct translocation . In SKW6 . 4 and HeLa cells , Vm measurement was much more variable than in Raji cells . Nevertheless , a trend of increased Vm was observed when KCNN4 or KCNK5 were invalidated genetically or pharmacologically ( Figure 3A , middle and right graphs ) and this was accompanied by reduced peptide uptake ( Figure 3B , middle and right graphs ) . As the CRISPR/Cas9 screens performed in various cell lines identified a variety of potassium channels required for efficient CPP internalization , we conclude that it is the Vm maintenance activity of these channels that is important for CPP direct translocation and not some specific features of the channels . If the reason why invalidation of the KCNQ5 , KCNN4 , and KCNK5 potassium channels inhibits TAT-RasGAP317-326 cellular entry is cell depolarization , a similar response should be obtained by artificially depolarizing cells . Indeed , depolarizing cells with gramicidin ( Eisenman et al . , 1978 ) ( making non-specific 0 . 4 nm pores [Kelkar and Chattopadhyay , 2007] in cell membranes ) or by increasing the extracellular concentration of potassium ( dissipating the potassium gradient ) totally blocked cytosolic peptide acquisition into the three studied cell lines ( Figure 3B ) but not peptide endocytosis ( Figure 3—figure supplement 1B ) . Hence , cellular depolarization in itself inhibits TAT-RasGAP317-326 direct translocation into the cytosol . Next , we determined whether hyperpolarization could reverse the inability of potassium channel knock-out cells to take up TAT-RasGAP317-326 . Cells were either incubated in the presence of valinomycin ( Rimmele and Chatton , 2014 ) , which leads to formation of potassium-like channels , or transfected with KCNJ2 channel that also provokes potassium efflux and membrane hyperpolarization ( Xue et al . , 2014 ) . Figure 3B shows that hyperpolarization of cells lacking CRISPR/Cas9-identified potassium channels fully restored peptide translocation . Moreover , hyperpolarization increased peptide cytosolic acquisition in wild-type cells ( Figure 3B ) . Similar effect , albeit to a lesser extent , was observed by ectopically expressing KCNQ5 in wild-type and KCNN4 knock-out SKW6 . 4 and HeLa cells as well as by ectopically expressing KCNN4 in wild-type and KCNQ5 knock-out Raji cells ( Figure 3—figure supplement 1C ) . Additionally , cells such as primary rat cortical neurons that naturally have a low Vm ( –48 mV ) take up the CPP in their cytosol more efficiently than cells with higher Vm such as HeLa cells ( –25 mV ) ( Figure 3—figure supplement 1D ) . Altogether , these results demonstrate that the Vm modulates internalization of TAT-RasGAP317-326 in various cell lines . This internalization can be manipulated through cellular depolarization to block it and through hyperpolarization to increase it , confirming earlier results obtained for the R8 CPP in Jurkat cells ( Rothbard et al . , 2004 ) . We then assessed whether the entry of TAT , nanomeric arginine ( R9 ) , and Penetratin ( Figure 3—figure supplement 2A ) , three commonly used cationic CPPs in biology and medicine , was regulated by the plasma membrane potential as shown above for TAT-RasGAP317-326 . Similarly to TAT-RasGAP317-326 , these CPPs are taken up by HeLa cells by both direct translocation and endocytosis ( Figure 3—figure supplement 2B-C ) . Depolarization , induced by either gramicidin or high extracellular potassium concentrations ( Figure 3D ) , led to decreased cytosolic fluorescence of these CPPs , while valinomycin-mediated hyperpolarization favored their translocation in the cytosol ( Figure 3C , upper graphs , Figure 3E , and Figure 3—figure supplement 2D-E ) . Although the cellular membrane composition of neurons may differ from the other cell lines used in this study , the Vm also controlled peptide translocation in non-transformed rat primary cortical neurons ( Figure 3—figure supplement 3 ) . In contrast , depolarization had no impact on the ability of the cells to endocytose these CPPs ( Figure 3C , bottom graphs ) , further confirming that CPP endocytosis is not affected by Vm . Finally , we note that CPP membrane binding was only minimally affected by depolarization ( Figure 3—figure supplement 4 ) . Hence , the reason why depolarized cells do no take up CPPs is not a consequence of reduced CPP binding to cells , confirming our earlier observation obtained with TAT-RasGAP317-326 ( Figure 2—figure supplement 3D ) . Altogether the data presented in Figure 3 show that direct translocation of cationic CPPs is modulated by the Vm of cells and that specific potassium channels are involved in this modulation . To further study the mechanism of CPP cellular entry through direct translocation , we took advantage of coarse-grained molecular dynamics ( MD ) technique and MARTINI force field 2 . 2p ( Marrink et al . , 2007; Marrink and Tieleman , 2013 ) . In our simulations we have used TAT-RasGAP317-326 , TAT , R9 and Penetratin in presence of a natural cell membrane-like composition ( for both inner and outer leaflets ) while earlier studies have employed simpler membrane composition ( Gao et al . , 2019; Herce et al . , 2009; Lin and Alexander-Katz , 2013; Moghal et al . , 2020; Via et al . , 2018; Zhang et al . , 2009 ) . Membrane hyperpolarization was achieved by setting an ion imbalance ( Delemotte et al . , 2008; Gao et al . , 2019; Gurtovenko and Vattulainen , 2007; Herrera and Pantano , 2009 ) through a net charge difference of 30 positive ions ( corresponding to a Vm of ~2 V ) between the intracellular and extracellular spaces . The use of very high Vm values , typically used in computational studies , is required to capture nanosecond occurring events . This protocol ( Figure 4—figure supplement 1A ) allowed us to observe CPP translocation across membranes within a few tens of nanoseconds ( Figure 4A and Video 5 ) . In presence of ~2 V Vm , the CPPs approached the membrane on the extracellular side and this led to the formation of a water column within the membrane that the CPP then used to move to the intracellular space ( Video 5 ) . The movement of the positive charges carried by the CPPs , as well as extracellular cations , to the intracellular compartment via the water pore induced membrane depolarization . This depolarization provoked the collapse of the water pore and membrane resealing . Even though CPPs play an active role in their internalization , the mere presence of the CPP in the absence of a sufficiently low Vm was not enough to trigger water pore formation ( Figure 4B , right graph and Video 6 ) . These data confirm earlier work describing the role of the Vm in CPP penetration into or through bilipidic membranes ( Gao et al . , 2019; Herce and Garcia , 2007; Herce et al . , 2009; Lin and Alexander-Katz , 2013; Moghal et al . , 2020; Via et al . , 2018; Zhang et al . , 2009 ) . TAT , R9 , and Penetratin all translocated into the intracellular compartment but with different propensities ( Figure 4B , left graph ) and with different kinetics ( Figure 4—figure supplement 1B ) that appeared to be related to the positive charges they carry ( Figure 3—figure supplement 2A ) : the more positively charged a CPP , the higher probability to translocate across cell membranes and the faster kinetics of water pore formation at a given Vm . We also applied a metadynamics protocol to estimate the impact of the Vm on the free energy landscape of R9 translocation . The free energy barriers recorded in depolarized membranes ( Vm = 0 ) and polarized membranes ( Vm = –80 mV ) were similar ( Figure 4C–D ) . The obtained value of about 200 kJ/mol is in line with recent estimation of the free energy barrier associated with CPP translocation at a Vm = 0 ( Gao et al . , 2019 ) . Only at much lower Vm values ( –150 mV ) was a marked decrease in free energy barrier recorded . This indicates that hyperpolarization values found in resting cells ( down to about –80 mV in neurons and higher in many other cells types; Yang and Brackenbury , 2013 ) are not more favorable than fully depolarized membranes to establish conditions for the formation of water pores . It appears therefore that cells need to decrease their Vm to much lower values ( e . g . –150 mV or lower ) to reach conditions compatible with water pore formation . This in silico observation may appear contradictory with our results obtained in live cells showing direct translocation at –25 mV ( Figure 2 ) , as well as with the experiment demonstrating that CPP cytosolic internalization was more efficient in cortical neurons in comparison to less negatively charged HeLa cells ( Figure 3—figure supplement 1D ) . We therefore postulate that the presence of CPPs on the cell surface induces locally a substantial voltage drop from the resting Vm . To test this assumption , we analyzed the electrostatic potential map in a molecular system composed of the R9 peptide in contact with the plasma membrane in the absence of an external electrostatic field ( Figure 4E ) . This simulation indicated that the presence of CPPs at the cell surface is sufficient to decrease locally the transmembrane potential to about –150 mV ( Figure 4E ) . This was not observed in the absence of the CPP . In conclusion , our data support a model where CPPs further decrease the Vm of resting cells to very low values ( equal to or less than –150 mV ) that are compatible with spontaneous water pore formation and that we coin megapolarization . Our model also predicts that the electric force exerted on CPPs when cells are megapolarized permit CPPs to accumulate in the cytosol and reach concentrations that are higher than in the extracellular milieu . Figure 4—figure supplement 1C shows indeed that cells can concentrate TAT-RasGAP317-326 in the cytosol of Raji and HeLa cells , up to 100 times in extreme cases . Propidium iodide ( PI ) , with a diameter of 0 . 8–1 . 5 nm ( Bowman et al . , 2010 ) or fluorophore-labeled 3 , 10 , and 40 kDa dextrans , with diameters ( provided by Thermo Fisher ) of 2 . 3 ± 0 . 38 ( Thorne and Nicholson , 2006 ) , 4 . 5 , and 8 . 6 nm , respectively ( Figure 5—figure supplement 1A ) , were used to estimate the size of the water pores formed in the presence of CPPs in live cells . These molecules by themselves did not translocate in the cytosol of cells ( Figure 5A and Figure 5—figure supplement 1B ) . They were then co-incubated with different FITC-labeled CPPs and their uptake monitored by confocal microscopy . While PI and CPPs efficiently co-entered cells ( Figure 5B and Figure 5—figure supplement 1C-D ) , there was only marginal co-entry of the dextrans with the CPPs ( Figure 5B ) . The marginal dextran co-entry was inversely correlated with the dextran diameters ( inset in Figure 5B ) : ~2 . 3-nm-wide dextrans entered cells better than ~4 . 5-nm-wide dextrans and ~8 . 6-nm-wide dextrans mostly remained outside cells . The entry of PI and CPPs in cells occurred with identical kinetics ( Figure 5—figure supplement 1D ) , further supporting the notion that they enter cells together . The PI/CPP co-entry was prevented by cell depolarization ( Figure 5—figure supplement 1B ) , which is expected if PI accesses the cytosol via the megapolarization-induced pores used by CPP to enter cells . CPPs do not need to be labeled with a fluorophore to allow PI co-entry into cells ( Figure 5B , ‘unlabeled TAT+ PI’ condition ) , ruling out phototoxicity as a confounding effect . Similar results were obtained in primary rat cortical neurons , where PI cytosolic signal was observed in cells that took up the selected CPPs through direct translocation ( Figure 5C ) . These data are compatible with the notion that water pores triggered by CPPs allow molecules up to ~2 nm in diameter to efficiently enter cells . They are also in line with the in silico prediction of the water pore diameter of 1 . 6±0 . 26 nm obtained by analyzing the structure of the pore at the transition state ( i . e . when the CPP is crossing the cell membrane; see Figure 4A ) . Molecules in the 2–5 nm diameter range , such as 3 and 10 kDa dextrans , can still use this entry route to a limited extent . In this context , the Cre recombinase , with a diameter of 5 nm ( estimated from its crystal structure; NDB:PD0003 ) , can be transported by TAT into cells ( Figure 2—figure supplement 4B ) , another indication that the pores used by cationic CPPs to enter cells can allow the passage of molecules up to 5 nm . Despite identical net positive charges ( Figure 3—figure supplement 2A ) , and as reported earlier ( Mitchell et al . , 2000 ) , the K9 peptide made of nine lysine residues was less capable of translocating into cells compared to R9 ( Figure 5B and Figure 3—figure supplement 2C , right graph ) . This may be due to the deprotonation of K9 once in the plasma membrane ( see Discussion ) . However , in the few cases when cells have taken up K9 , PI co-internalized as well ( middle graph of Figure 5B ) . This indicates that K9 has a reduced capacity compared to R9 to trigger water pore formation but when they do , PI can efficiently translocate through the pores created by K9 . Modeling experiments indicate that water pores are created in membranes subjected to sufficiently high ( absolute values ) Vm . We therefore tested whether the mere hyperpolarization of cells ( i . e . in the absence of CPPs ) could trigger the translocation of PI into cells , indicative of water pore formation . Figure 5D ( left ) shows that the hyperpolarizing drug valinomycin significantly increased PI cell permeability . In contrast , depolarization , mediated by gramicidin , reduced PI internalization ( Figure 5D , left ) . Cells incubated with CPPs took up PI in their cytosol to a much greater extent than when cells were treated with valinomycin ( Figure 5D , left ) , as expected if CPPs participate in setting plasma membrane megapolarization . Figure 5E shows the correlation between cytosolic PI accumulation over time and Vm . Based on this correlation , we estimated the Vm of cells incubated with a CPP to be in the order of –150 mV ( Figure 5F ) . In accordance with the modeling experiments , these data further support the notion ( i ) that water pore formation in cells is favored by cell hyperpolarization and inhibited by depolarization and ( ii ) that CPPs themselves ( Rao et al . , 2014; Wallbrecher et al . , 2017 ) further contribute to the establishment of local megapolarization in the plasma membrane . We investigated whether it was possible to experimentally manipulate the Vm to favor CPP internalization in in vivo situations . Systemic exposure of zebrafish embryos to valinomycin in Egg water led to cell hyperpolarization ( Figure 6—figure supplement 1A ) and improved internalization of a TAT-based CPP ( Figure 6—figure supplement 1B ) . This systemic treatment , while not acutely toxic , halted development ( Figure 6—figure supplement 1C-E ) . However , local valinomycin injection did not affect long-term viability ( Figure 6—figure supplement 1F ) and efficiently increased CPP cellular internalization ( Figure 6A ) . Subcutaneous injections of valinomycin in mice induced tissue hyperpolarization ( Figure 6—figure supplement 1G ) and boosted the CPP delivery in skin cells ( Figure 6B ) . These results demonstrate that hyperpolarizing drugs can be used to ameliorate CPP internalization in animal tissues . Multiple models , mostly inferred from artificial experimental paradigms , have been proposed to explain CPP direct translocation . These include the formation of pores made of the CPPs themselves that they use for their own entry , the formation of inverted micelles in the plasma membrane that translocate the CPPs , or diffusion of the CPPs across the plasma membrane ( Bechara and Sagan , 2013; Futaki et al . , 2013; Guidotti et al . , 2017; Koren and Torchilin , 2012; Trabulo et al . , 2010 ) . Our simulation and cellular data , while providing no evidence for such models , demonstrate that CPP cellular internalization is potassium channel- and Vm-dependent in vitro and in vivo . Potassium channels are required to establish a basal low Vm , subsequently permissive for CPP direct translocation . Hyperpolarizing drugs , such as valinomycin , enhance permissiveness . When CPPs come into contact with the plasma membrane , they decrease even more the Vm , resulting in a locally megapolarized membrane . This increases the likelihood of water pore formation that the CPPs then use to penetrate into cells according to their electrochemical gradient ( Figure 7 ) . Water pores are created by a combination of lipid head group reorientation coupled to intrusion of a column of water in the membrane bilayer . Water movement plays therefore an active role in the formation of the pore and is not merely occurring once the pores are formed . The movement of the positive charges carried by the CPPs into the cell , as well as the transport of extracellular cations ( e . g . Na+ ) , dissipates the Vm , resulting in the collapse of the water pores and sealing of the plasma membrane . CPP-mediated formation of water pores is therefore transient and does not affect cell viability . Multiple rounds of CPP-driven water pore formation and CPP translocation into cells can lead to intracellular accumulation of the CPP to concentrations higher than found outside cells ( Figure 4—figure supplement 1C ) . It has not been possible to measure directly the precise values of the Vm that allow the formation of water pores used by CPPs to enter cells . Using an indirect calculation mode based on the uptake of PI alongside CPPs , we have estimated that a Vm in the order of –150 mV is required for water pores to be formed ( Figure 5F ) . This might be an underestimation however as modeling data indicate that , at –150 mV , the free energy barrier , while being markedly diminished compared to those calculated at –80 or 0 mV , is not fully abrogated ( Figure 4D ) . Possibly therefore , the local Vm where CPPs interact with the plasma membranes is much lower than –150 mV and/or changes in CPP structures occur ( e . g . refolding , aggregation ) leading to further reduction in free energy barrier . It is worth mentioning that the applied coarse-grained MARTINI force field , as any other model , has a number of known limitations ( Marrink et al . , 2019; Marrink and Tieleman , 2013 ) such as the chemical and spatial resolution , which are both limited compared to atomistic models . There is also a shifted balance between entropy and enthalpy due to the reduced number of degrees of freedom . Moreover , the secondary structure is an input parameter of the model , which implies that secondary structure elements remain fixed during the simulation ( Monticelli et al . , 2008 ) . However , the coarse-grained approach has provided reliable results in the context of protein-membrane interactions and peptide translocation ( Castillo et al . , 2013; Koch et al . , 2019; Marrink et al . , 2003; Monticelli et al . , 2008; Monticelli et al . , 2010; Periole et al . , 2009; Periole et al . , 2007; Ramadurai et al . , 2010; Yesylevskyy et al . , 2010 ) . Moreover , the ability of MARTINI coarse-grained force field to model realistic and heterogeneous membranes has been repeatedly reported in literature , as summarized in a recent review paper ( Marrink et al . , 2019 ) . Our model posits that the number of positively charged amino acids influence the ability of CPPs to hyperpolarize cells and hence to form water pores that they take to translocate into cells . CPP hydropathy strongly correlates with penetration of water molecules in the lipid bilayer , thus supporting the hypothesis that the amount of water each CPP can route inside the membrane is modulated by the hydrophobic and hydrophilic character of the peptide ( Grasso et al . , 2018 ) . The nature of cationic amino acids in peptides determines their translocation abilities . It is known for example that peptides made of nine lysines ( K9 ) poorly reaches the cytosol ( Figure 5B and Figure 3—figure supplement 2C ) and that replacing arginine by lysine in Penetratin significantly diminishes its internalization ( Amand et al . , 2012; Mitchell et al . , 2000 ) . According to our model , K9 should induce megapolarization and formation of water pores that should then allow their translocation into cells . However , it has been determined that , once embedded into membranes , lysine residues tend to lose protons ( Armstrong et al . , 2016; Li et al . , 2013; MacCallum et al . , 2008 ) . This will thus dissipate the strong membrane potential required for the formation of water pores and prevent lysine-containing CPPs to cross the membrane . In contrast , arginine residues are not deprotonated in membranes and water pores can therefore be maintained allowing the arginine-rich CPPs to be taken up by cells . This phenomenon was not modeled in our coarse-grained in silico simulations because the protonation state was fixed at the beginning of the simulation runs and was not allowed to evolve . An additional potential explanation for the internalization differences observed between arginine- and lysine-rich peptides is that even though both arginine and lysine are basic amino acids , they differ in their ability to form hydrogen bonds , the guanidinium group of arginine being able to form two hydrogen bonds ( Fromm et al . , 1995 ) while the lysyl group of lysine can only form one . Compared to lysine , arginine would therefore form more stable electrostatic interactions with the plasma membrane . According to previously published studies ( Kosuge et al . , 2008; Mitchell et al . , 2000 ) , the optimal length of consecutive arginine residues appears to be between 9 and 16 amino acids , resulting in optimal CPP cytosolic acquisition . Shorter and longer peptides have decreased internalization efficiencies . The role of the Vm presented in our model is consistent with the reduced uptake of short polyarginine peptides but the Vm parameter of our model cannot explain why longer polyarginine peptides are less efficiently taken up by cells . Our work however also indicates that the water pores created by megapolarization have a diameter of about 2 ( –5 ) nm . Molecules larger than 2 nm are therefore less efficiently transported through these water pores and if polyarginine peptides reach that size their internalization will be hindered . The efficiency of direct translocation of peptides is therefore likely modulated by their sizes , the secondary structures they adopt , and the number of positive charges they carry . Cationic residues are not the only determinant in CPP direct translocation . The presence of tryptophan residues also plays important roles in the ability of CPPs to cross cellular membranes . This can be inferred from the observation that Penetratin , despite only bearing three arginine residues , can penetrate cells with similar propensities compared to R9 or TAT that contain 9 and 8 arginine residues , respectively . The aromatic characteristics of tryptophan is not sufficient to explain how it favors direct translocation as replacing tryptophan with the aromatic amino acid phenylalanine decreases the translocation potency of the RW9 ( RRWWRRWRR ) CPP ( Derossi et al . , 1994 ) . Rather , differences in the direct translocation promoting activities of tryptophan and phenylalanine residues may come from the higher lipid bilayer insertion capability of tryptophan compared to phenylalanine ( Christiaens et al . , 2002; Jobin et al . , 2015; MacCallum et al . , 2008 ) . There is a certain degree of interchangeability between arginine and tryptophan residues as demonstrated by the fact that replacing up to four arginine residues with tryptophan amino acids in the R9 CPP preserves its ability to enter cells ( Walrant et al . , 2020 ) . Therefore , despite the importance of the membrane potential for CPP direct translocation into cells , other factors also appear to play a role in this process . While the nature of the CPPs likely dictate their uptake efficiency as discussed in the previous paragraph , the composition of the plasma membrane could also modulate how CPPs translocate into cells . In the present work , we have recorded CPP direct translocation in transformed or cancerous cell lines as well as in primary cells . These cells display various abilities to take up CPPs by direct translocation and the present work indicates that this is modulated by their Vm . But as cancer cells display abnormal plasma membrane composition ( Szlasa et al . , 2020 ) , it will be of interest in the future to determine how important this is for their capacity to take up CPPs . We propose , based on the work described here , that hyperpolarization induced by drugs such as valinomycin represents a simple alternative or parallel approach to optimize CPP internalization . However , hyperpolarizing drugs may be toxic when systemically applied . For example , valinomycin at the concentrations used to induce hyperpolarization ( 10 μM ) would be lethal if systemically injected in mice ( LD50 in the low micromolar range; Daoud and Juliano , 1986 ) . On the other hand , local administration of valinomycin is far less toxic ( Gad et al . , 1985; Waksman , 1953 ) as confirmed here in zebrafish and mice . Hyperpolarizing agents may therefore be preferentially used for local or topical applications , which is incidentally the case for the clinically approved CPPs ( Abes et al . , 2007; Abraham et al . , 2015 ) . Strategies to improve CPP delivery are becoming increasingly elaborate through the use of nanoparticles ( Bansal et al . , 2018 ) , double-coated nanoparticles ( Khalil et al . , 2018 ) , liposome-polycation-DNA complexes ( Wu et al . , 2018 ) , branched peptides ( Jeong et al . , 2016 ) , etc . Our molecular characterization of the process of CPP direct translocation can be taken advantage of to ( i ) improve or optimize ‘old’ CPPs , ( ii ) design new CPPs , ( iii ) help explain the behavior of newly discovered CPPs ( Du et al . , 2011; Kauffman et al . , 2018; Yin et al . , 2009 ) , ( iv ) discriminate between target cells and cells that should be left unaffected based on Vm , and ( v ) distinguish between direct translocation and endosomal escape . The present work indicates that the impact on megapolarization should be evaluated when chemical modifications are performed on cationic CPPs to augment their delivery capacities . Puromycin 10 mg/ml ( Thermo Fisher , ref no . A11138-02 ) was aliquoted and stored at –20°C . Blasticidin ( Applichem , ref no . A3784 ) was dissolved at 1 mg/ml in water and stored at –20°C . XE-991 and TRAM-34 ( Alomone Labs , ref no . X-100 and T-105 , respectively ) was dissolved in DMSO at 100 mM and stored at –20°C . Cells were pre-incubated with 10 μM of these inhibitors for 30 min and then kept throughout the experiments . Live Hoechst 33342 ( Sigma , ref no . CDS023389 ) was aliquoted and stored at –20°C . Trypan Blue 0 . 4% ( Life Technologies , ref no . 15250061 ) was stored at room temperature . AlexaFluor488-labeled human transferrin was dissolved in PBS at 5 mg/ml and stored at 4°C ( Thermo Fisher , ref no . 13342 ) . TexasRed-labeled neutral 3000 and 40 , 000 Da dextran was dissolved in PBS at 10 mg/ml and stored at –20°C ( Thermo Fisher , ref no . D3329 and D1829 , respectively ) . TMR-labeled 10 , 000 neutral dextran was dissolved in PBS at 10 mg/ml and stored at –20°C ( Thermo Fisher , ref no . D1816 ) . The rabbit polyclonal anti-V5 ( Bethyl , ref no . A190-A120 ) , mouse monoclonal anti-FLAG antibody was from Sigma-Aldrich ( ref no . F1804 ) , rabbit monoclonal anti-actin ( Cell Signaling , ref no . 4970 ) , and rat monoclonal anti-γ-tubulin ( Santa Cruz , ref no . sc-51715 ) antibodies were used for Western blotting . All cell lines were culture in 5% CO2 at 37°C . Raji ( kind gift from the laboratory of Aimable Nahimana , ATCC: CCL-86 ) , SKW6 . 4 ( kind gift from the laboratory of Pascal Schneider , ATCC: TIB-215 ) , and HeLa ( ATCC: CCL-2 ) cells were cultured in RPMI ( Invitrogen , ref no . 61870 ) supplemented with 10% heat-inactivated FBS ( Invitrogen , ref no . 10270–106 ) . HEK293T cells ( ATCC: CRL-3216 ) were cultured in DMEM supplemented with 10% FBS and were used here only for lentiviral production . All cell lines were mycoplasma-negative and authenticated via Microsynth cell authentication service . Unless , otherwise indicated , experiments were performed in RPMI with 10% FBS . Zebrafish ( Danio rerio ) from AB line were bred and maintained in our animal facility under standard conditions ( Marrink et al . , 2019 ) , more specifically at 28 . 5°C and on a 14:10 hr light:dark cycle at the Zebrafish facility of the Faculty of Biology and Medicine , University of Lausanne ( cantonal veterinary approval VD-H21 ) . Zebrafish of 20 hr post fertilization were collected and treated with 0 . 2 mM phenylthiourea ( Sigma , St Louis , MO ) to suppress pigmentation . Embryos were raised at 28 . 5°C in Egg water ( 0 . 3 g sea salt/l reverse osmosis water ) up to 4 days post fertilization . C57BL/6NCrl were acquired from Charles River laboratories , which were then housed and bred in our animal facility . All experiments were performed according to the principles of laboratory animal care and Swiss legislation under ethical approval ( Swiss Animal Protection Ordinance; permit number VD3374 . a ) . Sprague-Dawley rat pups ( from Janvier , France ) were euthanized in accordance with the Swiss Laws for the protection of animals , and the procedures were approved by the Vaud Cantonal Veterinary Office ( permit number VD1407 . 9 ) . Primary neuronal cultures from cortices of 2-day-old rats were prepared and maintained at 37°C with a 5% CO2-containing atmosphere in neurobasal medium ( Life Technologies , 21103–049 ) supplemented with 2% B27 ( Invitrogen , 17504044 ) , 0 . 5 mM glutamine ( Sigma , G7513 ) , and 100 μg/ml penicillin-streptomycin ( Invitrogen , 15140122 ) as described previously ( Vaslin et al . , 2007 ) . Neurons were plated at a density of ~3 × 105 cells on 12 mm glass coverslips coated with 0 . 01% poly-L-lysine ( Sigma , P4832 ) . Half of the medium was changed every 3–4 days and experiments were performed at 12–13 days in vitro . Confocal microscopy experiments were done on live 300 , 000 cells . Cells were seeded for 16 hr onto glass bottom culture dishes ( MatTek , corporation ref no . P35G-1 . 5–14C ) in 2 ml RPMI , 10% FBS and treated as described in the figures in 1 ml media , 10% FBS . For nuclear staining , 10 μg/ml live Hoechst 33342 ( Molecular Probes , ref no . H21492 ) was added in the culture medium 5 min before washing cells twice with PBS . After washing , cells were examined with a plan Apochromat 63× oil immersion objective mounted on a Zeiss LSM 780 laser scanning fluorescence confocal microscope equipped with gallium arsenide phosphide detectors and three lasers ( a 405 nm diode laser , a 458-476-488-514 nm argon laser , and a 561 nm diode-pumped solid-state laser ) . Time-lapse experiments were done using an incubation chamber set at 37°C , 5% CO2 and visualized with a Zeiss LSM710 Quasar laser scanning fluorescence confocal microscope equipped with either Neofluar 63× , 1 . 2 numerical aperture ( NA ) or plan Neofluar 100× , 1 . 3 NA plan oil immersion objective ( and the same lasers as above ) . Visual segregation of cells based on types of CPP entry , associated with either vesicular or diffuse cytosolic staining , was performed as shown in Figure 1—figure supplement 1A . Cell images were acquired at a focal plane near the middle of the cell making sure that nuclei were visible . Image acquisition was performed using identical settings for the data presented in a given panel and the related supplementary information . Flow cytometry experiments were performed using a Beckman Coulter FC500 instrument . Cells were centrifuged and resuspended in PBS prior to flow cytometry . Data analysis was done with Kaluza Version 1 . 3 software ( Beckman Coulter ) . With the exception of neurons , cell death was quantitated with 8 μg/ml PI ( Sigma , ref no . 81845 ) . Unless otherwise indicated , cell death was assessed after 16 hr of continuous incubation in Raji and SKW6 . 4 cells and 24 hr in HeLa cells . Prior to treatment , 300 , 000 cells were seeded in six-well plates for 16 hr in 2 ml media , 10% FBS . Treatment was done in 1 ml media with 10% FBS . Cell death and peptide internalization were analyzed by flow cytometry . Internalization measurements were done after 1 hr of incubation . Peptide internalization in primary cortical neurons was assessed by confocal microscopy . Cell-associated fluorescence was quantitated with ImageJ . When cytosolic fluorescent was recorded with ImageJ , the regions of interest that were analyzed were chosen so as not to contain labeled endosomes ( Figure 1—figure supplement 2D , circle ) . Recombinant lentiviruses were produced as described ( Marrink et al . , 2019; Melikov et al . , 2015 ) with the following modification: the envelope plasmid pMD . G and the packaging vector pCMVΔR8 . 91 were replaced by pMD2 . G and psPAX2 , respectively . Two methods were used to assess cellular membrane potential in vitro . With the first method , the membrane potential was determined by incubating 300 , 000 cells for 40 min with 100 nM of the fluorescent probe DiBAC4 ( 3 ) ( Thermo Fisher , ref no . B438 ) in six-well plates in 1 ml media , 10% FBS , and the median fluorescence intensity was then assessed by flow cytometry . Calculation of the actual membrane potential in mV based on the DiBAC4 ( 3 ) signals was performed as described earlier ( Klapperstück et al . , 2009; Krasznai et al . , 1995 ) . The second method relied on electrophysiology recordings . To perform these , the bath solution composition was ( in mM ) : 103 . 9 NaCl , 23 . 9 NaHCO3 , 2 CaCl2 , 1 . 2 MgCl2 , 5 . 2 KCl , 1 . 2 NaH2PO4 , 2 glucose , and 1 . 7 ascorbic acid . The pipet solution was composed of ( in mM ) : 140 KMeSO4 , 10 HEPES , 10 KCl , 0 . 1 EGTA , 10 phosphocreatine , and 4 MgATP . The patch pipets had a resistance of 2 . 4–3 . 6 MΩ . Perforated patch recordings were performed as previously described ( Cueni et al . , 2008 ) . Briefly , freshly prepared gramicidin D ( Sigma , ref no . G5002 ) , at 2 . 8 μM final concentration , was added to prefiltered patch pipet solution and then sonicated for three consecutive times during 10 s . Cell-attached configuration was achieved by applying negative pressure on patch pipet until seal resistance of over 1 GΩ was reached . After gaining cell access through gramicidin created pores , membrane potential measurements were done in current clamp at 0 pA for at least 3 min . Since primary rat neurons are killed following full depolarization induced by gramicidin , the standard curve from membrane potential calculations ( Klapperstück et al . , 2009; Krasznai et al . , 1995 ) was performed using gramicidin-treated Raji cells incubated with increasing concentrations of DiBac4 ( 3 ) for 40 min . Images of the cells were then taken using an LSM780 confocal microscope and the cell-associated fluorescence quantitated with ImageJ . Zebrafish embryos in Egg water ( see ‘Zebrafish’ section ) were incubated for 40 min in the presence or in the absence of various concentrations of valinomycin together with 950 nM DiBac4 ( 3 ) . The embryos were then fixed and visualized under a confocal LSM710 microscope ( Adams and Levin , 2012 ) . DiBac4 ( 3 ) -associated fluorescence of a region of interest ( ROI ) of about 0 . 0125 mm2 in the tail region was quantitated with ImageJ . The values were normalized to the control condition ( i . e . in the absence of valinomycin ) . Mice were intradermally injected with 10 μl of a 950 nM DiBac4 ( 3 ) PBS solution containing or not 10 μM valinomycin and sacrificed 1 hr later . The skin was excised , fixed in 4% formalin , paraffin-embedded , and used to prepare serial histological slices . Pictures of the slices were taken with a CYTATION3 apparatus . The DiBac4 ( 3 ) -associated fluorescence in the whole slice was quantitated with ImageJ . The slice in the series of slices prepared from a given skin sample displaying the highest fluorescence signal was considered as the one nearest to the injection site . The signals from such slices are those reported in the figures . RPMI-like media made without potassium chloride and without sodium chloride was from Biowest ( Supplementary file 1 ) . Varying concentrations of potassium chloride were added to this medium containing 10% FBS . Sodium chloride was also added so that the sum of potassium and sodium chloride equaled 119 mM ( also taking into account the concentrations of sodium and potassium in FBS ) . Three-hundred thousand cells were pre-incubated in 1 ml media , 10% FBS containing different concentrations of potassium for 20 min , then different CPPs at a 40 μM concentration were added and cells were incubated for 1 hr at 37°C in 5% CO2 . Cells were washed once in PBS and CPP internalization was measured by flow cytometry . The corresponding membrane potential was measured with DiBac4 ( 3 ) . An asymmetric multi-component membrane was constructed and solvated using CHARMM-GUI ( Jo et al . , 2008; Qi et al . , 2015 ) . Each layer contained 100 lipids ( Supplementary file 2 ) , in a previously described composition ( Ingólfsson et al . , 2014 ) . The membrane was solvated with 2700 water molecules , obtaining a molecular system of 10 , 200 particles . The MARTINI force field 2 . 2p ( Marrink et al . , 2003; Wassenaar et al . , 2015 ) was used to define phospholipids’ topology through a coarse-grained approach . The polarizable water model has been used to assess the water topology ( Yesylevskyy et al . , 2010 ) . Each peptide ( R9 , TAT , and TAT-RasGAP317-326 ) structural model has been obtained by PEPFOLD-3 server ( Lamiable et al . , 2016 ) , as done in previous studies in the field ( Grasso et al . , 2015; Grasso et al . , 2018; Serulla et al . , 2020 ) . For each molecular system , one CPP was positioned 3 nm away from the membrane outer leaflet , in the water environment corresponding to the extracellular space . Then , the system was equilibrated through four MD simulations of 100 ps , 200 ps , 500 ps , and 100 ns under the NPT ensemble . Position restraints were applied during the first three MD simulations and gradually removed , from 200 to 10 kJ/molnm2 . Velocity rescaling ( Bussi et al . , 2007 ) , temperature coupling algorithm , and time constant of 1 . 0 ps were applied to keep the temperature at 310 . 00 K . Berendsen ( Berendsen et al . , 1984 ) semi-isotropic pressure coupling algorithm with reference pressure equal to 1 bar and time constant 5 . 0 ps was employed . Then , all systems were simulated for the production run in the NPT ensemble with a time step of 20 fs . Electrostatic interactions were calculated by applying the particle-mesh Ewald ( Darden et al . , 1993 ) method and van der Waals interactions were defined within a cut-off of 1 . 2 nm . Periodic boundary conditions were applied in all directions . Trajectories were collected every 10 ps and the Visual Molecular Dynamics ( VMD ) ( Humphrey et al . , 1996 ) package was employed to visually inspect the simulated systems . Three different transmembrane potential values were considered: 0 , 80 , and 150 mV . In the MD simulations , an external electric field Eext was applied parallel to the membrane normal z , that is , perpendicular to the bilayer surface . This was achieved by including additional forces Fi = qEext acting on all charged particles i . In order to determine the effective electric field in simulations , we applied a computational procedure reported in literature ( Gumbart et al . , 2012 ) . A well-tempered metadynamics protocol ( Barducci et al . , 2008 ) was applied to estimate the free energy landscape of CPP translocation . Two collective variables were considered: the lipid/water density index and the CPP-membrane distance . Gaussian deposition rate of 2 . 4 kJ/mol every 5 ps was initially applied and gradually decreased on the basis of an adaptive scheme . Gaussian widths of 0 . 5 and 0 . 2 nm were applied following a well-established scheme ( Deriu et al . , 2016; Granata et al . , 2013; Grasso et al . , 2017; Laio and Gervasio , 2008 ) . In particular , the Gaussian width value was of the same order of magnitude as the standard deviation of the distance CV , calculated during unbiased simulations . The well-tempered metadynamics simulations were computed using GROMACS 2019 . 4 package ( Abraham et al . , 2015 ) and the PLUMED 2 . 5 open-source plug-in ( Tribello et al . , 2014 ) . The reconstruction of the free-energy surface was performed by the reweighting algorithm procedure ( Tiwary and Parrinello , 2015 ) allowing the estimation of the free energy landscape . The comparison between the water pore formation free energy estimated by our MARTINI coarse-grained simulations and previous estimations available in literature is reported in Supplementary file 3 . Each system was simulated ( with a 20 fs time step ) until convergence was reached . The electrostatic potential maps were computed by the APBS package ( Baker , 2004 ) on the molecular system composed of R9 peptide in contact with the cell membrane , without any applied external electrostatic field . In detail , the non-linear Poisson-Boltzmann equation was applied using single Debye-Huckel sphere boundary conditions on a 97 × 97 × 127 grid with a spacing of 1 Å centered at the COM of the molecular system . The relative dielectric constants of the solute and the solvent were set to 2 . 5 and 78 . 4 , respectively . The ionic strength was set to 150 mM and the temperature was fixed at 310 K ( Baker , 2004; Grasso et al . , 2017 ) . The average and standard deviation values of the local transmembrane potential were computed considering 10 different trajectory snapshots taken from the molecular trajectory . The same molecular system previously constructed and equilibrated was investigated to estimate the water pore formation kinetics by applying a constant electrostatic potential ( Böckmann et al . , 2008; Fernández et al . , 2010; Gao et al . , 2019; Gumbart et al . , 2012; Gurtovenko and Lyulina , 2014; Kirsch and Böckmann , 2016; Tieleman , 2004; Ziegler and Vernier , 2008 ) . Each peptide ( R9 , TAT , and TAT-RasGAP317-326 ) was positioned 3 nm away from the membrane at the beginning of each simulation . The relatively small size of the molecular system and the application of coarse-grained MARTINI force field allowed us to study the pore formation kinetics , requiring many simulations at varying field strengths . In detail , 25 simulations were performed for each molecular system at different external electric field strengths from 0 . 0055 to 0 . 090 V/nm . In the MD simulations , an external electric field Eext was applied parallel to the membrane normal z , that is , perpendicular to the bilayer surface . This was achieved by including additional forces Fi = qEext acting on all charged particles i . In order to determine the effective electric field in simulations , we applied a computational procedure reported in literature ( Gumbart et al . , 2012 ) . The results are reported in Figure 4—figure supplement 1 . The translocation mechanism of each CPP was studied by ion imbalance in a double-bilayer system ( Delemotte et al . , 2008; Gao et al . , 2019; Gurtovenko and Vattulainen , 2007; Herrera and Pantano , 2009 ) . The same asymmetric membrane considered to perform the single-bilayer simulations ( Supplementary file 2 ) was used to build up the double-bilayer system . The double-membrane system was solvated with 4300 water molecules , obtaining a molecular system of 20 , 000 particles . The MARTINI force field 2 . 2p ( Marrink et al . , 2003; Wassenaar et al . , 2015 ) was used to define phospholipids’ topology through a coarse-grained approach . The polarizable water model was used to model the water topology ( Yesylevskyy et al . , 2010 ) . The elastic network ELNEDYN ( Periole et al . , 2009 ) was applied to reproduce the structural and dynamic properties of the CPPs . For each molecular system , one CPP was positioned in the middle of the double-bilayer system , 2 nm away from the membrane outer leaflets , in the water environment corresponding to the extracellular space . Then , the system was equilibrated through four MD simulations of 100 ps , 200 ps , 500 ps , and 100 ns under the NPT ensemble . Position restraints were applied during the first three MD simulations and gradually removed , from 200 to 10 kJ/molnm2 . Velocity rescaling ( Bussi et al . , 2007 ) , temperature coupling algorithm , and time constant of 1 ps were applied to keep the temperature at 310 K . Berendsen ( Berendsen et al . , 1984 ) semi-isotropic pressure coupling algorithm with reference pressure equal to 1 bar and time constant 5 ps was employed . Then , all systems were simulated for the production run in the NPT ensemble with the time step of 20 fs with Parrinello-Rahman pressure coupling ( Parrinello and Rahman , 1981 ) . Membrane hyperpolarization was achieved through a net charge difference of 30 positive ions between intracellular and extracellular space , considering all charged ions of the system and fulfilling the full system electroneutrality . Ten different replicas of each molecular simulation were performed until the water pore formation and closure events were observed . The visual inspection of the simulated molecular systems is reported in Figure 4—figure supplement 1A . To analyze whether the CPPs were able to cross the membrane and reach the intracellular compartment , their trajectories were studied in the last 5 ns of each simulation replica . Considering the CPP position with respect to the membrane bilayers and the CPP’s solvent accessible surface area , three different compartments were defined: intracellular space , lipid bilayer ( cell membrane ) , and extracellular space . The radius of the water pores within the membrane was calculated as previously done ( Gurtovenko and Vattulainen , 2007; Leontiadou et al . , 2004 ) . We assumed that the central part of the cylindrical water pore contains N water molecules at the same density as outside of the water flux . Three-hundred thousand wild-type HeLa cells or primary rat cortical neurons were incubated with 32 µg/ml PI ( 0 . 8–1 . 5 nm diameter 75 ) or 200 µg/ml dextran of different molecular weight in the presence or in the absence of the indicated CPPs in normal , depolarizing ( 2 µg/ml gramicidin ) , or hyperpolarizing ( 10 µM valinomycin ) conditions in 1 ml media , 10% FBS . Time-lapse images were acquired by confocal microscopy every 10 s . The percentage of cells where direct CPP translocation occurred , as well as the percentage of cells positively stained for PI , were manually quantitated using ImageJ based on snap shot images taken after 30 min of incubation , as shown in Figure 5—figure supplement 1C . Cytosolic PI fluorescence was assessed with ImageJ , by selecting a region within the cell's cytosol devoid of endosomes . Cells were permeabilized with 0 . 1% saponin ( Sigma , ref no . 4706 , diluted in PBS weight:volume; 30 min incubation at 37°C in a 5% CO2 incubator ) to determine the maximal PI uptake cell capacity . Three fitting models were obtained: These equations fitted equally well the PI uptake/Vm curve in Figure 5E . For the calculations used in Figure 5F , the exponential decline equation was used . FITC-TAT-RasGAP317-326 ( W317A ) internalization in zebrafish was assessed either by adding the peptide directly in Egg water or by injection . Experiments in which the peptide was added in the water were performed on fish between 4 and 24 hr post fertilization . Viability assays were done on embryos of 4 , 6 , and 24 hr post fertilization to determine a maximal nonlethal dose of the peptide that can be used . Different concentrations of the peptide were added to 500 µl water per well in 24-well plate , with between eight and eleven embryos per well . Fish viability was visually assessed at 20 hr post incubation with the peptide . Hyperpolarization-associated viability was visually assessed at different time points in the presence or in the absence of the peptide with or without various valinomycin concentrations . Zebrafish were visualized with binocular microscope and CYTATION3 apparatus . Survival was visually assessed under a binocular microscope by taking into consideration the embryo transparency ( as dead embryos appear opaque ) , general development characteristics , and motility . To assess peptide internalization in zebrafish , two methods were used: ( i ) addition of the peptide directly in 500 µl of Egg water in 24-well plates containing between eight and twelve embryos per well and ( ii ) intramuscular injections . In the case of the first method , after the indicated treatments , zebrafish were washed , fixed in 4% PFA/PBS for 1 hr at room temperature . Whole embryos were mounted on slides with Fluoromount-G ( cBioscience , ref no . 00-4958-02 ) . Zebrafish were then visualized under an LSM710 confocal microscope . Experiments where the peptide was added directly to the water were performed on zebrafish at 18 hr post fertilization to limit cuticle development that would hinder peptide access to the cells . In the case of the second method , 8 nl injections ( containing the various combinations of peptide and valinomycin and 0 . 05% ( vol:vol ) phenol red as an injection site labeling agent ) were done on 48 hr post fertilization embryos into the tail muscle around the extremity of yolk extension , after chorion removal and anesthesia with 0 . 02% ( w:vol ) tricaine ( Schroeder et al . , 2000 ) buffered with sodium bicarbonate to pH 7 . 3 . At this age , zebrafish already have well developed tissues that can be easily visually distinguished . Injections were done with an Eppendorf Microinjections FemtoJet 4i apparatus . After the indicated treatments , embryos were fixed in 4% PFA/PBS and visualized under a confocal microscope . Some embryos were kept alive for viability evaluation post injection until the age of 4 days . Experiments with mice were performed in 10- to 14-week-old C57BL/6NCrl mice anesthetized with ketasol/xylasol ( 9 . 09 mg/ml ketasol and 1 . 82 mg/ml xylasol in water; injection: 10 μl per g of body weight ) . The back of the mice was shaved and intradermic injections were performed ( a total of 10 μl was injected ) . Mice were kept under anesthesia for 1 hr and Artificial tears ( Lacryvisc ) were used to avoid eye dryness . Mice were then sacrificed by CO2 inhalation , skin was cut at injection sites , fixed in 4% formalin , and paraffin embedded for histology analysis . For each sample , 10–15 slides were prepared and peptide internalization was visualized with a CYTATION3 apparatus . Fluorescence intensity was quantitated with ImageJ . The slices displaying the highest fluorescence signal were considered as those nearest to the injection site and the fluorescent values from these slides were used in Figure 6B . Three hundred thousand cells were seeded onto glass-bottom dishes in 2 ml RPMI , 10% FBS for 16 hr . Quantitation of cytosolic fluorescence was performed within live HeLa cells pre-incubated with 80 µM TAT-RasGAP317-326 for 30 min at 37°C in 1 ml media , with 10% FBS and then incubated for the indicated periods of time in the presence ( i . e . no wash after the pre-incubation ) or in the absence ( i . e . following three consecutive washes with RPMI supplemented with 10% FBS ) of extracellular labeled peptide . Endosomal escape from lysosomes was induced by 1 mM LLOME ( L-leucyl-L-leucine methyl ester ) ( Repnik et al . , 2017 ) , added in the 1 ml media , 10% FBS 30 min after the CPPs were washed out ( cells were exposed to LLOME for 100 min ) . Confocal images were taken every 5 min after the 30 min pre-incubation with the CPPs . For each cell , the fluorescence intensity of one ROI devoid of labeled endosomes throughout the experiment was quantitated over time using ImageJ Time Series Analyzer V3 . The surface of the ROI was identical for all cells . Only cells displaying labeled endosomes after the 30 min pre-incubation ( that is cells that had taken up the CPPs by endocytosis ) were analyzed . Note that the washing steps , for reasons unclear at this time , induced a slightly higher initial ROI intensity signal . Wild-type HeLa cells were plated in 12-well plates ( 200 , 000 cells per well ) for 16 hr in 1 ml RPMI ( Invitrogen , ref no . 61870 ) , supplemented with 10% heat-inactivated FBS ( Invitrogen , ref no . 10270–106 ) . Cells were then incubated with 20 µg/ml AlexaFluor488-conjugated transferrin for 20 min at 37°C in 5% CO2 . Cells were washed with PBS and pelleted after trypsinization . To quench membrane-bound transferrin fluorescence , cells were resuspended in 0 . 2% trypan blue diluted in PBS . Transferrin internalization was quantitated by flow cytometry using Beckman Coulter FC500 instrument . Data analysis was done with Kaluza Version 1 . 3 software ( Beckman Coulter ) . The LeGOiG2-LUC705 lentiviral construct ( plasmid #975 ) encodes a luciferase gene interrupted by a mutated human beta globin intron 2 . This mutation creates a new aberrant splicing site at position 705 that , when used , produced an mRNA that encodes a truncated non-functional luciferase ( Guidotti et al . , 2017 ) . In the presence of the TAT-peptide nucleic acid ( TAT-PNA ) CPP described below , the aberrant splice site is masked allowing the production of a functional luciferase enzyme . Lentiviruses produced using this construct were employed to infect cells . The doses used resulted in >90% cells infected ( based on GFP expression from the lentiviral vector ) . The infected cells ( 200 , 000 cells in 12-well plates containing 1 ml of RPMI , 10% FBS ) were treated or not with 5 μM TAT-PNA ( GRKKRRQRRR-CCTCCTACCTCAGTTACA ) . TAT-PNA is made of TAT48-57 and an oligonucleotide complimentary to a sequence containing the aberrant splice site . After 16 hr incubation , cells were washed twice in HKR buffer ( 119 mM NaCl , 2 . 5 mM KCl , 1 mM NaH2PO4 , 2 . 5 mM CaCl2 , 1 . 3 mM MgCl2 , 20 mM HEPES , 11 mM dextrose , pH 7 . 4 ) and lysed in 40 μl HKR containing 0 . 1% Triton X-100 for 15 min at room temperature . Luciferase activity was measured with a GLOMAXTM 96 Microplate Luminometer ( Promega ) using a Dual-Luciferase Reporter Assay ( Promega ) and normalized to the protein content . Results are displayed as the ratio between the protein-normalized luciferase signal obtained in TAT-PNA-treated cells and the signal obtained in control untreated cells . Raji cells were infected with a lentivirus encoding a Cre-reporter gene construct ( D’Astolfo et al . , 2015 ) . TAT-Cre recombinase was produced as described ( Wadia et al . , 2004 ) . Briefly , Escherichia coli BL21 transformed with the pTAT-Cre plasmid ( #917 , Addgene plasmid #35619 ) were grown for 16 hr in LB containing 100 µg/ml kanamycin . Protein production was induced at OD600 of 0 . 6 with 500 µM IPTG ( isopropyl β-D-1-thiogalactopyranoside ) for 3 hr . Bacteria were collected by centrifugation at 5000× g and kept at –20°C . Purification was performed on Äkta prime ( GE , Healthcare , Chicago , IL ) equipped with a 1 ml HisTrap FF column equilibrated with binding buffer ( 20 mM sodium phosphate , 500 mM NaCl , 5 mM imidazole pH 7 . 4 ) . The day of the purification , bacterial pellet was resuspended in lysis buffer ( binding buffer with protease inhibitors; Roche , ref no . 4693132001; one tablet per 50 ml ) , 0 . 025 mg/ml DNase I ( Roche , ref no . 04716728001 ) , and 2 mg/ml lysozyme ( Roche , ref no . 10 837 059 001 ) and sonicated six times for 30 s . After 20 min centrifugation at 5000× g , the supernatant was filtered through Steriflip 0 . 45 µm and loaded on the column . Elution buffer ( 20 mM sodium phosphate , 500 mM NaCl , 500 mM imidazole pH 7 . 4 ) was used to detach His-tagged proteins from the column . Imidazole was removed from collected fractions by overnight dialysis using 10 K MWCO cassette ( Thermo Scientific , ref no . 66807 ) in PBS . Raji cells encoding the Cre-reporter were treated for 48 hr with 20 µM TAT-Cre-recombinase . Fluorescence was imaged using a Nikon Eclipse TS100 microscope . Three-hundred thousand cells were incubated for 60 s in 1 ml RPMI supplemented with 10% FBS and 10 mM HEPES in Eppendorf tubes at 37°C in the presence of increasing concentrations of FITC-TAT-RasGAP317-326 . Half of the cells were then immediately placed on ice , pelleted at 4°C , and resuspended in 1 ml of ice-cold PBS and then split into two tubes , one of which receiving a final concentration of 0 . 2% ( w:w ) trypan blue to quench surface-associated FITC signals . The cells ( still kept at 4°C ) were then analyzed by flow cytometry . The surface associated peptide signal was calculated by subtracting total fluorescence measured in PBS and fluorescence measured after trypan blue quenching . The other half of the cells after the 60 s peptide incubation was incubated at 37°C for 1 hr at which time the cellular internalization of the labeled peptide was assessed by flow cytometry . Calcium phosphate-based transfection of HeLa cells was performed as previously described ( Jordan et al . , 1996 ) . Briefly , cells were plated overnight in DMEM ( Invitrogen , ref no . 61965 ) medium supplemented with 10% heat-inactivated FBS ( Invitrogen , ref no . 10270–106 ) , 2 . 5 µg of total plasmid DNA of interest was diluted in water , CaCl2 was added and the mixture was incubated in presence of HEPES 2× for 60 s before adding the total mixture drop by drop to the cells . Media was changed 10 hr after . Isothermal titration calorimetry ( ITC ) was performed using MicroCal ITC200 ( Malvern Panalytical ) at 37°C with 600 µM FITC-R9 in the cell ( total volume 300 µl ) and consecutive injections ( 2 . 5 µl/injection , except for the first injection of 0 . 4 µl ) of 6 mM PI from the syringe ( total volume 40 µl ) with 2 min delay between injections and 800 rotations/min rotation speed . Differential power was set to 7 , as we had no prior knowledge of the expected reaction thermodynamics . The results in Figure 5—figure supplement 2A are represented as a: thermogram ( measurement of thermal power need to ensure that there is no temperature difference between reference and sample cells in the calorimeter as a function of time ) and a binding isotherm ( normalized heat per peak as a function of molar ratio ) . Three-hundred thousand wild-type HeLa cells were plated overnight in RPMI with 10% FBS in six-well plates . Cells were then treated for 1 hr with the indicated concentrations of CPP , PI , and membrane potential modulating agents ( gramicidin or valinomycin ) in 1 ml RPMI . As control , cells were either left untreated or incubated with DMSO , the vehicle used to dilute gramicidin and valinomycin . Cells were then washed , trypsinized , and plated on 10 cm dishes at a density of 300 cells per condition . Colonies were counted at day 14 after 100% ethanol fixation for 10 min and Giemsa staining . Washes were done with PBS . The human GeCKO v2 library ( two plasmid system ) ( Addgene plasmid #1000000049 ) was amplified by electroporation using a Bio-Rad Gene Pulser II electroporation apparatus ( Bio-Rad #165–2105 ) and the Lucigen Endura bacteria ( Lucigen ref no . 60242 ) . Cells were plated on LB agar plates containing 100 µg/ml ampicillin . After 14 hr at 32°C , colonies were scrapped and plasmids recovered with the Plasmid Maxi kit ( Qiagen , ref no . 12162 ) . To produce lentivirus library , 12 T-225 flasks were seeded with 12 × 106 HEK293T cells per flask in 40 ml DMEM , 10% FBS . The following day , 10 µg pMD2 . G , 30 µg psPAX2 , and 25 µg GeCKO plasmid library in 1 . 8 ml H2O were mixed with 0 . 2 ml 2 . 5 M CaCl2 ( final calcium concentration: 250 mM ) . This solution was mixed ( v/v ) with 2× HEPES buffer ( 280 mM NaCl , 10 mM KCl , 1 . 5 mM Na2HPO4 , 12 mM D-glucose , 50 mM HEPES ) , incubated for 1 min at room temperature , added to the culture medium , and the cells placed back in a 37°C , 5% CO2 incubator for 7 hr . The culture medium was then removed and replaced by DMEM supplemented with 10% FBS containing 100 U/ml penicillin and 100 µg/ml streptomycin . Forty-eight hours later , the medium was collected and centrifuged 5 min at 2000× g to pellet the cells . The remaining cell-free medium ( 12 × 40 ml ) was then filtered through a 0 . 45 µm HV/PVDF ( Millipore , ref no . SE1M003M00 ) and concentrated ~100 times by resuspending the viral pellet obtained by ultracentrifugation at 70000× g for 2 hr at 4°C in ~5 ml ice-cold PBS . The concentrated viruses were aliquoted in 500 µl samples and stored at –80°C . To express the Cas9 endonuclease , cells ( e . g . Raji or SKW6 . 4 ) were infected with Cas9-expressing viruses that were produced in HEK293T cells transfected with the lentiCas9-Blast ( #849 , Addgene plasmid #52962 ) , pMD2 . G , and psPAX2 plasmids as described under ‘Lentivirus production’ . The infected cells were selected with 10 µg/ml blasticidin for a week . The multiplicity of infection ( MOI ) of the GeCKO virus library was determined as follows . Different volumes of the virus library were added to 3 × 106 Cas9-expressing cells plated in 12-well plates . Twenty-four hours later , the cells were split into two wells of 12-well plates . One well per pair was treated with 10 µg/ml puromycin for 3 days ( the other cultured in normal medium ) . Cell viability was determined by trypan blue exclusion and MOI was calculated as the number of cells in the well treated with puromycin divided by the number of cells in the control well . The virus volume yielding to a MOI ~ 0 . 4 was chosen to perform large-scale infection of 12 × 107 cells that was carried out in 12-well plates with 3 × 106 cells per well . After 24 hr , the infected cells were collected and pooled in a T-225 flask and selected with 10 µg/ml puromycin for a week . Thirty millions of these were frozen ( control untreated cells ) and 60 million others were treated with 40 µM TAT-RasGAP317-326 for 8 days ( Raji ) or for 17 days ( SKW6 . 4 ) with a medium and peptide renewal every 2–3 days . Thirty million of the peptide-treated cells were then also frozen . Genomic DNA was extracted from the control and the peptide-treated frozen cells using the Blood & Cell Culture DNA Midi Kit according to manufacturer’s instructions ( Qiagen , ref no . 13343 ) . A first PCR was performed to amplify the lentiCRISPR sgRNA region using the following primers: A second PCR ( see Supplementary file 4 for the primers used ) was performed on 5 µl of the first PCR reaction to attach Illumina adaptors with barcodes ( nucleotides highlighted in green ) and to increase library complexity ( using the sequences highlighted in red ) to prevent signal saturation when the sequencing is performed . The blue sequences are complementary to the extremities of the first PCR fragments . Both PCRs were performed in 100 µl with the 2 µl of the Herculase II Fusion DNA Polymerase from Agilent ( ref no . 600675 ) according the manufacturer’s instructions . Amplicons were gel extracted , quantitated , mixed , and sequenced with an MiSeq ( Illumina ) . Raw FASTQ files were demultiplexed and processed to contain only unique sgRNA sequences . The number of reads of each sgRNA was normalized as described ( Tribello et al . , 2014 ) . The MAGeCK algorithm ( Tsoutsou et al . , 2017 ) was used to rank screening hits by the consistent enrichment among multiple sgRNAs targeting the same gene . Single guide RNAs targeting an early exon of the gene of interest were chosen in the sgRNA library ( Wallbrecher et al . , 2017 ) and are listed in Supplementary file 5 . LentiCRISPR plasmids specific for a gene were created according to the provided instructions ( Vasconcelos et al . , 2013 ) . Briefly , oligos were designed as follows: Forward 5’-CACCGnnnnnnnnnnnnnnnnnnnn-3’; Reverse-3’-CnnnnnnnnnnnnnnnnnnnnCAA-5’ , where nnnnnnnnnnnnnnnnnnnn in the forward oligo corresponds to the 20 bp sgRNA . Oligos were synthetized , then phosphorylated , and annealed to form oligo complexes . LentiCRISPR vector was BsmBI digested and dephosphorylated . Linearized vector was purified and gel extracted and ligated to oligo complexes . The lentiCRISPR vector containing the sgRNA was then used for virus production . Recombinant lentiviruses were produced as described ( Mitchell et al . , 2000 ) with the following modification: pMD . G and pCMVDR8 . 91 were replaced by pMD2 . G and psPAX2 , respectively . Cells were infected and selected with the appropriate dose of puromycin ( 2 µg/ml for HeLa cells ) . Clone isolation was performed by limiting dilution in 96-well plate . TA cloning is a subcloning technique that allows integration of a PCR-amplified product of choice into a PCR2 . 1 vector based on complementarity of deoxyadenosine added onto the PCR fragment by Taq polymerase . This approach is useful to distinguish between several alleles and to determine whether the cells are heterozygous or homozygous at a given locus . TA cloning kit ( Life Technologies , ref no . K202020 ) was used according to manufacturer’s instructions to sequence DNA fragment containing the region targeted by a given sgRNA . Briefly , DNA was isolated and the fragment of interest was PCR-amplified using primers listed in Supplementary file 6 , then ligated into PCR2 . 1 vector . E . coli competent cells were then transformed and at least 15 colonies were selected per condition for DNA isolation and sequencing . The hKCNN4-V5 . lti ( #953 ) lentiviral plasmid encoding a V5-labeled version of the KCNN4 potassium channel was from DNASU ( ref no . HsCD00441560 ) . The hKCNK5-FLAG . dn3 ( #979 ) plasmid encoding the human KCNK5 potassium channel ( NCBI reference sequence NM_003740 . 3 ) , Flag-tagged at the C-terminus , was purchased from GenScript ( ref no . OHu13506 ) . The Myc-mKCNJ2-T2A-IRES-tdTomato . lti ( #978 ) lentiviral vector encoding the mouse Kir2 . 1 ( KCNJ2 ) potassium channel and tdTomato ( separated by an IRES ) was generated by subcloning myc-mKCNJ2-T2A-Tomato . pCAG plasmid ( #974 , Addgene plasmid #60598 ) into the LeGo-iT2 lentiviral backbone ( #809 ) , a gift from Boris Fehse ( Addgene plasmid #27343 ) , through ligation of both plasmids after digestion with BamHI ( NEB , reg . no . R313614 ) . The pMD2 . G plasmid ( #554 , Addgene plasmid #12259 ) encodes the envelope of lentivirus . The psPAX2 plasmid ( #842 , Addgene plasmid #12260 ) encodes the packaging system . Both pMD2 . G and psPAX2 plasmids were used for lentiviral production . The Flag-hKCNQ5 ( G278S ) -IRES-NeoR plasmid ( #938 ) codes for the N-terminal Flag-tagged G278S human KCNQ5 inactive mutant and a neomycin resistant gene separated by an IRES sequence . It was generated by subcloning a BamHI/XmaI digested PCR fragment obtained by amplification of pShuttle-Flag-hKCNQ5 ( G278S ) -IRES-hrGFP2 ( #937 , kind gift from Dr Kenneth L Byron ) using forward primer #1397 ( CAT CGG GAT CCG CTA TAC CGG CCA CCA TGG ATT ACA AGG A ) and reverse primer #1398 ( CAT CGC CCG GGG CTA TAC CGT ACC GTC GAC TGC AGA ATT C ) into the lentiviral vector TRIP-PGK-IRES-Neo ( #350 ) opened with the same enzyme . The Flag-hKCNQ5 ( SM , G278S ) -IRES-Neo ( #939 ) plasmid is identical to Flag-hKCNQ5 ( G278S ) -IRES-NeoR except that the sequence targeted by the sgKCNQ5 . 1 sgRNA ( Supplementary file 5 ) was mutated with the aim to decrease Cas9-mediated degradation . Silent mutations ( SM ) , at the protein level , were introduced using the QuikChange II XL Site-Directed Mutagenesis Kit ( ref no . 200522 ) according to manufacturer’s instructions using forward primer #1460 ( AAA TAA GAA CCA AAA ATC CTA TGT ACC ATG CCG TTA TCA GCT CCT TGC TGT GAG CAT AAA CCA CTG AAC CCA G ) and reverse primer #1461 ( CTG GGT TCA GTG GTT TAT GCT CAC AGC AAG GAG CTG ATA ACG GCA TGG TAC ATA GGA TTT TTG GTT CTT ATT T ) . The Flag-hKCNQ5 ( SM ) -IRES-NeoR ( #940 ) lentiviral construct codes for a Flag-tagged wild-type version of human KCNQ5 . It was made by reverting the G278S mutation in Flag-hKCNQ5 ( SM , G278S ) -IRES-Neo ( #939 ) using the QuikChange II XL Site-Directed Mutagenesis Kit with the #1462 forward primer ( TTT TGT CTC CAT AGC CAA TAG TTG TCA ATG TAA TTG TGC CCC ) and the #1463 reverse primer ( GGG GCA CAA TTA CAT TGA CAA CTA TTG GCT ATG GAG ACA AAA ) . The pLUC705 ( Kang et al . , 1998 ) ( #876 , gift from Dr Bing Yang ) plasmid encodes a luciferase gene interrupted by a mutated human beta globin intron 2 . This mutation creates a new aberrant splicing site at position 705 that when used produced an mRNA that encodes a truncated non-functional luciferase ( Kang et al . , 1998 ) . To introduce this construct into a lentiviral vector , the pLUC705 plasmid was digested with HindIII/XhoI , blunted with T4 DNA polymerase , and ligated into StuI-digested and dephosphorylated LeGO-iG2 ( #807 , Addgene plasmid #27341 ) , resulting in plasmid pLUC705 . LeGO-iG2 ( #875 ) . The pTAT-Cre ( #917 , Addgene plasmid #35619 ) bacterial plasmid encodes a histidine-tagged TAT-Cre recombinase . The Cre reporter lentiviral vector ( #918 , Addgene plasmid #62732 ) encodes a LOXP-RFP-STOP-LOXP-GFP gene construct . Cells expressing this plasmid appear red but once recombination has occurred when TAT-Cre is translocated into cells , the RFP-STOP fragment will be excised , GFP but not RFP will now be produced , and cells will appear green . TAT-RasGAP317-326 is a retro-inverso peptide ( i . e . synthesized with D-amino-acids in the opposite direction compared to the natural sequence ) labeled or not with FITC or TMR . The TAT moiety corresponds to amino acids 48–57 of the HIV TAT protein ( RRRQRRKKRG ) and the RasGAP317–326 moiety corresponds to amino acids from 317 to 326 of the human RasGAP protein ( DTRLNTVWMW ) . These two moieties are separated by two glycine linker residues in the TAT-Ras-GAP317-326 peptide . FITC-bound peptides without cargo: TAT , Penetratin ( RQIKWFQNRRMKWKK ) , R9 ( RRRRRRRRR ) , and K9 ( KKKKKKKKK ) were synthesized in D-amino acid conformation . All peptides were synthesized in retro-inverso conformation ( over the years different suppliers were used with routine checks for activity of TAT-RasGAP317-326 derived peptides , Biochemistry Department of University of Lausanne , SBS Genetech , China and Creative Peptides , USA ) and resuspended to 1 mM in water . Statistical analysis was performed on non-normalized data , using GraphPad Prism 7 . ANOVA multiple comparison analysis to wild-type condition was done using Dunnett’s correction ( Figure 3A–C and Figure 3—figure supplement 1B , C and Figure 3C , top panel , and PI internalization in Figure 5C , as well as TAT-PNA internalization in Figure 2—figure supplement 4A ) . ANOVA multiple comparison analysis between several conditions was done using Tuckey’s correction ( Figure 3—figure supplement 2E ) . Comparison between two conditions was done using two-tailed paired t-test for the CPP internalization experiments described in Figure 3C ( bottom panel ) , Figure 6A–B , Figure 1—figure supplement 1B , and Figure 3—figure supplement 4 . All measurements were from biological replicates . Unless otherwise stated , the horizontal bars in the graph represent the median , the height of columns correspond to averages , and the dots in the figures correspond to values derived from independent experiments .	Before a drug can have its desired effect , it must reach its target tissue or organ , and enter its cells . This is not easy because cells are surrounded by the plasma membrane , a fat-based barrier that separates the cell from its external environment . The plasma membrane contains proteins that act as channels , shuttling specific molecules in and out of the cell , and it also holds charge , with its inside surface being more negatively charged than its outside surface . Cell-penetrating peptides are short sequences of amino acids ( the building blocks that form proteins ) that carry positive charges . These positive charges allow them to cross the membrane easily , but it is not well understood how . To find out how cell-penetrating peptides cross the membrane , Trofimenko et al . attached them to dyes of different sizes . This revealed that the cell-penetrating peptides enter the cell through temporary holes called water pores , which measure about two nanometres across . The water pores form when the membrane becomes ‘megapolarized’ , this is , when the difference in charge between the inside and the outside of the membrane becomes greater than normal . This can happen when the negative charge on the inside surface or the positive charge on the outer surface of the membrane increase . Megapolarization depends on potassium channels , which transport positive potassium ions outside the cell , making the outside of the membrane positive . When cell-penetrating peptides arrive at the outer surface of the cell near potassium channels , they make it even more positive . This increases the charge difference between the inside and the outside of the cell , allowing water pores to form . Once the peptides pass through the pores , the charge difference between the inside and the outside of the cell membrane dissipates , and the pores collapse . Drug developers are experimenting with attaching cell-penetrating peptides to drugs to help them get inside their target cells . Currently there are several experimental medications of this kind in clinical trials . Understanding how these peptides gain entry , and what size of molecule they could carry with them , provides solid ground for further drug development .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2021	Genetic, cellular, and structural characterization of the membrane potential-dependent cell-penetrating peptide translocation pore
70-kDa Heat shock proteins are ATP-driven molecular chaperones that perform a myriad of essential cellular tasks . Although structural and biochemical studies have shed some light on their functional mechanism , the fundamental issue of the role of energy consumption , due to ATP-hydrolysis , has remained unaddressed . Here we establish a clear connection between the non-equilibrium nature of Hsp70 , due to ATP hydrolysis , and the determining feature of its function , namely its high affinity for its substrates . Energy consumption can indeed decrease the dissociation constant of the chaperone-substrate complex by several orders of magnitude with respect to an equilibrium scenario . We find that the biochemical requirements for observing such ultra-affinity coincide with the physiological conditions in the cell . Our results rationalize several experimental observations and pave the way for further analysis of non-equilibrium effects underlying chaperone functions . ATP-driven molecular chaperones play a central role in protecting cells against proteins that could unfold or misfold because of mutations , various stresses or fluctuations and ultimately result in cytotoxic aggregates ( Bukau et al . , 2006; Hartl et al . , 2011 ) . 70-kDa Heat shock proteins ( Hsp70s ) stand out for several reasons: they are possibly the most ubiquitous , they function as monomers , and they supervise a plethora of diverse cellular processes ( Mayer and Bukau , 2005; Zuiderweg et al . , 2013 ) such as protein translation ( Kramer et al . , 2009 ) , protein trafficking ( Matlack et al . , 1999; Neupert and Brunner , 2002 ) , the disassembly of protein complexes ( Böcking et al . , 2011 ) , signaling ( Pratt and Toft , 2003 ) and protein degradation ( Hohfeld et al . , 2001 ) . All these tasks crucially depend on the high-affinity binding of Hsp70s to substrate proteins during a complex ATP-driven conformational cycle ( Mayer , 2013 ) . The ATP- and ADP-bound states of Hsp70 ( Hsp70⋅ATP and Hsp70⋅ADP respectively ) and their interconversion play a major role in the chaperone functional cycle: the nature of the bound nucleotide affects the affinity of the chaperone for its substrates , with Hsp70⋅ADP binding the substrate more stably than Hsp70⋅ATP ( Schmid et al . , 1994; Theyssen et al . , 1996; Packschies et al . , 1997; Gisler et al . , 1998; Russell et al . , 1998; Laufen et al . , 1999; Mayer et al . , 2000 ) . Intriguingly , several experimental evidences suggested that the effective affinity of Hsp70 for substrates when the chaperone was running through its ATP-hydrolysis driven cycle was significantly higher than both the ones of Hsp70⋅ATP and Hsp70⋅ADP ( Laufen et al . , 1999; Wittung-Stafshede et al . , 2003 ) . Because this remarkable result would not be possible within the boundaries of thermodynamic equilibrium , it is therefore necessary to clarify how ATP hydrolysis , and thus energy consumption , can affect the binding strength of Hsp70s to their substrates . According to the consensus Hsp70 cycle ( Figure 1 ) substrate binding/unbinding takes place with rates that depend on the state of the bound nucleotide ( kATPon , kATPoff , kADPon , kADPoff ) ( Schmid et al . , 1994; Gisler et al . , 1998; Mayer et al . , 2000 ) , with Hsp70⋅ATP exchanging the substrate two to three orders of magnitude faster than Hsp70⋅ADP . The conversion from Hsp70⋅ADP to Hsp70⋅ATP occurs through a nucleotide exchange process which is described here at an effective level as a simple first-order reaction with rate kDT , or kDTS in the presence of a bound substrate ( Figure 1 , and ‘Materials and methods’ for a full derivation ) . The conversion from Hsp70⋅ATP to Hsp70⋅ADP can occur by means of two different processes: nucleotide exchange ( dashed arrows in Figure 1 , with rate kTDex , and kTDex , S in the presence of the substrate ) and ATP hydrolysis ( red arrows in Figure 1 ) , whose rate depends on the absence or presence of a bound substrate ( kh and khS respectively ) . The total conversion rate from Hsp70⋅ATP to Hsp70⋅ADP is thus kTD=kTDex+kh ( and analogous expressions in the presence of a substrate ) . In the cell , several cochaperones tune the exchange and hydrolysis rates: J-domain proteins ( JDPs ) enhance the rate of ATP hydrolysis , and nucleotide exchange factors ( NEFs ) catalyze nucleotide release ( Youker and Brodsky , 2007; Kampinga and Craig , 2010 ) . Within the present description , cochaperones are not taken into account explicitely . Rather , their action is captured as a modulation of the cycle timescales . In particular , JDPs are known to bind the substrate and subsequently interact with Hsp70 , enhancing ATP hydrolysis . Consequently here only the hydrolysis rate in the presence of the substrate , khS , is affected by the action of JDPs . 10 . 7554/eLife . 02218 . 003Figure 1 . Canonical Hsp70 biochemical cycle . The model takes into account four states in Hsp70 ( NBD is schematically represented here in green , SBD in orange ) , which are defined by substrate binding and by the nature of the bound nucleotide ( ADP or ATP ) . The rates of the substrate binding/unbinding process ( horizontal blue lines ) are influenced by the nucleotide ( kATPon , kATPoff , kATPon , kATPoff ) . ADP-bound states are converted to ATP-bound states through a nucleotide exchange process ( vertical solid blue lines ) with rates kDT , kDTS . The ATP to ADP conversion can occur by means of either a nucleotide exchange process ( dashed blue lines ) with rates kTDex , kTDex , S or ATP-hydrolysis ( red lines ) with rates kh , khS . DOI: http://dx . doi . org/10 . 7554/eLife . 02218 . 003 A significant difference between hydrolysis and nucleotide exchange must be outlined here: ATP-hydrolysis , at variance with exchange , results into a net production of ADP and a loss of ATP . In a cellular perspective , the ATP and ADP concentrations are kept fixed by energy-consuming chemostats . In vivo , ATP hydrolysis is therefore an intrinsically non-equilibrium process . Because we aim here at elucidating the relation between energy consumption and substrate affinity , we determine the effective dissociation constant of the system , Keff , which provides a coarse-grained measure of how well Hsp70s can bind their substrates through their cycle . Keff is defined in the usual way as Keff = [S][Hsp70]/[Hsp70⋅S] , where [Hsp70] is the total concentration of Hsp70 not bound to a substrate ( [Hsp70] = [Hsp70⋅ATP]+[Hsp70⋅ADP] ) , [Hsp70⋅S] is the total concentration of substrate-bound chaperone ( [Hsp70⋅S] = [Hsp70⋅ATP⋅S]+[Hsp70⋅ADP⋅S] ) and [S] is the concentration of free substrate . In the absence of hydrolysis , no energy is consumed and all the reactions of the Hsp70 cycle are at equilibrium . In this scenario , where all the reactions are driven by thermal fluctuations , the detailed balance rule holds and each branch of the biochemical cycle is individually balanced ( ‘Materials and methods’ ) . In fact , in this case , the ratio between the forward and backward rates for each reaction is completely determined by the free energy difference between the two states , for example kDTS/kTDS=exp ( GHsp70·ADP·S−GHsp70·ATP·S ) . The equilibrium effective dissociation constant ( Keffeq ) can be easily determined asKeffeq=kDTSKD ( ATP ) +kTDSKD ( ADP ) kDTS+kTDSwhere KD ( ATP ) and KD ( ADP ) are the dissociation constants of the Hsp70⋅ATP-substrate and Hsp70⋅ADP-substrate complexes , respectively . Not surprisingly , Keffeq corresponds to a weighted average of KD ( ATP ) and KD ( ADP ) , and it cannot be lower than KD ( ADP ) , usually the lowest of the two ( Schmid et al . , 1994; Gisler et al . , 1998; Mayer et al . , 2000 ) . As a matter of fact , in vivo Keffeq would be close to its typical upper bound , namely KD ( ATP ) , because of the cellular excess of ATP over ADP . The equilibrium dissociation constant can be in principle measured in experiments where Hsp70 hydrolysis deficient mutants are used ( McCarty and Walker , 1991 ) , at varying ratios of the concentrations of the two nucleotides . At variance with equilibrium , when ATP hydrolysis is turned on , the energy budget along the cycle is not restricted to the free-energy differences between the different states . Rather , the dissipated energy Ediss , which is related to ATP hydrolysis , must be taken into account . Thus detailed balance is broken and pairwise reactions cannot be solved individually as in the equilibrium case . Nonetheless , even in non-equilibrium conditions a steady-state solution of the cycle exists ( ‘Materials and methods’ ) , and it provides an expression for the non-equilibrium dissociation constant ( Keffneq ) . Keffneq= ( kDTSkATPoff+kTDSkADPoff+kATPoffkADPoff ) ( kDT+kTD ) + ( kDTSkATPoffkADPon+kTDSkADPoffkATPon ) [S]kDT ( kADPoff+kDTS+kTDS ) kATPon+kTD ( kATPoff+kDTS+kTDS ) kADPon+ ( kDTS+kTDS ) kATPonkADPon[S] In this scenario , ATP hydrolysis is controlled by the fixed basal hydrolsysis rate kh and the substrate-enhanced rate khS , which is further modulated in cellular conditions by JDP co-chaperones . The ratio khS/kh , that measures the hydrolysis acceleration , is thus a natural parameter to characterize the behavior of the system . In order to prove the intimate relation between this quantity and the total energy consumption , we report in Figure 2A the experimentally-measurable hydrolysis flux , Pdiss , defined asPdiss=kh[Hsp70·ATP]+khS[Hsp70·ATP·S]as a function of khS/kh . Here we consider the experimentally determined parameters for the Escherichia coli DnaK-DnaJ system ( see Table 1 ) , and concentrations that roughly mimic cellular conditions ( [Hsp70]tot = 40 μM and substoichiometric substrate , here [S]tot = 4 μM ) . Not surprisingly , the hydrolysis flux increases with khS/kh , before saturation , corresponding to a regime dominated by the rate-limiting exchange process ( Gassler et al . , 2001 ) . 10 . 7554/eLife . 02218 . 004Figure 2 . Effect of ATP-hydrolysis on Keffneq . Total energy consumption ( A ) and effective non-equilibrium dissociation constant of the Hsp70-substrate complex ( B ) is plotted as a function of the hydrolysis acceleration ratio khs/kh , for the DnaK/DnaJ/substrate system with concentrations [Hsp70]tot = 40 μM and [S]tot = 4 μM ( see ‘Materials and methods’ for the parameters ) , The approximate dissociation constant Keff , 0neq is also plotted for comparison ( black dashed line ) . The green region comprised between KD ( ATP ) and KD ( ADP ) corresponds to the range of affinities accessible at equilibrium ( no hydrolysis ) . The red-to-yellow region corresponds to the values of the dissociation constants that are exclusively accessible to the non-equilibrium regime . The region where red fades to yellow ( 103 ≤ khs/kh ≤ 104 ) corresponds to the transition from physiological to non-physiological values of hydrolysis acceleration . DOI: http://dx . doi . org/10 . 7554/eLife . 02218 . 00410 . 7554/eLife . 02218 . 005Table 1 . Parameters of the modelDOI: http://dx . doi . org/10 . 7554/eLife . 02218 . 005kh0 . 0006 s−1 ( McCarty et al . , 1995 ) kh . maxs1 . 8 s−1 ( Laufen et al . , 1999 ) kATPon4 . 5 × 105 s−1 M−1 ( Schmid et al . , 1994; Gisler et al . , 1998 ) kATPoff2 s−1 ( Schmid et al . , 1994; Gisler et al . , 1998 ) kADPon1000 s−1 M−1 ( Mayer et al . , 2000 ) kADPoff4 . 7 × 10−4 s−1 ( Mayer et al . , 2000 ) kATP−1 . 33 × 10−4 s−1 ( Russell et al . , 1998 ) kATP+1 . 3 × 105 s−1 M−1 ( Russell et al . , 1998 ) kADP−0 . 022 s−1 ( Theyssen et al . , 1996; Russell et al . , 1998 ) kADP+2 . 67 × 105 s−1 M−1 ( Russell et al . , 1998 ) Parameters used in the model , from various sources . Notable cases are:This corresponds to the reaction Hsp70⋅ATP⋅DnaJ2⋅S → Hsp70⋅ADP⋅DnaJ2⋅S . In Figure 2B we report Keffneq , as a function of the acceleration ratio khS/kh . As khS/kh increases and more energy is consumed , Keffneq decreases , until it becomes lower than KD ( ADP ) by several orders of magnitude . Non-equilibrium conditions lead thus to a dramatic increase of the affinity of Hsp70s for their substrates , that could not be achieved at equilibrium , where the effective dissociation constant would be bounded between the ones of the ATP-bound and ADP-bound states . We dub such effect ultra-affinity in analogy with energy-consuming ultrasensitivity observed in many enzymatic systems ( Goldbeter and Koshland , 1981 ) . The lower bound of the non-equilibrium dissociation constant is Keffneq=kADPoff/kATPon , which can be achieved for extremely high values of khS/kh ( Figure 2B ) . This regime corresponds to the limiting case of substrate binding exclusively to Hsp70⋅ATP , which has the fastest binding rate , and being released exclusively from the ADP-bound state , which has the slowest unbinding rate . Our analysis indicates that this theoretical limit , recently hinted at ( Zuiderweg et al . , 2013 ) , likely pertains to a regime that is not accessible to Hsp70s . It must be stressed here that this lower bound , as well as ultra-affinity , depends on the kinetic properties of the cycle and does not rely on the dissociation constants of any nucleotide-bound state . In order to better elucidate this point , we move beyond the experimentally measured rates for the DnaK/DnaJ system , and we explore the theoretical dependence of Keffneq on the time-scale separation between the binding/unbinding kinetics in the two states . To this aim , in Figure 3A we report Keffneq as a function of both khS/kh and the ratio kADPoff/kATPoff , which measures the time-scale separation in the unbinding kinetics , while keeping KD ( ADP ) unchanged . For kADPoff≥kATPoff the non-equilibrium dissociation constant is bound to the equilibrium range . Ultra-affinity can be achieved only for kADPoff/kATPoff<1 , and is more pronounced for larger time-scale separations . In the limit kADPoff≪kATPoff , both the binding and the unbinding processes of the ADP-state become negligible and the non-equilibrium dissociation constant reduces toKeff , 0neq=KD ( ATP ) kDTS ( kDT + kTDex + kh ) kDT ( kDTS + kTDex , S + khS ) which is reported in Figure 2B ( dashed line ) and provides a good approximation of the exact behavior in the physiologically-accessible range of khS/kh . 10 . 7554/eLife . 02218 . 006Figure 3 . Dependence of Keffneq on time-scale separation and on stoichiometric ratio . ( A ) Non-equilibrium dissociation constant as a function of the hydrolysis acceleration ratio khs/kh and of the time-scale separation between the ATP- and ADP-state , expressed as the ratio between the substrate unbinding rates between the ATP- and ADP-state . ( B ) Non-equilibrium dissociation constant as a function of the hydrolysis acceleration ratio khs/kh and of the stoichiometric ratio between the total substrate and Hsp70 concentrations . The color codes are the same as in Figure 2 , green for the region accessible in equilibrium , and red-to-yellow for the region accessible in non-equilibrium . The blue line is the non-equilibrium dissociation constant reported in Figure 2B . DOI: http://dx . doi . org/10 . 7554/eLife . 02218 . 006 Strikingly , ultra-affinity depends also on the concentration of the substrate as can be inferred from the explicit expression for Keffneq . This is another effect intrinsically tied to the non-equilibrium nature of the cycle . Indeed , at equilibrium , dissociation constants do not depend on the total species concentration since they simply encode the difference of free-energy between the bound and unbound states . We thus explore in Figure 3B the dependence of Keffneq on both khS/kh and the ratio between the total substrate and the total chaperone concentrations ( for [Hsp70]tot = 40 μM ) . Two distinct regimes can be observed here with a sharp transition occuring at [S]tot/[Hsp70]tot = 1 . In the excess of chaperone ( [S]tot/[Hsp70]tot<1 ) we observe ultra-affinity and the previously described behavior , whereas in the excess of substrate ( [S]tot/[Hsp70]tot>1 ) the gain in affinity as a function of khS/kh is limited and Keffneq never exceeds its equilibrium range . This effect can be easily rationalized considering that in the latter condition , substrate binding to ADP-bound state becomes dominant and the system cannot exploit the time-scale separation in binding/unbinding kinetics to achieve ultra-affinity . Quite surprisingly , the experimental characterization of the Hsp70-substrate dissociation constant in physiological , non-equilibrium conditions is extremely limited . However , ultra-affinity was implicitely suggested in a series of works assessing the binding of different substrates to Hsp70s , always in the presence of a co-localized JDP , thus ensuring maximal hydrolysis acceleration upon substrate binding ( Misselwitz et al . , 1998; Laufen et al . , 1999; Sullivan et al . , 2001 ) . In all these assays the substrate was observed to bind more efficiently in the presence of ATP , that drives the chaperone through its cycle , than in the presence of ADP , that instead blocks the system in the Hsp70⋅ADP state , the one with the lowest thermodynamic dissociation constant . We know of only one case where ultra-affinity has been carefully measured for the Hsp70 system interacting with a substrate peptide fused to a J-domain ( Wittung-Stafshede et al . , 2003 ) . The dissociation constant observed with this setup ( ≈0 . 22 nM ) was two orders of magnitude smaller than both the measured KD ( ATP ) and KD ( ADP ) ( ≥30 nM ) . If we combine our model with the specific rates provided in ( Wittung-Stafshede et al . , 2003 ) and values of khS/kh compatible to what reported in the literature , we obtain a predicted Keffneq in the range 0 . 15–0 . 6 nM , which is in excellent agreement with the experimental value . All these findings strongly suggest that ultra-affinity becomes manifest when substrate binding is coupled with enhanced hydrolysis acceleration by the colocalization of the Hsp70 binding region and of a J-domain . Our analysis of the cycle has unveiled the conditions that Hsp70 must satisfy to exhibit ultra-affinity: ( i ) the substrate-exchange rates of the ADP-bound state must be significantly slower than the ATP-state so that extremely different timescales can be exploited; ( ii ) ATP hydrolysis must be enhanced by orders of magnitude to fully enter the ultra-affinity regime; ( iii ) the chaperone must be in excess over the substrate . All these conditions are typically met in the cell by canonical Hsp70 chaperones , such as bacterial DnaK and cytosolic Hsc70 in eukaryotes: Hsp70⋅ADP is known to have limited exchange kinetics; Hsp70s are known to work only in partnership with JDPs; Hsp70s are highly abundant and typically in excess over JDPs ( Finka and Goloubinoff , 2013 ) . Cellular conditions seem thus to be optimal for ultra-affinity . Our results provide an additional example of how ATP hydrolysis can be exploited by cells to overcome the constraints set by equilibrium thermodynamics . The key role of energy consumption in driving biochemical cycles for performing a variety of functions is well established ( Schnakenberg 1976; Hill , 2005; Ge et al . , 2012 ) and recently it has been recognized in cellular processes such as sensing , signaling and adaptation ( Qian and Reluga , 2005; Lan et al . , 2012; Mehta and Schwab , 2012 ) . Notably , the ultra-affinity concept proposed here shares some similarities with the well-known kinetic proofreading ( Hopfield , 1974 ) since in both cases chemical energy consumption is used to increase the binding affinity of specific molecules beyond their equilibrium value . However , a significant dissimilarity can be outlined: kinetic proofreading exploits multiple non-equilibrium steps to enhance a pre-existing difference in binding affinity among various substrates , whereas ultra-affinity is achieved due to the specific ability of the substrate to induce upon binding a non-equilibrium transition in the receptor ( i . e . , the chaperone ) . We expect that a similar approach may be applicable to more specialized Hsp70s , such as HscA ( Hesterkamp and Bukau , 1998 ) , and to other chaperones , with possibly more complex cycles , such as the GroEL/S system , Hsp100s and Hsp90s . As well , it would not be surprising to discover other molecular machines working along the same principles , so that the present study might also provide a general scheme applicable beyond Hsp70 chaperones . The exchange process corresponds to the reactionsHsp70·ATP+ADP→kATP−←kATP+Hsp70+ATP+ADP→kADP+←kADP−Hsp70·ADP+ATP The corresponding rate equations ared[Hsp70·ATP]dt=−kATP−[Hsp70·ATP]+kATP+[ATP][Hsp70]d[Hsp70]dt=− ( kATP+[ATP]+kADP+[ADP] ) [Hsp70]+kATP−[Hsp70·ATP]+kADP−[Hsp70·ADP]d[Hsp70·ADP]dt=−kADP−[Hsp70·ADP]+kADP+[ADP][Hsp70]where kATP− and kADP− are the ATP and ADP release rates , respectively , and kATP+ and kADP+ are the ATP and ADP binding rates , respectively . In steady state we can obtain the concentration of the nucleotide-free state from the second equation , and substitute it in the equations for the nucleotide-bound state concentrations , obtaining , for example ( from the first equation ) d[Hsp70·ATP]dt=−kATP−kADP+[ADP]kATP+[ATP]+kADP+[ADP][Hsp70·ATP]+kADP−kATP+[ATP]kATP+[ATP]+kADP+[ADP][Hsp70·ADP]and analogously for the equation for [Hsp70·ADP] . As a consequence , the expressions for the effective exchange rates kTDex and kDTex arekTDex=kATP−kADP+[ADP]kATP+[ATP]+kADP+[ADP]kDTex=kADP−kATP+[ATP]kATP+[ATP]+kADP+[ADP] Typically the nucleotide release rates are much slower than the nucleotide binding rates ( Table 1 ) , and the above expressions correctly capture that nucleotide release sets upper bounds for the rates , which are then further modulated by the partitioning between ATP and ADP binding . Importantly , the equilibrium constant between the ATP-bound and ADP-bound states is correctly reproduced by the ratio kTDex/kDTex . Analogous expressions can be obtained for the exchange process in the presence of a bound substrate . The acceleration of hydrolysis by JDPs is described here as a modulation of khs in the range kh<khs<kh , maxS , where kh , maxS is the maximal experimentally determined JDP/substrate accelerated hydrolysis rate . In this work we extend our analysis also to larger values of khs for completeness . A more complete description of the underlying co-chaperone binding/unbinding reactions would only overburden the present model without providing further insights . The dimerization of DnaJ into DnaJ2 has also been considered as implicit . The rates that we have used to solve the model equations and for the data in Figure 2 , B have been taken from studies of the DnaK/DnaJ/GrpE system , consistently with previous modeling ( Hu et al . , 2006 ) , and are reported in Table 1 . The value of kTDex , S has been obtained from the other parameters , using the relationkTDex , S=kTDexKD ( ATP ) kDTex , SKD ( ADP ) kDTexwhich holds for a cycle at thermodynamic equilibrium because the free energy difference accumulated over a cycle is ΔG = 0 ( Ge et al . , 2012 ) . The upper bound for the hydrolysis acceleration induced by JDPs ( kh , maxS ) corresponds to the hydrolysis rate experimentally observed in saturation of DnaJ concomitant with the presence of the substrate ( Laufen et al . , 1999 ) . The rate of synthesis of ATP from ADP ( the microscopic reverse of hydrolysis ) is neglected here because it was experimentally proven to be below the level of detectability even in the presence of the hydrolysis-accelerating JDP cochaperones ( Russell et al . , 1998 ) , that is ksynth≤10−6 s−1 . In all the calculations the chaperone concentration is 40 μM . The ratio [ATP]/[ADP] = 10 has been used throughout the calculations , approximately matching the physiological ratio . At equilibrium the detailed balance rule holds , which implies that the scheme in Figure 1 can be solved by balancing each branch of the cycle individually . Keffeq corresponds to the solution of the system of equationskATPon[S][Hsp70·ATP]=kATPoff[Hsp70·ATP·S]kTDex[Hsp70·ATP]=kDT[Hsp70·ADP]kADPon[S][Hsp70·ADP]=kADPoff[Hsp70·ADP·S]khS[Hsp70·ATP·S]=kDTS[Hsp70·ADP·S][S]=[S]tot− ( [Hsp70·ATP·S]+[Hsp70·ADP·S] ) where [S]tot is the total substrate concentration . In non-equilibrium the detailed balance rule does not hold anymore , and individual branches do not lead to a solution of the full cycle . Yet , the steady-state solution still exists and can be found solving the steady-state mass-balance equations for the scheme in Figure 1 , namelykATPon[S][Hsp70·ATP]− ( khS+kATPoff ) [Hsp70·ATP·S]+kDTS[Hsp70·ADP·S]=0kATPoff , S[Hsp70·ATP·S]− ( kh+kATPon[S] ) [Hsp70·ATP]kDT[Hsp70·ADP]=0kADPon[S][Hsp70·ADP]− ( kDTS+kADPoff ) [Hsp70·ADP·S]+khS[Hsp70·ATP·S]=0kADPoff , S[Hsp70·ADP·S]− ( kDT+kADPon[S] ) [Hsp70·ADP]+khs[Hsp70·ATP]=0[S]=[S]tot− ( [Hsp70·ATP·S]+[Hsp70·ADP·S] )	Proteins perform numerous essential tasks in cells . Most of these tasks require the protein to have a very specific structure , which is maintained by a balance of chemical and physical interactions . However , this delicate balance is vulnerable to excessive heat , changes in the pH of the cell , and certain chemicals . As a consequence , proteins could lose their specific structure and stop working . Cells employ a group of specialized proteins—called chaperones—to check that other proteins have the correct structure , and to ‘refold’ those that do not . The Hsp70 chaperone family needs energy to do its job , and it gets this energy from a molecule called ATP . However , the exact way that Hsp70s work and use this energy is not fully understood . One major puzzle is how Hsp70 binds to a protein to fold it up . Previous experiments suggested that this binding is particularly effective if Hsp70 can adopt different structures as part of a complex cycle governed by ATP . Now , De Los Rios and Barducci reveal that the energy released from breaking down ATP molecules enables this extra-efficient binding to occur . According to the proposed model , this is possible under some conditions that are normally found in cells . These include having many more Hsp70 proteins than target proteins , and producing energy at extremely high rates from ATP . The specific kinetic properties of the different structures Hsp70 can form are also crucial . More generally , the principle that energy consumption enhances binding could be extended beyond chaperone proteins and represent a general mechanism for other biomolecular systems .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "physics", "of", "living", "systems" ]	2014	Hsp70 chaperones are non-equilibrium machines that achieve ultra-affinity by energy consumption
Regulation of rod gene expression has emerged as a potential therapeutic strategy to treat retinal degenerative diseases like retinitis pigmentosa ( RP ) . We previously reported on a small molecule modulator of the rod transcription factor Nr2e3 , Photoregulin1 ( PR1 ) , that regulates the expression of photoreceptor-specific genes . Although PR1 slows the progression of retinal degeneration in models of RP in vitro , in vivo analyses were not possible with PR1 . We now report a structurally unrelated compound , Photoregulin3 ( PR3 ) that also inhibits rod photoreceptor gene expression , potentially though Nr2e3 modulation . To determine the effectiveness of PR3 as a potential therapy for RP , we treated RhoP23H mice with PR3 and assessed retinal structure and function . PR3-treated RhoP23H mice showed significant structural and functional photoreceptor rescue compared with vehicle-treated littermate control mice . These results provide further support that pharmacological modulation of rod gene expression provides a potential strategy for the treatment of RP . Retinitis pigmentosa ( RP ) is an inherited retinal degenerative disease with a prevalence of 1 to 3 , 000-5 , 000 births ( Hartong et al . , 2006; Parmeggiani , 2011; Boughman et al . , 1980 ) . More than 3 , 000 mutations in about 60 genes have been identified to be associated with RP ( Hartong et al . , 2006; Daiger et al . , 2013 ) . Most of these mutations are in genes essential for rod photoreceptor development and function ( Hartong et al . , 2006 ) . There is currently no approved medical therapy that slows or prevents rod degeneration in these individuals . One emerging approach to treating retinal degeneration is through targeting the factors that regulate rod gene expression . Studies of retinal development have identified several transcription factors that regulate photoreceptor gene expression . For example , loss of function mutations in the rod-specific transcription factors Nrl or Nr2e3 cause rods to acquire a more cone-like identity ( Mears et al . , 2001; Montana et al . , 2013; Haider et al . , 2006; Haider et al . , 2000; Cheng et al . , 2011; Cheng et al . , 2004; Corbo and Cepko , 2005; Peng et al . , 2005; Chen et al . , 2005; Zhu et al . , 2017; Yu et al . , 2017 ) . Knockout/down strategies have shown that Nrl and Nr2e3 are necessary even in mature rods to maintain their normal levels of gene expression ( Montana et al . , 2013; Zhu et al . , 2017; Yu et al . , 2017 ) . Moreover , the reductions in rod gene expression from deletion of Nrl or Nr2e3 , with either conditional deletion or CRISPR-Cas9 deletion , were sufficient to promote the survival of photoreceptors in multiple models of recessive and dominant RP ( Montana et al . , 2013; Zhu et al . , 2017; Yu et al . , 2017 ) . We have recently reported that this regulatory pathway can be modulated using small molecule modulators of rod gene expression that we have named Photoregulins ( Nakamura et al . , 2016 ) . Treatment of developing or mature retina with Photoregulin1 ( PR1 ) reduces rod gene expression and increases the expression of some cone genes . In addition , treatment of two mouse models of RP ( mice with the RhoP23H and the Pde6bRd1 mutations ) with PR1 slows rod degeneration in vitro ( Nakamura et al . , 2016 ) . However , in vivo analyses of PR1 were limited by the compound's potency , solubility , and stability in vivo . In this study , we have identified a structurally unrelated compound , Photoregulin3 ( PR3 ) that also significantly represses rod gene expression , but is more amenable for in vivo studies . With PR3 treatment , we show anatomical and functional preservation of the retina in RhoP23H mice , providing in vivo proof-of-concept of this novel therapeutic strategy for the treatment of RP . In order to identify compounds that may target Nr2e3 , we searched PubChem ( https://pubchem . ncbi . nlm . nih . gov/ ) for top-scoring hit compounds previously identified to interact with Nr2e3 in transfected CHO-S cells in a luciferase-based assay ( PubChem Assay IDs: 602229 , 624378 , 624394 , and 651849 ) . As a secondary screen for the initial hits we used primary retinal cell cultures and assayed Rhodopsin expression because it is a well-defined target of Nr2e3 signaling and is expressed at high levels exclusively in rod photoreceptors ( Cheng et al . , 2004; Peng et al . , 2005; Haider et al . , 2009 ) . We dissociated retina from postnatal day 5 ( P5 ) mice and cultured them in media containing the small molecules . After treatment for 2 days , we assessed Rhodopsin expression with an immunofluorescence-based assay ( Nakamura et al . , 2016 ) . One compound , PR3 ( Figure 1A ) , showed robust reduction in Rhodopsin compared to DMSO and PR1 treatment ( Figure 1B ) . We confirmed this finding with qPCR analysis using intact retinal explant cultures from P4 mice treated with a 0 . 3 μM dose of PR1 or PR3 . Similar to the immunofluorescence assay , treatment with PR3 resulted in reduced Rhodopsin but not Otx2 , a rod transcription factor upstream of Nr2e3 , compared to DMSO and PR1 ( Figure 1C ) . Mutations in Nr2e3 result in an increased number of S Opsin+ photoreceptors as well as a reduction in rod gene expression ( Haider et al . , 2006; Haider et al . , 2000; Cheng et al . , 2011; Peng et al . , 2005; Chen et al . , 2005; Cheng et al . , 2006 ) . To determine if PR3 treatment also affects cone gene expression , we explanted intact retinas from P11 wild type mice in media containing DMSO or PR3 for 3 days . We used intact retinas for this experiment to assess changes in dorsal and ventral retina independently . After fixation and whole mount immunostaining , we counted S Opsin+ cells in the dorsal and ventral retina . Similar to Nr2e3 mutations , treatment with PR3 resulted in an increase in the number of S Opsin+ cells in the ventral , but not dorsal retina ( Figure 1D–E ) . PR3 was initially identified as a chemical modulator of Nr2e3 in a luciferase-based assay that identified ligands by disruption of the Nr2e3-NCoR dimer complex and had a calculated IC50 of 0 . 07 μM in this assay ( PubChem Assay IDs: 602229 , 624378 , 624394 , and 651849 ) . To confirm a direct Nr2e3-PR3 interaction , we used isothermal titration calorimetry ( ITC ) . Consistent with our other assays , ITC qualitatively showed a direct interaction between PR3 and Nr2e3 ( Figure 1F; estimated Kd of 67 μM using a one site model ) . The initial screen that identified PR3 demonstrated its effects on Nr2e3 in a co-repression assay with NCoR; however , the effects we observed on rod gene expression suggested that PR3 also inhibits the co-activator function of Nr2e3 . To explore this possibility , we assessed the effects of PR3 on the ability of Nr2e3 to cooperate with Nrl and Crx , two other transcription factors known to form a co-activation complex for rod gene expression . We first transfected HEK cells with Nr2e3 , Nrl , and Crx and measured activation of the Rhodopsin promoter in the presence of DMSO or PR3 . PR3 strongly reduced Rhodopsin promoter activation ( Figure 1—figure supplement 1A ) . We next used co-immunoprecipitation to assess whether PR3 affects the interaction between Nr2e3 and Nrl in HEK 293 cells . We found that PR3 caused an increase in the binding of Nr2e3 to Nrl ( Figure 1—figure supplement 1B–C ) . Several reports have shown that ligand binding to nuclear receptors stabilizes interactions with co-activators by stabilizing the structure of the ligand-binding domain and co-activator binding sites ( Onishi et al . , 2010; Jia et al . , 2009; Forrest and Swaroop , 2012; Fu et al . , 2014 ) . It is not clear why stabilizing the complex would reduce its activity; however , it may be that stabilizing interactions with Nr2e3 prevents Nrl and Crx from interacting with consensus sites on the DNA , or alternatively prevents the recruitment of other components of the transcriptional machinery . Nevertheless , our results indicate that PR3 affects the formation or stabilization of the complex among these critical rod gene regulators . Although PR3 was identified in an assay for Nr2e3 , it is also possible that it interacts with other nuclear receptors . Two related nuclear receptors that are expressed in photoreceptors are Errb and Rorb . Deletion of these genes in mice has effects on rod gene expression ( Onishi et al . , 2010; Jia et al . , 2009; Forrest and Swaroop , 2012; Fu et al . , 2014 ) . However , small molecule modulators of ERRb induce rapid rod death ( Onishi et al . , 2010 ) , and so it is unlikely that PR3 acts through this pathway . Loss of function Rorb mutations reduce rod gene expression and lead to a phenotype more like what we observe with PR3 treatment . To determine whether PR3 might also act through antagonism of RORb we transfected HEK 293 cells with RORb and Crx and measured activation of the S Opsin promoter , a known target of RORb . In transfected HEK 293 cells , RORb and Crx synergistically activate the S Opsin promoter ( Srinivas et al . , 2006; Liu et al . , 2017 ) . Treatment with PR3 did have a small effect on RORb-Crx driven activation of the S Opsin promoter ( Figure 1—figure supplement 1D ) ; however , this effect was much smaller than what we found for PR3 on activation of the Rhodopsin promoter by Nr2e3 , Nrl , and Crx ( Figure 1—figure supplement 1A ) . To further evaluate a role for PR3 as an RORb antagonist , we analyzed downstream RORb target gene expression in developing retina . One target that is specific to RORb , but not regulated by Nr2e3 is Prdm1 ( Liu et al . , 2017; Wang et al . , 2014; Mills et al . , 2017 ) . Prdm1 expression was unchanged in P0 retinal explants treated with PR3 ( Figure 1—figure supplement 1E ) , supporting our hypothesis that the effects of PR3 are primarily mediated through modulation of Nr2e3 . To further evaluate whether PR3 has other targets in the retina , we tested PR3 in Rd7 retinas , which harbor a spontaneous mutation in Nr2e3 ( Akhmedov et al . , 2000; Haider et al . , 2001; Chen et al . , 2006 ) . We explanted adult ( P23-35 ) retinas from wild type and Rd7 mice in media containing DMSO or PR3 and measured Rhodopsin expression by qPCR . Consistent with the hypothesis that the effects of PR3 on rod gene expression are mediated through Nr2e3 , PR3 caused a significant reduction in Rhodopsin expression in wild type retinas but not in retinas from Rd7 mice ( Figure 1G ) . While this experiment confirms Nr2e3 specificity of PR3 , it remains unclear why Rd7 retinas exhibit only moderate reductions in rod gene expression ( Corbo and Cepko , 2005; Peng et al . , 2005; Chen et al . , 2005; Haider et al . , 2001 ) , while PR3 treatment or CRISPR-Cas9 deletion of Nr2e3 ( Zhu et al . , 2017 ) result in substantial decreases . This suggests that there is some developmental compensation that is not present when the gene is deleted or inhibited in postmitotic rods; alternatively , PR3 may be acting in a dominant-negative manner , and inhibiting the ability of the Nr2e3-Nrl-Crx complex to function properly . Nr2e3 signaling is important for rod photoreceptor cell fate , development and maturation , and maintenance of expression . To determine the effect of PR3 treatment on gene expression in postmitotic retinal cells , we systemically treated ( intraperitoneal injection; IP ) wild type mice with PR3 or vehicle at P12 . At P13 , 24 hr after the injection , we collected the retinas for global transcriptome analysis by RNA sequencing . As hypothesized , we observed a decrease in most rod photoreceptor-specific transcripts ( Figure 2A–B ) . Similar to conditional knockout of Nrl in adult mice and knockdown of Nrl by CRISPR/Cas9 in postmitotic photoreceptors , we did not observe global increases in cone gene expression ( Montana et al . , 2013; Yu et al . , 2017 ) . Nevertheless , we did observe an increase in the number of S Opsin+ cells in the ventral retina with PR3 treatment in vivo ( Figure 2C–D ) . Genes expressed in both rod and cone photoreceptors , eg . Crx and Otx2 , showed no difference in expression between control and PR3 treatment . We next used electron microscopy ( EM ) to perform an ultrastructural analysis of photoreceptor morphology after PR3 treatment . We carried out IP injections of vehicle or PR3 in wild type mice for three consecutive days starting at P11 . At P14 , 24 hr after the third injection , we euthanized the mice and processed their retinas for EM . Photoreceptor outer segments begin to form during the second and third postnatal week; mutations in Nr2e3 lead to an impairment in rod outer segment formation , and we predicted that PR3 treatment would affect their development in a similar way ( Haider et al . , 2006 ) . As predicted , PR3 treatment prevented outer segment development; outer segments of PR3-treated photoreceptors were strikingly truncated compared to controls ( Figure 2E ) . We did not observe any indication of photoreceptor apoptosis induced by PR3 treatment upon examination of outer nuclear layer ( ONL ) nuclei ( Figure 2E ) , further indicating that the effects on rod development were not due to an increase in cell death . We have recently shown that reductions in rod gene expression caused by treatment with PR1 were sufficient to slow the degeneration of RhoP23H photoreceptors in vitro ( Nakamura et al . , 2016 ) . The RhoP23H mutation causes misfolding of Rhodopsin in rod photoreceptors , which leads to activation of the unfolded protein response and eventually results in rod and cone death ( Nguyen et al . , 2014 ) . In RhoP23H mice , most rod photoreceptors undergo apoptosis by the end of the third postnatal week ( Sakami et al . , 2011 ) . To determine whether we could prevent photoreceptor degeneration in this model , we treated RhoP23H mice with PR3 or vehicle from P12-P14 until P21 , during the period of rod photoreceptor death ( Figure 3A ) . At P21 , we assessed visual function with ERGs and euthanized the mice for histological and qPCR analyses . At P21 , control RhoP23H mice had only 2–3 rows of photoreceptors remaining in their ONL ( Figure 3B–C ) . Rods were sparse and there were few remaining cones ( S Opsin+ and Cone Arrestin+ photoreceptors ) . Control RhoP23H mice had minimal scotopic and photopic b-wave amplitudes by ERG analysis ( Figure 4A , B , D and E ) . By contrast , retinas from PR3-treated RhoP23H mice had several rows of rod and cone photoreceptors in the ONL ( Figure 3B–C ) . The surviving cones in the PR3-treated retinas were more elongated and healthier than in the DMSO control retinas . We confirmed our histological results with qPCR on retinas from control and PR3 RhoP23H mice and found that treated mice had more expression of Recoverin and Rhodopsin , indicating greater photoreceptor cell survival ( Figure 3D ) . ERG analysis of PR3-treated RhoP23H mice showed significantly elevated scotopic and photopic b-wave amplitudes at most stimulation intensities compared to littermate controls ( Figure 4A , C , D and F ) . Together , these data support the conclusion that PR3 treatment prevented structural and functional degeneration of photoreceptors in this model of RP . In this study we successfully prevented photoreceptor degeneration in the RhoP23H mouse , the first report of successful treatment of this RP model with a small molecule in vivo . Our strategy was to reduce the expression of photoreceptor genes by targeting the rod-specific nuclear receptor Nr2e3 with a small molecule modulator . Treatment with PR3 decreased rod gene expression , and was sufficient to functionally and structurally preserve photoreceptors in the RhoP23H mouse . Previous studies have shown that genetic manipulation of the rod photoreceptor differentiation pathway may be useful for the treatment of multiple RP models . Conditional deletion of Nrl in adult mouse rods prevents degeneration in the Rho−/− model of recessive RP ( Montana et al . , 2013 ) . More recently , knockdown of Nrl or Nr2e3 by AAV-CRISPR/Cas9 gave long-term histological and functional preservation of photoreceptors in numerous RP models ( Zhu et al . , 2017; Yu et al . , 2017 ) . Our report now shows that a small molecule targeting this same regulatory pathway is also effective at slowing rod degeneration in a particularly aggressive RP model and provides a novel target for medical therapy of retinal degeneration . C57Bl/6 ( Jackson Stock No: 000664 ) , RhoP23H ( Sakami et al . , 2011 ) ( Jackson Stock No: 017628 ) , and Nr2e3Rd7 ( Jackson Stock No: 004643 ) mice were used at the indicated ages . All mice were housed by the Department of Comparative Medicine at the University of Washington and protocols were approved by the University of Washington Institutional Animal Care and Use Committee . The research was carried out in accordance with the ARVO statement for the Use of Animals in Ophthalmic and Vision Research . For all experiments , a sample size of at least four mice per condition was chosen to ensure adequate power to detect a pre-specified effect size . From each litter , half of the animals were randomly assigned to the control group and the other half to the experimental group , and no animals were excluded . Photoregulin3 was identified by searching previous small molecule screens with PubChem for Nr2e3 interacting molecules . It was initially obtained from ChemDiv and then synthesized and purified in large quantities in the lab after initial screening . For in vivo experiments , mice were injected intraperitoneally with PR3 dissolved in DMSO at 10 mg/kg . Retinas were dissected from postnatal day 5 ( P5 ) mice and dissociated by treatment with 0 . 5% Trypsin diluted in calcium- and magnesium-free HBSS for 10 min at 37°C . Trypsin was inactivated by adding an equal volume of FBS and cells were pelleted by centrifugation at 4°C and resuspended in media ( Neurobasal-A containing 1% FBS , 1% N2 , 1% B27 , 1% Pen/Strep , and 0 . 5% L-Glutamine ) . For the immunofluorescence assay , cells were plated into 96-well black walled , clear bottom tissue culture plates at a density of 1 retina/5 wells . Small molecules were diluted in media and were added the day following dissociation . After two days of treatment , cells were fixed with 4% PFA for 20 min at room temperature , blocked with blocking solution ( 10% Normal Horse Serum and 0 . 5% Triton X-100 diluted in 1X PBS ) for 1 hr at room temperature , and incubated overnight at 4°C with primary antibodies generated against Rhodopsin ( 1:250; Rho4D2 , Gift from Dr . Robert Molday , UBC ) diluted in blocking solution . The following day , wells were washed with 1X PBS and then incubated with species appropriate , fluorescently labeled secondary antibodies diluted in blocking solution for 1 hr at room temperature . Wells were washed three times , counterstained with ToPro3 , and the entire plate was imaged using a GE Typhoon FLA 9400 imager . Optical density measurements were obtained from the plate scans using ImageJ software and Rhodopsin expression was normalized to ToPro3 nuclear stain . RNA from retinas was isolated using TRIzol ( Invitrogen ) and cDNA was synthesized using the iScript cDNA synthesis kit ( Bio-Rad ) . SSO Fast ( Bio-Rad ) was used for quantitative real-time PCR . For analysis , values were normalized to Gapdh ( ΔCT ) and ΔΔCT between DMSO and compound-treated samples was expressed as percent of DMSO treated controls ( 100*2^ΔΔCt ) . t-tests were performed on ΔCT values . The following primer sequences were used: Gapdh ( F: GGCATTGCTCTCAATGACAA , R: CTTGCTCAGTGTCCTTGCTG ) , Rhodopsin ( F: CCCTTCTCCAACGTCACAGG , R: TGAGGAAGTTGATGGGGAAGC ) , Opn1sw ( F: CAGCATCCGCTTCAACTCCAA , R: GCAGATGAGGGAAAGAGGAATGA ) , Recoverin ( F: ACGACGTAGACGGCAATGG , R: CCGCTTTTCTGGGGTGTTTT ) , Otx2 ( F: CCGCCTTACGCAGTCAATG , R: GAGGGATGCAGCAAGTCCATA ) , and Prdm1 ( F: TTCTCTTGGAAAAACGTGTGGG , R: GGAGCCGGAGCTAGACTTG ) . Intact retinas without RPE from mice at the indicated ages and genotypes were explanted on 0 . 4 μm pore tissue culture inserts in media ( Neurobasal-A containing 1% FBS , 1% N2 , 1% B27 , 1% Pen/Strep , and 0 . 5% L-Glutamine ) containing DMSO or 0 . 3–1 . 0 μM PR3 . Full media changes were performed every other day . Explants or eye cups were fixed with 4% PFA for 20 min at room temperature , blocked with blocking solution ( 10% normal horse serum and 0 . 5% Triton X-100 diluted in 1X PBS ) for 1 hr at room temperature , and incubated overnight at 4°C with primary antibodies generated against S Opsin ( 1:400 , SCBT , sc-14363 ) . The following day , the retinas were washed with 1X PBS , and then incubated with a species appropriate , fluorescently-labeled secondary antibody diluted in blocking solution overnight , followed by washing with 1X PBS and DAPI staining . The retinas were transferred to slides and coverslipped with Fluoromount-G ( SouthernBiotech ) . An Olympus FluoView FV1000 was used for confocal microscopy . Cells were counted from single plane confocal images taken at fixed settings . Eyecups were fixed in 4% PFA in 1X PBS for 20 min at room temperature and then cryoprotected in 30% sucrose in 1X PBS overnight at 4°C . Samples were embedded in OCT ( Sakura Finetek ) , frozen on dry ice , and then sectioned at 16–18 μm on a cryostat ( Leica ) . Slides were blocked with a solution containing 10% normal horse serum and 0 . 5% Triton X-100 in 1X PBS for 1 hr at room temperature and then stained overnight at 4°C with primary antibodies ( Rho4D2 at 1:250 from Dr . Robert Molday , S Opsin at 1:400 from SCBT: sc-14363 , Cone Arrestin at 1:1000 from Millipore: AB15282 , Otx2 at 1:200 from R and D Systems: BAF1979 ) diluted in blocking solution . Slides were washed three times with 1X PBS the following day and then incubated in fluorescently labeled secondary antibodies diluted in blocking solution for 2 hr at room temperature , stained with DAPI , washed , and coverslipped using Fluoromount-G ( SouthernBiotech ) . An Olympus FluoView FV1000 was used for confocal microscopy . Cells were counted from single plane confocal images taken at fixed settings . Counts in the central retina were taken adjacent to the optic nerve head ( 50 μm from the nerve head on the ventral side ) and counts in the peripheral retina were taken 50 μm from the peripheral edge on the ventral side . NR2E3 protein ( aa 90–410 ) was expressed as an His8-MBP-TEV fusion protein from the expression vector pVP16 ( DNASU Plasmid ID: HsCD00084154 ) . E . coli BL21 ( DE3 ) cells were grown to an OD600 of 1 , and then induced with 0 . 2 mM IPTG at 16°C overnight . Cells were harvested , resuspended in extract buffer ( 20 mM Tris pH 8 , 200 mM NaCl , 10% glycerol , 5 mM 2-mercaptoethanol , and saturated PMSF diluted 1:1 , 000 ) , and then lysed by sonication on ice . Lysates were centrifuged at 4°C and the supernatant was loaded onto an equilibrated column containing 5 mL of Ni-NTA agarose ( Qiagen ) . The column was washed with 20 mM Tris pH 8 , 1M NaCl , 5 mM 2-mercaptoethanol , and 40 mM imidazole , and then the protein was eluted with 20 mM Tris pH 8 , 200 mM NaCl , 5 mM 2-mercaptoethanol , and 100 mM imidazole . The fusion protein was incubated with TEV overnight at 4°C and then the His8-MBP tags were separated from NR2E3 by ion exchange chromatography . For isothermal titration calorimetry , 100 μM PR3 was injected into 20 μM NR2E3 in 10 mM Sodium Phosphate buffer pH 8 with 50 mM NaCl and 0 . 5% DMSO in a MicroCal ITC-200 ( Malvern ) and the data was analyzed with Origin 7 . 0 software . RNA from retinas was isolated using TRIzol ( Invitrogen ) and total RNA integrity was checked using an Agilent 4200 TapeStation and quantified with a Trinean DropSense96 spectrophotometer . RNA-seq libraries were prepared from total RNA using the TruSeq RNA Sample Prep kit ( Illumina ) and a Sciclone NGSx Workstation ( PerkinElmer ) . Library size distributions were validated using an Agilent 4200 TapeStation . Additional Library quality control , blending of pooled indexed libraries , and cluster optimization were performed using Life Technologies’ Invitrogen Qubit Fluorometer . RNA-seq libraries were pooled ( 4-plex ) and clustered onto a flow cell lane . Sequencing was performed using an Illumina HiSeq 2500 in rapid mode employing a paired-end , 50 base read length ( PE50 ) sequencing strategy . Mice were euthanized by CO2 , and then perfused with 0 . 9% saline followed by 4% glutaraldehyde in 0 . 1 M sodium cacodylate buffer . Eye cups were fixed in 4% glutaraldehyde in 0 . 1 M sodium cacodylate buffer , washed with 0 . 1 M sodium cacodylate buffer , and then post-fixed in 2% osmium tetroxide . After fixation , eye cups were washed with water , dehydrated through a graded series of ethanol , incubated in propylene oxide and then epon araidite , polymerized overnight at 60°C , and then sectioned at a thickness of 70 nm . Images were obtained using a JEOL JEM-1230 electron microscope . Mice were dark adapted overnight ( 12–18 hr ) . All subsequent steps were carried out under dim red light . Mice were placed in an anesthesia chamber and anesthetized with 1 . 5–3% isoflurane gas . Mice were transferred from the anesthesia chamber to a heated platform maintained at 37 ˚C and positioned in a nose cone to maintain a constant flow of Isoflurane . Drops of 1% tropicamide and 2 . 5% phenylephrine Hydrochloride were applied to each eye . A reference needle electrode was placed subdermally on the top of the head and a ground needle electrode was placed subdermally in the tail . Drops of 1 . 5% methyl cellulose were applied to each eye and contact lens electrodes were placed over each eye . Dim red light was turned off and the platform was positioned inside of an LKC Technologies UTAS BigShot ganzfeld and a series of flashes of increasing intensity were delivered scotopically . A series of photopic flashes were performed immediately following the series of scotopic flashes . HEK293 cells were transfected with 1 μg of the luciferase reporter BR-225Luc ( Dr . Shiming Chen ) or S Opsin 600 pGL3 ( Dr . Douglas Forrest ) , 1 ng of the control pRL-CMV ( Promega ) and 100 ng of hNRL-pCMVSport6 ( Open Biosystems ) , hCRX-pCMVSport6 ( Open Biosystems ) , hNR2E3-pcDNA3 . 1/HisC ( Dr . Shiming Chen ) , or hRORB-pLenti6 . 2/V5-DEST ( DNASU ) in 24-well plates using Lipofectamine 3000 reagent ( Thermo Fisher Scientific ) . Transfection reagents were removed the following day and replaced with media containing DMSO or 1 μM PR3 for 2 days . Cells were lysed and firefly and renilla luciferase activity was measured with the Dual-Luciferase Reporter Assay System ( Promega ) using a 1420 Multilabel Victor3V plate reader . HEK293 cells were transfected in 6-well tissue culture plates with lipofectamine 3000 ( Thermo Fisher Scientific ) and 800 ng of each hNRL-pCMVSport6 ( Open Biosystems ) and hNR2E3-pcDNA3 . 1/HisC ( Dr . Shiming Chen ) in Opti-MEM media . Transfection reagents were removed after 24 hr and replaced with media containing DMSO or 1 μM PR3 for 2 days . Cells were lysed with Co-IP lysis buffer ( 25 mM Tris-HCl pH 7 . 5 , 150 mM NaCl , 1 mM EDTA , 1% Triton X-100 , 5% glycerol and 1X protease inhibitor cocktail ) . Sheep anti-mouse IgG magnetic Dynabeads were incubated with anti-Nr2e3 antibody ( 5 μg/precipitation ) diluted in Co-IP buffer for 2 hr at 4°C . Equal volume of lysate were then added to the antibody-coated beads and incubated overnight at 4°C . The following day , beads were washed four times with Co-IP buffer and then incubated at 85°C for 15 min in 1X sample buffer diluted in Co-IP buffer . Samples were loaded and run in a 4–20% SDS gel ( Bio-Rad ) . Protein was transferred to a PVDF membrane ( Thermo Fisher Scientific ) , blocked ( 5% BSA and 0 . 1% Tween 20 in 1X PBS ) for at least 1 hr at room temperature and stained with primary antibodies anti-Nrl ( Chemicon ) or Anti-Nr2e3 ( R and D systems ) diluted in blocking solution overnight at 4°C . Membranes were washed with 0 . 1% Tween 20 in 1X PBS and then incubated with Clean-Blot IP Detection Reagent ( Thermo Fisher Scientific ) diluted in blocking solution for 1 hr at room temperature . Signals were visualized on X-ray film with SuperSignal West Dura Extended Duration Substrate ( Thermo Fisher Scientific ) and quantified using ImageJ software .	There are several diseases that cause people to lose their eyesight and become blind . One of these diseases , called retinitis pigmentosa , kills cells at the back of the eye known as rod cells . At first , it affects vision in low light and peripheral vision , but later it affects vision during the daytime as well . There are no effective treatments for patients with retinitis pigmentosa . Yet previous genetic studies have shown that disrupting the activity of genes in rod cells can slow the progression of the disease and preserve vision in mice . As for all genes , proteins called transcription factors regulate the activity of rod cell genes . Nakamura et al . now report the discovery of a small drug-like molecule , that they name Photoregulin3 , which alters the activity of a transcription factor that regulates rod genes . In follow-up experiments , mice with a mutation that replicates many of the features of retinitis pigmentosa were given Photoregulin3 to see if it could slow the progression of the disease . Indeed , Photoregulin3 could stop many of the rod cells from degenerating in the treated mice . At the end of the experiment , the mice treated with this small molecule had about twice as many rods as the control mice . The treated mice also responded better to flashes of light . Nakamura et al . hope that the findings will one day benefit patients with retinitis pigmentosa . But first more research needs to be done before testing Photoregulin3 in humans . For example , the drug-like molecule needs to be made more potent , and if possible adapted to work when given orally , meaning patients could take it as a pill .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "short", "report", "neuroscience" ]	2017	Small molecule Photoregulin3 prevents retinal degeneration in the RhoP23H mouse model of retinitis pigmentosa
Wapl induces cohesin dissociation from DNA throughout the mitotic cell cycle , modulating sister chromatid cohesion and higher-order chromatin structure . Cohesin complexes containing meiosis-specific kleisin subunits govern most aspects of meiotic chromosome function , but whether Wapl regulates these complexes remains unknown . We show that during C . elegans oogenesis WAPL-1 antagonizes binding of cohesin containing COH-3/4 kleisins , but not REC-8 , demonstrating that sensitivity to WAPL-1 is dictated by kleisin identity . By restricting the amount of chromosome-associated COH-3/4 cohesin , WAPL-1 controls chromosome structure throughout meiotic prophase . In the absence of REC-8 , WAPL-1 inhibits COH-3/4-mediated cohesion , which requires crossover-fated events formed during meiotic recombination . Thus , WAPL-1 promotes functional specialization of meiotic cohesin: WAPL-1-sensitive COH-3/4 complexes modulate higher-order chromosome structure , while WAPL-1-refractory REC-8 complexes provide stable cohesion . Surprisingly , a WAPL-1-independent mechanism removes cohesin before metaphase I . Our studies provide insight into how meiosis-specific cohesin complexes are regulated to ensure formation of euploid gametes . Structural maintenance of chromosome ( SMC ) proteins take part in complexes that associate with DNA to promote key events of the cell cycle , such as chromosome condensation and segregation , DNA repair , and gene expression ( Jeppsson et al . , 2014 ) . The cohesin complex , which mediates sister chromatid cohesion ( SCC ) between S-phase and chromosome segregation at anaphase ( Michaelis et al . , 1997 ) , consists of two SMC proteins ( Smc1 and Smc3 ) plus a kleisin subunit ( Scc1/Rad21 ) , forming a tripartite structure that topologically embraces DNA molecules ( Haering et al . , 2002; Haering et al . , 2008 ) . A fourth cohesin subunit ( Scc3 ) binds to the kleisin and is also required for the functionality of the complex , while other proteins associate temporarily with cohesin to regulate its binding to DNA ( Haarhuis et al . , 2014 ) . Cohesin is loaded to chromosomes by the Scc2/4 complex ( Ciosk et al . , 2000 ) , and SCC is established during DNA replication in a process that involves acetylation of Smc3 ( Ben-Shahar et al . , 2008; Unal et al . , 2008; Zhang et al . , 2008 ) . SCC is ultimately dissolved at anaphase onset , when cleavage of the kleisin subunit by the protease separase triggers the segregation of sister chromatids to opposite poles of the spindle ( Uhlmann et al . , 1999 ) . Proper establishment and release of SCC is also essential for chromosome segregation during meiosis , the specialized cell division program that produces haploid gametes from diploid germ cells ( Petronczki et al . , 2003 ) . In addition to the separase-dependent removal of cohesin at anaphase onset , a pathway dependent on the Wapl protein removes cohesin from chromosomes at earlier stages of the cell cycle in somatic cells ( Gandhi et al . , 2006; Kueng et al . , 2006 ) . Wapl is thought to destabilize the interaction between the kleisin and Smc3 subunits , allowing the release of cohesin from DNA without catalytically cleaving any subunit ( Chan et al . , 2012; Eichinger et al . , 2013; Huis in 't Veld et al . , 2014 ) . Cohesin complexes in which the Smc3 subunit is acetylated during DNA replication become resistant to Wapl and remain stably bound to DNA , thereby providing persistent SCC ( Ben-Shahar et al . , 2008; Nishiyama et al . , 2010; Lopez-Serra et al . , 2013 ) . However , acetylated cohesin is also removed from chromosome arms during prophase and early prometaphase in mammalian cells , when phosphorylation of Scc3 and Sororin , a protein that antagonizes Wapl , renders these complexes sensitive to Wapl ( Hauf et al . , 2005; Nishiyama et al . , 2010; Nishiyama et al . , 2013 ) . This mode of cohesin removal is known as the prophase pathway and its failure in cells lacking Wapl causes increased arm cohesion in metaphase chromosomes and defects in chromosome segregation ( Waizenegger et al . , 2000; Haarhuis et al . , 2013; Tedeschi et al . , 2013 ) . Removal of Wapl before S-phase causes a large increase in chromatin-associated cohesin and dramatic changes in chromosome organization ( Tedeschi et al . , 2013 ) , demonstrating that Wapl is a key regulator of SCC and chromatin organization throughout the mitotic cell cycle . Cohesin is an essential component of meiotic chromosomes , not only by mediating SCC , but also by promoting the acquisition of structural features required for meiotic chromosome function ( McNicoll et al . , 2013 ) . Meiotic chromosomes are organized as linear arrays of chromatin loops , which are attached at their base to cohesin-containing proteinaceous axial elements ( Kleckner , 2006 ) . Proper assembly of axial elements during early prophase is required for subsequent pairing and recombination between homologous chromosomes . Crossovers formed during recombination , together with SCC , form attachments between homologous chromosomes ( chiasmata ) that are responsible for the correct orientation of chromosomes on the first meiotic spindle and ultimately for their correct partitioning during the meiotic divisions ( Petronczki et al . , 2003 ) . Crucially , these events require the formation of axial elements containing meiosis-specific cohesin complexes in which the mitotic kleisin Scc1 is substituted by Rec8 ( Klein et al . , 1999; Watanabe and Nurse , 1999 ) . Moreover , additional meiosis-specific kleisins beyond Rec8 have been identified in mouse ( Rad21L ) ( Herrán et al . , 2011; Ishiguro et al . , 2011; Lee and Hirano , 2011 ) and in C . elegans , where the highly homologous and functionally redundant COH-3 and COH-4 kleisins associate with SMC-1 and SMC-3 to form cohesin complexes that associate with meiotic chromosomes independently of REC-8 cohesin ( Severson et al . , 2009; Severson and Meyer , 2014 ) . Although Rad21L and COH-3/4 are not essential for SCC , these kleisins are required for pairing and recombination between homologous chromosomes ( Ishiguro et al . , 2014; Severson and Meyer , 2014 ) . Interestingly , large amounts of Rad21L and COH-3/4 are removed from chromosomes before metaphase I ( Herrán et al . , 2011; Ishiguro et al . , 2011; Lee and Hirano , 2011; Severson and Meyer , 2014 ) , and a prophase pathway has recently been proposed to operate during late meiotic prophase in plants ( De et al . , 2014 ) . However , whether Wapl induces cohesin removal at any stage of meiotic prophase in animals , and whether Wapl may regulate some of the functions of different meiosis-specific cohesin complexes is not known . Using the C . elegans germ line , which contains a complete time course of meiotic prophase , we demonstrate that WAPL-1 antagonizes cohesin binding from the onset of meiosis , and show that cohesin complexes containing the COH-3/4 kleisins are specifically targeted by WAPL-1 . By antagonizing the binding of COH-3/4 complexes to axial elements , WAPL-1 acts as a regulator of meiotic chromosome structure and SCC . Moreover , we also show that SCC is modulated by WAPL-1 and recombination during the chromosome remodeling process that starts at the end of pachytene , and report that a WAPL-1-independent mechanism removes cohesin during the oocyte maturation process preceding metaphase I . In order to investigate the role of WAPL-1 during meiotic prophase , we used a deletion allele , wapl-1 ( tm1814 ) , that removes the first two exons of the C . elegans Wapl homolog ( Figure 1A ) . Western blot analysis on whole-worm protein extracts showed that the WAPL-1 protein is absent in wapl-1 ( tm1814 ) mutants , confirming that the tm1814 deletion is a null allele of wapl-1 ( Figure 1B ) . Homozygous wapl-1 ( tm1814 ) mutants ( referred from now on as wapl-1 mutants ) are viable , but display a reduction in brood size and high levels of embryonic lethality ( Figure 1C ) . In addition to these reproductive defects , wapl-1 mutants also displayed somatic defects , as demonstrated by the high incidence of larval arrest amongst the hatched wapl-1 embryos ( Figure 1C ) and by the presence of an egg laying defect in adult worms . In order to prevent the accumulation of somatic defects , all the analysis presented here was performed in homozygous wapl-1 worms derived from heterozygous mothers . 10 . 7554/eLife . 10851 . 003Figure 1 . WAPL-1 localizes to germ line nuclei and promotes viability . ( A ) Structure of the wapl-1 gene , red bar indicates the region deleted in the tm1814 allele . ( B ) Western blot demonstrates that wapl-1 ( tm1814 ) is a null allele and that a protein of the expected size is present in worms carrying a GFP::wapl-1 transgene . ( C ) wapl-1 mutants display reduced fertility and larval lethality , numbers in parenthesis indicate total number of embryos analysed per genotype . ( D ) Projections of whole-mounted germ lines stained with DAPI , the different stages of meiotic prophase are noted above the WT germ line , with transition zone containing nuclei in leptotene and zygotene . Note that overall germ line organization in wapl-1 mutants is similar to WT . ( E ) Projections of diakinesis oocytes stained with DAPI , six bivalents are present in both WT and wapl-1 mutants . ( F ) Whole-mounted germ line from a transgenic worm homozygous for the wapl-1 ( tm1814 ) deletion and for a GFP::wapl-1 single copy transgene stained with DAPI and anti-GFP antibodies . Note that the intensity of GFP::WAPL-1 decreases in transition zone and peaks again during late pachytene . ( G ) Insets from germ line shown in F showing GFP::WAPL-1 staining in transition zone and pachytene nuclei , note that GFP::WAPL-1 intensity is very high in transition zone nuclei that do not display chromosome clustering ( arrowheads ) . Figure 1—figure supplement 1 shows quantification of GFP::WAPL-1 intensities along the germ line . Scale bars in E and G = 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 00310 . 7554/eLife . 10851 . 004Figure 1—figure supplement 1 . Quantification of GFP::WAPL-1 intensity . Y axis depicts average nuclear intensity of GFP signal per nucleus ( calculated from maximum intensity projections made from non-deconvolved Z-stacks acquired with the same settings ) , X axis indicates stages at which intensity was measured , and error bars indicate standard deviation . Note the drop in GFP::WAPL-1 fluorescence intensity between pre-leptotene and leptotene nuclei . Differences between pre-leptotene and leptotene nuclei , and between leptotene and late pachytene nuclei are significant ( p<0 . 0001 , t-test ) . A total of 58 ( pre leptotene ) , 61 ( leptotene ) , and 42 ( late pachytene ) nuclei from 6 different germ lines were included in the fluorescence intensity analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 004 The overall organization of wapl-1 mutant germ lines appears largely normal , with clearly defined mitotic and meiotic compartments in which the different stages of meiotic prophase can be easily identified ( Figure 1D ) . In fact , observation of diakinesis oocytes ( the last stage of meiotic prophase ) showed that both wild type and wapl-1 mutant oocytes displayed 6 DAPI-stained bodies , demonstrating that WAPL-1 is not required for chiasma formation ( Figure 1E ) . Nonetheless , the reduced fertility of wapl-1 mutants suggested that WAPL-1 may play important roles in the germ line . Thus , we investigated the staining pattern of WAPL-1 during meiosis by creating transgenic worms homozygous for the tm1814 deletion and for a single-copy insertion of a transgene that expresses a GFP::WAPL-1 fusion protein using the 5’ and 3’ UTRs from the wapl-1 locus . Expression of this transgene largely rescued the fertility defects of wapl-1 ( tm1814 ) mutants ( Figure 1C ) , and western blot analysis confirmed the presence of a band of the expected molecular weight for the GFP::WAPL-1 fusion protein , although the overall intensity of this band was reduced compared to the endogenous WAPL-1 protein ( Figure 1B ) . Staining of germ lines from these transgenic worms with anti-GFP antibodies demonstrated that the GFP::WAPL-1 protein is present , with a diffuse staining pattern , in both mitotic and meiotic nuclei ( Figure 1F ) . Interestingly , the intensity of the GFP::WAPL-1 signal decreased drastically as transition zone nuclei acquired the chromosome clustering characteristic of early meiotic prophase stages ( leptotene and zygotene ) , peaking again at late pachytene and then remaining at similar high levels in diplotene and diakinesis oocytes ( Figure 1F–G and Figure 1—figure supplement 1 ) . Despite normal formation of chiasmata , the high incidence of embryonic lethality among the progeny of wapl-1 mutants ( Figure 1C ) suggested the existence of meiotic defects . Moreover , the presence of developmental defects among the progeny of wapl-1 mutants could be a consequence of defects in DNA repair during meiosis , so we monitored the progression of meiotic recombination by visualizing the appearance and disappearance of the RAD-51 recombinase , which labels early meiotic recombination intermediates ( Colaiácovo et al . , 2003 ) , and of COSA-1 and ZHP-3 foci , two proteins that are required for crossover formation and that localize specifically to crossover-fated recombination events in late pachytene nuclei ( Bhalla et al . , 2008; Yokoo et al . , 2012 ) . We observed 6 COSA-1 and 6 ZHP-3 foci in both wapl-1 mutants and wild-type controls ( Figure 2A and Figure 2—figure supplement 1 ) , consistent with the normal presence of chiasmata in wapl-1 diakinesis oocytes . Despite this , RAD-51-positive recombination intermediates accumulate in mid pachytene nuclei of wapl-1 mutants ( Figure 2B ) . Furthermore , while RAD-51 foci were no longer detected in 98% of late pachytene nuclei of wild-type controls , 52% percent of nuclei in the same region of wapl-1 mutant germ lines displayed RAD-51 foci . Thus , although crossover precursors are successfully formed , the repair of a subset of DSBs is delayed in wapl-1 mutants . 10 . 7554/eLife . 10851 . 005Figure 2 . WAPL-1 affects DNA repair during meiotic prophase . ( A ) Projections of late pachytene nuclei from worms expressing COSA-1::GFP , note that both wapl-1 mutants and WT controls display 6 COSA-1 foci per nucleus . Graph showing quantification of COSA-1 foci ( 126 nuclei from wapl-1 mutants and 100 nuclei from WT ) . ( B ) Projections of pachytene nuclei stained with anti-RAD-51 antibodies and DAPI , note increased RAD-51 foci in wapl-1 mutant panel . Quantification of RAD-51 foci in germ lines of WT and wapl-1 mutants . Each germ line was divided into 7 equal-sized regions , with regions 4 to 7 representing early to late pachytene . The X axis indicates the seven regions along the germ line , while the Y axis indicates the percentage of nuclei with a given number of RAD-51 foci ( as indicated in the color key ) . wapl-1 mutants accumulate RAD-51 in mid and late pachytene nuclei . Number of nuclei analyzed ( WT , wapl-1 mutant ) : Zone 1 ( 133 , 138 ) , zone 2 ( 246 , 195 ) , zone 3 ( 137 , 135 ) , zone 4 ( 154 , 101 ) , zone 5 ( 122 , 90 ) , zone 6 ( 114 , 69 ) , zone 7 ( 93 , 61 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 00510 . 7554/eLife . 10851 . 006Figure 2—figure supplement 1 . wapl-1 mutants form normal numbers of ZHP-3 foci . Projections of late pachytene nuclei from worms expressing a ZHP-3::GFP transgene , note that both wapl-1 mutants and WT controls display 6 ZHP-3 foci per nucleus . Graph showing quantification of ZHP-3 foci ( 124 nuclei from wapl-1 mutants and 119 nuclei from WT ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 006 Next , we investigated if the meiotic divisions proceeded normally in the absence of WAPL-1 , as defects in this process could induce aneuploidy even if chiasma formation is not affected . During oogenesis , each meiotic division results in the formation of a polar body that contains a full complement of homologs ( meiosis I ) or sister chromatids ( meiosis II ) . Polar bodies are extruded away from the egg pronucleus , localizing on the cortex and not contributing to the genetic content of the developing embryo ( Figure 3A ) . All wapl-1 mutant embryos analyzed formed two polar bodies , however , only 22% of post meiotic embryos ( up to the two-cell embryo ) displayed both polar bodies at the cortex , with most embryos ( 61% ) displaying one polar body located away from the cortex and the remaining ( 11% ) displaying both polar bodies away from the cortex ( Figure 3A–B ) . In addition , we noticed that chromatin morphology in the polar bodies was altered in wapl-1 mutants , often displaying separated DNA masses within the polar body ( Figure 3—figure supplement 1 ) . To gain better understanding of the effect of WAPL-1 on polar body extrusion , we performed live imaging of the meiotic and early mitotic divisions in wild type and wapl-1 mutant embryos expressing histone H2B::mcherry ( Figure 3D; Videos 1–3 ) . These experiments confirmed that most wapl-1 mutant embryos display defects in the extrusion of the second polar body ( Figure 3C–D; Videos 2–3 ) . We also observed that in 2 out of 11 wapl-1 mutant embryos , the second polar body , which failed to migrate to the cortex , underwent chromatin decondesation and condensation cycles , mimicking the changes observed in the mitotic nuclei of the embryo ( Figure 3D and Videos 2–3 ) . Moreover , in 1 out of 11 filmed embryos the second polar body eventually fused with one of the mitotic nuclei generated after the first mitotic division ( Figure 3D and Video 3 ) , demonstrating that the failure in polar body extrusion of wapl-1 mutants can lead to aneuploidy in the embryo . We also used fluorescence in situ hybridization ( FISH ) to investigate if chromosome non-disjunction occurs during the meiotic divisions of wapl-1 mutants . Labeling of the 5S rDNA locus on chromosome V in early embryos demonstrated that the oocyte pronucleus contained one single signal for the 5S rDNA in all embryos analyzed from wild-type ( 17 ) and wapl-1 mutants ( 17 ) ( Figure 3E ) , suggesting that chromosome V segregates properly during the meiotic divisions in the absence of WAPL-1 . This analysis demonstrates that WAPL-1 is required to ensure proper polar body extrusion during the meiotic divisions , and suggests that defects in this process may contribute to the embryonic lethality observed in wapl-1 mutants . 10 . 7554/eLife . 10851 . 007Figure 3 . WAPL-1 is required for polar body extrusion . ( A ) Projections of fixed embryos at the 1- or 2-cell stage stained with DAPI , arrowheads point to the position of polar bodies generated during the meiotic divisions . Polar bodies are found near the cortex in WT control , but one ( middle panel ) or both ( right-hand side panel ) polar bodies localize away from the cortex in wapl-1 mutant embryos . ( B ) Quantification of the percentage of embryos with zero , one , or two polar bodies localized at the cortex ( 36 embryos scored in wapl-1 and 34 in WT ) . ( C ) Quantification of polar body behavior in videos from live WT and wapl-1 mutant embryos expressing a histone H2B::mcherry fusion protein . Note that 7 out of 11 wapl-1 mutant embryos displayed defects in polar body extrusion . Examples of videos used for the quantification are shown in Video 1 ( WT ) , Videos 2–3 ( wapl-1 ) . ( D ) Selected frames from the WT embryo shown in Video 1 and from the wapl-1 mutant embryo shown in Videos 2–3 . Time is indicated on top-left corner , starting from metaphase I . Abbreviations: PB I ( first polar body ) , PB II ( second polar body ) , OP ( oocyte pronucleus ) , SP ( sperm pronocleus ) , P0 ( first mitotic metaphase following fusion of OP and SP ) , P1 and AB ( cells resulting from the first mitotic division ) . Note that in the WT embryo PB II remains highly condensed and locates close to PB I on the cortex . In the wapl-1 mutant embryo , PB II becomes decondensed and fails to move to the cortex , first remaining close to the OP and then close to the AB cell produced after the first mitotic division . Chromosomes from AB and PB II appear to mix together before the second mitotic division of the embryo . ( E ) Projections of fixed embryos following the completion of the second meiotic division and labeled with a FISH probe against the 5S rDNA locus on chromosome V and DAPI . Note that in both WT and wapl-1 mutant embryos the oocyte pronucleus ( OP ) and the second polar body ( PB II ) contain a single FISH signal , even when PB II is not localized on the cortex and chromatin appears decondensed ( wapl-1 example on right-hand side ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 00710 . 7554/eLife . 10851 . 008Figure 3—figure supplement 1 . Example of abnormal chromatin condensation in polar bodies of wapl-1 mutant embryos . Projection of a wapl-1 mutant 2-cell embryo stained with DAPI . Polar bodies are labeled with arrows , and bottom polar body is magnified in the inset , note the presence of separated chromatin masses within the polar body . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 00810 . 7554/eLife . 10851 . 009Video 1 . Live imaging of a WT embryo expressing histone H2B::mcherry . Filming covers the interval between the first meiotic metaphase and the end of the first mitotic division . Each meiotic division results in the production of a polar body and both polar bodies remain highly condensed and located at one end of the embryo , away from the oocyte and sperm pronuclei . Figure 3D contains individual images from this video in which specific meiotic and mitotic events are labeled ( images in Figure 3D are rotated with respect to the video ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 00910 . 7554/eLife . 10851 . 010Video 2 . Live imaging of a wapl-1 mutant embryo expressing histone H2B::mcherry . Filming covers the interval between the first meiotic division and the end of the first mitotic division . Note that the second polar body does not migrate to the cortex , instead it follows the movement of the oocyte pronucleus towards the middle of the embryo and the chromosomes appear decondensed . As chromosomes in the sperm and oocyte pronuclei condense in preparation for the first mitotic division , condensation of chromatin also occurs in the second polar body . Figure 3D contains individual images from this video in which specific meiotic and mitotic events are labeled , and Video 3 shows continued filming from the same embryo . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 01010 . 7554/eLife . 10851 . 011Video 3 . Continuation of live imaging of the wapl-1 mutant embryo shown in Video 2 . Filming covers the interval between the end of the first mitotic division and up to the four cell embryo . Following the completion of the first mitotic division , chromatin in the second polar body ( PB II ) undergoes decondensation . PB II localizes to the vicinity of the AB mitotic nucleus , and as chromosomes in the AB nucleus condense so do chromosomes in PB II . At this point condensed chromosomes from the AB nucleus and PB II appear to mix up before dividing into two daughter nuclei . Figure 3D contains individual images from this video in which specific mitotic events are labeled . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 011 During the initial observation of wapl-1 mutant germ lines we noticed that pachytene chromosomes appeared more widely spaced within the nucleus and somewhat thicker than in wild-type controls , suggesting that WAPL-1 may regulate the shape of meiotic chromosomes , as it does in mitotic cells ( Tedeschi et al . , 2013 ) . Meiotic chromosomes are organized around axial elements containing cohesin and meiosis-specific HORMA-domain proteins ( Kleckner , 2006 ) . Thus , we used antibodies against HTP-3 , a HORMA-domain protein that is an essential component of axial elements in C . elegans ( Goodyer et al . , 2008; Severson et al . , 2009 ) , to investigate chromosome organization in wapl-1 mutants . Projections made from late pachytene nuclei of wild-type germ lines demonstrated ample overlap between different HTP-3-labeled axial elements , making it difficult to follow individual HTP-3 tracks along their whole length ( Figure 4A and Figure 4—figure supplement 1 ) . In contrast , the overlap of HTP-3 tracks was reduced in wapl-1 mutants , with some nuclei displaying six distinctive HTP-3 tracks ( one per homolog pair ) that could be clearly traced along their full length ( Figure 4A and Figure 4—figure supplement 1 ) . Measuring of total HTP-3 track length per nucleus demonstrated a 28% decrease in axial element length in late pachytene nuclei of wapl-1 mutants ( Figure 4B ) . 10 . 7554/eLife . 10851 . 012Figure 4 . WAPL-1 regulates chromosome organization during meiotic prophase . ( A ) Projections of late pachytene nuclei stained with anti-HTP-3 antibodies ( axial elements ) and DAPI . Axial elements are shorter in wapl-1 mutants , arrowheads point to 6 HTP-3 tracks that can be individually traced along their whole length ( each track represents a pair of aligned homologous chromosomes ) . ( B ) Quantification of total HTP-3 length per late pachytene nucleus in WT controls and wapl-1 mutants . Error bars represent standard deviation , differences are significant ( p<0 . 0001 , t-test ) . ( C ) Projections of late pachytene nuclei stained with HTP-3 antibodies ( axial elements ) and DAPI . The number of HTP-3 tracks appears larger in syp-1 mutants than in wapl-1 syp-1 double mutants , where some short HTP-3 tracks are seen . ( D ) Projections of mid pachytene nuclei stained with anti-HTP-3 antibodies ( axial elements ) , DAPI , and labeled with a probe against the 5S rDNA locus on chromosome V . Most nuclei in syp-1 and wapl-1 syp-1 mutants display 2 5S foci , showing that homologs are not associated ( quantification shown in Figure 4—figure supplement 2 ) . ( E ) Projection of a late pachytene nucleus stained with DAPI , anti-HTP-3 antibodies and anti-HIM-8 antibodies ( binding to a single end of the X chromosome ) . Both HIM-8 signals are located at the end of a short HTP-3 track , each one representing a highly compacted X chromosome . Shortening of X chromosome axial elements in late pachytene nuclei was seen in 3 out of 3 syp-1 wapl-1 germ lines . ( F ) Projections of transition zone nuclei from worms carrying a GFP tag on the endogenous smc-1 gene ( generated by CRISPR ) stained with DAPI , anti-GFP antibodies , and anti-PLK-2 antibodies . Appearance of PLK-2 aggregates on the nuclear envelope marks the onset of meiotic prophase ( leptotene stage ) . In the WT germ line , SMC-1::GFP tracks are only observed in nuclei with PLK-2 aggregates , while pre-leptotene nuclei display diffuse SMC-1::GFP staining ( arrows ) . SMC-1::GFP tracks are present in pre-leptotene nuclei of wapl-1 mutants ( arrows ) . ( G ) Projections of late pachytene and diplotene nuclei from worms carrying a GFP tag on the endogenous smc-1 gene ( generated by CRISPR ) stained with DAPI and anti-GFP antibodies . A large accumulation of nuclear soluble SMC-1::GFP is present in wild-type nuclei , but not in wapl-1 mutants ( see quantification on Figure 4—figure supplement 5 ) . Note that axial elements become elongated , twisted and with a more diffuse appearance in wild-type nuclei compared with wapl-1 mutant nuclei ( insets show magnification of the indicated nuclear region ) . Scale bar = 5 µm in all panels . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 01210 . 7554/eLife . 10851 . 013Figure 4—figure supplement 1 . Examples of HTP-3 tracking in projections of late pachytene nuclei stained with anti-HTP-3 antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 01310 . 7554/eLife . 10851 . 014Figure 4—figure supplement 2 . Quantification of homolog pairing in germ lines of syp-1 and wapl-1 syp-1 double mutants . Germ lines were labeled with a FISH probe against the 5S rDNA locus on chromosome V , or with anti-HIM-8 antibodies to visualize the pairing center end of the X chromosomes . Y axis indicates percentage of nuclei with paired signals and X axis indicates regions along the germ line: zone-1 ( premeiotic nuclei ) , zone 2 ( transition zone ) , zones 3 to 6 ( early to late pachytene ) . In mid- and late-pachytene nuclei ( zones 4-6 ) 5S rDNA signals are separated in most nuclei of both syp-1 and wapl-1 syp-1 double mutants . Number of nuclei per zone: syp-1 ( 98 , 180 , 295 , 293 , 180 , 185 ) , wapl-1 syp-1 ( 77 , 114 , 142 , 173 , 126 , 69 ) . X chromosomes are unpaired in ~50% of late pachytene nuclei in both syp-1 and wapl-1 syp-1 double mutants . Number of nuclei per zone: WT ( 83 , 113 , 157 , 114 , 96 ) , wapl-1 ( 105 , 101 , 121 , 87 , 78 ) , syp-1 ( 143 , 195 , 108 , 149 , 111 ) , wapl-1 syp-1 ( 182 , 158 , 215 , 161 , 149 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 01410 . 7554/eLife . 10851 . 015Figure 4—figure supplement 3 . Shortening of X chromosome axial elements in wapl-1 syp-1 double mutants . Projections of late pachytene nuclei stained with anti-HTP-3 and anti-HIM-8 antibodies , and counterstained with DAPI . Note overall reduction in axis overlap in syp-1 wapl-1 nucleus compared to syp-1 nucleus . Arrowheads in syp-1 wapl-1 panels point to two short stretches of HTP-3 corresponding to the axial elements of the X chromosomes ( they colocalize with HIM-8 signals ) . Axial elements of the X chromosomes can not be easily traced on the syp-1 panel due to extensive overlap with axial elements from other chromosomes . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 01510 . 7554/eLife . 10851 . 016Figure 4—figure supplement 4 . WAPL-1 affects nuclear organization during early meiosis . Projections show nuclei about to enter meiotic prophase ( left of image ) progressing to early pachytene ( right-hand side of image ) stained with anti-GFP antibodies ( to visualize SMC-1::GFP ) and DAPI . Arrows indicate nuclei at preleptotene stages , note that these nuclei contain linear tracks of SMC-1::GFP in wapl-1 mutants but no in WT . Arrowheads in wapl-1 panel point to early pachytene nuclei in which it is possible to distinguish 6 linear structures ( each one representing paired homologous chromosomes ) . Note that in WT nuclei at the same stages there is extensive overlap of SMC-1::GFP tracks , making it difficult to trace individual tracks along their entire length . Scale bar = 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 01610 . 7554/eLife . 10851 . 017Figure 4—figure supplement 5 . WAPL-1 induces accumulation of nuclear soluble SMC-1::GFP in diplotene nuclei . Images are maximum intensity projections from non-deconvolved Z stacks acquired with the same settings and adjusted with the same display scale to allow a direct visual comparison of fluorescence intensities in WT and wapl-1 mutants . Note that in WT nuclei a large fraction of SMC-1::GFP staining decorates nuclear regions lacking chromosomes ( identified by DAPI staining ) , while in wapl-1 diplotene nuclei most SMC-1::GFP signal overlaps with chromosomes . Asterisks in the DNA panel indicate mitotic nuclei of the germ line sheath . Graph shows quantification of SMC-1::GFP average nuclear fluorescence intensity in regions lacking chromosomes ( measured in maximum intensity projections from Z stacks of non-deconvolved diplotene nuclei acquired with the same settings ) . Note that average SMC-1::GFP intensity in chromosome-free regions is much higher in WT than wapl-1 mutants ( p<0 . 0001 , t-test ) . Three areas of 0 . 63 x 0 . 63 µm ( not containing chromosomes ) were measured per nucleus and a total of 10 nuclei from three different germ lines were included in the analysis . Scale bar = 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 01710 . 7554/eLife . 10851 . 018Figure 4—figure supplement 6 . SC disassembly is delayed in wapl-1 mutant germ lines . Projections of diplotene nuclei stained with anti-HTP-1/2 antibodies , anti-SYP-1 antibodies and counterstained with DAPI . SYP-1 is restricted to a very short line per pair of homologs in WT germ lines , while long tracks of SYP-1 are observed in wapl-1 mutants . Note also that coiling of axial elements ( HTP-1/2 staining ) is more evident in WT nuclei compared with wapl-1 mutants . Scale bar = 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 018 The organization of meiotic chromosomes in pachytene nuclei is not only determined by axial elements , but also by the synaptonemal complex ( SC ) , a proteinaceous structure that glues together homologous axial elements to promote stable pairing and inter-homolog recombination ( MacQueen et al . , 2002 ) . In order to understand better the effect of WAPL-1 on the organization of axial elements , we investigated chromosome structure in pachytene nuclei of wapl-1 syp-1 double mutants , since lack of SYP-1 prevents SC assembly without affecting the formation of axial elements ( MacQueen et al . , 2002 ) . Late pachytene nuclei of syp-1 mutant germ lines displayed many long and thin HTP-3 tracks , corresponding to individualized and elongated axial elements ( Figure 4C ) . In contrast , late pachytene nuclei of wapl-1 syp-1 double mutants contained fewer HTP-3 tracks , which also appeared bulkier ( Figure 4C ) . This suggested that axial elements were much shorter , and/or the presence of some kind of SC-independent association between axial elements , in wapl-1 syp-1 double mutants . Visualization of the 5S rDNA locus on chromosome V by FISH showed that this region was unpaired in most pachytene nuclei of both syp-1 and wapl-1 syp-1 double mutants , demonstrating that increased association of homologous axial elements is not responsible for the overall reduction in total axial element length observed in wapl-1 syp-1 mutants ( Figure 4D and Figure 4—figure supplement 2 ) . Moreover , visualization of the X chromosome pairing center region using HIM-8 antibodies ( Phillips et al . , 2005 ) demonstrated that at late pachytene , when this region becomes separated in syp-1 mutants ( MacQueen et al . , 2002 ) ( Figure 4—figure supplement 2 ) , each HIM-8 signal is associated with a short track of axial element in syp-1 wapl-1 double mutants ( Figure 4E and Figure 4—figure supplement 3 ) . Thus , removal of WAPL-1 induces dramatic changes in the overall organization of axial elements in pachytene nuclei . Since cohesin loading during early meiosis is an essential step in the assembly of axial elements ( Severson et al . , 2009; Lightfoot et al . , 2011; Llano et al . , 2012 ) , and Wapl antagonizes stable cohesin binding before S-phase in mitotic cells ( Tedeschi et al . , 2013 ) , we investigated whether WAPL-1 regulates the morphogenesis of axial elements at meiosis onset . In order to visualize all cohesin complexes in the germ line we used CRISPR to add a C-terminal GFP tag on the endogenous smc-1 gene . Homozygous smc-1::GFP worms are viable and healthy and , as expected , the SMC-1::GFP protein is present in all germline and somatic nuclei . To clearly define the onset of meiotic prophase ( leptotene stage ) , we stained germ lines of SMC-1::GFP wild type and wapl-1 mutant worms with PLK-2 antibodies , as PLK-2 forms aggregates on the nuclear envelope of leptotene nuclei to promote homolog pairing ( Labella et al . , 2011 ) . In wild-type germ lines , the presence of elongated SMC-1::GFP structures ( axial elements ) coincided with the appearance of PLK-2 aggregates ( 8 out 8 germ lines ) , while nuclei preceding the formation of PLK-2 aggregates displayed diffuse SMC-1::GFP staining ( Figure 4F and Figure 4—figure supplement 4 ) . In contrast , the appearance of elongated SMC-1::GFP structures preceded the formation of PLK-2 aggregates in transition zone nuclei of wapl-1 mutant germ lines ( 7 out of 8 germ lines , average of 5 nuclei per germ line ) ( Figure 4F ) . Interestingly , transition zone nuclei in which chromosome clustering has not yet occurred display the highest intensity of WAPL-1 staining in the whole germ line ( Figure 1F–G and Figure 1—figure supplement 1 ) , suggesting that high levels of WAPL-1 activity may be present in these nuclei . These observations suggest that WAPL-1 regulates the timing of axial element assembly by antagonizing cohesin association with chromosomes during early meiosis . During late pachytene meiotic chromosomes initiate a major remodeling process that includes the ordered disassembly of the SC , chromosome condensation and changes in axial element composition ( Chan et al . , 2004; Nabeshima et al . , 2005; de Carvalho et al . , 2008; Martinez-Perez et al . , 2008 ) . Whether the early steps of this process involve major changes in the association of cohesin with chromosomes is not known , but imaging of SMC-1::GFP in wild-type germ lines demonstrated that a diffuse pool of nuclear SMC-1::GFP starts accumulating in late pachytene , becoming very prominent in diplotene and diakinesis oocytes ( Figure 4G and Figure 4—figure supplement 5 ) . This staining pattern suggests that there is strong expression of smc-1 during late prophase , but that most of this SMC-1 remains unbound to chromosomes , and/or that cohesin is actively removed from axial elements during these stages . In contrast , a diffuse pool of SMC-1::GFP is not observed in late prophase nuclei of wapl-1 mutants ( Figure 4G and Figure 4—figure supplement 5 ) . Moreover , while axial elements labeled by SMC-1::GFP remained as easily traceable linear structures in diplotene nuclei of wapl-1 mutants , they appeared as longer and more diffuse structures in wild-type oocytes ( Figure 4G ) . This suggests that WAPL-1 antagonizes cohesin binding during late pachytene and diplotene stages . Interestingly , the intensity of nuclear GFP::WAPL-1 undergoes an increase during late pachytene and diplotene compared with earlier prophase ( Figure 1F and Figure 1—figure supplement 1 ) , suggesting that WAPL-1 activity may be increased during these stages . In addition to the increased amount of cohesin associated with diplotene chromosomes , wapl-1 mutants display other defects in the chromosome remodeling process . First , diplotene nuclei in wapl-1 mutants do not show the coiling of axial elements that is characteristic of this stage in wild-type germ lines ( Nabeshima et al . , 2005 ) and instead axial elements visualized by SMC-1::GFP or anti-HTP-1 antibodies display a more linear appearance ( Figure 4G and Figure 4—figure supplement 6 ) . Second , wapl-1 mutants display altered SC disassembly , as evidenced by the persistence of long tracks of SC component SYP-1 in diplotene oocytes ( Figure 4—figure supplement 6 ) . These observations suggest that cohesin removal by WAPL-1 is an intrinsic feature of chromosome remodeling during late pachytene and diplotene . The evidence presented above strongly suggests that wapl-1 mutants undergo meiotic prophase with increased levels of axis-associated cohesin . During C . elegans meiosis cohesin complexes carrying different kleisin subunits , either REC-8 or the highly homologous and functionally redundant COH-3 and COH-4 , associate with axial elements to ensure efficient SCC and crossover formation ( Pasierbek et al . , 2001; Severson et al . , 2009; Severson and Meyer , 2014 ) . Thus , we used anti-REC-8 and anti-COH-3/4 antibodies ( Figure 5—figure supplement 1 ) to determine the staining pattern of the different meiotic cohesin complexes in the presence and absence of WAPL-1 . Upon initial observation , the intensity of REC-8 staining appeared very similar in wapl-1 mutant germ lines and wild-type controls ( Figure 5A ) . However , the COH-3/4 signal was clearly increased in wapl-1 mutants ( Figure 5A ) . A quantitative analysis of the mean nuclear intensity of REC-8 and COH-3/4 staining between leptotene and late pachytene confirmed that wapl-1 mutant germ lines display increased levels of COH-3/4 throughout meiotic prophase , with the highest differences detected in late pachytene nuclei ( Figure 5B ) . The same results were obtained using a COH-3-specific antibody , instead of the COH-3/4 used above , and anti-GFP antibodies to determine the staining of REC-8::GFP expressed from a single copy , and fully functional , transgene ( Figure 5—figure supplements 2–4 ) . In order to confirm that removal of WAPL-1 causes an increase in the levels of axis-associated COH-3/4 , we also performed a line profile analysis of REC-8 , COH-3/4 and SMC-3 intensity across late pachytene nuclei . This analysis allows a direct comparison of the signal intensity associated with axial elements versus the intensity observed in the inter-chromosomal regions of the same nucleus . The average difference between axis-associated and inter-chromosomal signal for COH-3/4 was increased by 114% in wapl-1 mutants compared with wild-type controls , while the increase observed for REC-8 and SMC-3 was 27% and 59% respectively ( Figure 5C ) . The intermediate increase of axis-associated SMC-3 levels is to be expected since SMC-3 should form part of both REC-8 and COH-3/4 cohesin complexes . Separation of soluble and DNA-bound protein fractions from whole-worm extracts confirmed that the amount of COH-4 associated with DNA is increased in wapl-1 mutants compared to wild-type controls ( Figure 5D–E and Figure 5—figure supplement 5 ) . Therefore , in the absence of WAPL-1 there is a large increase in the amount of cohesin associated with axial elements , and most of this increase corresponds to cohesin complexes containing the COH-3/4 kleisins . 10 . 7554/eLife . 10851 . 019Figure 5 . WAPL-1 antagonizes binding of COH-3/4 cohesin to axial elements . ( A ) Projections of pachytene nuclei stained with the indicated antibodies and DAPI . In all cases , WT and wapl-1 examples were acquired with the same exposure settings and images are non-deconvolved projections adjusted with same settings to allow visual comparisons in staining intensity . The intensities of SMC-3 and COH-3/4 are increased in wapl-1 mutants , while REC-8 staining appears similar in WT and wapl-1 nuclei . Scale bar = 5 µm . ( B ) Quantification of mean fluorescence intensity per nucleus of REC-8 , COH-3/4 and SMC-3 in transition zone ( TZ ) , early pachytene ( EP ) and late pachytene ( LP ) nuclei . Between 15 and 20 nuclei per germ line , from a minimum of 5 germ lines were analyzed per genotype and stage . Differences indicated with asterisks are significant ( p<0 . 0001 , t-test ) , error bars= SEM . ( C ) Line profile quantification to compare the intensity of cohesin at axial elements versus inter-chromosome domains . Top left-hand side panel: example of a SMC-3 and DAPI-stained pachytene nucleus showing the intensities of DAPI ( blue ) and SMC-3 ( green ) along the depicted line . The line profile of SMC-3 intensity is shown in the top right-hand panel , ΔF indicates the increment in staining between the peak ( axial element ) and the valley ( inter chromosome domain as determined by lack of DAPI staining ) . Graph: Plotting of individual ΔF values from late pachytene nuclei of wapl-1 mutants and WT stained with REC-8 , COH-3/4 or SMC-3 antibodies . Between 12 and 32 nuclei from different germ lines were analyzed per genotype . Mean value and SEM are indicated . Proportional increase between the mean ΔF value in wapl-1 and WT: 27% for REC-8 ( p= 0 . 0008 , t-test ) , 114% for COH-3/4 ( p<0 . 0001 , t-test ) , 59% for SMC-3 ( p<0 . 0001 , t-test ) . ( D ) Western blots of triton soluble and insoluble ( DNA bound ) protein fractions from WT and wapl-1 mutant worms probed with anti-COH-4 ( see Figure 5—figure supplement 5 for additional controls ) , anti-HAL-2 ( marker for soluble fraction ) , and anti-H3 ( marker for DNA-bound fraction ) antibodies . Asterisk indicates a non-specific band recognized by anti-COH-4 antibodies ( see Figure 5—figure supplement 5 ) . Note that COH-4 signal in WT extracts is higher in the soluble than in the DNA fraction , while in wapl-1 mutant extracts COH-4 intensity is higher in DNA-bound than in the soluble fraction . ( E ) Quantification of relative intensity of anti-COH-4 signal in the soluble and DNA-bound fractions . Three westerns were included in the analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 01910 . 7554/eLife . 10851 . 020Figure 5—figure supplement 1 . Controls showing specificity of anti-COH-3/4 antibodies . Projections of pachytene nuclei stained with the anti-COH-3/4 antibodies and DAPI . Note that signal from anti-COH-3/4 antibodies only disappears in germ lines from coh-3 coh-4 double mutants , demonstrating that the antibodies recognize both COH-3 and COH-4 . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 02010 . 7554/eLife . 10851 . 021Figure 5—figure supplement 2 . A different anti-COH-3 antibody also shows increased staining intensity in pachytene nuclei of wapl-1 mutant germ lines . WT and wapl-1 examples were acquired with the same exposure settings and images are non-deconvolved projections adjusted with same display settings to allow visual comparisons in staining intensity . Chromatin was stained with DAPI ( blue ) . Graph shows quantification of mean fluorescence intensity in pachytene nuclei stained with anti-COH-3 antibodies . Differences between WT and wapl-1 are significant ( p<0 . 0001 , t-test ) , error bars = SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 02110 . 7554/eLife . 10851 . 022Figure 5—figure supplement 3 . A REC-8::GFP transgene shows similar staining intensity in pachytene nuclei of WT and wapl-1 mutant germ lines . WT and wapl-1 examples were acquired with the same exposure settings and images are non-deconvolved projections adjusted with same display settings to allow visual comparisons in staining intensity . Chromatin was stained with DAPI ( blue ) . Graph shows quantification of mean fluorescence intensity in pachytene nuclei stained with anti-GFP antibodies . Differences between WT and wapl-1 are not significant , error bars = SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 02210 . 7554/eLife . 10851 . 023Figure 5—figure supplement 4 . Controls demonstrating the functionality of the REC-8::GFP transgene . Expression of the REC-8::GFP transgene shown in Figure 4—figure supplement 3 rescues the defects in chiasma formation of rec-8 mutants , demonstrating that the REC-8::GFP fusion protein provides full REC-8 function . Scale bar = 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 02310 . 7554/eLife . 10851 . 024Figure 5—figure supplement 5 . Control western blots for anti-COH-4 antibodies . Left-hand side western blot: whole worm extracts from indicated genotypes labeled with anti-COH-4 antibodies . A band of the expected molecular weight for COH-4 is present in WT and coh-3 mutant extracts , but not in coh-4 single or coh-3 coh-4 double mutant extracts , demonstrating that the antibodies specifically recognize COH-4 . Asterisks indicate two non-specific bands recognized by the antibodies as they are present in extracts from coh-4 single and coh-3 coh-4 double mutants . Actin antibodies are used as loading control . Right-hand side western blot: Protein fractionation ( soluble and DNA bound ) from WT and coh-4 mutant extracts labeled with anti-COH-4 antibodies . Note that the band corresponding to COH-4 is only detected in WT ( arrowheads ) , with the signal being stronger in the soluble fraction . Two non-specific bands are detected in the soluble fraction of both WT and coh-4 mutant extracts ( asterisks ) . Anti-HAL-2 antibodies were used as a marker for the soluble protein fraction and anti-histone H3 antibodies were used as a marker for the DNA-bound protein fraction . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 024 Following exit from pachytene , meiotic chromosomes undergo a condensation process that culminates with the formation of highly compacted and individualized chromatin structures seen in diakinesis oocytes , where 6 bivalents ( pairs of homologous chromosomes linked together by chiasmata ) are present in wild-type oocytes ( Figure 6A ) . The presence of between 7 and 12 chromatin masses in diakinesis oocytes indicates defects in the formation of chiasmata , while defects in SCC typically result in oocytes containing more than 12 chromatin bodies ( Figure 6A ) . Although gross loss of SCC in diakinesis oocytes requires the simultaneous depletion of REC-8 , COH-3 and COH-4 ( Severson et al . , 2009; Tzur et al . , 2012; Severson and Meyer , 2014 ) , REC-8 appears to play a more prominent role in mediating SCC than COH-3/4 . For example , removal of REC-8 from spo-11 mutants , which fail to form meiotic DSBs , induces extensive separation of sister chromatids in oocytes , and removal of REC-8 from mutants lacking the SC results in oocytes with separated sisters and extensive chromosome fragmentation ( Figure 6A ) ( Colaiácovo et al . , 2003; Severson et al . , 2009; Severson and Meyer , 2014 ) . In contrast , removing SPO-11 ( Severson and Meyer , 2014 ) or the SC from coh-3 coh-4 double mutants does not compromise SCC in diakinesis oocytes ( Figure 6A ) . These observations , together with the finding that WAPL-1 antagonizes the association of COH-3/4 with axial elements , led us to investigate if the antagonistic effect of WAPL-1 on COH-3/4 may be responsible for the compromised cohesion in mutants lacking REC-8 . We first tested if removing WAPL-1 from spo-11 rec-8 double mutants had an effect on SCC observed in oocytes . Strikingly , the presence of detached sister chromatids was dramatically reduced in the oocytes of wapl-1 spo-11 rec-8 triple mutants , which displayed an average of 11 . 2 DAPI-stained bodies , while diakinesis oocytes of spo-11 rec-8 double mutants displayed an average of 19 . 1 DAPI-stained bodies , demonstrating that WAPL-1 antagonizes cohesion in spo-11 rec-8 oocytes ( Figure 6B–C ) . Crucially , cohesion observed in wapl-1 spo-11 rec-8 oocytes is mediated by COH-3/4 , since sister chromatids were separated in oocytes of wapl-1 spo-11 rec-8 coh-3 coh-4 quintuple mutants ( Figure 6B–C ) . Given that DSBs are required for tethering sister chromatids in rec-8 oocytes ( Severson and Meyer , 2014 ) , we tested if the cohesion rescue observed in wapl-1 spo-11 rec-8 oocytes could be due to the presence of unscheduled DSBs in these triple mutants . However , RAD-51 foci were not observed in germ lines of wapl-1 spo-11 rec-8 triple mutants ( Figure 6—figure supplement 1 ) , suggesting that removal of WAPL-1 increases COH-3/4-mediated cohesion independently of DSBs . Introducing the GFP::wapl-1 transgene into wapl-1 spo-11 rec-8 triple mutants resulted in oocytes with separated sister chromatids ( Figure 6B–C ) , confirming that WAPL-1 antagonizes COH-3/4-mediated cohesion . 10 . 7554/eLife . 10851 . 025Figure 6 . WAPL-1 antagonizes COH-3/4 cohesion in diakinesis oocytes . ( A ) Projections of diakinesis oocytes of the indicated genotypes stained with DAPI . 6 DAPI-stained bodies ( WT ) indicates presence of 6 bivalents , 12 DAPI-stained bodies indicates absence of chiasmata , and the presence of more than 12 DAPI-stained bodies indicates separation of sister chromatids , with 24 demonstrating separation of all sisters . The average number of DAPI-stained bodies observed in each genotype is indicated on the bottom right of each panel , number of nuclei analyzed: WT ( 20 ) , rec-8 ( 29 ) , coh-3 coh-4 ( 16 ) , rec-8 coh-3 coh-4 ( 17 ) , rec-8 spo-11 ( 94 ) , coh-3 coh-4 spo-11 ( 26 ) , and coh-3 coh-4 syp-2 ( 26 ) . Arrowheads in rec-8 syp-1 panel point to chromosome fragments . Note that removal of SPO-11 or SYP-1/2 causes loss of cohesion in rec-8 mutants but not in coh-3 coh-4 double mutants . ( B ) Projections of diakinesis oocytes of the indicated genotypes stained with DAPI , quantification shown in ( C ) . Note the reduction of DAPI-stained bodies in rec-8 spo-11 wapl-1 compared with three other genotypes ( p<0 . 0001 in all cases , t-test ) . Error bars = standard deviation . Number of nuclei analyzed: rec-8 spo-11 ( 94 ) , rec-8 spo-11 wapl-1 ( 85 ) , rec-8 spo-11 wapl-1 coh-3 coh-4 ( 29 ) , rec-8 spo-11 wapl-1GFP::wapl-1 ( 27 ) . ( D ) Automated quantification ( CellProfiler ) of area sizes corresponding to chromatin bodies in projections of diakinesis oocytes of indicated genotypes stained with DAPI . Values on the X axis represent area in pixels and binning of the different categories was adjusted using oocytes of known phenotypes: WT ( bivalents ) , coh-3 coh-4 ( univalents ) spo-11 rec-8 ( detached sisters ) . Number of oocytes analyzed: WT ( 40 ) , coh-3 coh-4 ( 61 ) spo-11 rec-8 ( 116 ) . ( E ) Automated quantification of area sizes corresponding to chromatin bodies in projections of diakinesis oocytes , note that removing WAPL-1 from rec-8 syp-1 double mutants causes a large decrease of chromosome fragments and an increase in univalents . Number of nuclei analyzed: rec-8 syp-1 ( 52 ) , rec-8 syp-1 wapl-1 ( 36 ) . ( F ) Example of DAPI-stained oocyte from rec-8 syp-1 wapl-1 demonstrating absence of chromosome fragments , compare with rec-8 syp-1 example shown in A . ( G ) Projections of diakinesis oocytes stained with DAPI , note that chromosome fragments are present in rec-8 coh-4 ( arrowheads ) but not in rec-8 coh-4 wapl-1 oocytes . ( H ) Automated quantification of area sizes corresponding to chromatin bodies in projections of diakinesis oocytes , note that removing WAPL-1 from rec-8 coh-4 double mutants causes a large decrease of chromosome fragments . 26 diakinesis oocytes were analyzed for both rec-8 coh-4 and rec-8 coh-4 wapl-1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 02510 . 7554/eLife . 10851 . 026Figure 6—figure supplement 1 . Control demonstrating that unscheduled DSBs are not formed in wapl-1 spo-11 rec-8 triple mutants . Projections of pachytene nuclei from wapl-1 spo-11 rec-8 triple mutants stained with anti-RAD-51 antibodies and DAPI , note the absence of RAD-51 foci ( compare with RAD-51 staining in the wapl-1 panel shown in Figure 2B ) . Scale bar = 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 02610 . 7554/eLife . 10851 . 027Figure 6—figure supplement 2 . Examples of automated area analysis using CellProfiler . Images show projections of diakinesis oocytes of the indicated genotypes stained with DAPI . Red lines represent the boundaries of chromatin bodies as determined by CellProfiler . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 027 Next , we tested if WAPL-1 contributes to the chromosome fragmentation observed in oocytes of rec-8 syp-1 double mutants . The extensive presence of small chromosome fragments with irregular shapes in rec-8 syp-1 oocytes made manual quantification of the number of DAPI-stained bodies impractical . Thus , we used CellProfiler to determine the boundaries and area ( in pixels ) of each DAPI-stained body in projections of diakinesis oocytes . By imaging diakinesis oocytes from wild type , coh-3 coh-4 double mutants and rec-8 spo-11 double mutants we were able to calibrate the area range corresponding to bivalents , univalents and sister chromatids respectively . Despite the fact that some overlap of chromosomes occurs in projections of diakinesis oocytes , this automated method clearly identifies the predominant types of chromatin bodies known to be present in each of the three genotypes ( Figure 6D and Figure 6—figure supplement 2 ) . Based on this analysis , chromosome fragments were defined by an area smaller than 120 pixels and 49% of chromatin bodies in syp-1 rec-8 oocytes fell within this category ( Figure 6E ) . In contrast , chromosome fragments only accounted for 10% of chromatin bodies in wapl-1 rec-8 syp-1 oocytes ( Figure 6E–F ) . This reduction in chromosome fragments was accompanied by a large increase in the number of chromatin bodies corresponding to univalents . Thus , WAPL-1 is largely responsible for the chromosome fragmentation and sister separation seen in rec-8 syp-1 oocytes , likely by antagonizing COH-3/4 cohesin . We further tested this possibility by building a coh-4 rec-8 double mutant , in which we hypothesized that the presence of a single meiotic kleisin ( COH-3 ) should result in a worsening of the cohesion defects observed in rec-8 single mutants . In agreement with this , 25% of chromatin bodies in coh-4 rec-8 oocytes had an area corresponding to chromosome fragments , while this number was reduced to 8% in wapl-1 coh-4 rec-8 oocytes ( Figure 6G–H ) . These observations demonstrate that WAPL-1 antagonizes SCC mediated by COH-3/4 both in the presence and absence of SPO-11 DSBs . Sister chromatids remain attached in the diakinesis oocytes of rec-8 mutants , but the contact between them is often limited to a very small region , giving rec-8 univalents a bilobed appearance that contrasts with the rounded appearance of univalents observed in recombination-deficient mutants such as spo-11 ( Figure 6A ) ( Severson et al . , 2009; Collette et al . , 2011 ) . Why sisters are weakly attached in rec-8 oocytes remains unknown , but these attachments require SPO-11 DSBs ( Severson et al . , 2009 ) , leading to the proposal that DSBs induce phosphorylation of COH-3/4 , which in turn make these complexes cohesive in a pathway similar to the damage-induced cohesion observed in mitotic yeast cells ( Heidinger-Pauli et al . , 2008; Severson and Meyer , 2014 ) . However , we considered the possibility that the formation of inter-sister recombination events during the repair of SPO-11 DSBs leads to the formation of sister attachments in rec-8 oocytes . To test this hypothesis we built worms lacking REC-8 and COSA-1 , a protein required for the late stages of crossover formation but not for the induction of DSBs ( Yokoo et al . , 2012 ) . Strikingly , oocytes from rec-8 cosa-1 double mutants display a dramatic increase in the number of DAPI-stained bodies ( average 23 . 89 ) compared with rec-8 ( average 13 . 06 ) and cosa-1 ( average 11 . 83 ) single mutants , suggesting extensive separation of sister chromatids ( Figure 7A–B ) . FISH experiments confirmed extensive separation of sister chromatids in diakinesis oocytes of rec-8 cosa-1 double mutants , while also demonstrating that SCC is not affected in pachytene nuclei of these same mutants ( Figure 7C–D ) . Thus , loss of cohesion must occur during the chromosome remodeling process that starts at late pachytene . These results suggest that attachments between sister chromatids in diakinesis oocytes of rec-8 mutants require the presence of a recombination event formed by the COSA-1-dependent crossover pathway . Moreover , similar to the situation in spo-11 rec-8 oocytes , removing WAPL-1 from rec-8 cosa-1 mutants restored cohesion in diakinesis oocytes ( Figure 7A–B ) , confirming that WAPL-1 antagonizes cohesion mediated by COH-3/4 during late prophase . 10 . 7554/eLife . 10851 . 028Figure 7 . Tethering of sister chromatids in rec-8 oocytes requires crossover precursors . ( A ) Projections of diakinesis oocytes stained with DAPI . Sisters are detached in rec-8 cosa-1 oocytes , but not in rec-8 cosa-1 wapl-1 oocytes . ( B ) Quantification of the number of DAPI-stained bodies in diakinesis oocytes of indicated genotype , note significant increase of DAPI-stained bodies in rec-8 cosa-1 compared with three other genotypes ( p<0 . 0001 in all cases , t-test ) . Error bars= standard deviation . Number of nuclei analyzed: cosa-1 ( 65 ) , rec-8 ( 29 ) , rec-8 cosa-1 ( 39 ) , rec-8 cosa-1 wapl-1 ( 42 ) . ( C ) Projection of pachytene nuclei labeled by FISH with 5S rDNA probe and stained with DAPI . The presence of two signals per nucleus ( one per homolog ) indicates that sister chromatids are not separated . ( D ) Projection of a diakinesis oocyte labeled by FISH with 5S rDNA probe and stained with DAPI . The four signals indicate separation of sister chromatids . Scale bar in all panels = 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 028 Imaging of germ lines from wild-type worms expressing SMC-1::GFP demonstrated that the soluble pool of SMC-1 that starts accumulating at late pachytene persists during diakinesis , and that the oocyte about to be ovulated ( -1 oocyte ) displayed a reduction in the intensity of chromosome-associated cohesin compared to the -2 oocyte of the same germ line ( Figure 8A ) . Since Wapl is required for the prophase pathway that removes cohesin before metaphase in mitotic cells ( Haarhuis et al . , 2014 ) and a similar pathway operates during meiosis in plants ( De et al . , 2014 ) , we tested whether WAPL-1 is required for cohesin removal in late diakinesis oocytes . Surprisingly , a clear reduction in chromosome-associated cohesin occurs in late diakinesis oocytes of wapl-1 mutant germ lines ( Figure 8A ) . In most cases ( 21 out of 26 analyzed germ lines ) this reduction is evident between the -2 and -1 oocytes , while in a few germ lines the reduction was more evident between the -3 and -2 oocytes ( Figure 8C and Figure 8—figure supplement 1 ) . Similar to what we observed in diplotene , diakinesis oocytes of wapl-1 mutants lack the accumulation of nuclear soluble SMC-1 observed in WT oocytes ( Figure 8A ) . We also imaged diakinesis oocytes using super resolution structural illumination microscopy , which allowed us to observe in greater detail changes in chromosome-associated cohesin . These experiments also demonstrated that a reduction in chromosome bound SMC-1::GFP occurs in late diakinesis oocytes of both wild-type and wapl-1 mutant germ lines ( Figure 8B ) . Thus , a WAPL-1-independent wave of cohesin removal occurs during late diakinesis . 10 . 7554/eLife . 10851 . 029Figure 8 . Cohesin is removed by a WAPL-1-independent mechanism in late diakinesis oocytes . Projections of diakinesis oocytes from worms expressing SMC-1::GFP ( tagged by CRISPR ) stained with anti-GFP antibodies and DAPI , and imaged with a Delta Vision system ( A ) or a structured illumination microscope ( B ) . WT oocytes accumulate a large amount of nuclear soluble SMC-1::GFP that is lacking in wapl-1 mutant oocytes , but a reduction in chromosome-associated SMC-1::GFP occurs in -1 oocytes of both WT and wapl-1 mutants . -1 and -2 oocytes shown for each genotype were part of the same germ line and were acquired on the same image . Scale bars = 5 µm . ( C ) Table showing analysis of reduction in chromosome-associated SMC-1::GFP staining in late diakinesis oocytes of WT and wapl-1 mutant worms . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 02910 . 7554/eLife . 10851 . 030Figure 8—figure supplement 1 . WAPL-1-independent reduction in SMC-1::GFP staining during late meiotic prophase . Projection of diakinesis oocytes from wapl-1 mutants expressing SMC-1::GFP and stained with anti-GFP antibodies and DAPI . The numbers at the top of the panel indicate the position of oocytes within the germ line ( -1 oocyte is about to be ovulated and enter the first meiotic division ) . Note the decrease in chromosome-associated SMC-1::GFP signal between oocytes -3 and -2 . Scale bar = 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 10851 . 030 Similar to mitotic prophase chromosomes , meiotic chromosomes are organized as linear arrays of sister-chromatid loops that are attached at their base to a proteinaceous axial element containing cohesin ( Kleckner , 2006; Liang et al . , 2015 ) . Under this organization , the shape of chromosomes is largely determined by the size of chromatin loops and by the compaction exerted by axial elements on loop bases , two parameters that must be regulated by proteins localizing to axial elements . In fact , the meiosis-specific subunit Smc1β , which localizes to axial elements , regulates the organization of chromatin loops and the length of axial elements in mouse pachytene oocytes ( Novak et al . , 2008 ) . This suggests that a key role of meiosis-specific cohesin complexes may be to modulate the higher-order organization of chromosomes . Our studies provide strong support for this hypothesis . First , the formation of SMC-1-labelled axial elements is observed in nuclei before the onset of meiotic prophase ( leptotene ) in wapl-1 mutants , but not in wild-type germ lines . Interestingly , the intensity of GFP::WAPL-1 is at its highest in nuclei preceding leptotene , suggesting that WAPL-1 antagonizes cohesin binding from the onset of meiosis and that the regulation of WAPL-1 levels , and presumably its activity , may be coupled to meiotic progression . Second , in pachytene nuclei removal of WAPL-1 causes a dramatic increase in the levels of chromosome-associated COH-3/4 cohesin and shortening of axial elements , consistent with COH-3/4 complexes acting as regulators of axis compaction . Shortening of axial elements is also observed in yeast cells lacking wpl1 ( A . Shinohara personal communication ) , consistent with our findings . Interestingly , we observed that the shortening of axial elements caused by the absence of WAPL-1 is exacerbated in mutants lacking the SC , hinting that the SC is somehow capable of antagonizing cohesin’s ability to induce compaction of axial elements . In contrast to COH-3/4 , REC-8 levels remain similar to wild-type controls throughout meiotic prophase of wapl-1 mutant germ lines . The specific effect of WAPL-1 on COH-3/4 complexes is consistent with the finding that REC-8 and COH-3/4 display clear differences in their timing and mechanism of loading during early prophase ( Severson and Meyer , 2014 ) , and may help to explain some of the functional differences between REC-8 and COH-3/4 cohesin . For example , the fact that REC-8 complexes are refractory to the removal activity of WAPL-1 is consistent with these complexes playing a prominent role in providing cohesion , while COH-3/4 promote SC assembly in a cohesion-independent fashion ( Severson and Meyer , 2014 ) . Similarly , mouse Rad21L promotes homolog pairing without playing a direct role in cohesion , presumably by regulating higher-order structure of chromosomes during early prophase ( Ishiguro et al . , 2014 ) . A clear precedent for cohesin affecting chromosome topology independently of cohesion is observed when Wapl is removed from mammalian cells before S-phase , which causes unscheduled condensation of unreplicated chromosomes ( Tedeschi et al . , 2013 ) . We propose that by antagonizing the stable association of COH-3/4 cohesin to axial elements , WAPL-1 acts as a major regulator of meiotic chromosome structure . The finding that WAPL-1 has a strong antagonistic effect on the chromosomal association of COH-3/4 , but not REC-8 , suggests that some property of REC-8 cohesin must make these complexes refractory to the removal activity of WAPL-1 . In mitotic cells , Wapl is thought to mediate cohesin removal by triggering opening of an 'exit gate' present at the interface delimited by the interaction between the N-terminal region of Smc3 and the Scc1 kleisin ( Chan et al . , 2012; Gligoris et al . , 2014; Huis in 't Veld et al . , 2014 ) . Opening of this interface by Wapl is antagonized by acetylation of two conserved lysines in Smc3 , a process that is mediated by the acetyl transferase Eco1 ( Ben-Shahar et al . , 2008; Unal et al . , 2008; Zhang et al . , 2008 ) . Under unchallenged growth conditions , Smc3 acetylation only occurs during DNA replication , since Eco1 is degraded after S-phase ( Lyons and Morgan , 2011 ) . Thus , cohesin that associates with chromosomes during G2 is not acetylated and shows a rapid turnover mediated by Wapl ( Kueng et al . , 2006 ) . Interestingly , REC-8 is the only meiosis-specific kleisin subunit that appears to be loaded to chromosomes during S-phase in mice and worms , while Rad21L ( mice ) and COH-3/4 ( worms ) are loaded post S-phase ( Ishiguro et al . , 2014; Severson and Meyer , 2014 ) . Thus , a plausible explanation for our observations is that only REC-8 forms part of cohesin complexes in which SMC-3 is acetylated , while COH-3/4 associate with non-acetylated SMC-3 and therefore are sensitive to WAPL-1 . However , mechanisms that protect cohesin from the removal activity of Wapl independently of Smc3 acetylation have been described in yeast and human mitotic cells . For example , DSBs induce the establishment of Wpl1-resistant cohesion during G2 by a mechanism involving Eco1-mediated acetylation of Scc1 ( Heidinger-Pauli et al . , 2009 ) . Interestingly , depletion of Eco1 after meiotic S-phase induces chromosome segregation defects in Drosophila oocytes , suggesting that Eco1 is functional during meiotic prophase , although the targets of Eco1 in this case remain unknown ( Weng et al . , 2014 ) . In addition to acetylation by Eco1 , the presence of DNA damage also induces Scc1 sumoylation , which is required for establishment of cohesion and sister chromatid-mediated homologous repair during G2 ( McAleenan et al . , 2012; Wu et al . , 2012 ) . If posttranslational modifications on REC-8 , or another subunit of REC-8 complexes , are required to antagonize the releasing activity of WAPL-1 , untimely loss of these modifications could lead to premature release of REC-8 cohesin , compromising chromosome segregation during the meiotic divisions . Elucidating the molecular mechanisms that prevent removal of REC-8 complexes by WAPL-1 during meiotic prophase remains an important goal for future studies . Our results indicate that cohesin removal by WAPL-1 is an important aspect of the chromosome remodeling process that starts during late pachytene and that involves SC disassembly and changes in chromosome condensation . In organisms as diverse as Sordaria and mouse ( oocytes ) , visualization of cohesin during late prophase demonstrates that the emergence of compacted chromosomes attached by chiasmata is preceded by the large disappearance of axial elements as continuous linear structures , and by the accumulation of diffuse cohesin staining in the nucleus ( Prieto et al . , 2004; Storlazzi et al . , 2008 ) . We report that WAPL-1 promotes loosening of axial element structure and accumulation of soluble cohesin during late prophase in C . elegans , suggesting that a Wapl-dependent weakening of cohesin-mediated chromosome organization may be a general feature of late prophase chromosome remodeling . Importantly , this wave of WAPL-1-mediated cohesin removal has a direct impact on the ability of COH-3/4 complexes to provide cohesion following the completion of chromosome remodeling . In the absence of REC-8 , attachment of sister chromatids in diakinesis oocytes requires COH-3/4 cohesin and SPO-11 ( Severson et al . , 2009 ) , but since sister chromatids are attached in pachytene nuclei of spo-11 rec-8 double mutants ( Severson and Meyer , 2014 ) , loss of cohesion must occur during the process of chromosome remodeling via an unknown mechanism . By showing that removal of WAPL-1 causes a rescue of cohesion in oocytes of spo-11 rec-8 double mutants and that this rescue requires COH-3/4 cohesin , we clearly identify WAPL-1 as a factor that antagonizes COH-3/4-mediated cohesion during late prophase . Our investigation of the mechanisms that mediate cohesion in diakinesis oocytes of rec-8 mutants has uncovered an important role for recombination in the modulation of cohesion during late meiotic prophase . The requirement of DSBs for cohesion in diakinesis oocytes of rec-8 mutants has been explained by the existence of a mechanism similar to the break-induced cohesion observed in mitotic yeast cells ( Heidinger-Pauli et al . , 2008; Severson and Meyer , 2014 ) . Under this model , DSBs trigger establishment of COH-3/4 cohesion via a mechanism involving CHK-2-mediated phosphorylation of COH-3/4 . However , we show that removal of COSA-1 , a protein that localizes to chromosomes during late pachytene and that is not required for the formation and early processing of DSBs ( Yokoo et al . , 2012 ) , causes loss of cohesion in diakinesis oocytes of rec-8 mutants . Therefore , the presence of DSBs alone is not sufficient to induce COH-3/4 cohesion that persists in diakinesis oocytes . Instead , tethering of sister chromatids in rec-8 mutant oocytes appears to require the presence of crossover-fated recombination events . This requirement would explain the lack of cohesion in diakinesis oocytes of spo-11 rec-8 , which fail to initiate recombination , and in oocytes of syp-1 rec-8 double mutants , since SYP-1 is required for crossover formation ( MacQueen et al . , 2002 ) . Interestingly , SC proteins are thought to load between sister chromatids in mouse and worm mutants lacking REC-8 ( Xu et al . , 2005; Rog and Dernburg , 2013; Severson and Meyer , 2014 ) , suggesting that SC-promoted recombination intermediates may form between sister chromatids in rec-8 mutants . In agreement with this possibility , crossover designation appears to occur normally in Sordaria rec8 mutants , which also display SC formation between sister chromatids ( Storlazzi et al . , 2008 ) . Since sister chromatids are not separated in pachytene nuclei of spo-11 rec-8 , syp-1 rec-8 ( Severson and Meyer , 2014 ) or cosa-1 rec-8 ( this study ) , and removal of WAPL-1 from these three mutants restores cohesion in diakinesis oocytes , the antagonistic effect of WAPL-1 on COH-3/4 cohesion must occur after pachytene exit . Our studies identify chromosome remodeling as a key stage in the modulation of cohesion during meiotic prophase , and suggest that a complex interplay between late crossover precursors and WAPL-1 plays an important role in this process . An essential aspect of meiotic chromosome segregation is ensuring that chiasmata remain intact until the onset of anaphase I . This issue is particularly relevant to human fertility , as oocytes establish chiasmata during fetal development , but then arrest at the dictyate stage , with the first meiotic division only taking place at ovulation , up to several decades later . In fact , cohesion deterioration in arrested oocytes has been proposed as a contributing factor to explain why the incidence of aneuploidy increases dramatically with maternal age ( Nagaoka et al . , 2012; Herbert et al . , 2015 ) . Data from different mouse models offer support for this possibility . First , partial loss of cohesin before metaphase I in oocytes of mice lacking Smc1β ( a meiosis-specific Smc1 subunit ) causes the dissolution of chiasmata ( Hodges et al . , 2005 ) . Second , Rec8-mediated cohesion is not regenerated during late stages of meiotic prophase ( Tachibana-Konwalski et al . , 2010 ) and Smc1β expressed before birth is sufficient to ensure chiasma maintenance throughout adult age ( Revenkova et al . , 2010 ) . Third , age-related deterioration of bivalents is observed in oocytes from Smc1β mutant mice ( Revenkova et al . , 2010 ) , as well as in oocytes from old ( long lived ) wild-type mice that also display defects in chromosome segregation and depletion of cohesin from chromosomes before anaphase onset ( Lister et al . , 2010 ) . These observations are consistent with the view that cohesin deterioration could be an important contributor to human aneuploidy . However , the mechanisms responsible for cohesin depletion in aged oocytes are not known , with possible explanations including decay of cohesin molecules over time and leaky separase activity ( Jessberger , 2012 ) . We have found that two different mechanisms actively remove cohesin during late prophase as part of normal meiotic progression: First , WAPL-1 antagonizes cohesin during chromosome remodeling and second , a WAPL-1-independent wave of cohesin removal occurs during oocyte maturation . These findings could be relevant to human aneuploidy , as excessive cohesin removal by the WAPL-1-dependent or independent mechanisms identified here could compromise maintenance of chiasmata before the onset of anaphase I , inducing errors in chromosome segregation during the meiotic divisions . All strains were maintained at 20°C and the N2 strain was used as the wild-type control . The following mutant alleles were used: wapl-1 ( tm1814 ) , syp-1 ( me17 ) , rec-8 ( ok978 ) , coh-3 ( gk112 ) , coh-4 ( tm1857 ) , spo-11 ( ok79 ) , cosa-1 ( tm3298 ) , syp-2 ( ok307 ) . The following transgenes were used: meIs8[unc-119 ( + ) pie-1promoter::GFP::cosa-1] ( Yokoo et al . , 2012 ) , [unc-119 ( + ) pie-1promoter::zhp-3::GFP] ( Jantsch et al . , 2004 ) , itIs37[unc-119 ( + ) pie-1promoter::mcherry::his-58] . Transgenic strains carrying single copy insertions of the GFP::wapl-1 and rec-8::GFP transgenes in the ttTi5605 locus on chromosome II were created following the protocol described in ( Frøkjær-Jensen et al . , 2008 ) . The GFP::wapl-1 transgene carried 494 bp upstream of the starting codon , a GFP cDNA containing 3 artificial introns and the entire sequence of the wapl-1 locus plus 1505 bp of downstream sequence . The rec-8::GFP:: transgene carried 536 bp upstream of the starting codon , the entire sequence of the rec-8 locus , a GFP cDNA containing 3 artificial introns , and 665 bp of downstream sequence . Tagging of the endogenous smc-1 locus by CRISPR was performed with an sgRNA targeting the 'GTTGCAATCGATGGTGTTGG' sequence at the 3’ end of smc-1 and a repair template containing GFP cDNA with 3 artificial introns flanked by 1400 bp of upstream and downstream sequence from the site of the DSB . Expression of sgRNA and Cas9 was performed using the protocols described in ( Friedland et al . , 2013 ) . Germ lines from young adult hermaphrodites were dissected , fixed and processed for immunostaining and FISH as described in ( Martinez-Perez and Villeneuve , 2005 ) . All images were acquired using a Delta Vision Deconvolution system equipped with an Olympus 1X70 microscope . Primary antibodies: rabbit anti-COH-3/4 were generated against residues 286-338 of the COH-3 protein; rabbit anti-HTP-1/2 were generated against residues 37-93 of HTP-1; chicken anti-SYP-1 were generated against a peptide including the first 23 residues of SYP-1; rabbit anti-HTP-3 ( Goodyer et al . , 2008 ) ; rabbit anti-RAD-51 , anti-HIM-8 , anti-REC-8 and anti-COH-3 antibodies were all purchased from Novus Biologicals; rabbit anti-SMC-3 ( Millipore AB3914 ) ; rabbit anti-GFP conjugated to Alexa488 ( Invitrogen ) . Mean whole nuclear fluorescence was quantified with an ImageJ macro written by D . Dormann and K . Hng . Quantification was performed on unprocessed raw images acquired as three-dimensional stacks using identical exposure settings in a Delta Vision system . Briefly , an oval defining the area of a nucleus was drawn and the mean fluorescence intensity within that area was quantified on each Z section and automatically averaged across the entire stack . Fluorescence intensity line profiles were calculated on maximum intensity projections from unprocessed raw images acquired using identical exposure settings in a Delta Vision system using the SoftWoRx Line Profile tool . Lines were drawn to intersect with at least three axial elements and peak to trough fluorescence intensity ranges ( ΔF ) were calculated for all peaks within a nucleus using DAPI signal as the reference for the position of chromosomes . Images of DAPI-stained -1 and -2 diakinesis oocytes were acquired using a Delta Vision system equipped with an Olympus 1X70 microscope . Deconvolved image stacks were either used directly for visual counting , or were converted into maximum intensity projections , cropped to the size of an individual nucleus using ImageJ and analysed in CellProfiler to define the boundaries of each DAPI-stained body and to calculate its area in pixels . Following analysis of oocytes from control genotypes ( Figure 5D ) , the following areas were used to classify DAPI-stained bodies: bivalents 500–1000 pixels , univalents 250–500 pixels , sister chromatids 120–250 pixels . DAPI-stained bodies with an area smaller than 120 pixels were counted as chromosome fragments . Immunostaining was performed as described above , but slides were mounted using ProLong Gold mounting media instead of vectashield . Images were acquired using a Zeiss Elyra microscope . 24 hrs post L4 hermaphrodites were dissected to release embryos in a drop of 60% v/v Leibowitz-15 media , 20% fetal bovine serum , 25 mM HEPES pH 7 . 4 , 5 mg/ml inulin . Embryos were then mounted on 2% agarose pads and imaged with a Delta Vision system equipped with an Olympus 1X70 microscope . Images of the meiotic divisions were acquired as series of 1 µm-spaced Z stacks ( 9–12 section ) with a regular time lapse of 5 s intervals . Videos of these time series were created using SoftWoRx . 100 young hermaphrodites were collected in 1X TE ( 10 mM Tris-HCl pH 8 , 1 mM EDTA ) supplied with complete protease inhibitor ( Roche ) and freeze-thawed three times in liquid nitrogen . Worms were then resuspended in 40 µl of 1X Laemli buffer and boiled for 10 min . Extracts were run on a 10% acrylamide gel and transferred on nitrocellulose for 1 hr at 4°C , blocked for 1 hour in 5% milk TBST ( 1x TBS 0 . 1% Tween ) and then incubated with anti-WAPL-1 ( 1:3000 ) and goat anti-actin ( Santa Cruz , 1:3000 ) antibodies . Anti-WAPL-1 antibodies were generated against residues 2-101 of the WAPL-1 protein . In order to separate soluble and DNA-bound protein fractions from whole-worm extracts we followed the methods described in ( Silva et al . , 2014 ) . Briefly , 150 young hermaphrodite worms were picked into a 1 . 5 ml tube containing 40 µl of extraction buffer ( 10 mM Tris-HCl pH 7 . 5 , 1 mM EDTA , 2X Complete Protease Inhibitor and 2X Phospho-STOP ) , snap-frozen in liquid nitrogen , thawed , and grinded with a plastic pestle . Triton-X was added to a final concentration of 0 . 25% and tubes were placed on a thermomixer at 4°C in mild shaking ( 750 rpm ) for 20 min . Tubes were then centrifuged for 10 min at 14 , 800 rpm at 4°C and the supernatant from this step , which represents the soluble fraction , was collected into a new tube and centrifuged once more under the same conditions to remove any debris . The remaining pellet , containing the non-soluble fraction , was resuspended with 40 µl of extraction buffer and washed twice before being resuspended in a final volume of 40 µl of extraction buffer . Laemmli buffer was added to a 1X final concentration and equal volumes of the protein extracts were run on a 7 . 5% acrylamide gel and transferred on nitrocellulose for 1 hr at 4°C , blocked for 1 hr in 5% milk TBST ( 1x TBS 0 . 1% Tween ) and then incubated with the following primary antibodies over night at 4°C: rabbit anti-COH-4 ( 1:1000 ) ( antibodies raised against residues 289–341 of the COH-4 protein ) , rabbit anti-HAL-2 ( 1:10 , 000 ) ( Zhang et al . , 2012 ) , and rabbit anti-Histone H3 ( 1:10 , 000 , AbCam ) . The following HRP-conjugated secondary antibodies were used in 5% milk in TBST for one hour at room temperature: donkey anti-goat ( 1:8000 , Sigma ) and goat anti-rabbit ( 1:5000 , Millipore ) . Quantification of band intensities was performed using ImageJ software .	Most of the genetic material of plant and animal cells is stored in structures called chromosomes . Nearly all the cells in the body contain two copies of each chromosome , one inherited from the mother and the other from the father , but sex cells – such as egg and sperm – contain just one copy of each . If eggs or sperm contain the wrong number of copies of a chromosome , genetic disorders such as Down syndrome can occur . New sex cells form in a process called meiosis , which begins with a cell that contains two copies of each chromosome duplicating each of these copies . The duplicated copies are known as sister chromatids , and are held together by a ring-like protein complex called cohesin . In addition to tethering sister chromatids , cohesin affects the ‘higher-order’ organization of chromosome structure and promotes the recruitment of other proteins that are essential for different aspects of chromosome behavior during meiosis . Therefore , regulating cohesin binding during meiosis is key to ensuring that sex cells contain the correct number of chromosomes . Cohesin is ultimately removed from chromosomes in two steps during the consecutive cell divisions at the end of meiosis , resulting in the formation of sex cells containing a single copy of each chromosome . However , whether cohesin is actively removed from chromosomes during early meiosis , when chromosomes undergo dramatic structural changes , is not known . Using a combination of microscopy and genetic techniques to study the developing egg cells of the worm Caenorhabditis elegans , Crawley et al . investigated how a protein called WAPL-1 affects cohesin binding to chromosomes during early meiosis . This revealed that WAPL-1’s effects depend on the identity of a particular subunit of the cohesin complex . If this subunit is a protein called COH-3 or COH-4 , WAPL-1 reduces the ability of cohesin to bind to chromosomes during the early stages of meiosis . However , WAPL-1 does not affect cohesin complexes that instead feature a protein called REC-8 as this subunit . By preventing excessive binding of COH-3 and COH-4 cohesin , WAPL-1 regulates chromosome structure and sister chromatid cohesion during early meiosis . Crawley et al . further observed that during the stage preceding the first meiotic division , cohesin is removed from chromosomes by a mechanism that does not involve WAPL-1 . The next challenge is to work out why cohesin containing the REC-8 protein is protected from being released by WAPL-1 . Whether defects in this protection can trigger the premature separation of sister chromatids is also an important question to answer .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression" ]	2016	Cohesin-interacting protein WAPL-1 regulates meiotic chromosome structure and cohesion by antagonizing specific cohesin complexes
Gene-environment interactions impact the development of neuropsychiatric disorders , but the relative contributions are unclear . Here , we identify gut microbiota as sufficient to induce depressive-like behaviors in genetically distinct mouse strains . Daily gavage of vehicle ( dH2O ) in nonobese diabetic ( NOD ) mice induced a social avoidance behavior that was not observed in C57BL/6 mice . This was not observed in NOD animals with depleted microbiota via oral administration of antibiotics . Transfer of intestinal microbiota , including members of the Clostridiales , Lachnospiraceae and Ruminococcaceae , from vehicle-gavaged NOD donors to microbiota-depleted C57BL/6 recipients was sufficient to induce social avoidance and change gene expression and myelination in the prefrontal cortex . Metabolomic analysis identified increased cresol levels in these mice , and exposure of cultured oligodendrocytes to this metabolite prevented myelin gene expression and differentiation . Our results thus demonstrate that the gut microbiota modifies the synthesis of key metabolites affecting gene expression in the prefrontal cortex , thereby modulating social behavior . Despite the diffuse prevalence of mental illness and the large efforts spent in identifying genetic elements of susceptibility , there is a need to define the role of environment—gene interactions . In addition to genetic predisposition , there is extensive epidemiologic literature emphasizing the role of environmental exposure in the development of mild to severe mood disorders . The aftermath of traumatic life events , for instance , is often characterized by the onset of severe depression or post-traumatic stress disorder ( Shalev et al . , 1998 ) . The interplay between genes and environmental variables has gained recent attention , and several immunologic and lifestyle contributors have been proposed to modulate depressive symptoms . The detection of high levels of serum cytokines and the higher incidence of depression in patients with autoimmune disorders ( Postal and Appenzeller , 2015; Walker et al . , 2011; Moll et al . , 2011; van Hees et al . , 2015; Feinstein et al . , 2014 ) has suggested a role for neuroinflammation ( Godbout et al . , 2008; Menard et al . , 2016; Audet et al . , 2014 ) . Deficiency of specific nutrients such as omega-3 fatty acids has been reported in subsets of patients with mental illnesses ( Ohara , 2005; Patrick and Ames , 2015; Poudel-Tandukar et al . , 2009; Panagiotakos et al . , 2010 ) , highlighting the link between mood disorders and the bioavailability of metabolites . There is evidence that bioactive metabolites act as mediators of gut—brain communication , as shifts in gut microbial composition impact brain neurochemistry ( Cryan and Dinan , 2012; Collins et al . , 2012; Desbonnet et al . , 2014; Bercik et al . , 2010; 2011 ) . Indeed , psychiatric comorbidities often accompany conditions characterized by an aberrant gut microbiota composition , such as irritable bowel syndrome , functional gastrointestinal disorder , and inflammatory bowel disease ( Gevers et al . , 2014; Morgan et al . , 2012; Haberman et al . , 2014; Carroll et al . , 2011; Addolorato et al . , 1997 ) . Conversely , altered gut microbiota composition and function have been reported in patients with major depressive disorders and children with autism ( Jiang et al . , 2015; De Angelis et al . , 2015; De Angelis et al . , 2013; Parracho et al . , 2005 ) . The gut microbiota is a complex microbial ecosystem that rapidly responds to environmental changes and can modulate brain development , function , and behavior ( Cryan and Dinan , 2012; Collins et al . , 2012; Desbonnet et al . , 2014; Bercik et al . , 2010; 2011; Wu et al . , 2011; Daniel et al . , 2014; Lax et al . , 2014 ) . These studies suggest that social behavior may be affected by abnormal interactions between gut microbiota and the brain , though the underlying mechanisms remain only partially understood . One hypothesis for the pathogenesis of depressive-like behaviors has been suggested through studies on social isolation in mice ( Liu et al . , 2012; 2016; Makinodan et al . , 2012 ) , which revealed a reduction of myelinated fibers in the prefrontal cortex ( PFC ) , associated with changes in the oligodendrocyte transcriptome ( Liu et al . , 2012; 2016 ) . Myelination is a dynamic process that continues into adulthood and contributes to physiologic brain function ( Liu et al . , 2012; 2016; Makinodan et al . , 2012; Sánchez et al . , 1998; Gibson et al . , 2014; McKenzie et al . , 2014 ) . Oligodendrocytes are the myelinating cells of the central nervous system ( CNS ) , and neuropathologic and transcriptomic studies have reported downregulated oligodendroglial transcripts and reduced myelin thickness in the brains of patients with schizophrenia , major depression , and bipolar disorder ( Tkachev et al . , 2003; Aston et al . , 2005; Katsel et al . , 2005 ) . These data highlight the role of myelin in mental illness and depressive-like behaviors , though it remains to be established whether myelination in the adult PFC and social behavior are affected by alterations in gut microbiota composition . This study characterizes the gut microbiota in mice with social avoidance behavior and demonstrates that transfer of specific bacterial taxa is sufficient to alter adult PFC myelination and results in behavioral changes consistent with a depressive-like phenotype . Although gastric gavage and subcutaneous injections are routine , daily procedures used to administer drugs or special diet to rodents , the potential behavioral effects they may induce in mice have not been investigated . Daily gastric gavage with vehicle for two weeks ( Figure 1A ) was sufficient to induce social avoidance behavior in NOD mice ( Figure 1B ) , without affecting their overall locomotor activity ( Figure 1C ) . This depressive-like behavior induced by daily gavage was dependent on the specific mouse strain , as C57BL/6 mice were not affected ( Figure 1D , E ) ( Moy et al . , 2008 ) . Subcutaneous injection of vehicle did not elicit any behavioral effect in either strain ( Figure 1—figure supplement 1 ) . Daily gastric gavage with an antibiotic cocktail proven to deplete the gut microbiota ( Reikvam et al . , 2011 ) failed to induce the social avoidance behavior in NOD mice ( Figure 1B ) , and similarly had no effect on the C57BL/6 mice ( Figure 1D ) . The antibiotic regimen was well tolerated by both NOD and C57BL/6 mice , did not impact body weight or glucose levels , and did not result in any gastric hemorrhage or visible stomach damage ( Figure 1—figure supplement 2 and Figure 1—figure supplement 3 ) . Consistent with previous reports , only chronic oral antibiotic treatment ( but not subcutaneous delivery ) induced enlargement of the large intestine ( Figure 1—figure supplement 2 ) , a macroscopic sign associated with microbiota depletion ( Reikvam et al . , 2011 ) . Interestingly , daily gavage also induced an anxiety-like behavior in both NOD and C57BL/6 mice , as revealed by the elevated plus maze ( EPM ) ( Figure 2B , D ) . However , the anxiety-like behavioral change displayed in response to daily gavage was not affected by oral antibiotic treatment ( Figure 2B , D ) , suggesting that only the depressive-like behavior is mediated by alterations in gut microbiota . To further validate this hypothesis , we conducted the forced swim test ( FST ) , which is considered a measure of despair-like behavior , in NOD and C57BL/6 mice after daily gavage with either vehicle or antibiotics . The despair-like behavior was induced by vehicle gavage in the NOD strain , and was prevented by oral antibiotic treatment ( Figure 2C ) , but was not detected in the C57BL/6 mice ( Figure 2E ) . Together , these results indicate that daily gavage of vehicle induces social avoidance and despair-like behaviors in NOD mice , but not in C57BL/6 mice , and that this effect is not observed when gavaging antibiotics orally and not subcutaneously . 10 . 7554/eLife . 13442 . 003Figure 1 . The strain-specific social avoidance behavioral response to daily gavage is affected by oral antibiotic treatment . ( A ) Experimental timeline: vehicle or antibiotic mix were administered daily by gastric gavage ( g . g . ) for 14 days . Behavioral testing was performed before ( baseline ) and after treatment . ( B–D ) Results of the Social Interaction ( SI ) test for NOD ( B ) and C57BL/6 ( D ) mice . Oral antibiotic treatment did not affect locomotor activity measured during the social interaction test ( C , E ) ( 3 independent experiments with 8 mice per group/experiment for a total of n=23–24 mice per condition ) . Data are mean ± S . E . M; p<0 . 05 , p<0 . 01 based on one-way ANOVA with Bonferroni’s post hoc test; n . s . indicates not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 13442 . 00310 . 7554/eLife . 13442 . 004Figure 1—figure supplement 1 . The subcutaneous delivery of vehicle or antibiotic did not induce social avoidance behavior . ( A ) Experimental timeline: vehicle or antibiotic mix were administered daily ( s . c . ) for 14 days . Behavioral testing was performed before ( baseline ) and after treatment . ( B–C ) . Results of the Social Interaction ( SI ) test for NOD ( B ) and C57BL/6 ( C ) mice . ( D–E ) Locomotor activity measured during the Social interaction test ( 2 independent experiments with 10 mice per group/experiment for a total of n=20 mice per condition ) . Data are mean ± S . E . M; p<0 . 05 , **p<0 . 001 based on one-way ANOVA with Bonferroni’s post hoc test; n . s . indicates not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 13442 . 00410 . 7554/eLife . 13442 . 005Figure 1—figure supplement 2 . Effect of subcutaneous or oral antibiotic treatment on body weight and macroscopic appearance of large intestine . ( A ) Experimental timeline . ( B , E ) Representative pictures of the intestine from NOD and C57BL/6 mice treated with vehicle or antibiotic ( subcutaneous [s . c . ] or oral administration [g . g . ] ) ; scale bar: 1 cm . Graphs represent the gut weight relative to the mouse total body weight . ( C , D , F , G ) Body weight monitoring in NOD ( C , D ) and C57BL/6 ( F , G ) mice ( n=10 per group ) . Data are mean ± S . E . M; p<0 . 05 , p<0 . 01 , p<0 . 001 based on one-way ANOVA followed by Bonferroni’s post hoc test . n . s . indicates not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 13442 . 00510 . 7554/eLife . 13442 . 006Figure 1—figure supplement 3 . Oral antibiotic treatment is well tolerated by recipients . ( A ) Representative pictures of stomachs from C57BL/6 mice treated with vehicle or antibiotic . ( B ) Blood glucose levels were measured after 14 days of oral treatment ( antibiotic or vehicle ) ( n=6 per group ) . Normoglycemic levels were considered below 220 mg/dL . DOI: http://dx . doi . org/10 . 7554/eLife . 13442 . 00610 . 7554/eLife . 13442 . 007Figure 2 . The strain-specific anxiety- and despair-like behavioral responses to daily gavage are differentially affected by oral antibiotic treatment . ( A ) Experimental timeline: vehicle or antibiotic mix were administered daily by gastric gavage ( g . g . ) for 14 days . Behavioral testing was performed before ( baseline ) and after treatment . Figure shows the results for the Elevated Plus Maze ( EPM ) and Forced Swim Test ( FST ) for NOD ( B , C ) and C57BL/6 ( D , E ) mice after oral treatment ( g . g . ) . Baseline measurements for FST were performed in a separate cohort of mice ( n=10 ) to avoid carryover effects ( 3 independent experiments with 8 mice per group/experiment for a total of n=24 mice per condition ) . Data are mean ± S . E . M; p<0 . 05 , p<0 . 01 , p<0 . 001 based on one-way ANOVA followed by Bonferroni’s post hoc test; n . s . indicates not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 13442 . 00710 . 7554/eLife . 13442 . 008Figure 2—figure supplement 1 . Anxiety and despair-like behaviors after subcutaneous ( s . c . ) vehicle or antibiotic treatment . ( A ) Experimental timeline: vehicle or antibiotic mix were administered daily ( s . c . ) for 14 days . Behavioral testing was performed before ( baseline ) and after treatment . Figure shows the results for Elevated plus maze ( EPM ) and Forced Swim Test ( FST ) for NOD ( B–C ) and C57BL/6 ( D–E ) mice after s . c . treatment . Baseline measurements for FST were performed in a separate cohort of mice ( n=10 ) to avoid carryover effects of the FST ( 2 independent experiments with 10 mice per group/experiment for a total of n=20 mice per condition ) . Data are mean ± S . E . M; p<0 . 05 , *p<0 . 001 based on one-way ANOVA followed by Bonferroni’s post hoc test; n . s . indicates not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 13442 . 008 To further characterize the effect of vehicle and antibiotic treatment on gut microbiota composition , we conducted 16S rRNA sequencing analysis of cecal and fecal samples collected after behavioral testing and after 14 days of treatment ( Figure 3A ) . Unweighted UniFrac distances ( Lozupone and Knight , 2005 ) were calculated between all pairs of fecal samples based on their microbiota composition . Based on these distances , Principal coordinate analysis ( PCoA ) , an ordination method conceptually similar to principal component analysis , revealed a clear separation between vehicle-gavaged and baseline NOD ( Figure 3 ) and between vehicle and antibiotic treated NOD and C57BL/6 mice ( Figure 3—figure supplement 1A ) . PCoA analysis revealed clear differences between NOD mice before ( “baseline” ) and after oral treatment with antibiotics ( Figure 3B ) , with differences also observed between samples before and after treatment with vehicle ( Figure 3C ) . Since the depressive-like behavior was only observed in oral vehicle-treated NOD mice , we focused on identifying the specific microbiota that differ in these animals before and after treatment . Analysis of Operational Taxonomic Units ( OTUs , defined as groups of 16S rRNA gene sequences with high similarity and that broadly correspond to a bacterial species ) identified several taxonomic groups that were exclusively found in the vehicle-treated mice ( Figure 3D and Gacias et al . , 2016 ) . These taxa represent potential candidates associated with the depressive-like phenotype observed in NOD mice . Linear discriminant analysis effect size ( LEfSe ) ( Segata et al . , 2011 ) , a biomarker discovery method based on the Kruskal–Wallis and Wilcoxon tests , was used to identify key bacterial taxa enriched in vehicle-treated versus antibiotic-treated animals in each strain ( Figure 3—figure supplement 1B–E ) . As expected , Proteobacteria were enriched in antibiotic-treated animals , while vehicle-treated mice had enrichment in Bacteroidetes and Firmicutes ( Figure 3—figure supplement 1C , E ) . 10 . 7554/eLife . 13442 . 009Figure 3 . Enrichment of bacterial OTUs induced by gastric gavage ( g . g . ) in NOD mice . ( A ) Experimental timeline indicating time points of fecal collection ( arrows ) relative to behavioral testing and treatment . ( B , C ) Principal coordinate analysis plots of unweighted UniFrac distances of microbiota in fecal samples at baseline and after 14 days of daily g . g . of antibiotics or vehicle in NOD mice . Each dot represents the microbiota of a sample ( 1 sample = feces pooled from 3–5 mice ) , color-coded by treatment ( vehicle or antibiotic ) and time-point . The percentage of variation explained by each principal coordinate ( PC ) is shown in parentheses . All samples were rarefied at 5000 sequences . ( D ) Analysis of unique Operational Taxonomic Units ( OTUs ) present in NOD vehicle-treated mice compared to their fecal microbiota at baseline . Figure shows representative taxa enriched in fecal samples of NOD vehicle-treated mice compared to their baseline samples . Each bar represents the microbiota of an individual sample ( 1 sample = 3–5 mice per cage ) . See Gacias et al . ( 2016 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13442 . 00910 . 7554/eLife . 13442 . 010Figure 3—figure supplement 1 . Oral antibiotic treatment effectively modifies the microbiota composition in NOD and C57BL/6 mice . ( A ) Principal coordinate analysis plots of unweighted UniFrac distances of microbiota from the fecal samples after 14 days of daily gastric gavage ( g . g . ) in NOD ( left ) and C57BL/6 ( right ) mice . Each dot represents the microbiota of a sample ( 1 sample = feces pooled from 3–5 mice ) , colored and shaped by treatment ( vehicle or antibiotic ) . The percentage of variation explained by each principal coordinate ( PC ) is shown in parentheses . All samples were rarefied at 5000 sequences . ( B , D ) Cladogram generated from LEfSe analysis showing the most abundant taxa enriched in antibiotic- ( green ) or vehicle-treated ( red ) NOD ( B ) and C57BL/6 ( D ) mice . ( C , E ) Linear discriminant analysis ( LDA ) scores of the differentially abundant taxa in fecal pellets after oral antibiotic treatment compared to vehicle for NOD ( C ) and C57BL/6 ( E ) mice . Graphs show taxa-enriched microbiota from mice treated with antibiotic ( green ) or vehicle ( red ) with a positive or negative LDA score , respectively ( significant taxa [p<0 . 05 , Kruskal–Wallis] with LDA score >2 are shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13442 . 010 Analysis of tissue samples revealed similar differences between vehicle- and antibiotic-treated mice in both strains ( p<0 . 01 , adonis with 999 permutations ) . No significant changes in gut microbiota composition were detected when antibiotics were administered subcutaneously . To identify possible CNS transcriptional signatures associated with the behavioral outcomes described in the vehicle-gavaged , but not antibiotic-gavaged , NOD mice , we performed an unbiased transcriptomic analysis of the medial prefrontal cortex ( mPFC ) using RNA sequencing . This analysis revealed decreased expression of genes related to myelination ( Figure 4 and Figure 4—figure supplements 1 , 2 and Gacias et al . , 2016 ) in vehicle-gavaged NOD mice - characterized by social avoidance behavior - compared to antibiotic-treated mice , whose behavior was comparable to baseline controls ( Figure 1 ) . The differences in myelin gene transcripts in the mPFC of vehicle-gavaged NOD compared to antibiotic-treated mice were validated by quantitative real-time qPCR ( Figure 4B ) and immunohistochemistry ( Figure 4C ) . These differences were detected only in NOD mice , and not in C57BL/6 mice that showed no change in social behavior with oral gavage ( Figure 4D , E ) . The differences in myelin gene expression in the mPFC could not be attributed to a nonspecific effect of antibiotic treatment , as there were no differences observed after subcutaneous delivery ( Figure 4—figure supplement 1 ) . The regional specificity of the transcriptional changes was also assessed in NOD mice by evaluating samples from a distinct brain region , the nucleus accumbens ( NAc ) , revealing no difference in the two treatment groups ( Figure 4—figure supplement 1 ) . These data provide further support for the relationship between defective mPFC adult myelination and depressive-like behavior , as indicated by the lower levels of myelin transcripts and reduced area of MBP immunostaining in vehicle-gavaged NOD mice exhibiting social avoidance . The results also demonstrate that the transcriptional and behavioral effects were prevented by oral antibiotic treatment . 10 . 7554/eLife . 13442 . 011Figure 4 . Myelin transcripts and myelinated fibers in the medial prefrontal cortex ( mPFC ) of adult NOD mice with social avoidance behavior . ( A ) Experimental timeline: vehicle or antibiotic mix were administered daily by gastric gavage ( g . g . ) for 14 days and mPFC was harvested for RNA extraction and quantitative real-time qPCR or immunohistochemsitry ( B , D ) qPCR of myelin transcripts after vehicle or antibiotic treatment of NOD ( B ) and C57BL/6 ( D ) mice . Values were normalized to 36b4 mRNA levels and are referred as fold change relative to vehicle-treated values ( n=6 mice per group ) . ( C , E ) Representative confocal images and quantification of MBP+ fibers ( red ) in mPFC of NOD ( C ) and C57BL/6 ( E ) mice after vehicle or antibiotic treatment . DAPI ( blue ) was used as nuclear counterstain . Scale bar , 50 μm . Graph represents quantification of MBP+ fibers per surface area ( n=3 for NOD; n=4 for C57BL/6 ) . Data are mean ± S . E . M; p<0 . 01 , **p<0 . 001 based on unpaired t test . n . s . indicates not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 13442 . 01110 . 7554/eLife . 13442 . 012Figure 4—figure supplement 1 . Regional specificity of myelin changes in response to antibiotic treatment . ( A ) Experimental timeline: vehicle or antibiotic mix were administered daily by gastric gavage ( g . g . ) for 14 days and nucleus accumbens ( NAc ) was harvested for RNA extraction and quantitative real-time qPCR or immunohistochemsitry . ( B ) qPCR of myelin transcripts in the NAc after oral treatment in NOD mice . Values were normalized to 36b4 mRNA levels and are referred as fold change relative to vehicle-treated values ( n=6 mice per group ) . ( C ) Representative confocal images and quantification of MBP+ myelinated fibers ( red ) in the NAc of NOD mice after vehicle or antibiotic treatment . DAPI ( blue ) was used as nuclear counterstain . Scale bar , 20 μm . Graph represents quantification of MBP+ fibers per surface area ( n=3 mice per group ) . ( D ) Experimental timeline of subcutaneous treatment ( vehicle or antibiotic ) . ( E ) qPCR of myelin transcripts in the mPFC after 14 days of subcutaneous treatment ( s . c . ) . Values were normalized to 36b4 mRNA levels and referred as fold change relative to vehicle-treated values ( n=6 mice per group ) . Data are mean ± S . E . M; statistical differences were determined using unpaired t tests . n . s . indicates not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 13442 . 01210 . 7554/eLife . 13442 . 013Figure 4—figure supplement 2 . Effect of oral antibiotic treatment on the transcriptional profile in medial prefrontal cortex ( mPFC ) . Unbiased genome-wide transcriptomic analysis of mPFC was performed after 14 days of either oral antibiotic or vehicle treatment ( NOD and C57BL/6; n=2 mice per group ) . ( A ) Experimental timeline: vehicle or antibiotic mix were administered daily by gastric gavage ( g . g . ) for 14 days . ( B ) Venn diagram representing up- and downregulated genes after antibiotic treatment in both mouse strains . ( C ) Graph shows the results of DAVID gene ontology analysis using uniquely differentially expressed genes between vehicle and antibiotic treated NOD mice . ( D ) qPCR validation of the transcriptional changes detected by RNA-sequencing . Values were normalized to 36b4 mRNA levels and are referred as fold change relative to vehicle-treated values ( n=6 mice per group ) . Data are mean ± S . E . M; p<0 . 05 , p<0 . 01 , p<0 . 001 based on unpaired t test . n . s . indicates not significant . See Gacias et al . ( 2016 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13442 . 013 To determine whether the social avoidance behavior and mPFC transcriptional changes induced by daily gavage of vehicle in NOD mice were caused by the enrichment of specific gut bacteria , we transferred the cecal content of vehicle-treated or antibiotic-treated NOD mice into C57BL/6 recipients , whose endogenous flora had been depleted by antibiotic treatment ( Figure 5A ) . Social behavior in C57BL/6 depleted recipients was assessed before and after transplantation with microbiota from either vehicle-gavaged ( Group I ) or antibiotic-gavaged ( Group II ) NOD donors . The behavior of the C57BL/6 recipients resembled that of the donors: Social avoidance behavior was detected in Group I recipients , and was not observed in Group II recipients ( Figure 5B , C ) . Intriguingly , transplantation of vehicle-gavaged NOD microbiota also transferred the transcriptional changes in the mPFC , but not in the NAc , as shown by the lower levels of myelin gene transcripts ( Mag , Mog , Plp1 , Mobp ) in Group I mice compared to Group II recipients ( Figure 5D ) . The functional consequences of the transcriptional changes in myelin genes were further validated by electron microscopy , and ultrastructural analysis revealed decreased myelin thickness in Group I recipients displaying a social avoidance behavior ( Figure 5E ) . Quantification of myelin thickness relative to axonal diameter ( g ratio ) revealed that Group I recipients transplanted with vehicle-gavaged NOD microbiota , presented thinner myelin than Group II , recipients of antibiotic-treated NOD donors . No significant differences between the two groups were observed in the NAc ( Figure 5E ) . The transfer of depressive-like behavior from donor to recipient was further validated by the detection of increased immobility at the FST in Group I mice compared to Group II ( Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 13442 . 014Figure 5 . Social avoidance behavior transfer from NOD donors to microbiota depleted C57BL/6 by fecal transplantation . ( A ) Experimental timeline for donor ( NOD ) and transplant-recipient ( C57BL/6 ) mice . ( B , C ) Results from Social Interaction ( SI ) tests conducted in C57BL/6 recipients before and after transplantation with either microbiota from vehicle-treated ( Group I; B ) or antibiotic-treated ( Group II , C ) NOD mice . Graphs represent the amount of time spent ( seconds ) in the interaction zone when a target is present . Red dashed bar represents the interaction time of the NOD donors . Data are mean ± S . E . M; p<0 . 05 , *p<0 . 01 based on a two-way ANOVA ( n=12 mice per experiment , 2 replicates of 12 for a total of 24 mice per condition ) . ( D ) Graphs indicate the relative levels of myelin gene transcripts in mPFC and NAc of C57BL/6 recipients displaying ( Group I ) or not displaying ( Group II ) social avoidance behavior after transplantation with NOD microbiota ( n=6–8 mice per group; p<0 . 05 , p<0 . 01 , p<0 . 001 based on unpaired t test ) . ( E ) Electron micrographs and quantified g-ratios of myelinated axons in mPFC and NAc in Group I and Group II C57BL/6 recipients after transplantation with the NOD microbiota . Scale bar , 1 μm . ( n=3 per treatment and condition; statistical differences between groups were determined using two-tailed t-test; n . s . indicates not significant ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13442 . 01410 . 7554/eLife . 13442 . 015Figure 5—figure supplement 1 . Effect of NOD vehicle-treated microbiota on the despair-like behavior of C57BL/6 recipients . ( A ) Experimental timeline for donor ( NOD ) and transplanted recipient mice ( C57BL/6 ) . Despair-like behavior in colonized C57BL/6 mice was tested after transplantation . ( B ) Effect of NOD cecal microbiota transfer on despair-like behavior measured as immobility time in the Forced Swim Test ( FST ) after transplantation . Data are mean ± S . E . M; *p<0 . 001 based on a unpaired t-test ( n=12 mice per experiment , 2 replicates of 12; total of 24 mice per condition ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13442 . 015 Collectively , these findings suggest that the gut microbiota of vehicle-gavaged NOD donors was sufficient to transfer the depressive-like behavior , modulate transcript levels in the mPFC , and impact region-specific adult myelination in microbiota-depleted C57BL/6 recipients . The genomic DNA content was measured in fecal pellets of C57BL/6 recipients to validate the depletion of the gut microbiota with 14 days of antibiotic treatment , and to evaluate the effectiveness of recolonization after transplantation ( Figure 6B , D ) . Analysis of alpha diversity ( the number of bacterial taxa present in a sample or group of samples ) further confirmed the microbiota depletion ( Figure 6C , E ) . In both groups , diversity was significantly reduced from baseline after antibiotic treatment ( Figure 6C , E; p<0 . 01 ANOVA with Tukey’s honest significant difference ( HSD ) post-hoc analysis ) . As expected , after transplantation Group II mice still exhibited a significantly depleted diversity compared to baseline ( Figure 6E; p<0 . 01 ANOVA with Tukey’s HSD ) , while bacterial diversity in Group I had recovered to levels similar to baseline and was not significantly different ( Figure 6C; p=0 . 09 , ANOVA with Tukey’s HSD ) . These results suggested that transfer of behavioral traits was associated with restoration of bacterial diversity to baseline levels . In order to determine the differences in microbiota compositions associated with the behavioral phenotype , we conducted PCoA analysis based on unweighted UniFrac analysis ( Figure 6F ) . Although all pooled fecal samples from NOD donors and C57BL/6 recipients clustered together at baseline ( samples on the right side of the plot ) , treatment with antibiotics resulted in a drastic reshaping of the bacterial communities of both NOD ( middle of the plot ) and C57BL/6 ( bottom-left side ) mice . The microbiota composition of Group II mice after transplant ( which did not display social avoidance behavior ) was distinct from baseline , similar to antibiotic-treated animals pre-transplant ( top-left side ) . However , Group I recipients which displayed social avoidance behavior ( #19 and #18 on the plot ) , had compositions that were close to those of their vehicle-treated NOD donors . In contrast , Group I recipients which did not display social avoidance behavior ( #17 on the plot ) , clustered with Group II recipients . This result suggests that the transplant procedure was not equally effective in all Group I mice . The distance in microbiota composition between vehicle-gavaged donors and recipients was significantly correlated with the social avoidance behavior , as measured by social interaction time ( Figure 6G; p=0 . 01 ) . This result suggests that the ability to successfully transfer the gut microbiota from vehicle-gavaged NOD donors was significantly correlated with the transmission of the depressive-like behavior . LEfSe analysis revealed a number of taxa that were significantly different between Group I and Group II C57BL/6 recipients ( Figure 6—figure supplement 1 and Gacias et al . , 2016 ) . We further refined this analysis , by identifying the specific OTUs transferred from vehicle-gavaged NOD donors to Group I recipients ( Figure 6—figure supplement 2 and Gacias et al . , 2016 ) . Members of the Clostridiales order , including Lachnospiraceae and Ruminococcaceae , were among those present in equal proportions both in the donors and the recipients in Group I recipients displaying a depressive-like behavior ( i . e . samples #18 and #19 ) , while absent in Group I recipients that did not exhibit such behavior ( i . e . sample#17; Figure 6F and Gacias et al . , 2016 ) . We further confirmed these taxa as potentially responsible for this phenotype by qPCR using primers specific to these bacterial groups ( Figure 6—figure supplement 2C ) . In order to identify differences undetectable at the OTU level , we performed oligotype analysis in those OTUs established as potentially responsible for the depressive-like behavior ( Segata et al . , 2011; Eren et al . , 2014 ) . Oligotype analysis is an entropy-based method to identify single nucleotide differences in sequences from closely related organisms . We found that most OTUs were composed of a single high-abundance oligotype ( Figure 6—figure supplement 3A–C , E , G–O ) and therefore support the conclusions from the OTU-level analysis . However , we identified three OTUs that had two oligotypes with similar abundances and distribution across samples: OTU 183849 ( Blautia producta , a member of the Lachnospiraceae , Figure 6—figure supplement 3D ) , 188840 ( unidentified member within Lachnospiraceae , Figure 6—figure supplement 3F ) , and 4418586 ( unidentified member within Clostridiales , Figure 6—figure supplement 3P ) . Additional inspection of these sequences revealed the oligotypes GTT and TTT from the Blautia producta OTU , as well as the TG and TT oligotypes from the Lachnospiraceae OTU , had B . producta JCM 1471 as the closest reference sequence in NCBI; the oligotypes from Clostridiales had no close reference sequence . Overall , these results show that either a single oligotype or a combination of two oligotypes with similar abundance distributions were dominant within the analyzed OTUs , which suggested they might drive the observed social phenotypes . 10 . 7554/eLife . 13442 . 016Figure 6 . Effect of fecal transplantation on bacterial mass and biodiversity in microbiota depleted C57BL/6 recipients . ( A ) Experimental timeline for donors ( NOD ) and transplanted recipients ( C57BL/6 ) . ( B , C ) Graphs represent fecal biomass ( µg of gDNA relative to total fecal weight ) of C57BL/6 recipients prior to transplantation ( #1 before and #2 after 14 days of antibiotic treatment ) and at end point after-transplantation ( #3 ) with donor microbiota ( n=3 pooled samples per time-point , each sample represents 1 sample = pooled feces from 3–5 mice . Data are mean ± S . E . M; p<0 . 05 , *p<0 . 01 based on one-way ANOVA with Bonferroni’s post hoc test ) . ( C , E ) Rarefaction curves comparing alpha diversity of fecal microbiota samples from C57BL/6 recipients at different experimental time-points ( #1 , #2 , and #3 ) . ( F ) Principal coordinate analysis plot of unweighted UniFrac distances of fecal samples from NOD donors and C57BL/6 mice at different time-points . ( #1 , #2 , and #3 ) . Each dot represents the microbiota of a sample , colored by group , treatment , and time-point ( n=3 pooled samples per time-point; each sample corresponds to pooled feces from 3–5 mice ) . The percentage of variation explained by each principal coordinate ( PC ) is shown in parentheses . ( E ) Relationship between social interaction time and unweighted UniFrac distance to NOD donor mice ( n=3 ) for all C57BL/6 recipients ( n=10 ) . Each point represents a single C57BL/6 animal , colored by group ( light blue: Group_I , transplanted with NOD-vehicle microbiota; pink: Group_II , transplanted with NOD-antibiotic microbiota ) . Linear regression analysis indicates a significant correlation ( p=0 . 0103 ) between the variables . DOI: http://dx . doi . org/10 . 7554/eLife . 13442 . 01610 . 7554/eLife . 13442 . 017Figure 6—figure supplement 1 . Transfer of social avoidance behavior is associated with altered colonic composition of the microbiota . ( A ) Cladogram generated from LEfSe analysis showing the most differentially abundant taxa enriched in C57BL/6 recipients with ( Group I , red ) or without ( Group II , green ) social avoidance behavior . ( B ) Linear discriminant analysis ( LDA ) scores of the differentially abundant taxa in cecal tissue from C57BL/6 recipients with ( Group I , red ) or without ( Group II , green ) social avoidance behavior . Graphs shows taxa enriched with a positive or negative LDA score ( significant taxa [p<0 . 05 , Kruskal–Wallis] with LDA score >2 are shown ) ( n=10–12 samples per group ) . See Gacias et al . ( 2016 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13442 . 01710 . 7554/eLife . 13442 . 018Figure 6—figure supplement 2 . Social avoidance behavior is associated with enrichment of specific OTUs . ( A ) Schematic representation of microbiota and Operational Taxonomic Unit ( OUT ) analysis . ( B ) Relative abundance of the OTUs enriched in mice with social avoidance behavior ( vehicle-treated NOD donors ) and C57BL/6 Group I recipients ( samples #18–19 ) . Note that the sample #17 was from mice without the behavioral phenotype . See Gacias et al . ( 2016 ) . ( C ) Quantitative real-time PCR analysis of genomic DNA extracted from gut tissue of C57BL/6 mice transplanted with microbiota from vehicle- or antibiotic-treated NOD mice ( Group I and Group II , respectively ) to quantify total bacteria of the order Clostridiales , and the families of Lachnospiraceae and Ruminococcaceae ( n=6 mice per group ) . Data are mean ± S . E . M; p<0 . 05 based on unpaired t test; n . s . indicates not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 13442 . 01810 . 7554/eLife . 13442 . 019Figure 6—figure supplement 3 . Oligotype analysis of gut tissue samples . Each panel represents the counts per sample for different oligotypes , identified by nucleotide sequence , within a specific Operational Taxonomic Unit ( OTU ) , named by its Greengenes 13–8 identifier . Individual samples are represented in x axis and colored by group; Black: NOD_vehicle donors ( NODv ) , Red: NOD_antibiotic donord ( NODa ) ; Gray: C57BL/6 Group I ( transplanted with NOD vehicle-treated microbiota ) ; light red: C57BL/6 Group II ( transplanted with NOD antibiotic-treated microbiota ) ( A ) OTU 167509 g__Oscillospira; s__ ( B ) OTU 176118 g__Oscillospira; s__ ( C ) OTU 179657 f__Lachnospiraceae; g__; s__ ( D ) OTU 183849 g__Blautia; s__producta ( E ) OTU 187223 g__Ruminococcus; s__ ( F ) OTU 188840 f__Lachnospiraceae; g__; s__ ( G ) OTU 234121 o__Clostridiales; f__; g__; s__ ( H ) OTU 259006 o__Clostridiales; f__; g__; s__ ( I ) OTU 263337 g__Oscillospira; s__ ( J ) OTU 267689 f__Ruminococcaceae; g__; s__ ( K ) OTU 661055 o__Clostridiales; f__; g__; s__ ( L ) OTU 1571092 o__Clostridiales; f__; g__; s__ ( M ) OTU 3694603 f__Lachnospiraceae; g__; s__ ( N ) OTU 4008606 f__Lachnospiraceae; g__; s__ ( O ) OTU 4390755 g__Anaeroplasma; s__ ( P ) OTU 4418586 o__Clostridiales; f__; g__; s__ . DOI: http://dx . doi . org/10 . 7554/eLife . 13442 . 019 The gut metabolome is altered in microbiota-transplanted C57BL/6 mice displaying altered social and despair-like behaviors . Several studies have demonstrated that gut metabolites can impact the homeostatic host-microbiota interactions and affect behavior ( Daniel et al . , 2014; Hsiao et al . , 2013 ) . To determine whether altered taxa in the gut microbiota could impact the levels of metabolites , which in turn drive behavioral and transcriptional changes observed in the mPFC , we performed an unbiased metabolomic analysis of gut tissue from C57BL/6 recipients with ( Group I ) and without ( Group II ) social avoidance behavior ( Figure 7 ) . The analysis included non-targeted and targeted protocols and gas chromatography combined with time-of-flight high-resolution mass spectrometry , hydrophilic liquid chromatography coupled with high-resolution mass spectrometry and hydrophilic interaction chromatography with liquid chromatography and tandem mass-spectrometry for the study of monoamine to neurotransmitters ( Tolstikov et al . , 2014; Danaceau et al . , 2012 ) . After statistical corrections and normalization , we conducted Partial Least Squares-Discriminant Analysis ( PLS-DA ) , a method that incorporates elements from principal component analysis , regression , and linear discriminant analysis , which revealed a clear separation of the overall gut metabolites between Group I and Group II ( Figure 7B ) . A total of 382 metabolites were detected in the guts of C57BL/6 transplant recipients ( Gacias et al . , 2016 ) A first pathway impact analysis provided a visual representation of the most dramatically affected pathways between the two groups , and identified the linoleic/linolenic acid and phenylalanine/tryptophan synthetic pathways as differentially represented in the two sets of samples ( Figure 7C , Table 1 and Gacias et al . , 2016 ) . Further evaluation of the metabolome using a volcano plot representing individual differences in metabolites revealed increased levels of cresol , stearamide , N-acetylasparagine , and oleamide in Group I recipients , which displayed social avoidance and despair-like behaviors ( Figure 7D , E ) . As cresol is a highly permeable compound that was detected at high levels in the guts of Group I mice characterized by behavioral changes and impaired mPFC myelination , we treated primary cultured oligodendrocyte progenitors with increasing concentrations of cresol and tested for myelin gene expression ( Figure 8 ) . Expression of Mag , Mog , Mbp , and Cnp transcripts and the number of double-positive CNP+/OLIG2+ cells were reduced incresol-treated cultures compared to controls ( Figure 8A–C ) . However , this effect was not due to toxicity , but rather to impaired differentiation , as indicated by the increased transcripts of immature progenitor markers ( Pdgfra ) and the stable OLIG2+ cell counts ( Figure 8D , E ) . 10 . 7554/eLife . 13442 . 020Figure 7 . Metabolomic analysis of gut tissue from microbiota-transplanted C57BL/6 mice . ( A ) Experimental timeline . ( B ) 3D plot of scores between selected components generated by PLS-DA analysis comparing Group I ( transplanted with microbiota from vehicle-treated NOD mice; filled circles ) and Group II ( transplanted with microbiota from antibiotic-treated NOD mice; open circles ) . ( C ) Metabolic pathway impact overview generated with MetaboAnalyst 3 . 0 . Unaltered pathways have a score of 0 , and the most impacted pathways have higher scores . Pathways having the least statistical significance score are uncolored , whereas pathways having a high statistical significance score are colored in red . See Gacias et al . ( 2016 ) . ( D ) Metabolites with the greatest differential between mice with ( Group I ) and without ( Group II ) behavioral phenotype , were selected by volcano plot with a fold-change threshold of 1 . 5 ( x axis ) and t test threshold of 0 . 1 ( y axis ) . Red circles represent metabolites above the threshold ( Group II vs Group I ) ; see Table 1 . ( E ) One-way analysis of variance box and whisker plots illustrating the metabolite changes observed in Groups I and II . The y axis illustrates normalized , log transformed , and scaled peak area . Horizontal lines within the boxes represent the group means . Open circles represent excluded levels ( outliers ) ( n=6 mice per group ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13442 . 02010 . 7554/eLife . 13442 . 021Table 1 . Summary of trends in levels of cecal metabolites in C57BL/6 transplanted mice ( Group II vs Group I ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13442 . 021Super PathwaySub-pathwayMetaboliteFold change ( Group II vs I ) p valueAmino acidPhenylalanine metabolismBenzoic Acid1 . 010 . 031786Amino acidAlanine , aspartate and Glutamate metabolismN-acetylasparagine0 . 520 . 066953Amino acidTryptophan metabolismXanthurenic acid7 . 550 . 072748Amino acidUrea cycleHomocitrulline2 . 70 . 075261Amino acidArginine and proline metabolismN-acetyl-glutamate1 . 680 . 08886Amino acidPhenylalanine metabolismphenylpyruvate3 . 30 . 094054CarbohydratePentose phosphate pathwaySedoheptulose-7- phosphate0 . 990 . 05242Cofactors and vitaminsMicrobial metabolism in diverse environmentscresol0 . 130 . 019692LipidFatty acidsHexanedioic acid42 . 320 . 0053711LipidLong chain fatty acidLinoleic acid1 . 950 . 011101LipidLong chain fatty amideOleamide0 . 460 . 030305LipidLong chain fatty aciddihydroxystearic acid1 . 020 . 042845LipidLong chain fatty amideStearamide0 . 090 . 068105NucleotidePurine metabolismcAMP0 . 990 . 077892Data were analyzed using comprehensive global mass spectrometry-based metabolomics analysis . Additional details are provided in Experimental Procedures . 10 . 7554/eLife . 13442 . 022Figure 8 . Cresol treatment decreases myelin gene expression . ( A ) Transcript levels of oligodendrocyte lineage ( Olig2 ) , progenitor ( Pdgfrα , Cspg4 ) and differentiation ( Mag , Mog , Mbp , Cnp , Sox10 ) markers in oligodendrocyte progenitors cultured in differentiating conditions and treated with increasing concentrations of cresol ( 0 , 10 , 50 μM ) . DMSO was used as vehicle and negative control . Values were normalized to 36b4 mRNA levels and are referred as fold change relative to the control group ( n=3 independent primary cultures ) . ( B , C ) Representative confocal images and quantification of early differentiated oligodendrocytes ( CNP+/OLIG2+ ) after treatment with increasing concentrations of cresol ( 0 , 10 , 50 μM ) for 24 hr . ( D , E ) Representative confocal images and quantification of oligodendrocytes ( OLIG2+/DAPI+ ) treated with increasing concentrations of cresol ( 0 , 10 , 50 μM ) for 24 hr . Scale bars , 20 μm; 10–15 fields ( 20× ) per condition/experiment; n=2 independent primary cultures . Data are mean ± S . E . M; p<0 . 05 , **p<0 . 001 based on one-way ANOVA with Dunnett's Multiple Comparison Test; n . s . indicates not significantDOI: http://dx . doi . org/10 . 7554/eLife . 13442 . 022 Our results provide strong evidence that manipulations of gut microbiota are sufficient to induce depressive-like behaviors in adult mice . The behavioral changes were detected in mice with gut microbiota enriched for the taxa Clostridiales , including the Lachnospiraceae and Ruminococcaceae families , and with increased levels of highly permeable metabolites ( such as cresol ) with the ability to impair oligodendrocyte differentiation and myelin gene transcription . The observation that behavioral traits were only detected in transplant recipients with effective colonization of these taxa highlights the potential molecular mechanisms by which gut microbiota impacts CNS homeostasis . To date , several studies have focused on the relationship between microbiota composition and the development of anxiety-like behaviors . Dysbiotic microbiota induced by either pathogenic infections or antibiotic treatment has been shown to increase anxiety-like behavior in conventionally raised mice ( Bercik et al . , 2010; 2011; Lyte et al . , 2006 ) , while germ-free mice show reduced levels of anxiety-like behaviors compared to normal mice ( Diaz Heijtz et al . , 2011; Neufeld et al . , 2011 ) . In our study , anxiety-like behaviors were also shown to be affected by daily gastric manipulations affecting microbiota composition . However , social avoidance and despair-like behaviors were differentially induced by gavage in two genetically distinct strains of mice , which could be prevented by the administration of a broad-spectrum antibiotic cocktail . Daily gavage of NOD mice induced significant changes of gut bacterial communities and depressive-like behavior ( social and despair-like behaviors ) , which was associated with enrichment of bacteria within the Clostridiales . Antibiotic treatment decreased the overall bacterial diversity and prevented the behavioral effects . Subcutaneous administration of the same antibiotic treatment failed to induce significant changes either in the microbiota composition or behavior , further highlighting the importance of a local effect of oral antibiotic treatment on these intestinal microbial communities . To prove causality and understand whether behavioral changes observed in vehicle-gavaged NOD mice were in fact modulated by the intestinal microbiota , we transferred the cecal content of these mice ( displaying a social avoidance behavior ) or antibiotic-treated ( with normal social behavior ) NOD donors into the microbiota-depleted guts of C57BL/6 recipients . Our results demonstrate that only recipients with successful recolonization of the taxa enriched in the vehicle-treated NOD mice ( e . g . Clostridiales , Lachnospiraceae and Ruminococcaceae ) exhibited the social avoidance and despair-like behaviors , as well as the myelin gene expression in the mPFC of the donors . These transcriptional changes resulted in decreased adult PFC myelination in mice with transferred behavior . The microbiota of these C57BL/6 recipients showed significant differences in the abundance of several of the bacterial populations identified in the donors . Interestingly , alterations of some Lachnospiraceae and Ruminococcaceae spp . have been associated with behavioral deficits in mice ( Bruce-Keller et al . , 2015 ) . Our results did not identify a single bacterium responsible for the behavioral changes induced by vehicle-gavage in the NOD mice or by transplantation in the C57Bl6 animals , suggesting that specific communities enriched in taxa from the Lachnospiraceae and Ruminococcaceae are responsible for the observed phenotype . Community-driven effects have also been reported in the induction of colonic regulatory T cells by specific mixtures of Clostridia strains in models of colitis or in cognitive and stereotypic behavioral changes induced by high-fat diet microbiota in non-obese mice ( Bruce-Keller et al . , 2015; Atarashi et al . , 2015 ) . Our results also show that alterations of the microbial composition modified gut-produced metabolites and transcriptomic profiles in the mPFC , subsequently affecting behavior ( Daniel et al . , 2014; Hsiao et al . , 2013 ) . Microbiota composition has previously been shown to modulate anxiety-like behaviors in adult mice via changes in levels of brain-derived neurotrophic factor in the hippocampus ( Bercik et al . , 2011 ) . The results of our untargeted transcriptomic analysis of the mPFC , the region responsible for the integration of external stimuli and complex behaviors ( Regenold et al . , 2006 ) , identified a signature characterized by genes regulating transcription , circadian rhythm , protein phosphorylation , synapses , and myelin . Altered expression of genes related to myelin and circadian rhythm is consistent with reported white matter changes and sleep disruption in human patients with major depression ( Liu et al . , 2015; Landgraf et al . , 2014; Lavebratt et al . , 2010; Kishi et al . , 2009 ) as well as with the reported behavioral changes detected on myelin mutant mice ( Hagemeyer et al . , 2012 ) . The association of diminished myelination in mPFC with the observed social avoidance behavior is supported by recent studies describing decreased myelin gene expression and fewer myelinated fibers in the mPFC of mice after prolonged social isolation ( Liu et al . , 2012; 2016; Makinodan et al . , 2012 ) . Importantly , adoptive transfer of gut microbiota from NOD mice was able to recapitulate the mPFC transcriptional changes detected in recipient mice , thereby directly implicating gut microbiota as a causal factor for the induced behavioral and transcriptional changes . One mechanism by which the gut microbiota may regulate such alterations is through the production of selective metabolites . Several recent studies have shown that a dysbiotic gut microbiota can produce neurotoxic metabolites directly impacting behavior ( Hsiao et al . , 2013; Persico and Napolioni , 2013; Shaw , 2010 ) . For instance , in a mouse model for autism spectrum disorders during development , characterized by dysregulation of Lachnospiraceae , Ruminococcaceae the anxiety-like phenotype correlated with the levels of the metabolite 4-ethylphenylsulfate ( 4-EPS ) ( Hsiao et al . , 2013 ) . In our study , social avoidance behavior in adult mice was significantly associated with enrichment in Lachnospiraceae , Ruminococcaceae and Clostridiales and thedetection of high levels of cresol . This highly permeable metabolite was detected only in the gut of mice with social avoidance behavior , and was capable of preventing myelin gene expression and differentiation of oligodendrocyte progenitors into myelin-forming cells . These results suggest a potential mechanism linking CNS transcriptional changes to gut microbial homeostasis . Thereby increased intestinal production of cresol could be responsible for the behavioral changes observed in Group I transplant recipients by impacting adult myelination in the mPFC , possibly because this brain region is still capable of generating myelin after development . Several species of Clostridia have been shown to be producers of 4-EPS and cresol ( Persico and Napolioni , 2013; Nicholson et al . , 2012 ) , consistent with our findings that the microbiota of transplant recipients displaying an altered behavior ( social avoidance and increased despair-like behavior ) was enriched with members of the Lachnospiraceae , Ruminococcaceae , and other unidentified families within the Clostridiales order . We also detected a disruption in gut biosynthesis of tryptophan , tyrosine , and phenylanine in recipient mice with behavioral changes after transplantation . This might result in changes in the systemic/CNS levels of serotonin and other neurotransmitters , as almost 90% of serotonin production occurs within the gastrointestinal tract from its precursor tryptophan ( Berger et al . , 2009; Yano et al . , 2015 ) . Intestinal serotonin could cross through the blood brain barrier into the brain to regulate the observed social and despair-like behaviors . Additionally , accumulating evidence suggests that alterations in the glutamatergic system impact the pathophysiology of major depressive disorders ( Tokita et al . , 2012 ) . Interestingly , another metabolite that was significantly downregulated in affected transplant recipients was hexanedioic acid , also known as adipic acid , which can impact glutamate signaling by inhibiting the L-glutamate decarboxylase in the brain ( Wu and Roberts , 1974 ) . Recent work has demonstrated that epsilon toxin produced by Clostridium perfringens Type B is able to bind CNS endothelial cells and white matter tracts , inducing blood brain barrier disruption and oligodendrocyte apoptosis ( Linden et al . , 2015; Rumah et al . , 2013; 2015 ) . Although in our studies we could detect C . perfringens , its low abundance and the lack of demyelination at the ultrastructural level suggests that other members of the Clostridiales might be driving the behavioral outcome . In conclusion , our data support the concept that myelinating oligodendrocytes play a pivotal role in the pathogenic process underlying social avoidance , and define the intestinal microbiota as a potential regulator of such behavioral alterations in adult mice . Seven-week-old male C57BL/6 and NOD mice were purchased from Jackson Laboratories ( Bar Harbor , ME ) and housed in specific pathogen-free facilities at Mount Sinai . All procedures were performed in accordance with the Institutional Animal Care and Use Committee guidelines of the Icahn School of Medicine at Mount Sinai ( #08–0676 , #08–0675; LA10-00398; LA12-00193; LA12-00146 ) . A cocktail consisting of vancomycin ( 50 mg/kg ) , neomycin ( 100 mg/kg ) , metronidazole ( 100 mg/kg ) , and amphotericin B ( 1 mg/kg ) was administered daily by gastric gavage or subcutaneous injection within a volume of 200 μL and 100 μL , respectively . Control mice received dH2O ( gastric gavage ) or saline ( s . c . ) as vehicle . Ampicillin ( 1 g/L ) was supplemented in drinking water in the antibiotic-treated group ( Reikvam et al . , 2011 ) . Antibiotics were administered daily for 14 days prior to behavioral testing . During behavioral testing , antibiotics were administered every other day and always after the behavioral tests . All behavioral tests were recorded and tracked using Ethovision 3 . 0 ( Noldus , Netherlands ) for unbiased quantification . Overall anxiety behavior was assessed using Elevated plus maze . Social and despair-like behaviors were assessed using Social interaction and Forced swim tests . To limit carryover effects , behavioral tests were assessed in the order listed below over 14 days . Locomotor activity , Open field , Elevated plus maze , and Social interaction tests were conducted during the first week of testing with 24 hr of recovery between each task , while the Forced swim test was tested the following week . All mice used for 16S rRNA sequencing were co-housed per group ( 3–5 mice per cage ) in specific pathogen-free conditions . Fecal pellets were collected directly into sterile 1 . 5 mL tubes and immediately frozen and stored at -80°C . Cecal content was harvested at the end of each experiment and immediately frozen and stored at -80°C until further analysis . Fecal pellets from co-housed mice were weighted and pooled , and gDNA isolated using Powersoil DNA Isolation kit ( Mo Bio Laboratories , Inc . , Carlsbad , CA ) . DNeasy blood and tissue kit ( Qiagen , Venlo , Netherlands ) was used to isolate gDNA from gut tissue ( cecum ) . For gut microbiome characterization , the V4 hypervariable region of the bacteria 16S rRNA gene was amplified using the universal primers F515 ( 50-CACGGTCGKCGGCGCCATT-30 ) and R806 ( 50-GGACTACHVGGGTWTCTAAT-30 ) . A 12 bp GOLAY error-correcting barcode was added to the reverse primer to enable sample multiplexing . Reactions were performed in triplicate using the AccuPrime Taq DNA Polymerase High Fidelity system ( Thermo Fisher Scientific , Waltham , MA ) . Unless noted , all the analyses were performed using QIIME 1 . 8 . 0 as previously described ( Clemente et al . , 2015 ) . Linear discriminant analysis effect size was performed using default parameters ( Segata et al . , 2011 ) . Raw data presented in Gacias et al . ( 2016 ) ( doi:10 . 5061/dryad . 31v06 ) . Oligotype analysis was performed on OTUs belonging to the Lachnospiraceae , Anaeroplasmataceae , and Ruminococcaceae families , and the Clostridiales order . Singletons and OTUs of low prevalence ( <80% of the samples ) were removed , and the sequences from the five most abundant OTUs were picked for further analysis . Entropy analysis was performed on this set of sequences to look for highly variable positions within all sequences in each OTU , and the number of oligotypes was chosen based on the entropy peaks generated ( Eren et al . , 2014 ) . qPCR analysis of genomic DNA extracted from tissue of C57BL/6 mice transplanted with microbiota ( from vehicle- and antibiotic-treated NOD mice ) was performed to quantify the total bacteria , the order of Clostridiales , and the families of Lachnospiraceae , and Ruminococcaceae in both set of animals . The primer sequences and their features are reported in Gacias et al . ( 2016 ) . The reaction mixture contained 1× PerfeCTa SYBR Green FastMix , ROX ( #101414–278; Quanta Biosciences , Inc . , Gaithersburg , MD ) , 200 nM each primer , 1 μL gDNA in a total volume of 12 . 5 μL . Each SYBR Green PCR assay was performed in triplicate using the ABI 7900HT Real-Time PCR System ( Applied Biosystems of Thermo Fisher Scientific ) , with the following cycling program: 5 min at 95°C , 30 s at 95°C , 45 s at 55°C/60°C , and 45 s at 72°C for 40 cycles . PCR results were analyzed using RQ Manager softwere 1 . 2 . 2 ( Applied Biosystems ) . The annealing temperature was 55°C for all set of primers apart from Ruminococcaceae ( 60°C ) . The genome of Blautia producta ATCC 27 , 340 and Ruminococcus bromii ATCC 27 , 255 were used as reference genomes to construct the standard curves and to calculate the unknown numbers of bacterial gDNA copies in both set of animals as described previous in Tamburini et al . ( Tamburini et al . , 2013 ) . Tissue punches were taken from the mPFC or NAc and flash frozen for subsequent processing . RNA was extracted using Trizol ( #15596–018; Invitrogen of Thermo Fisher Scientific ) and purified with the RNeasy Micro kit ( #74004; Qiagen ) following the manufacturer’s protocol . RNA was reverse transcribed with qScript cDNA Supermix ( #95048; Quanta Biosystems , Inc . ) and qPCR was performed using Perfecta Sybr Fast Mix Rox 1250 ( Quanta Biosystems , Inc . ) at the Mount Sinai Shared Resource Facility ( primers listed in Gacias et al . , 2016 ) . Each transcript value was calculated as the average of triplicate samples from several mice per experimental condition ( typically 6–12 ) . After normalization to 36b4 , the average value for each transcript was calculated based on the values obtained in all the samples included for each experimental condition . RNA from the mPFC was flash frozen for subsequent processing . RNA was extracted using Trizol ( Invitrogen ) , purified with RNeasy Micro kit ( Qiagen ) . RNA was then used for deep sequencing analysis ( RNA Seq ) . Samples were mapped at a rate of 79–80% . After filtering out adaptor and low-quality reads , reads were mapped using TopHat ( version 2 . 0 . 8 ) to the mm10 mouse genome ( Trapnell et al . , 2009 ) . The Cufflinks/Cuffdiff suite was used to estimate gene-level expression values as fragments per kilobase of exon model per million mapped fragments and detect differentially expressed genes at a FDR <10% and subjected to Gene Ontology enrichment . Primary oligodendrocytes were prepared by sequential immunopanning and kept in undifferentiating conditions as described earlier ( Watkins et al . , 2008 ) until the onset of experiments . Briefly , oligodendrocyte progenitor cells ( OPCs ) were isolated from one P6 mouse pup brain using an immunopanning system enabling a purity of 95% . The dissected cortex was chopped in papain buffer , incubated for 20 min at 37°C and titrated in ovomucoid solution ( CellSystems GmbH , Troisdorf , Germany ) . The single cell solution was centrifuged at 1000 rpm for 10 min and resuspended in panning buffer and transferred to a bacterial culture plate coated with Anti-BSL1 Griffonia simplificonia lectin ( L-1100; Vector Labs , Inc . , Burlingame , CA ) , for negative selection for 15 min , followed by a positive selection step with rat anti-mouse CD140a ( 10R-CD140AMS; Research Diagnostics , Inc . , Flanders , NJ ) as primary antibody and AffiniPure goat anti-rat IgG ( H+L ) ( 112-005-003; Dianova ) as the secondary antibody for 45 min . The supernatant was aspirated , and the positive selection plate was washed with DPBS . The adherent OPCs were removed using trypsin , centrifuged for 10 min at 1000 rpm , resuspended in mouse OPC Sato medium ( Watkins et al . , 2008 ) and plated in a p100 culture plates coated with poly-d-lysine ( P7886; Sigma-Aldrich , St . Louis , MO ) . The OPCs were cultured in a humidified incubator at 5% CO2 and 37°C with media changes every 2 d . OPCs were maintained proliferating in the presence of bFGF ( 20 ng/mL ) and PDGF ( 10 ng/mL ) , while oligodendrocyte differentiation was induced by culturing the cells in the absence of mitogens and adding 60 nM T3 ( T5516; Sigma-Aldrich ) to Sato medium . Stock solutions of Cresol ( C85751; Sigma-Aldrich ) were prepared in DMSO ( 1000-fold concentrated ) and then diluted in differentiation media ( SATO+T3 ) to give final concentrations of 10 μM and 50 μM of Cresol . Primary oligodendrocytes were plated on 0 . 1 mg/mL poly-d-lysine coated 6-well plates in proliferating conditions ( SATO + bFGF and PDGF ) . Twenty-four hours after plating , cell differentiation was induced by changing the medium to SATO+T3 . At this point cells were treated for 24 hr with Cresol at 10 μM or 50 μM as well as DMSO as a control . Cells were gently washed with PBS after completion of the treatment and fixed with 4% paraformaldehyde for 15 min at room temperature for immunocytochemistry experiments . Experimental animals were anesthetized and then perfused with 4% ( w/v ) paraformaldehyde in 0 . 1 M phosphate buffer . Whole brains were cryopreserved in 30% ( w/v ) sucrose , embedded in OCT and sectioned ( 14 μm ) . Permeabilization in blocking buffer ( PGBA , 10% [v/v] normal goat serum [Vector Laboratories] and 0 . 5% [v/v] Triton X-100 ) followed by overnight incubation with primary antibody anti-MBP ( clone SMI99 , 1:500; BioLegend , San Diego , CA ) at 4°C . After incubation with secondary fluorescent antibodies ( Donkey anti-mouse Alexa Fluor 594 ) and nuclear counterstaining with DAPI ( 1:10 , 000; Molecular Probes of Thermo Fisher Scientific ) , immunoreactivity was visualized using LSM780 Meta confocal laser scanning microscope ( Carl Zeiss Micro-Imaging , Jena , Germany ) . Immunohistochemistry of cultured cells with CNP and OLIG2 antibodies was performed on fixed cells . Cells were grown on CC2-coated 8 well chambers ( Lab-Tek , Scotts Valley , CA ) for all immunocytochemistry . For staining oligodendrocyte lineage ( OLIG2 ) and differentiation markers ( CNP ) , cells were rinsed gently with PBS and were then fixed with 4% PFA for 15 min at room temperature . Fixed cells were first incubated with blocking/permeabilization solution ( PGBA plus 10% normal goat serum , and 0 . 5% Triton X-100 ) for 1 hr at room temperature . For co-staining experiments , cells were incubated with additional primary antibodies against OLIG2 ( AB9610 , 1:1000; Millipore , Darmstadt , Germany ) and CNP ( SMI91R , 1:500; Covance , Princeton , NJ ) overnight at 4°C . One-hour incubation with secondary fluorescent antibodies ( Alexa Fluor 594 ) was performed the following day with counterstaining for DAPI ( 1:10000 ) to visualize cell nuclei . Images were captured with a 20× objective using an LSM 780 Metaconfocal laser scanning microscope ( Carl Zeiss MicroImaging , Inc . , Jena , Germany ) . For OLIG2 and CNP cell counts , 10–15 fields were taken per condition . For MBP area quantification , four fields were taken per mouse . Three to four mice were included per treatment condition . MBP+ area and OLIG2+/CNP+ cell counts were quantified using ImageJ ( Liu et al . , 2016; Rusielewicz et al . , 2014 ) . An unpaired Student’s t test or one-way ANOVA was performed to assess statistical differences between conditions as indicated in figure legends . mPFC and NAc samples were prepared from standard electron microscopic analysis as previously described ( Liu et al . , 2012; 2016 ) . Briefly , mice were transcardially perfused with 0 . 1 M Millonigs buffer containing 4% paraformaldehyde and 5% glutaraldehyde and post-fixed for 2 wk . Brains were harvested and the region spanning from bregma to 2 . 5 mm anterior to bregma was vibratome sectioned at 40 μm . Comparable sections ~1 . 5 mm anterior to bregma and at the level of the forceps ( Liu et al . , 2012 ) minor of the corpus callosum were selected and embedded in PolyBed resin ( Polysciences ) , thick sectioned ( 1 μm ) and stained with toluidine blue . Using these sections , the mPFC and the core of the NAc were identified , and both regions were thin sectioned ( 90 nm ) and stained with uranyl acetate and lead citrate . For quantitation of myelin thickness , 10 electron micrographs were collected at 10 , 000× per region using a JEOL JEM 1230EX transmission electron microscope equipped with a Gatan Orius SC1000 side mount CCD camera . Using NIH Image J , the g-ratio of a minimum of 100 myelinated axons per region was calculated using the collected electron micrographs . Blood samples were collected by tail snip , and blood glucose was measured using glucose strips ( 7080G; Bayer Contour ) . At the time of transplantation , microbiota was freshly harvested from the cecum of 8–9-wk-old NOD mice treated with either vehicle or antibiotic . Cecal content was harvested , pooled , homogenized in a 1:4 in sterile solution ( 1x PBS: 80% glycerol , ratio 1:1 ) , centrifuged at 800 rpm and the supernatant was collected , aliquoted , and stored at -80°C . Recipient 8-wk-old C57BL/6 mice received an oral cocktail of antibiotics ( describe above in this section ) once daily for 14 consecutive days prior to the transplantation . Recipients were then randomized in two groups ( Group I and II ) , tested for social behavior , and then immediately started on the re-colonization protocol . To re-colonize the gut of C57BL/6 mice , recipient mice were orally gavaged every other day with 200 μL of cecal content isolated from the vehicle-treated or antibiotic-treated NOD mice over the subsequent 14 d ( for a total of 7 times ) . Behavioral testing was repeated after 15 d of first transplantation , and group-blinded analysis of the results was performed . Cecal and mPFC samples were harvested at the end of the behavioral testing ( 14 d post-transplantation ) and immediately stored at -80°C for further processing and analysis . Frozen tissues ( 30 mg ) were placed in pre-chilled ( -80°C ) 2 mL round bottom Eppendorf tubes having a stainless steel ball in it . Next , 400 mL of a pre-chilled ( -20°C ) mixture of acetonitrile , isopropanol , and deionized water in proportion 3:3:2 ( v/v/v ) was added . Samples were homogenized using Tissue Lyser ( Qiagen ) at 25 Hz speed for 5 min . Samples were further centrifuged at 4°C at 12 , 000 rpm for 3 min . Clean supernatant was transferred into vials or 0 . 5 mL Eppendorf tubes ( to be dried for gas chromatography combined with time-of-flight high-resolution mass spectrometry ) . Tissue extracts were divided in to three parts: 75 μL for gas chromatography combined with time-of-flight high-resolution mass spectrometry , 150 μL for hydrophilic liquid chromatography coupled with high-resolution mass spectrometry , and 150 μL for hydrophilic interaction chromatography with liquid chromatography and tandem mass-spectrometry . Metabolomic analyses were performed using non-targeted and targeted protocols as previously described ( Tolstikov et al . , 2014; Urayama et al . , 2010; Zou and Tolstikov , 2008 ) . A standard quality control sample containing a mixture of amino and organic acids was injected daily to monitor mass spectrometer response . A pooled quality control sample was obtained by taking an aliquot of the same volume of all samples from the study and injected daily with a batch of analyzed samples and to determine the optimal dilution of the batch samples and to validate metabolite identification and peak integration . Identified metabolites were subjected to pathway analysis with MetaboAnalyst 3 . 0 , which consists of an enrichment analysis relying on measured levels of metabolites and pathway topology , and provides visualization of the identified metabolic pathways . Accession numbers of detected metabolites ( HMDB , PubChem , and KEGG Identifiers ) were generated , manually inspected , and utilized to map the canonical pathways . Behavioral and biochemical data were analyzed by unpaired , two-tailed Student’s t tests or one-way ANOVA followed by Bonferroni post hoc test , as appropriatem using Prism software ( GraphPad Software , Inc . , La Jolla , CA ) . Microbiome data were analyzed using QIIME 1 . 8 . 0 with default parameters . Statistical significance was assessed using R 3 . 0 . 2 . Statistical significance for all analyses was accepted at p<0 . 05 . Metabolomic data was analyzed as previously described in Tolstikov et al . ( Tolstikov et al . , 2014 ) .	A combination of genes and environmental factors underlie an individual’s risk of developing a mental illness . Among the environmental factors , it is becoming clear that communication between the gut and the brain is involved , but we do not understand how these two organs communicate . Our gut contains a variety of bacteria that help us to digest food and there is some evidence that changes in these bacterial communities can influence our mental health . Transplanting feces from one individual to the gut of another is one way to alter the communities of bacteria in the gut . Here , Gacias et al . investigated whether fecal transplants are sufficient to induce social avoidance behavior – a symptom of depression – in mice . The experiments show that introducing specific combinations of bacteria into the gut is indeed able to cause healthy adult mice to avoid social interactions . This effect was caused by changes in the “myelin” sheath that surrounds many nerve fibers and could be prevented by giving the mice antibiotics , which decreased the number of bacteria in the gut . Further experiments revealed that the mice that became depressed after fecal transplants had higher levels of a molecule called cresol , which is produced by certain gut bacteria . Gacias et al . found that cresol is able to reduce the amount of myelin produced by brain cells . Therefore , these findings show that changing the communities of bacteria in the gut can result in the accumulation of molecules that influence social behavior . Future work will aim to identify bacteria that can reduce the amount of cresol produced in the gut , which may have the potential to treat depression .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Microbiota-driven transcriptional changes in prefrontal cortex override genetic differences in social behavior
Anatomical and physiological studies have led to the assumption that the dorsal striatum receives exclusively excitatory afferents from the cortex . Here we test the hypothesis that the dorsal striatum receives also GABAergic projections from the cortex . We addressed this fundamental question by taking advantage of optogenetics and directly examining the functional effects of cortical GABAergic inputs to spiny projection neurons ( SPNs ) of the mouse auditory and motor cortex . We found that the cortex , via corticostriatal somatostatin neurons ( CS-SOM ) , has a direct inhibitory influence on the output of the striatum SPNs . Our results describe a corticostriatal long-range inhibitory circuit ( CS-SOM inhibitory projections → striatal SPNs ) underlying the control of spike timing/generation in SPNs and attributes a specific function to a genetically defined type of cortical interneuron in corticostriatal communication . It is very well established that cortical neurons regulate the activity of spiny projection neurons ( SPNs ) in the striatum through long-range glutamatergic/excitatory projections ( Landry et al . , 1984; Wilson , 1987; 2004; Graybiel et al . , 1994; Lovinger and Tyler , 1996; Reiner et al . , 2003; Kress et al . , 2013 ) , while inhibition is mediated by local feed-forward and feed-back circuits ( for review [Tepper et al . , 2008] ) . The feed-forward circuit is characterized by GABAergic striatal interneurons that receive excitatory inputs from the cortex and monosynaptically inhibit SPNs . The feed-back circuit is characterized by SPNs and their interconnections via local axon collaterals ( Calabresi et al . , 1991; Kawaguchi , 1993; Kita , 1993; 1996; Kawaguchi et al . , 1995; Mallet et al . , 2005; Planert et al . , 2010; Ibanez-Sandoval et al . , 2011 ) . Because striatal neuronal activity has been shown to be involved in movement , learning , and goal-directed behavior ( Graybiel , 1995; Schultz et al . , 2003; Barnes et al . , 2005; Kreitzer and Malenka , 2008 ) , it is crucial to understand the cortical connectivity pattern and dynamics that shape the flow of information in the striatum . Anatomical studies using retrograde tracers and immunohistochemistry have proposed that between 1–10% of the GABAergic 'interneurons' in rodents , cats , and monkeys also give rise to long-range corticocortical projections ( McDonald and Burkhalter , 1993; Tomioka et al . , 2005; Higo et al . , 2007; Tomioka and Rockland , 2007; Higo et al . , 2009 ) . A growing body of evidence suggests that many of these projections arise from somatostatin-expressing neurons ( Tomioka et al . , 2005; Higo et al . , 2007; Tomioka and Rockland , 2007; Higo et al . , 2009; McDonald et al . , 2012; Melzer et al . , 2012 ) . A previous study has demonstrated the presence of a corticostriatal GABAergic projection ( Tomioka et al . , 2015 ) , but the cells of origin and physiological function of this GABAergic projection from the cortex to the dorsal striatum were not explored . To test the hypothesis that the cortex has a direct inhibitory influence on the output neurons of the striatum , we measured the response of SPNs to optogenetic activation of corticostriatal somatostatin neuron ( CS-SOM ) axons . Our results describe a previously unknown corticostriatal direct inhibitory circuit ( CS-SOM inhibitory projections → striatal SPNs ) underlying the control of spike timing/generation in SPNs and attribute a specific function to a genetically defined type of cortical interneuron in corticostriatal communication . Overall this suggests that the timing and ratio of cortical excitatory and inhibitory inputs to the dorsal striatum , by shaping the activity pattern of SPNs , determines behavioral outcomes . To visualize long-range GABAergic projections originating in the cortex and terminating in the dorsal striatum , we conditionally expressed GFP in somatostatin-expressing ( SOM ) interneurons by injecting AAV . GFP . Flex into the right auditory cortex ( AC ) of SOM-Cre-tdTomato transgenic mice ( Figure 1a ) . GFP was colocalized with SOM/tdTomato-expressing neurons in the AC ( Figure 1b left ) and GFP-positive axons were visible in coronal sections of the right dorsal striatum in these mice ( Figure 1b middle and right ) . 10 . 7554/eLife . 15890 . 003Figure 1 . Morphological characteristics , axonal projections and electrical properties of long-range CS-SOM neurons in the mouse auditory cortex . ( a ) Schematic depicting injection site using the SOM-Cre-tdTomato transgenic mouse line to identify CS-SOM neurons and their projections to the dorsal striatum . Bottom , auditory cortex: AAV . GFP . Flex injection site; yellow CS-SOM somata coexpressing GFP and tdTomato . Top , dorsal striatum: green CS-SOM GFP-positive axons; red SOM tdTomato-positive interneurons . ( b ) Epifluorescence images of SOM GFP-positive neurons . Top , left: GFP-positive SOM neurons in the auditory cortex identified by viral injection of AAV . GFP . Flex in the SOM-Cre-tdTomato transgenic mouse line . Middle , left: tdTomato-expressing SOM neurons in the SOM-Cre-tdTomato transgenic mouse line . Bottom , left: overlay of GFP and tdTomato images . Middle , the dashed box indicates the location GFP-positive axons from CS-SOM neurons in the dorsal striatum and the location of image in the right panel . Right , higher magnification of GFP fluorescence of CS-SOM axons in the dorsal striatum . ( c ) Schematic depicting injection site using the SOM-Cre transgenic mouse line to identify CS-SOM neurons by anatomical retrograde transfection . Top , striatum: AAV . GFP . Flex injection site . Bottom , auditory cortex: green CS-SOM GFP positive somata . ( d ) Bright-field ( top left ) and epifluorescence ( bottom left ) images of striatal SOM interneurons transfected with AAV . GFP . Flex . Middle ( left and right ) , epifluorescence image of laminar distribution of CS-SOM neurons in the auditory cortex identified by anatomical retrograde transfection . Right , high-resolution image of a biocytin-labeled CS-SOM neuron . ( e ) Morphological reconstruction of one CS-SOM neuron ( dendrites , blue; axons , gray ) . ( f ) Bottom , train of action potentials recorded in a GFP-positive CS-SOM neuron during step current injection ( 1 . 0 s , 100 pA pulse ) . Top , single action potential from GFP-positive CS-SOM neuron ( black ) ; compare to an action potential from a fast-spiking interneuron ( red ) . ( g ) Summary plot of Vrest: resting membrane potential; Ri: input resistance; Tau: membrane time constant; Rheobase , the smallest current step evoking an action potential; AP thr: action potential threshold; AP height: action potential height; AP half-width: action potential half-width; and F/I slope from CS-SOM neurons ( n = 13 ) , including group averages ( ± s . e . m . ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15890 . 003 Next , to determine the layer of origin for the long-range corticostriatal SOM neurons ( from this point forward referred to as CS-SOM neurons ) , we injected the AAV . GFP . Flex virus into the right dorsal striatum of SOM-Cre transgenic mice ( Taniguchi et al . , 2011 ) ( Figure 1c ) . Using this virus , CS-SOM neurons were retrogradely labelled with GFP , and their somata were present primarily in layers 5 and 6 of the AC ( Figure 1d , middle ) . This approach allowed us to visually identify and record from layer 5 and 6 CS-SOM neurons using whole-cell patch clamp . Confocal images of biocytin-filled CS-SOM neurons showed that they are similar to SOM interneurons in their morphology and send an axonal projection into the subcortical white matter ( Figure 1d , right; Neurolucida-reconstructed CS-SOM neuron Figure 1e ) . We verified their identity based on the comparison with electrophysiological properties of SOM interneurons ( Ma et al . , 2006 ) . These properties include a wide action potential and low rheobase ( the smallest current step evoking an action potential ) ( Figure 1f , top; action potentials in CS-SOM neurons , shown in black , are wider than those from fast-spiking interneurons , example in red ) . The responses to current steps in CS-SOM neurons were typical for SOM interneurons ( Figure 1f , bottom; notice the sag from a hyperpolarizing current step ) . Basic electrophysiological properties for CS-SOM neurons ( n = 13 ) included ( Figure 1g ) : resting membrane potential , −72 . 93 ± 1 . 16 mV; input resistance , 240 ± 26 . 98 MΩ; membrane time constant , 6 . 84 ± 0 . 71 ms; rheobase , 100 ± 15pA; action potential threshold , −41 . 35 ± 1 . 33 mV; action potential height , 54 . 18 ± 2 . 21 mV; action potential width , 0 . 49 ± 0 . 05 ms; F-I slope , 0 . 24 ± 0 . 03 Hz/pA step . Although , in our experience , AAV1 . Flex viral vectors ( Atasoy et al . , 2008 ) exhibited both anterograde and retrograde ( Rothermel et al . , 2013 ) transfection capabilities , when we injected this virus in the cortex we only observed anterograde labeling of CS-SOM neurons ( i . e . no SOM somata transfected in the dorsal striatum; see Figure 1b , right ) . In contrast , when we injected this virus into the dorsal striatum , we observed transfected SOM somata both in the dorsal striatum ( Figure 1d , left ) and in the cortex ( Figure 1d , middle left and middle right ) . These data show that this inhibitory projection is unidirectional , arising preferentially from layer 5 and 6 SOM neurons in the cortex which project to the dorsal striatum . To determine the connectivity pattern of CS-SOM neurons onto neurons in the dorsal striatum , we used an optogenetic approach in which we conditionally expressed channelrhodopsin-2 ( ChR2 ) in SOM neurons by injecting AAV1 . ChR2 . Flex into the right AC of SOM-Cre transgenic mice ( Figure 2a , b; injection site Figure 2c ) . After 3–4 weeks , we recorded from the right dorsal striatum , in which ChR2-positive axons could be observed ( Figure 2d , right ) . These axons have been reported to remain photoexcitable even when severed from their parent somata ( Petreanu et al . , 2007; Rock and Apicella , 2015 ) . To determine the synaptic properties of CS-SOM projections onto striatal neurons , we photoactivated CS-SOM ChR2-positive axons by flashing blue light ( 470 nm ) for 2–5 ms during whole-cell recordings from striatal neurons . IPSCs ( Figure 2f , red trace ) were isolated by applying a command potential of 0 mV ( the calculated reversal potential for glutamatergic excitatory conductance ) . The IPSCs onset of the photo-evoked response was 2 . 2 ± 0 . 11 ms ( Figure 2g , left ) . This latency is consistent with the IPSCs being the result of a monosynaptic inhibitory input from the cortex and not a local striatal feedback inhibitory network recruited by cortical projections . Blocking excitatory neurotransmission by application of glutamate receptor antagonists NBQX and CPP did not abolish the CS-SOM-ChR2-evoked synaptic IPSCs ( Figure 2f , magenta trace ) . In contrast , blocking inhibitory neurotransmission by application of gabazine ( Figure 2f , black trace ) completely abolished the CS-SOM-ChR2-evoked synaptic IPSCs , confirming they were elicited by direct cortical inhibitory transmission . Basic biophysical properties for CS-SOM-ChR2-evoked synaptic IPSCs ( n = 16 ) included ( Figure 2g , h , i ) : peak , 122 ± 26 pA; charge , 3 . 85 ± 0 . 67 pC; rise time , 1 . 64 ± 0 . 13 ms; decay time , 28 . 87 ± 2 . 76 ms . Biocytin-filled neurons were morphologically identified as SPNs post-hoc by the presence of dendritic spines ( Figure 2e , right , white arrows ) . Eight out of 16 neurons were recovered after patching and were processed for imaging; all eight of these neurons showed a high density of dendritic spines at 40-63X magnification . These data reveal that a large proportion of striatal SPNs receive direct inhibitory input driven by CS-SOM neurons but does not exclude the possibility that other striatal neurons also receive inhibitory input from these projections . 10 . 7554/eLife . 15890 . 004Figure 2 . Photostimulation of auditory CS-SOM projections elicits direct inhibition and modulates action potentials in striatal SPNs . ( a ) Schematic depicting injection site using the SOM-Cre transgenic mouse line to transfect CS-SOM projections to the dorsal striatum with ChR2 . Bottom , auditory cortex: AAV . ChR2 . flex injection site . Top , dorsal striatum: red CS-SOM ChR2-tdTomato-positive axons . ( b ) Experimental paradigm for photostimulating ChR2-positive CS-SOM projections while recording from SPNs . ( c ) Bright-field ( left ) and epifluorescence ( right ) images of a slice containing the auditory cortex injection site for AAV . ChR2 . Flex . ( d ) Bright-field ( left ) and epifluorescence ( right ) images of a slice containing the dorsal striatum showing expression of ChR2-tdTomato following injection of AAV . ChR2 . Flex into the auditory cortex . ( e ) Left , bright-field image of neurons as seen in bright-field microscopy during patch recordings . Middle , high-resolution epifluorescence image of a biocytin-labeled SPN . The dashed box indicates the location of the image in the right panel . Right , high-resolution epifluorescence image of spines from the biocytin-labeled SPN . ( f ) Example of IPSCs recorded at 0 mV from an SPN before ( red trace ) and after application of ionotropic glutamate receptor antagonists ( NBQX 10 μM , CPP 5 μM: magenta trace ) and GABAA receptor antagonist ( gabazine 25 μM: black trace ) . ( g ) Left , plot of onset latencies recorded in SPNs ( n = 16 ) including group averages ( ± s . e . m . ) . Middle , plot of IPSCs peaks calculated for SPNs , including group averages ( ± s . e . m . ) . Right , plot of IPSCs charge transfer calculated for individual IPSCs for SPNs , including group averages ( ± s . e . m . ) . ( h ) Left , example of IPSCs ( black trace ) and rising time course ( red trace ) recorded at 0 mV from an SPN . Right , plot of IPSCs rising time course recorded in SPNs ( n = 16 ) including group averages ( ± s . e . m . ) . ( i ) Left , example of IPSCs ( black trace ) and decay time course ( amber trace ) recorded at 0 mV from an SPN . Right , plot of IPSCs decay time course recorded in SPNs ( n = 16 ) including group averages ( ± s . e . m . ) . ( j ) Left ( black trace ) , response of an SPN in the whole-cell current-clamp configuration to current injection ( 250 pA , 500 ms; n = 6 ) . Left ( blue trace ) , response of the SPN to current injection with photostimulation of CS-SOM projections ( blue bar , 5–20 ms ) . Right ( black trace ) , response of an SPN in the whole-cell current-clamp configuration to current injection ( 350 pA , 500 ms; n = 6 ) . Left ( blue trace ) , response of the SPN to current injection with photostimulation of CS-SOM projections ( blue bar , 5–20 ms ) . ( k ) Summary of ChR2-mediated delay of action potential generation in SPNs ( n = 6 ) during current injection combined with photostimulation of the ChR2 CS-SOM projections . Delay was relative to the onset of the first action potential measured during the current injection alone . DOI: http://dx . doi . org/10 . 7554/eLife . 15890 . 004 To determine how CS-SOM neurons affect the output of striatal neurons , we took advantage of the same viral ChR2 approach described above . We obtained whole-cell recordings from SPNs while injecting a step of current causing the neurons to spike between 2–7 action potentials ( Figure 2j , black traces ) . We then photoactivated CS-SOM ChR2-positive axons by flashing blue light ( 470 nm ) for 2–10 ms starting 10–50 ms before the first action potential . Combining current injection with photoactivation of CS-SOM ChR2-positive axons , we observed a delay of the first action potential ( Figure 2j , blue traces ) , with an average delay of 159 . 16 ± 66 . 45 ms ( n=6 ) ( Figure 2k ) . Overall , these results indicate that CS-SOM projections act directly on SPNs and have the ability to affect spike generation and timing in these neurons . The dorsal striatum is characterized by two parallel networks: the direct and indirect pathways ( Gerfen et al . , 1990; Kawaguchi et al . , 1990; Lei et al . , 2004; Gerfen and Surmeier , 2011; Calabresi et al . , 2014 ) . Direct pathway SPNs ( dSPNs ) express D1-type dopamine receptors and are suggested to promote behaviors that have previously been rewarded . On the other hand , indirect pathway SPNs ( iSPNs ) express D2-type dopamine receptors and are suggested to suppress behaviors that have not previously been rewarded ( Gong et al . , 2003; Ade et al . , 2011; Calabresi et al . , 2014 ) . We examined how activity of CS-SOM neurons regulates activity in the direct and indirect pathways . To address this question we used an optogenetic approach in SOM-Cre-D1/D2 transgenic mice ( Figure 3a ) in which dSPNs and iSPNs are labeled with td-Tomato and GFP , respectively ( Figure 3b left and middle ) . In these mice , AAV . ChR2 . Flex injected in the AC induces ChR2 expression in CS-SOM neurons . Although both D1-type dopamine receptors and our ChR2 virus are tagged with tdTomato , ChR2-tdTomato-positive axons were detectable in the area of our recordings ( Figure 3b , left and right panels , white arrows ) . To determine the relative strength of CS-SOM connections onto the two classes of SPNs , we recorded in coronal brain slices from a dSPN and a neighboring iSPN ( ~100 μm ) in the dorsal striatum ipsilateral to the injection site . It has been reported ( Ade et al . , 2011 ) that in Drd1a-tdTomato and Drd2-EGFP mice , fluorescent labeling is restricted to SPNs ( i . e . no fluorescent labeling of the three main classes of interneurons in the striatum: parvalbumin-expressing , somatostatin-expressing , and cholinergic ) . However , to ensure that the neurons we recorded from were in fact SPNs , we also identified them post-hoc by the presence of dendritic spines ( Figure 3c , right , white arrow ) . Six out of fourteen neurons were recovered after patching and were processed for imaging; all six of these neurons showed a high density of dendritic spines at 40-63X magnification . All recordings were performed in the presence of NBQX and CPP ( Figure 3d , red traces ) ; CS-SOM-ChR2-evoked synaptic IPSCs were completely blocked by application of gabazine ( Figure 3d , black traces ) . The peak and charge of the maximal IPSCs were used as measures of the strength of the CS-SOM projections to dSPNs ( n = 7 ) and iSPNs ( n = 7 ) . IPSCs peak amplitude ( dSPN: 149 ± 30 pA; iSPN: 130 ± 21 pA; p = 0 . 7 rank-sum test ) and charge ( dSPN: 6 . 26 ± 1 . 03 pC; iSPN: 5 . 24 ± 1 . 38 pC; p = 0 . 3 rank-sum test ) were similar in dSPNs and iSPNs ( Figure 3e ) . There were no differences in IPSCs rise times ( Figure 3g; example IPSCs from iSPN Figure 3f ) , suggesting the CS-SOM synapses had similar somatodendritic distributions on dSPNs and iSPNs . Together , these results indicate that CS-SOM neurons do not preferentially innervate dSPNs vs . iSPNs in the dorsal striatum . 10 . 7554/eLife . 15890 . 005Figure 3 . Auditory CS-SOM neurons innervate both dSPNs and iSPNs . ( a ) Experimental paradigm for photostimulating ChR2-positive CS-SOM projections while recording from genetically labeled dSPNs and iSPNs . ( b ) Left , epifluorescence image of dSPNs expressing D1-tdTomato ( white arrows indicate CS-SOM ChR2-tdTomato-positive axons ) . Middle , epifluorescence image of iSPNs expressing D2-EGFP . Right , overlay of D1-tdTomato ( dSPNs ) and D2-GFP ( iSPNs ) . Note that the two sub-types of SPNs have a different distribution with no overlap when located in the same region of the dorsal striatum ( white arrows indicate CS-SOM ChR2-tdTomato-positive axons ) . ( c ) Left , high-resolution epifluorescence image of a biocytin-labeled dSPN . The dashed box indicates the location of the image in right panel . Right , high-resolution epifluorescence image of spines from the biocytin-labeled dSPN . ( d ) Example of IPSCs recorded at 0 mV from a dSPN ( left ) and iSPN ( right ) after application of ionotropic glutamate receptor antagonists ( NBQX 10 μM , CPP 5 μM: red traces ) and GABAA receptor antagonist ( gabazine 25 μM: black traces ) . ( e ) Left , plot of IPSCs onset latencies recorded in dSPNs ( n = 7; red circles ) and iSPNs ( n = 7; green circles ) , including group averages ( ± s . e . m . ) . Middle , plot of IPSCs peaks calculated for dSPNs ( n = 7; red circles ) and iSPNs ( n = 7; green circles ) , including group averages ( ± s . e . m . ) . Right , plot of IPSCs charge calculated for dSPNs ( n = 7; red circles ) and iSPNs ( n = 7; green circles ) , including group averages ( ± s . e . m . ) . ( f ) Example of IPSCs ( black trace ) , rising time course ( red trace ) and decay time course ( amber trace ) recorded at 0 mV from a dSPN . ( g ) Left , plot of IPSCs rising time course calculated for dSPNs ( n = 7; red circles ) and iSPNs ( n = 7; green circles ) , including group averages ( ± s . e . m . ) . Right , plot of IPSCs decay time course calculated for dSPNs ( n = 7; red circles ) and iSPNs ( n = 7; green circles ) , including group averages ( ± s . e . m . ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15890 . 005 So far we had only observed this projection from the AC to the dorsal striatum . We next asked whether CS-SOM projections from the cortex to the dorsal striatum are a general feature of the corticostriatal circuit organization , or if they are specific to the AC . To test this , we used a similar optogenetic approach in the motor cortex ( MC ) , a cortical area which is not involved in sensory processing like the AC , but instead is involved in the planning , control and execution of voluntary movements . We used the same methods as in the AC to transfect CS-SOM neurons in the MC with AAV . ChR2 . Flex ( Figure 4a , b; injection site Figure 4c ) . As before , we observed local transfection in the cortex ( Figure 4c ) as well as ChR2-tdTomato-positive axons in the dorsal striatum ( Figure 4d , white arrows ) . Neurons in the dorsal striatum where ChR2-tdTomato-positive axons were present were chosen for recordings . We obtained whole-cell recordings from striatal neurons during photoactivation of CS-SOM ChR2-positive axons while recording IPSCs as in the AC ( Figure 4f , red trace ) . Responses in this area of the dorsal striatum were similar to those seen in the region of the dorsal striatum receiving auditory CS-SOM projections . Again , application of NBQX and CPP did not abolish the CS-SOM-ChR2-evoked synaptic IPSCs ( Figure 4f , magenta trace ) , while application of gabazine completely abolished the CS-SOM-ChR2-evoked synaptic IPSCs ( Figure 4f , black trace ) . Basic biophysical properties for CS-SOM-ChR2-evoked synaptic IPSCs ( n = 9 ) included ( Figure 4g , h , i ) : onset , 2 . 4 ± 0 . 11 ms; peak , 91 ± 17 pA; charge , 3 . 58 ± 0 . 59 pC; rise time , 1 . 76 ± 0 . 14 ms; decay time , 32 . 96 ± 4 . 98 ms . Neurons were identified as SPNs post-hoc by the presence of dendritic spines ( Figure 4e , right , white arrows ) . Three out of nine neurons were recovered after patching and were processed for imaging; all three of these neurons showed dendritic spines at 40-63X magnification . Overall , these results indicate that in MC , as well as in AC , CS-SOM projections act directly on SPNs . 10 . 7554/eLife . 15890 . 006Figure 4 . Photostimulation of motor CS-SOM projections elicits direct inhibition of striatal SPNs . ( a ) Schematic depicting injection site using the SOM-Cre transgenic mouse line to transfect CS-SOM projections to the dorsal striatum with ChR2 . Top , motor cortex: AAV . ChR2 . flex injection site . Bottom , dorsal striatum: red CS-SOM ChR2-tdTomato-positive axons . ( b ) Experimental paradigm for photostimulating ChR2-positive CS-SOM projections while recording from SPNs . ( c ) Bright-field ( left ) and epifluorescence ( right ) images of a slice containing the motor cortex injection site for AAV . ChR2 . Flex . ( d ) Bright-field ( left ) and epifluorescence ( right ) images of a slice containing the dorsal striatum showing expression of ChR2-tdTomato following injection of AAV . ChR2 . Flex into the motor cortex . ( e ) Left , bright-field image of neurons as seen in bright-field microscopy during patch recordings . Middle , high-resolution epifluorescence image of a biocytin-labeled SPN . The dashed box indicates the location of the image in the right panel . Right , high-resolution epifluorescence image of spines from the biocytin-labeled SPN . ( f ) Example of IPSCs recorded at 0 mV from an SPN before ( red trace ) and after application of ionotropic glutamate receptor antagonists ( NBQX 10 μM , CPP 5 μM: magenta trace ) and GABAA receptor antagonist ( gabazine 25 μM: black trace ) . ( g ) Left , plot of onset latencies recorded in SPNs ( n = 9 ) including group averages ( ± s . e . m . ) . Middle , plot of IPSCs peaks calculated for SPNs , including group averages ( ± s . e . m . ) . Right , plot of IPSCs charge transfer calculated for individual IPSCs for SPNs , including group averages ( ± s . e . m . ) . ( h ) Left , example of IPSCs ( black trace ) and rising time course ( red trace ) recorded at 0 mV from an SPN . Right , plot of IPSCs rising time course recorded in SPNs ( n = 9 ) including group averages ( ± s . e . m . ) . ( i ) Left , example of IPSCs ( black trace ) and decay time course ( amber trace ) recorded at 0 mV from an SPN . Right , plot of IPSCs decay time course recorded in SPNs ( n = 9 ) including group averages ( ± s . e . m . ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15890 . 006 In this study we test the hypothesis that the dorsal striatum is influenced by direct GABAergic projections from the cortex . Our results support this hypothesis and additionally conclude that both the motor and the auditory cortices send long-range GABAergic projections to the dorsal striatum , via CS-SOM neurons ( Figure 5 ) . Because of its presence in two such disparate cortical areas , this would suggest that the corticostriatal somatostatin projection connectivity pattern is likely a general feature of the corticostriatal network . We demonstrated that a class of corticostriatal long-range inhibitory neurons in the auditory and motor cortex of the mice are somatostatin-expressing neurons . Our results suggest that CS-SOM neurons provide a major GABAergic projection to the dorsal striatum , in agreement with a recent anatomical study in the mouse frontal cortex ( Tomioka et al . , 2015 ) . However previous results from retrospinal and somatosensory cortex ( Jinno and Kosaka , 2004 ) have provided anatomical evidence for expression of parvalbumin in glutamatergic and GABAergic pathways in mice suggesting that corticostriatal long-range inhibitory neurons are characterized predominantly by parvalbumin-expressing neurons . Moreover , recent results from prefrontal cortex ( Lee et al . , 2014 ) , show that corticostriatal long-range inhibitory neurons are characterized predominantly by parvalbumin-expressing or vasoactive intestinal peptide ( VIP ) -expressing but not somatostatin-expressing interneurons . Overall , these suggest that corticostriatal long-range GABAergic projections may be characterized by heterogeneous subtypes of interneurons , and that additional studies are necessary to determine the subtypes and the physiological impact of long-range corticostriatal GABAergic projections on striatal neurons . 10 . 7554/eLife . 15890 . 007Figure 5 . Summary diagram: CS-SOM neurons directly inhibit striatal SPNs . Auditory and motor CS-SOM projections modulate the activity of striatal SPNs by direct inhibition . Green lines: excitatory inputs from intratelencephalic ( IT-type ) and projecting-type ( PT-type ) layer 5 pyramidal neurons; red line: inhibitory input from CS-SOM neurons . DOI: http://dx . doi . org/10 . 7554/eLife . 15890 . 007 The main finding of the present study is that the cortex can exert a powerful direct inhibition of the dorsal striatum via CS-SOM neurons . Particularly , CS-SOM neurons target SPNs and have the ability to affect their spike timing and generation . In the cortex , interneurons target the somatodendritic compartment of pyramidal neurons ( Somogyi et al . , 1998 ) . SOM cortical and striatal interneurons in particular target the dendrites of pyramidal neurons and SPNs , respectively; although not specifically tuft dendrites as is characteristic of their hippocampal analogue , OLM cells ( Freund and Buzsaki , 1996; Kubota and Kawaguchi , 2000 ) . This connectivity pattern may allow the dendrites to sum incoming activity over a broader window of time ( synaptic integration ) ( Pouille and Scanziani , 2001 ) . We suggest that a similar inhibitory mechanism might be applied to the corticostriatal long-range inhibitory circuit , in which CS-SOM neurons , by targeting the dendrites of SPNs , play a role in the integration of incoming cortical excitatory inputs to the striatum . The dorsal striatum is characterized by two parallel networks: the direct and indirect pathways ( Gerfen et al . , 1990; Kawaguchi et al . , 1990; Lei et al . , 2004; Gerfen and Surmeier , 2011; Calabresi et al . , 2014 ) . Particularly , it is suggested that dSPNs promote behaviors that have previously been rewarded while iSPNs suppress behaviors that have not previously been rewarded ( Gong et al . , 2003; Ade et al . , 2011; Calabresi et al . , 2014 ) . It is very well established that long-range glutamatergic/excitatory projections regulate the activity of dSPNs and iSPNs . However , how activity of CS-SOM neurons influences dSPNs and iSPNs has not been established . Our results show that the cortex provides an inhibitory projection to both direct and indirect pathway neurons , influencing the output of both striatal networks involved in action selection and suppression of action . The functional role of inhibition from CS-SOM neurons might be not in separating the activity of direct and indirect striatal pathways , but rather in shaping the output of both . This is reminiscent of the finding by Kress et al . ( Kress et al . , 2013 ) that long-range excitatory corticostriatal connections from intratelencephalic and pyramidal tract neurons in motor cortex are made onto both dSPNs and iSPNs in the mouse dorsal striatum . The dorsal striatum is a brain structure where functionally distinct cortical regions ( e . g . motor , auditory , visual , somatosensory ) converge ( Wilson et al . , 1983; Schneider , 1991; Flaherty and Graybiel , 1993; Chudler et al . , 1995; Reig and Silberberg , 2014; Wilson , 2014 ) . Anatomical and physiological data have suggested that once this information reaches the dorsal striatum via the corticostriatal projection , it is either channeled into parallel functional segregated circuits ( Alexander et al . , 1986; Bordi and LeDoux , 1992; Bordi et al . , 1993; Parent and Hazrati , 1995; Middleton and Strick , 2000 ) or integrated within these circuits ( Nauta and Domesick , 1984; Hoffer and Alloway , 2001; Kolomiets et al . , 2001; Reig and Silberberg , 2014 ) . The existence of direct inhibitory cortical inputs to the dorsal striatum is likely to have profound consequences for corticostriatal mechanisms by which the AC regulates the output of the striatum and how auditory information is transformed into motor command ( Znamenskiy and Zador , 2013 ) . It is very well established that the dorsal striatum is characterized by activity-dependent plasticity ( for review , see [Kreitzer and Malenka , 2008] ) . Particularly , striatal plasticity can alter the transfer of information throughout the striatum and may serve as a neuronal substrate for the transformation of sensory information and motor planning for flexible motor actions and procedural memory . Recently , Xiong et al . ( Xiong et al . , 2015 ) found that auditory corticostriatal projections are involved in the learning of an auditory frequency discrimination task . They found that neurons in the dorsal striatum of the rat are selectively potentiated by auditory corticostriatal neurons to promote the learned transformation of sounds into motor actions . Moreover , Chen et al . ( Chen et al . , 2015 ) demonstrated that motor learning is characterized by structural rearrangements of presynaptic boutons of cortical SOM interneurons . Further experiments are needed to understand the functional significance of this direct cortical inhibition to the dorsal striatum , but in combination with the two above mentioned studies it invites speculation that CS-SOM neurons might play a role in mechanisms underlying spine dynamics by inducing a specific reorganization of dendritic excitatory synapses on SPNs and have consequences on learning and/or memory retrieval . Our results establish a previously unknown corticostriatal long-range inhibitory circuit ( CS-SOM inhibitory projections → striatal SPNs ) underlying the control of spike timing/generation in striatal spiny neurons and attribute a specific function to a genetically defined type of cortical neuron in corticostriatal communication . We have shown that the dorsal striatum receives not only glutamatergic excitatory inputs from the cortex , but also inhibitory inputs . This may suggest that the timing and ratio of excitation and inhibition , two opposing forces in the mammalian cerebral cortex , can dynamically affect the output of the dorsal striatum , providing a general mechanism for motor control driven by sensory stimuli . The following mouse lines were used in this study: SOM-Cre: Ssttm2 . 1 ( cre ) Zjh/J [The Jackson Laboratory , stock number 013044]; ROSA-tdTomato reporter: B6 . CG . Gt ( ROSA ) 26Sortm14 ( CAG-tdTomato ) Hze/J [The Jackson Laboratory , stock number 007914]; Drd1a-tdTomato/Drd2-GFP reporter: The two reporter lines , B6 . Cg-Tg ( Drd1a-tdTomato ) 6Calak/J [The Jackson Laboratory , stock number 016204; ( Ade et al . , 2011 ) ] and Tg ( Drd2-EGFP ) S118Gsat/Mmnc [MMRRC , stock number 000230-UNC] were crossed to generate a reporter for both D1- and D2-expressing neurons . A male mouse positive for both Drd1a-tdTomato/Drd2-EGFP was provided to us by Dr . Carlos Paladini at UTSA . SOM-Cre female mice were crossed with a ROSA-tdTomato reporter male mouse to generate a SOM-Cre-tdTomato line ( somatostatin-containing neurons expressed both Cre and tdTomato ) . SOM-Cre female mice were crossed with a Drd1a-tdTomato/Drd2-GFP reporter male mouse to generate a SOM-Cre-D1/D2 line ( somatostatin-containing neurons expressed Cre; D1 neurons expressed tdTomato; D2 neurons expressed GFP ) . CS-SOM neurons transfected with ChR2 showed ChR2-tdTomato-positive axons in the dorsal striatum related to the cortical area where the virus was injected ( auditory or motor ) . Because of variability both in ChR2 expression levels ( number of ChR2 molecules per transfected neuron ) and transfection efficiency ( number of ChR2-expressing neurons per animal ) , to minimize the variability from experiment to experiment we performed the electrophysiological recording in the same dorsal striatum slices ( identified by specific landmarks as slice 1 and 2 ) and with the highest density of ChR2 transfected axons . We recorded IPSCs from putative spiny projection neurons ( SPNs ) in the dorsal striatum during photoactivation of the CS-SOM ChR2-positive axon terminals . A 470nm wavelength blue LED ( CoolLED pE excitation system ) passed through a GFP filter cube ( Endow GFP/EGFP longpass , C-156625; Chroma ) and a 60X water-immersion objective was used to photoactivate CS-SOM ChR2-positive axon terminals . We recorded from putative SPNs in the dorsal striatum in an area containing ChR2-tdTomato-positive CS-SOM axons . In current-clamp configuration , a step of current was injected to cause the striatal neuron to fire 2–7 action potentials . To determine the effect of CS-SOM projections on the output of striatal neurons , we photoactivated CS-SOM ChR2-positive axons by flashing blue light ( 470 nm ) for 2–10 ms starting 10–50 ms before the first action potential . Combining current injection with photoactivation of CS-SOM projections delayed the current-evoked action potentials in striatal neurons . The action potential delay due to the combined current injection with photoactivation of CS-SOM projections was normalized to the onset of the first action potential measured during the current injection alone . D1- and D2-receptor-expressing neurons in the dorsal striatum were identified by the presence of fluorescent markers in the SOM-Cre-D1/D2 line of mice . Somatic expression of either tdTomato ( D1-receptor-expressing neurons ) or GFP ( D2-receptor-expressing neurons ) fluorescence was visualized under a 60X objective . During whole-cell recordings , neurons were filled with an internal solution containing 0 . 3–0 . 5% biocytin . Filled neurons were held for at least 20 min , and then the slices were fixed in a formalin solution ( neutral buffered , 10% solution; Sigma-Aldrich ) for several days at 4°C . The slices were washed well in PBS ( 6 times , 10 min per wash ) and placed in a 4% streptavidin ( Alexa Fluor 488 , 594 , or 680 conjugate; Life Technologies , Carlsbad , CA ) solution ( 498 μl 0 . 3% Triton X-100 in PBS , 2 μl streptavidin per slice ) . Slices were allowed to incubate in this solution at 4°C overnight , then washed well in PBS ( 6 times , 10 min per wash ) and mounted with Fluoromount-G ( SouthernBiotech ) on a glass microscope slide . Confocal images were taken with a Zeiss LSM-710 microscope at varying magnifications ( 3-63X ) . The identities of SPNs recorded in the dorsal striatum were confirmed by the presence of spines on their dendritic processes when imaged at 40-63X magnification . Individual high magnification images were stitched together , when necessary , using XuvStitch software ( XuvTools ) . Image adjustment was performed in ImageJ ( National Institutes of Health ) for brightness/contrast corrections and pseudocoloring . Some filled neurons were processed for light microscopy . In brief , after overnight incubation in ABC-Elite solution ( Vector Laboratories ) at 4°C , slices were pre-incubated in 3′3-diaminobenzidine ( DAB; Vector Laboratories ) for 20 min at 4°C and visualized by adding H2O2 to the DAB solution . The reaction was stopped when dendritic and axonal processes were visible under light microscopy examination . After washing well in 0 . 1 M PB ( 6 times , 10 min per wash ) , a DAB Enhancing Solution ( Vector Laboratories ) was applied for 10–20 s to intensify the DAB reaction product in the stained section . Slices were washed again in 0 . 1 M PB , then mounted with Fluoromount-G on a glass microscope slide for light microscopy . Neurons were morphologically reconstructed in three dimensions using Neurolucida ( MicroBrightField ) and an upright microscope fitted with a 63X/0 . 85 oil-immersion objective . Figure error bars represent SEM . Data analysis was performed offline using custom MATLAB ( MathWorks ) routines . Group comparisons were made using the Student’s t-test if data were normally distributed and the rank-sum test if not , with significance defined as p<0 . 05 .	The striatum is located beneath the cerebral cortex , where it contributes to processes including learning and movement . The Spanish anatomist Ramon y Cajal , working in the early 20th century , was the first to observe individual neurons extending from the cortex to the striatum . Cajal published drawings of these neurons in his now celebrated anatomical papers , but knew little about their properties . In the 1980s , advances in techniques for labeling individual cells made it possible to study these neurons in detail . The results suggested that the pathways are exclusively excitatory: that is , the cortical neurons always increase the activity of their partners in the striatum . However , this result made it difficult to explain why electrically stimulating the cortex can sometimes reduce or inhibit the activity of the striatum . To reconcile these facts , most people assumed that inhibition must occur when excitatory cortical neurons activate networks of inhibitory cells within the striatum itself . Rock et al . now challenge this view by providing anatomical and physiological evidence for the existence of long-range inhibitory pathways from the cortex to the striatum in the mouse brain . These inhibitory neurons project from the auditory and motor regions of the cortex , and contain a substance called somatostatin . These neurons form connections with a specific type of striatal neuron called medium spiny neurons , which in turn project to other brain regions outside the striatum . The inhibitory cortical neurons can alter the activity of the medium spiny neurons , and can therefore directly control the output of the striatum . The discovery that the striatum receives both excitatory and inhibitory inputs from cortex suggests that the timing and relative strength of these inputs can affect the activity of the striatum . Future experiments should examine whether this is a general mechanism by which sensory stimuli can influence the processes controlled by the striatum , such as movement .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	An inhibitory corticostriatal pathway
Small changes of Na/K pump activity regulate internal Ca release in cardiac myocytes via Na/Ca exchange . We now show conversely that transient elevations of cytoplasmic Ca strongly regulate cardiac Na/K pumps . When cytoplasmic Na is submaximal , Na/K pump currents decay rapidly during extracellular K application and multiple results suggest that an inactivation mechanism is involved . Brief activation of Ca influx by reverse Na/Ca exchange enhances pump currents and attenuates current decay , while repeated Ca elevations suppress pump currents . Pump current enhancement reverses over 3 min , and results are similar in myocytes lacking the regulatory protein , phospholemman . Classical signaling mechanisms , including Ca-activated protein kinases and reactive oxygen , are evidently not involved . Electrogenic signals mediated by intramembrane movement of hydrophobic ions , such as hexyltriphenylphosphonium ( C6TPP ) , increase and decrease in parallel with pump currents . Thus , transient Ca elevation and Na/K pump inactivation cause opposing sarcolemma changes that may affect diverse membrane processes . Na/K pumps establish the Na and K gradients used for ion transport and electrical signaling in animal cells ( Skou , 1990 ) , thereby accounting for a large fraction of total energy expenditure ( Milligan and McBride , 1985 ) . In cardiac myocytes , small changes of pump activity have large effects on excitation-contraction coupling by shifting the energetic equilibrium of Na/Ca exchangers ( Reuter et al . , 2002; Hilgemann , 2004 ) . Accordingly , regulatory mechanisms that control Na/K pump activity are of high biological significance . Surprisingly , cardiac Na/K pump regulation remains rather enigmatic . The FXYD protein , phospholemman , can inhibit Na/K pumps with analogy to phospholamban action at SERCA Ca pumps , and this inhibition is proposed to be released when phospholemman is phosphorylated by cAMP-dependent protein kinases ( PKAs ) ( Bibert et al . , 2008; Han et al . , 2010; Mishra et al . , 2015 ) . Interestingly , phosphorylation of phospholemman by PKAs appears to increase the affinity of pumps for cytoplasmic Na , while phosphorylation of phospholemman by protein kinase Cs ( PKCs ) increases maximal pump activity ( Han et al . , 2006 ) . These effects appear robust in optical experiments with Na-sensitive dyes , but without control of membrane potential , while the equivalent results in patch clamp-controlled experiments are remarkably inconsistent . During whole cell patch clamp of cardiac myocytes , activation of PKAs has been reported to duplicate results with dyes ( Despa et al . , 2005 ) , to mildly stimulate pump currents under all experimental conditions ( Kockskamper et al . , 2000 ) , to have no effect ( Ishizuka and Berlin , 1993; Main et al . , 1997; Fine et al . , 2013 ) , or to inhibit pump currents via an oxidative mechanism ( Galougahi et al . , 2013 ) . One analysis suggests that phosphorylation by PKA inhibits pump activity in the absence of Ca , but overcomes an inhibitory action of Ca and thereby enhances pump activity , albeit modestly , in the presence of cytoplasmic Ca ( Gao et al . , 1996 ) . Results for PKCs are similarly complex and presumably reflect complexities of PKC signaling in cardiac myocytes . Both small stimulatory effects ( Gao et al . , 1999; Han et al . , 2006 ) and inhibitory effects ( White et al . , 2009 ) are reported , the latter occurring through oxidative stress signaling mechanisms . Na/K pumps are evidently regulated by trafficking mechanisms in some cell types ( Al-Khalili et al . , 2003; Liu and Shapiro , 2007; Lecuona et al . , 2009; Alves et al . , 2015 ) , but there is little support for physiological regulation of cardiac pumps by these mechanisms . Nonconventional endocytic mechanisms can remove pumps from the sarcolemma during reperfusion injury ( Lin et al . , 2013 ) . Other mechanisms proposed to regulate the activity of Na/K pumps in cardiac myocytes include redox-dependent glutathionylation of the beta subunits of Na/K pumps ( Liu et al . , 2012 ) and the modulation of phospholemman function by palmitoylation ( Tulloch et al . , 2011 ) . Given this background , we initiated a new analysis of pump regulation in murine myocytes , starting with the observation that Na/K pump currents can run down during whole-cell patch clamp experiments . Attempting to stop and/or reverse this run-down , it became evident that a brief elevation of cytoplasmic Ca via reverse Na/Ca exchange had substantially larger effects on pump activity than the activation of either PKAs or PKCs . As described here , the stimulatory effects of transient Ca elevation , usually associated with spontaneous beating , occur as an apparent increase of the cytoplasmic Na affinity of Na/K pumps , the same functional effect proposed to occur with PKA activation ( Despa et al . , 2005 ) . Subsequent to a transient Ca elevation , peak pump currents activated by extracellular K are strongly enhanced , and the current decay that occurs during continued K application is attenuated . This decay has been ascribed previously to depletion of cytoplasmic Na in a restricted space ( Fujioka et al . ; Su et al . , 1998; Verdonck et al . , 2004 ) , and we explore here the possibility that current decay reflects instead , or additionally , an inactivation mechanism that bears similarity to the Na-dependent inactivation of Na/Ca exchangers ( Hilgemann et al . , 1992b ) . In contrast to Na/Ca exchangers , inactivation of Na/K pumps would not occur when cytoplasmic binding sites are occupied by Na . Rather , it would occur when one or more Na binding sites are unoccupied , while recovery from inactivation would be promoted by Na binding . We do not discern any effect of PKA activation in the absence of Ca elevations and the stimulatory effects of Ca elevation occur similarly in phospholemman-deficient myocytes . Further results suggest that physical changes of the membrane itself may be involved , rather than classical signaling mechanisms , and that Na transporters in close vicinity to one another may interact functionally by modifying physical properties of the adjacent bilayer . To facilitate the presentation of experiments and their interpretation , we present first in Figure 1 a cartoon of the Na/K pump cycle as it is widely thought to occur , together with our inactivation hypothesis: The normal cycle consists of a series of reactions in which 3 Na ions are bound from the cytoplasmic side when pumps are in the 'E1' configuration ( i . e . with binding sites open to the cytoplasmic side ) , followed by phosphorylation of the pump , Na occlusion and deocclusion to the outside in the 'E2' configuration , binding of 2 extracellular K ions , occlusion of the K ions , dephosphorylation of the pump , and deocclusion of K to the cytoplasmic side in the E1 configuration with renewed ATP and Na binding . We hypothesize that Na/K pumps in intact murine myocytes can enter into long-lived inactive states when they are in E1 configurations . Specifically , we hypothesize that pumps inactivate preferentially when the Na-selective binding site ( Kanai et al . , 2013; Vedovato and Gadsby , 2014 ) is not occupied by Na , while recovery from inactivation may be favored by Na binding to all sites . Inactivation could in principle be a process that hinders Na binding and/or hinders pump phosphorylation , therefore preventing the completion of a forward pump cycle . Transient cytoplasmic Ca elevations may act to attenuate Na/K pump inactivation via membrane modifications mediated by lipid-modifying enzymes or by long-lived conformational changes of high-density membrane-associated proteins . , 10 . 7554/eLife . 19267 . 003Figure 1 . Reaction scheme of the Na/K pump with inactivation from E1 states . E1 states are open to the cytoplasmic side and can bind 3 Na ions , Site III being a Na-selective site . Upon phosphorylation the Na ions are occluded and released to the outside in E2 states , followed by binding and occlusion of 2 K ions . Upon dephosphorylation pumps open to the cytoplasmic side in the E1 configuration and release K ions . Hypothetically , Na/K pump current decay occurs as an inactivation reaction taking place preferentially ( but probably not exclusively ) from 'E1' states in which the Na-selective ion binding site is not occupied by Na . Inactivation might occur with a lower probability from other states , including E2 states . Recovery from inactivation may be promoted by Na binding to inactive exchangers . DOI: http://dx . doi . org/10 . 7554/eLife . 19267 . 003 Figure 2 introduces the experimental conditions under which Na/K pump currents in murine cardiac myocytes are strongly activated by brief elevations of cytoplasmic Ca . Results in Figure 2A to D are at 37°C , while results in Figure 2E and F are at 23°C . The extracellular solution contains predominantly N-methyl-d-glucamine ( NMG ) as monovalent cation , and aspartate ( Asp ) is the major anion on both sides . Details of solution compositions , which minimize other currents , are given in Experimental Methods . Pump currents are activated and deactivated by moving the myocytes quickly back and forth between solution streams containing 7 mM Na and 7 mM K . By employing equal monovalent cation concentrations , changes of monvalent cation leak currents are minimized . As will be described in Figure 4 , we have verified that the concentrations of K which contaminate the solutions routinely employed are adequate to activate substantial pump activity in nominally Na- and K-free solutions ( Rakowski et al . , 1989 ) . Furthermore , it will be demonstrated in Fig . 4 that the 7 mM Na employed in our standard extracellular solution is adequate to effectively block this activity . 10 . 7554/eLife . 19267 . 004Figure 2 . Basic properties of Na/K pump currents in murine cardiac myocytes . Solutions contain 20 mM TEA on both sides to minimize K currents , 4 mM Mg on the outside to promote seal formation without Ca , and NMG as the major extracellular cation . Pump currents are initiated by replacing 7 mM Na for 7 mM K on the extracellular side . ( A ) Outward Na/K pump currents in the presence of 110 mM cytoplasmic Na and no cytoplasmic K are large ( >6 pA/pF ) and stable . Capacitance decreases and increases by ~1% within solution switch times when K is applied and removed , respectively . ( B ) Outward Na/K pump currents in the presence of 25 mM cytoplasmic Na and 90 mM cytoplasmic K decay over several seconds by more than 50% . Membrane capacitance decreases immediately during K application , as just described , and can then continued to decrease slowly but to a lesser extent during current decay; capacitance recovers slowly over 20 s after removing extracellular K . Typical for rodent Na/K pumps , currents are quite resistant to heart glycosides but can be rapidly and effectively blocked by a high concentration of ouabain ( 0 . 3 mM ) . ( C and D ) Cytoplsmic Na dependence of peak and steady state ( 10 s ) pump currents at 37°C with and without 90 mM cytoplasmic K . ( C ) In the presence of 90 mM K , the K50 for peak current is 19 mM Na and the Hill coefficient is 2 . 4 , while the 10 s current shows a K50 of 30 mM and Hill slope of 2 . 6 . ( D ) In the absence of cytoplasmic K , peak currents are best described by a Hill equation with a K50 of 6 mM cytoplasmic Na and a Hill slope of 1 . 6 . The 10 s current is shifted to a K50 of 37 mM with a Hill slope of 1 . 1 Although currents are smaller in the presence of cytoplasmic K , the fractional decay of current is greater at all Na concentrations . ( E ) Cytoplasmic Na dependence of peak and quasi steady state ( 15 s ) pump currents at 23°C without cytoplasmic K . Data points are the average of values from at least 5 separate myocytes . ( F ) The Na dependence of the fractional decay of pump current ( 1–15 s/peak ) is well described by the occupancy of a single Na binding site with an apparent Kd of 12 mM . The solid and dotted lines in panels E and F are predictions for overtly simplified Na depletion and inactivation models , described in Materials and methods . DOI: http://dx . doi . org/10 . 7554/eLife . 19267 . 004 Figure 2A is a representative current record when pumps are maximally activated by cytoplasmic Na ( 110 mM ) and extracellular K ( 7 mM ) . Maximal pump currents are at least 5 pA/pF . The pump currents typically activate within solution switch times ( ~100 ms ) and remain nearly constant during application of extracellular K . In contrast , Figure 2B is a representative current record using a cytoplasmic solution with 90 mM K and 25 mM Na . In the presence of cytoplasmic K and with this lower cytoplasmic Na concentration , the pump current activated by extracellular K ( 7 mM ) amounts to 2 . 4 pA/pF and immediately begins to decline , approaching a steady state over 10 to 15 s . The decaying current component typically amounts to more than one-half of the peak current . Importantly for our subsequent analysis , membrane capacitance typically decreases abruptly by 0 . 5 to 1% upon application of extracellular K and then declines further , but to a lesser extent , as pump current falls during K exposure . Upon removal of K , the capacitance signal returns toward its baseline with a slow multi-second time course that mimics in our experience the availability of pump currents for activation by renewed application of extracellular K . To test for sensitivity to heart glycosides , the myocyte was moved into two separate solution streams containing ouabain at a concentration that was high enough ( 0 . 3 mM ) to effectively block rodent Na/K pumps ( Herzig and Mohr , 1984 ) . The currents were very effectively inhibited by ouabain ( n = 5 ) , demonstrating that they indeed reflect Na/K pump activity . Importantly , the transient capacitive signals , as well as pump currents , are ablated by ouabain , indicating that they likely arise from Na/K pumps . As illustrated in the cartoon in Figure 1 , the absence of extracellular K will cause pumps to accumulate nearly exclusively in the E2 conformation because the E2 to E1 transition is blocked . Application of extracellular K then allows pumps to cycle between E2 and E1 conformations . However , all pumps should return to the E2 configuration immediately upon removing extracellular K . As evident in Figure 2A , the capacitive signal returns immediately to the pre-K value when cytoplasmic Na is high and pump currents are stable . However , the capacitive signals do not immediately return to baseline after removing extracellular K when cytoplasmic Na is submaximal , as in Figure 2B . This a first indication that pumps may become locked into states that cannot cycle during the time in which currents decay . In contrast , the idea that subsarcolemmal Na decreases during pump activity provides no explanation for this pattern . Concerning the source of these capacitance signals , we have shown previously that in the presence of low extracellular Na concentrations ( <20 mM ) , the rapid binding of Na to E2 configurations of the pump gives rise to significant capacitive signals that dissipate when pumps transition from the E2 to the E1 configuration ( Lu et al . , 1995 ) . We verified that Na binding is a major component of these signals as follows: Measurements were initiated with the standard extracellular solution containing 20 mM Na . Upon switching extracellular Na from 20 mM to lower concentrations and back to 20 mM , capacitance reversibly decreased and increased within solution switch times by about 0 . 5% , as in Figure 2A ( 5 observations ) . A significant but smaller involvement of both protons and contaminating K ions remains a possibility . Also , small contributions from still other sources are required to explain the small , gradual declines of capacitance that sometimes occur in the presence of K ( e . g . in Figure 2B , as compared to Figure 3B ) . 10 . 7554/eLife . 19267 . 005Figure 3 . Quantitative analysis of pump current decay in relation to predictions for cytoplasmic Na depletion . ( A ) The initial rate of decline of current should be predictable from the current magnitude , the number of Na moved per charge moved ( Na/e = 3 ) , the slope of the Na-current relation ( 'Hill slope' , ~2 ) , the cytoplasmic volume in which Na mixes , and the cytoplasmic Na concentration . Alternatively , the apparent cell volume can be calculated from the other equation parameters . ( B ) Pump currents at 37 and 23°C with 25 mM Na and 90 mM cytoplasmic K with current decay fitted to single exponential functions . ( C ) Composite results from experiments with 25 mM cytoplasmic Na at 37°C and 23°C with 90 mM K Na , at 37°C without K , and at 37°C with 40 mM Li and no K . Pump currents are four-fold smaller at 23°C , but the fractional decay is similar; currents are increased by removal of K , but fractional current decay is decreased; and currents are small with a high Li concentrations , while fractional current decay is increased . All results are inconsistent simple depletion models . DOI: http://dx . doi . org/10 . 7554/eLife . 19267 . 005 Figure 2C and D present the Na dependence of peak and quasi steady state ( 10 s ) Na/K pump currents in the presence and absence of 90 mM K . A somewhat larger Na range was used in experiments with K because the apparent Na affinity is lower than without K . Results were obtained one-myocyte and one-Na-concentration at a time . Peak and 10 s currents were averaged 3 to 4 times in individual myocytes , and the average individual myocyte results were then averaged from at least 4 myocytes under the same conditions for each data point presented . The composite results were fit to Hill functions , and values for the Hill parameters are given in the graphs . The results are similar to an equivalent data set presented by Su et al . ( Su et al . , 1998 ) , and the shift of Na dependence is reminiscent of changes that occur in response to rapid electrical stimulation in skeletal muscle ( Clausen , 2013 ) . The absence of cytoplasmic K shifts the half-maximal Na concentration of the peak currents from 19 to 6 mM , while the 10 s values shift from 30 to 37 mM . Accordingly , the peak pump current densities are substantially larger at submaximal cytoplasmic Na concentrations in the absence of cytoplasmic K . The Hill coefficient are substantially larger with cytoplamsic K ( 2 . 4 and 2 . 6 for peak and 10 s currents ) than without cytoplasmic K ( 1 . 6 and 1 . 1 for peak and 10 s currents ) . As noted in the Introduction , multiple studies suggest that the decay of Na/K pump currents during continued application of extracellular K in cardiac myocytes is caused by depletion of submembrane cytoplasmic Na . Assuming that diffusion is relatively temperature-independent , whereas pump activity is strongly temperature-dependent , we analyzed the dependence of peak and quasi steady state ( 15 s ) pump currents at 23°C with the expectation that current decay via Na depletion would be reduced . The experiments were perfomed without cytoplasmic K and are presented in Figure 2E . At the lower temperature , pump currents are at least 4-times smaller than at 37°C . Fitting the composite results for peak and 15 s currents to Hill functions , the half maximal pump currents occurred at 4 . 1 and 21 mM Na , respectively , and Hill slopes were 1 . 7 and 1 . 1 , respectively , for peak and 15 s currents . The overall pattern of current decay is not impressively different from results at 37°C with four-fold larger current . The fractional decay of pump currents , plotted in Figure 2F , increases monotonically with decreasing Na concentration and is reasonably described by a simple reciprocal hyperbola , Kd/ ( Kd+Nai ) , with a Kd of 12 mM . Instead of presenting Hill plots for this data set , we plot in both Figure 2E and F predictions from overtly simplified simulations of a depletion model ( dotted lines ) and an inactivation model ( solid lines ) that are described in Materials and methods . Clearly , the two postulates can account for these functional details equally well . Figure 3 presents multiple results that are not readily consistent with a Na depletion model , and Figure 3A highlights the major variables that would determine Na depletion in a restricted diffusion model . The initial rate of pump current decline ( s−1 ) will depend on the number of Na ions extruded ( i . e . the current magnitude in nA ∙ 6 ∙109times 3 , the number of Na ions that move per elementary charge moved ) , the slope of the Nai concentration-current relationship ( ~2 ) , the volume of the cytoplasm in which Na mixes , and the initial concentration of cytoplasmic Na . Rearranging the equation , one can calculate the apparent volume of cytoplasm in which Na is putatively depleting . As described next , values obtained for the cytoplasmic Na mixing volume are about 10% of the expected cytoplasmic volumes of murine myocytes ( Stagg et al . , 2004 ) , and the apparent exchange times to this restricted volume are on the order of 2 to 3 s . First , we performed a comparison of pump currents at 37° and 23°C with 25 mM Na and 90 mM K on the cytoplasmic side . Figure 3B presents individual experimental records together with their fits to single exponential functions . The fitted functions were used to calculate the fraction of current that decays ( Fdecay ) and the apparent Na mixing volumes . As tabulated in Figure 3C , the peak Na/K pump currents were about five-fold smaller at 23°C than at 37°C . The fractional decay of current was however very similar ( 0 . 67 versus 0 . 65 ) , while the decay time constant was larger and the apparent Na mixing volume was smaller at the lower temperature . Also tabulated in Figure 3C are two further results that seem inconsistent with the Na depletion hypothesis . First , results for pump currents in the absence of cytoplasmic K are compared to results with cytoplasmic K . With 25 mM cytoplasmic Na , peak currents are increased by about 30% , whereas the fractional decay of current is decreased from 0 . 65 to 0 . 52 . Second , the final column of the table in Figure 3C presents results of experiments using cytoplasmic lithium ( Li ) as a low affinity surrogate for cytoplasmic Na ( Hemsworth et al . , 1997; Hermans et al . , 1997 ) . With 40 mM Li in the pipette solution ( 30 mM TEA , 70 mM NMG , and no K ) , small pump currents ( <0 . 5 pA/pF ) could be activated by application of extracellular K . These currents showed unusually strong decay ( >90% ) in spite of their small magnitude , such that the apparent mixing volume of Li would be substantially less than a picoliter . This result is consistent with the inactivation model because more binding sites will be empty with a low affinity pump ligand , thereby promoting inactivation . In many other cases , we have observed that the fractional decay of pump currents increases , rather than decreases , as currents become smaller , and several clear examples will be considered with data presented in subsequent figures . The kinetics of current decay are an additional source of information that can help distinguish between models . In equivalent experiments using rat myocytes , Verdonck et al . described that as cytoplasmic Na is increased from 0 to 100 mM , the rate constant that best describes pump current decay increases to a maximum with nearly the same Na dependence as pump current itself ( Verdonck et al . , 2004 ) . This result seems inconsistent with the Na depletion model . If current decay were occurring as a result of Na depletion , the rate constant should have decreased as Na was increased above the half-maximal Na concentration of 8 mM and up to 100 mM . Furthermore , the fractional decay of current should have decreased markedly at the higher Na concentrations . Rather , the result supports an inactivation model in which the Na binding to inactive states promote recovery from inactivation . The kinetic results from our experiments were somewhat more complex: In the experiments described in Figure 2 at 23°C , the initial rate of pump current decay increased as cytoplasmic Na was decreased from 0 . 06 ± 0 . 02 s−1 at 40 mM Na to 0 . 14 ± 0 . 02 s−1 at 2 . 5 mM Na . This favors our suggestion that inactivation occurs preferentially from E1 states that are not fully occupied by Na . Nevertheless , as evident subsequently in Figure 6B at 37°C , current decay can also accelerate as the fractional current decay becomes smaller at high cytoplasmic Na concentrations . One specific prediction for the inactivation model was suggested by our previous work on Na-dependent inactivation of Na/Ca exchangers: The presence of either extracellular Na or Ca promotes exchanger transport reactions that move transporters from the equivalent E2 configurations to E1 configurations by the inward transport of bound ions . In other words , both extracellular Na and Ca promote accumulation of E1 states and favor inactivation ( Matsuoka and Hilgemann , 1994 ) . When experiments are performed in the presence of cytoplasmic Na and in the absence of extracellular Ca and Na , transporters initially accumulate in the E2 configuration that does not allow inactivation . Then , when Ca is applied to activate reverse transport ( i . e . Ca influx ) , transporters shift to the E1 configuration and inactivation is promoted . When extracellular Na is applied before applying Ca , large fractions of the exchangers shift to the E1 configuration and inactivate without transporting Ca . That inactivation then becomes evident when Ca is applied without Na . For Na/K pumps the equivalent experiments will not normally be possible because the reverse pump reactions , which move bound Na within E2 to E1 configurations , require that for cytoplasmic Na release the pump is dephosphorylated and that ATP is synthesized . These reverse reactions require , in addition to extracellular Na , high cytoplasmic concentrations of ADP . As described in Figure 4 , appropriate experiments are readily possible to test this prediction by selecting conditions that promote cytoplasmic accumulation of ADP . In describing the effects of extracellular Na on Na/K pump currents , we document first that just 5 mM Na in the extracellular solution without K suppresses for the most part an outward current that is very likely activated by contaminating K in nominally Na-free solutions , similar to results for squid axons ( Rakowski et al . , 1989 ) . At the start of the experimental record , pump current was activated twice by moving the myocyte from a solution with 5 mM Na to one with 5 mM K . When 5 mM Na is replaced with 5 mM NMG , a small outward current develops that decays partially over 20 s , as expected for a pump current , and application of 5 mM K then activates a maximal pump current that is less than half of the previous pump currents . Return to a solution with 5 mM Na restores the pump currents . Application of 15 mM Na enhances pump currents only about 10% beyond peak currents obtained after incubation with 5 mM Na . We conclude therefore that 5 mM Na suppresses most of the pump activity supported by residual K in these solutions . As indicated within Figure 4A , the half-maximal Na concentration ( Kd ) for this pump current rescue was 2 . 3 ± 0 . 3 mM ( n = 6 ) . This was determined in each experiment by designating the peak current in Na-free solution as the baseline for a rectangular hyperbola whose Kd and maximum were then determined by peak currents at 2 higher Na concentrations . We also tested in 6 experiments the ability of tetraethylammonium ( TEA ) to rescue pump current , since TEA binds to E2 potassium sites without being transported ( Peluffo et al . , 2004 ) . As expected , 120 mM TEA rescued pump activity equally well as 120 mM Na . However , much higher TEA concentrations were required than Na concentrations . The effect of 20 mM was very small , and the concentration dependence was concave upward with no evidence of saturation . 10 . 7554/eLife . 19267 . 006Figure 4 . Extracellular Na rescues Na/K pumps from steady state inactivation but promotes inactivation when reverse transport reactions are enabled by ADP . ( A ) Employing standard solutions , pump currents were initially activated twice by replacing 5 mM Na with 5 mM K . Removal of Na without application of K activates a small current , and the pump current activated subsequently with 5 mM K is less than one-half of the control pump current . Thus , pumps inactivate in nominally Na free solution , presumably because pump activity supported by contaminating K is adequate to allow inactivation . Pump currents recover in the presence of 5 mM Na , and the application of 15 mM has very little further rescue effect . From 6 experiments , employing 5 , 15 and 120 mM Na , the half-maximal Na concentration required to rescue pump currents was 2 . 3 mM . ( B ) Under standard conditions , 120 mM Na was applied , instead of NMG , for 30 s before applying K in the absence of Na . Peak pump currents were increased by 13% on average , compared to currents activated after incubation with 7 mM Na . ( B ) Na/K pump inactivation is promoted by conditions that allow backward pump reactions from E2 to E1 states . With 10 mM deoxyglucose in the pipette solution , to promote ATP hydrolysis and generation of ADP by hexokinsaes , pump currents ran down slowly over several minutes . After application of 120 mM Na for 30 s , peak pump currents activated by K were decreased to nearly their 15 s value , as expected if inactivation had occurred in the presence of extracellular Na . This result verifies a key prediction of the inactivation model presented in Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 19267 . 006 Figure 4B illustrates the protocol used to test whether extracellular Na ( 120 mM ) can force pumps to inactivate . Equivalent to the experiment of Figure 4A , we first activated pump current as usual by replacing 6 mM Na with 6 mM K . Then , we switched to 120 mM Na ( instead of NMG ) for 40 s and reactivated the pump current with 6 mM extracellular K . The potentiation of current by 120 mM Na versus 7 mM amounted to 13 ± 3% , a change that is in good agreement with the Kd of 2 . 3 mM . Thus , the activation of pump currents that occurs with increasing extracellular Na ( Garcia et al . , 2013 ) can be accounted for entirely by Na competition with contaminating K , thereby rescuing pumps from steady state inactivation . . Figure 4C illustrates , using the same protocol , that pump inactivation can be strongly promoted by extracellular Na under conditions that promote generation of cytoplasmic ADP , thereby allowing the reverse pump reactions to occur to the E1 state . Initially , we simply included 5 mM ADP in the pipette solution , together with 5 mM ATP . Peak pump currents then decreased significantly , but modestly , when extracellular Na was applied before pump current activation , as in Figure 4B . In this connection , we report that changes of nucleotides ( e . g . use of 0 . 5 mM ATP versus 12 mM ) in the pipette solutions in general had very little influence on pump currents , and in fact pump currents remained stable for periods of minutes with no added ATP . Clearly , ATP regenerating systems are very powerful in myocytes , and we therefore considered other means to decrease the ATP/ADP ratio . We included 10 mM deoxyglucose in the pipette solution with the expectation that hexokinases would rapidly phosphorylate deoxyglucose and thereby increase cytoplasmic ADP ( Russell et al . , 1992 ) . As apparent in Figure 4C , Na/K pump currents then ran down progressively over 5 min . After substitution of 120 mM extracellular NMG by 120 mM Na for 40 s , the peak pump currents activated by K in the absence of extracellular Na were decreased by 35% on average . Furthermore , currents activated during application of extracellular K showed very little decay , as expected if inactivation had already occurred in the presence of extracellular Na . We report finally that similar results were often obtained without deoxyglucose when myocytes were employed more than 4 hr after isolation , consistent with a time-dependent run-down of ATP-regenerating systems . From studies outlined in the Introduction , we expected that the ‘standard’ conditions of these experiments would be optimal to observe Na/K pump regulation by PKAs , and Figure 5 presents results for activation of both PKAs and PKCs . Pump currents were activated , as usual , by replacing 7 mM Na for 7 mM K on the extracellular side in the presence of 25 mM Na and 90 mM K on the cytoplasmic side . As shown in Figure 5A , currents decayed by ~70% with a time constant of ~ 2 s . As evident in the second half of the recording , the β-adrenergic agonist , isoproterenol ( 0 . 5 μM ) , applied by shifting the myocyte to two separate solution lines containing the drug , caused no evident change of the activation-decay pattern or current magnitudes . Figure 5B shows composite results for isoproterenol in individual myocytes from 5 different myocyte preparations . Similar to results for isoproterenol , we also did not observe significant effects of the adenylate cyclase activator , forskolin ( 20 μM ) . Furthermore , no significant changes of the current decay and run-down patterns were observed when cAMP ( 2 mM ) was included in the pipette solution . These results reiterate negative results using human iCell myocytes ( Fine et al . , 2013 ) . In contrast to results for PKAs , we did observe small stimulatory effects of the PKC activator , rac-1-oleyl-2-acetyl- glycerol ( OAG , 5 μM ) on peak pump currents ( ~25% ) , similar to results of others ( Han et al . , 2010 ) . These effects are however small in comparison to stimulatory effects of transient Ca elevations described next . 10 . 7554/eLife . 19267 . 007Figure 5 . Stimulation of Na/K pump currents by activating protein kinase A ( PKA ) and protein kinase C ( PKC ) is weak or absent . ( A ) Under standard conditions , Na/K pump currents were activated repeatedly by applying extracellular K , and then isoproterenol ( 0 . 5 μM ) was applied by moving the myocyte to separate solution lines containing the drug . ( B ) From left to right , composite results are presented for equivalent experiments in which isoproterenol ( 0 . 5 μM ) or forskolin ( 20 μM ) was applied to activate PKAs , in which 2 mM cAMP was included in pipette solutions , and in which 1-oleoyl-2-acetyl-sn-glycerol ( OAG , 5 μM ) was applied to activate PKCs . Only OAG has a significant effect to weakly increase the peak Na/K pump currents . DOI: http://dx . doi . org/10 . 7554/eLife . 19267 . 007 The weak Ca buffer power of our standard cytoplasmic solution ( 0 . 5 mM EGTA ) allows myocytes to readily initiate spontaneous contractions via Ca-activated Ca release . To do so , we activate Ca influx via reverse Na/Ca exchange current for 3 to 5 s by applying Ca ( 5 mM ) in the extracellular solution and then deactivate the current by returning myocytes to extracellular solution without Ca and containing 0 . 5 mM ETGA . Myocytes initiate a series of rapid contractions in this protocol , documented subsequently to involve Ca waves . Typically , contractile activity continues for a few seconds after exchange currents are deactivated . Contractions then stop and myocytes remain fully extended thereafter . Figure 6A illustrates the prominent after-effects transient Ca elevations when cytoplasmic Na is submaximal ( 25 mM with 90 mM K ) . During 15 to 40 s after the Ca elvation , membrane capacitance routinely increased by 3 to 6% . Within this same time frame , peak pump currents activated by extracellular K increased on average by 54% , and current decay during a 12 s K pulse was decreased from 70% to about 20% . Pump currents at 10 s were typically increased by three-fold . Over the course of 3 min , the Na/K pump currents returned to their initial wave form . Bar graphs to the right of the experimental record quantify representative results for 12 experiments , each using a myocyte from a different myocyte batch . We point out that these effects are the largest modulatory effects observed routinely on Na/K pump currents in the experience of this group . Further , we report that the myosin ii inhibitor , blebbistatin , at a high concentration ( 5 μM ) blocked the observed contractile activity but did not block these effects ( see Table 1 ) . Therefore , the stimulatory effects must be caused by Ca elevations per se rather than mechanical activity . 10 . 7554/eLife . 19267 . 008Figure 6 . Transient Ca elevations robustly activate Na/K pump currents when cytoplasmic Na is submaximal but do not enhance maximal pump currents ( A versus B ) . Na/K pump currents were activated repeatedly under standard conditions with 25 mM Nai ( A ) and with 40 mM Nai ( B ) for 15 s . Between two activation sequences , reverse Na/Ca exchange current was activated for 3 s by application of Ca ( 5 mM ) . After a few vigorous contractions , myocytes relaxed completely and then remained relaxed , while membrane capacitance typically increased by 2 to 5% over 30 s . With 25 mM Nai ( A ) , peak pump currents often increased by two-fold with an average of 54% , current decay during application of K was largely attenuated , and pump current at 10 s was increased three-fold on average . Thereafter , pump currents decreased nearly to baseline values with a time constant of ~150 s . With 40 mM Nai ( B ) , pump currents decayed rapidly by 19% on average , and this decay was fully attenuated after the Ca elevation . Peak currents were decreased by 18% on average after the Ca elevation . Average changes of peak and 10 s pump currents are given in bar graphs to the right of both recordings . DOI: http://dx . doi . org/10 . 7554/eLife . 19267 . 00810 . 7554/eLife . 19267 . 009Table 1 . Normalized results for interventions tested to modify pump current stimulation by Ca elevations using the standard protocol ( 25 mM Nai with 90 mM Ki ) . Currents are normalized to the peak current before application of Ca . Results are given for peak and 10 s currents before applying extracellular Ca , followed by the average fractional decay of that current , and for the first peak and quasi steady state ( 10 s ) current after applying Ca , followed by the average fractional decay of that current . See text for annotation of the major results . Abbreviations: PLM , phospholemman; BIM , Bisindolylmaleimide; DTT , dithiothreitol; SOD , Superoxide Dismutase . DOI: http://dx . doi . org/10 . 7554/eLife . 19267 . 009Pre-Ca2+Post-Ca2+Peak10 sFdecayPeak10 sFdecaynControl ( 25 mM Nai ) 10035 ± 20 . 65154 ± 12111 ± 100 . 28121 . Basic Mechanism Blebbistatin ( 5 μM ) 10044 ± 80 . 60138 ± 880 ± 150 . 406Thapsigargin ( 2 μM ) 10036 ± 50 . 74198 ± 70118 ± 500 . 414Ryanodine ( 2 μM ) 10034 ± 50 . 67154 ± 884 ± 100 . 4452 . Phospholemman ( PLM ) PLM -/-mice10074 ± 40 . 26125 ± 10114 ± 100 . 0993 . Membrane cytoskeleton Phalloidin , 3 μM10057 ± 100 . 43147 ± 25116 ± 200 . 195Latrunculin , 3 μM10037 ± 110 . 63135 ± 996 ± 220 . 2544 . Serine/threonine phosphorylation KN93 , 3 μM10054 ± 70 . 46121 ± 13112 ± 100 . 064CAMK Pep . , 15 μM10057 ± 60 . 43131 ± 993 ± 90 . 286CK59 , 20 µM10025 ± 60 . 78183 ± 3075 ± 20 . 544H7 , 0 . 3 mM10035 ± 40 . 65157 ± 1295 ± 100 . 396BIM , 4 μM 1 hr10027 ± 40 . 73131 ± 486 ± 50 . 346PKC19-31 , 10 μM10024 ± 10 . 77195 ± 58136 ± 460 . 349Cyclosporin A , 3 µM10048 ± 60 . 52126 ± 6101 ± 50 . 2555 . Redox and NOS signaling PRD6 -/- mice10033 ± 50 . 66189 ± 24134 ± 170 . 268Oxy-glutathione , 5 mM10031 ± 40 . 69195 ± 18120 ± 110 . 376Glutathione , 8 mM10038 ± 70 . 62130 ± 1489 ± 150 . 304DTT , 0 . 2 mM ( in & out ) 10028 ± 30 . 73148 ± 31100 ± 350 . 355SOD , 3 mg/ml10026 ± 20 . 74170 ± 18101 ± 120 . 675L-NAME , 0 . 1 mM10025 ± 60 . 75190 ± 41120 ± 270 . 3646 . Mitochondrial signaling CGP37157 , 20 μM10022 ± 40 . 78184 ± 2483 ± 150 . 554RU360 , 20 µM10044 ± 70 . 56111 ± 794 ± 150 . 1767 . Lipid metabolism PLD1-PLD2 -/-mice10034 ± 30 . 66114 ± 486 ± 90 . 259DHHC5-GT mice10031 ± 30 . 69115 ± 583 ± 50 . 276U73122 , 10 μM10039 ± 40 . 61165 ± 8117 ± 110 . 295Wortmannin , 10 μM10048 ± 60 . 51125 ± 8120 ± 100 . 0348 . Nonspecific membrane modifiers Genistein , 25 μM10016 ± 60 . 846 ± 00 ± 0--5Daidzein , 20 µM10040 ± 40 . 59104 ± 954 ± 50 . 486Quercetin , 100 μM10057 ± 110 . 4396 ± 5462 ± 310 . 315Methylene Blue , 3 μM10041 ± 40 . 5954 ± 1942 ± 150 . 228Triacsin C , 2 μM10043 ± 60 . 5743 ± 2131 ± 150 . 257 Figure 6B shows results for the equivalent experiments when cytoplasmic Na was nearly saturating for the activation of Na/K pump currents ( 40 mM with 90 mM K ) . Pump currents still routinely displayed a rapid but small decay phase , which was fully ablated after the Ca elevation . Bar graphs to the right of the experimental record give composite results for 8 similar experiments . We point out that the Ca elevation caused a small decrease of peak pump currents , on average , and we describe subsequently much larger inhibitory effects of repeated Ca elevations on pump currents . In summary , the results of Figure 6 document that the overall effect of transient Ca elevations is to increase the apparent affinity of Na/K pumps for cytoplasmic Na . To explain the effects of Ca elevation in terms of a Na depletion model , one might propose that Ca elevation enlarges the subsarcolemmal Na space or enhances the diffusion of Na through the restricted space . We demonstrate in Figure 7 , however , that much larger sustained Na fluxes can be generated by Na/Ca exchangers than are carried by Na pumps . For these experiments , we employed myocytes that overexpress NCX1 Na/Ca exchangers by at least five-fold in relation to WT expression levels ( Lin et al . , 2013 ) and that generate several nanoamperes of reverse exchange current in equivalent experiments . To limit the magnitude of Ca changes and to avoid excessive contraction , we employed heavily Ca- and pH-buffered pipette solutions without K ( 50 mM EGTA , 12 mM Ca , 50 mM HEPES , and 20 mM Na; resting free Ca , ~0 . 2 μM ) . Peak Na/K pump currents in these myocytes were not different in magnitude from WT myocytes ( ~2 pA/pF with 20 mM Na ) , but they decayed nearly completely . Peak exchange currents amounted to >20 pA/pF , and they decayed to a magnitude of 430 ± 33 pA after 20 s . Impressively , the maintained exchange current magnitudes approximate the maximal Na flux that can be expected for an access resistance of 3 MΩ and a pipette Na concentration that amounts to 7% of all diffusible ions ( i . e . 20 mM out of 300 mM with ion fluxes calculated from a constant field equation and assuming a 3Na/1Ca exchange stoichiometry ) . The total magnitude of Ca influx during each episode of exchange current amounts to more than 10 mmol/Liter of cytoplasmic space , assuming a cell volume of 30 pL . In response to these Ca elevations , peak Na/K pump currents doubled in magnitude and the 10 s current magnitudes increased from negligible values to 76 ± 6 pA ( n = 6 ) . Clearly , it becomes difficult to explain these current patterns in terms of a restricted subsarcolemmal Na space shared by Na/K pumps and Na/Ca exchangers . The total amount of Na moved during the initial Na pump current over 20 s is subseqeuntly moved by Na/Ca exchangers within just a few milliseconds . Therefore , one must assume that the Na space available to Na pumps is very small ( <1 pL ) , while that available to Na/Ca exchangers is very larger . Either the two transporters do not share the same Na space , or else Ca influx dramatically expands the space within milliseconds . 10 . 7554/eLife . 19267 . 010Figure 7 . Na/K pump and Na/Ca exchange currents in myocytes that overexpress NCX1 Na/Ca exchangers by more than five-fold . The cytoplasmic ( pipette ) solution is designed to highly buffer free Ca and protons . The magnitudes of exchange currents after 15 s are much greater than the 15 s pump currents . Thus , Na/K pumps and Na/Ca exchangers cannot be sharing a restricted subsarcolemmal space . Top panel . Complete current record in which Na/K pump currents and Na/Ca exchange curerrents were activated alternately . Bottom panel . Expanded view of the current record to accurately depict the Na/K pump current . Pump currents decay completely before activation of Na/Ca exchange . In response to Na/Ca exchange currents , peak pump currents double and a maintained current develops . These changes reverse completely over 3 min after the last Ca elevation . DOI: http://dx . doi . org/10 . 7554/eLife . 19267 . 010 Figure 8A presents optical measurements of cytoplasmic Ca changes that occur during two consecutive Ca elevation protocols using the protocol and conditions of Figure 6 . For these measurements myocytes were loaded with the AM ester of Fluor4 ( 2 μM ) ( Gee et al . , 2000 ) for 30 min . As necessary for optical measurements , the recordings were performed using mycoytes attached to cover slips , and solutions were changed with a device in which two solution lines shared a common volume outflow . Therefore , solution changes were not as abrupt as in routine experiments . Perhaps for this reason , or as a result of dye loading , myocytes did not visually contract as vigorously as in routine experiments . Nevertheless , the overall patterns were very similar . The whole-cell fluorescence is presented as the middle record in Figure 8A , and the upper trace is a time-expanded record for a cross-sectional optical slice taken roughly in the middle of the myocyte . The cross-section record documents the occurrence of profuse Ca waves which follow closely our routine observations of contraction . Ca waves become increasingly pronounced and rapid as the Ca signal declines . Thereafter , Ca waves decrease in magnitude and eventually in frequency . During application of extracellular Ca , the Fluor4 fluorescence increased on average 3 . 7 ± 0 . 4 fold ( n = 7 ) . This corresponds to a peak free Ca of 2 . 7 µM , calculated with Kd for Ca of 300 nM and a background free Ca of 0 . 13 μM . Free Ca decays with a time constant of ~15 s . Importantly , the cytoplasmic Ca changes can be reproduced with good precision after 3 min . In this experiment the first and second exchange currents were similar in magnitude , but we describe next that exchange currents can change substantially . We conclude from these measurements that during the routine Ca elevation protocols free Ca rises to at least 2 and probably 3 μM , and that Ca returns nearly to baseline within 15 to 30 s . 10 . 7554/eLife . 19267 . 011Figure 8 . Free cytoplasmic Ca changes and their effects on Na transport currents during repeated Ca elevations . ( A ) Free Ca measurements with Fluo4 . The center record presents the whole-cell free Ca signal measured with Fluo4 . Fluorescence increased by 3 . 7 ± 0 . 4 fold ( n = 7 ) during application of extracellular Ca and returned to baseline with a time constant of ~25 s . Correcting for dye nonlinearity , the fall of cytoplasmic Ca probably occurs with a time constant of ~15 s . The upper record shows in an expanded time scale fluorescence from a cross-sectional slice in the middle of themyocyte , revealing that extensive Ca waves are taking place in the myocytes . These waves become more rapid an large as free Ca declines , and finally they decline in both magnitude and frequency . ( B ) Representative for >400 observations , Na/Ca exchange current shows a time-dependent activation phase on first application of extracellular Ca . This secondary rise corresponds to the regulatory activation of exchangers that occurs as cytoplasmic Ca rises ( Matsuoka et al . , 1993 ) . On second application of Ca , the exchange current activates immediately and to a higher magnitude than on initial Ca application . Thereafter , pump current is increased beyond its peak after the first Ca elevation . ( C ) After three or more Ca elevations , and infrequently after the second repetition , both Na/K pump and exchange currents decreased and often became negligible . The loss of transporter currents was often accompanied by declining membrane capacitance during Ca elevations , indicative of massive endocytosis ( Lariccia et al . , 2011 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19267 . 011 Figure 8B illustrates that pump currents could usually be enhanced further by a second Ca elevation . As illustrated further in Figure 8B , Na/Ca exchange currents were often larger during the second application of Ca , and they activated immediately rather than with a multi-second time course . Importantly , the enhanced Ca-mediated current turned off immediately upon removal of extracellular Ca when the application period was brief . This behavior indicates that the current depends directly on extracellular Ca and therefore is indeed Na/Ca exchange current , rather than a cytoplasmic Ca-activated nonspecific current that would turn off with the slower time course of the cytoplasmic Ca decline . In conclusion , Ca-dependent mechanisms enhance both Na/K pump and Na/Ca exchange activities for periods of more than one minute after a transient Ca elevation . Potentially , therefore , the same mechanism might modulate both currents . It was important in these experiments to use rather short Ca influx episodes ( 3 to 5 s ) because longer Ca elevations , or repeated Ca elevations , caused pump currents to rapidly run down . When Ca elevations were repeated a second , third , and fourth time , pump currents decreased in magnitude with increasing likelihood and often became negligible . As illustrated in the record in Figure 8C , the outward Na/Ca exchange currents ran down in close parallel to pump currents . This parallel run-down was usually associated with a fall of membrane capacitance , reflecting Ca-activated endocytosis ( Lariccia et al . , 2011 ) . But even when the loss of membrane area was small , about 10% in Figure 8C , pump currents were strongly suppressed , usually ablated . Thus , transient Ca elevations have both facilitating and inhibiting effects on both Na/K pump currents and Na/Ca exchange currents . While the stimulatory effects invariably reversed over 3 min , reversal of inhibitory effects of repeated Ca elevations was highly variable . Since Na/Ca exchange currents change in parallel with pump currents in response to Ca elevation , we next tested whether Ca influx by mechanisms beside Na/Ca exchange could activate Na/K pump currents . Figure 9 demonstrates that Ca influx through Ca channels can indeed activate Na/K pump currents at a physiological cytoplasmic Na concentration ( 10 mM ) , and that cytoplasmic K can be replaced by cesium ( Cs ) in the protocol . To develop a large Ca current , we applied the Ca channel agonist FPL64176 ( 5 μM ) ( Baxter et al . , 1993 ) throughout the experiment . Application of 6 mM extracellular Ca then activated a robust Ca current that decayed within a few seconds . Thereafter , both peak and 15 s Na/K pump currents were increased by >50% in 5 similar observations . 10 . 7554/eLife . 19267 . 012Figure 9 . Stimulation of Na/K pump current in the presence of physiological ( 10 mM ) cytoplasmic Na and in the absence of cytoplasmic K . To evoke significant Ca influx in the presence of low cytoplasmic Na , FPL64176 was employed to lock L-type Ca channels open at 0 mV . On application of 6 mM extracellular Ca , inward Ca current amounted several hundred pA , and current inactivated with a time constant of about 1 s . Thereafter , Cm increased by 3% , similar to results with reverse Na/Ca exchange , and peak and 12 s pump currents were roughly doubled . Similar results were obtained in 4 similar experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 19267 . 012 Next , we attempted to determine by what molecular mechanism ( s ) Ca elevations enhance pump currents . Table 1 summarizes the extensive range of these experiments using the standard conditions and protocol of Figure 6 . All results are normalized to the peak current that occurred when pump currents were activated just before Ca was applied . From left to right , the columns give initial peak current ( 100% ) , the relative current magnitude after 10 s , and the fractional decay of current before the Ca elevation . These values are followed by the equivalent post-Ca values for peak current , current after 10 s , and the fractional decay at 10 s of the first pump current transient after the Ca elevation . The Control data ( 25 mM Nai with 90 mM Ki ) are averages from 12 representative experiments from 12 different batches of myocytes . Unless indicated otherwise , test compounds were included in both cytoplasmic and extracellular solutions . In short , we did not clearly implicate any classical signaling mechanism in the stimulation of Na/K pump activity by Ca elevations . Given our failure to clearly implicate classical signaling mechanisms , we next asked whether transient Ca elevations might act by changing physical-chemical properties of the sarcolemma that influence pump activity . We point out in this connection that large transient Ca elevations cause the plasmalemma of multiple cell types to become disordered , such that numerous hydrophobic compounds bind to the cell surface more avidly ( Hilgemann and Fine , 2011 ) . As described in Figure 1 , Ca elevations occurring in our standard protocol indeed enhance the binding of both hydrophobic cations and anions to the cardiac sarcolemma . Hexyltriphenylphosphonium ( HTPP ) is a hydrophobic cation that crosses membrane bilayers in the absence of transporters , owing to its hydrophobicity and delocalization of its positive charge ( see ( Hilgemann and Fine , 2011 ) for further explanations ) . Figure 10A presents parallel measurements of Na/K pump currents and inward currents generated by HTTP ( 20 µM ) , which was applied repeatedly for 3 s during the standard protocol . HTTP current amounted to −180 pA initially . After the transient Ca elevation , the peak Na/K pump current was doubled , and the HTTP current was increased to −270 pA ( i . e . by 50% ) . Thereafter , both the Na/K pump current and the HTTP current decreased toward baseline with similar 3 min time courses . Statistics for 5 experiments are given within Figure 10A . Notably , HTTP currents decreased below their initial values when Ca elevations were repeated and Na/K pump currents also decreased below baseline , as described in Figure 8C . Thus , both the stimulatory and inhibitory effects of Ca elevations are potentially occurring through changes of the membrane itself . 10 . 7554/eLife . 19267 . 013Figure 10 . Interactions of hydrophobic cations and anions with the sarcolemma track the stimulation and inhibition of pump currents by Ca elevations . ( A ) Na/K pump currents were activated repeatedly for 8 s and hexyltriphenylphosphonium ( C6-TPP , 15 μM ) currents were activated alternately with the pump currents for 3 s . Pump current increased and then decreased in the usual manner and rate in response to Ca elevation . Currents carried by C6-TPP increased by 39% on average and then decreased in parallel with pump currents . A second Ca elevation causes profound inhibition of pump current and a 60% decrease of C6-TPP current . As indicated , C6-TPP currents decreased on average by 62% during the time over which pump currents ran down almost completely . ( B ) Niflumic acid induces a capacitive binding signal , when applied at a concentration of 120 μM , which propably involves expansion of the membrane . In response to Ca elevation , the capacitive binding signal increases on average by 80% ( n = 5 ) . Thus , both cationic and anionic membrane probes bind more effectively after Ca elevations . DOI: http://dx . doi . org/10 . 7554/eLife . 19267 . 013 To test whether membrane changes caused by Ca transients might be electrostatic in nature , and therefore enhance specifically the binding of positively charged amphipaths , we examined capacitive signals that are detected when anionic amphipaths bind to cells and expand the membrane ( Fine et al . , 2011; Hilgemann and Fine , 2011 ) . Figure 10B shows results for the hydrophobic anion and Cl channel blocker , niflumic acid ( NFA ) . NFA causes a 1 to 3% increase of capacitance when applied at a concentration of 100 μM . This capacitive binding signal nearly doubled after transient Ca elevations that caused an approximate doubling of Na/K pump currents and a 5% increase of capacitance . Statistics for 5 similar experiments are given within Figure 10B . Given that both HTTP and NFA signals increase by at least 30 and 80% , respectively , while membrane capacitance increases by only 3 to 8% , we conclude that the membrane is becoming more disordered in response to Ca elevations . Next , we examined in similar experiments whether the run-down of Na/K pump currents , that is observed when deoxyglucose is included in the cytoplasmic solution ( see Figure 4C ) , also modifies C6-TPP currents . Using standard solutions ( 25 mM Nai ) with 10 mM deoxyglucose in Figure 11 , pump currents ran down completely in the course of 10 min , and the fractional decay of pump current ( Fdecay ) increased to essentially complete decay as currents decreased . We mention that pump currents did not run down significantly when 10 mM glucose-6-phosphate ( n = 4 ) or 10 mM deoxyglucose-6-phosphate was included in pipette solutions ( n = 3 ) . Therefore , the rapid phosphorylation of deoxyglucose by hexokinases , at the expense of ATP , is likely to be the cause of run-down in these experiments , rather than mechanisms downstream of deoxyglucose-6-phosphate . As summarized in the Table in Figure 11 , inward HTPP currents decreased in parallel with pump currents , on average by 69% . Furthermore , HTPP currents were stable when deoxyglucose was not included in the pipette , and HTPP currents decreased by 51% when the pipette solutions contained 140 mM K , instead of NMG , and no Na . Thus , the membrane changes that occur with deoxyglucose do not appear to depend on Na-dependent Na/K pump conformational changes . Rather , they appear to be caused by ATP/ADP changes , per se , exerting relatively rapid effects on proteins that in turn influence the bilayer . 10 . 7554/eLife . 19267 . 014Figure 11 . Parallel decay of pump currents and C6-TPP currents when deoxyglucose ( 10 mM ) is included in the pipette solution . Pump currents were activated for 10 s alternatively with C6-TPP currents for 2 s . Fractional decay of pump currents increases from 0 . 68 to > 0 . 9 as pump current declines , and the C6-TPP current decreases by 69% on average . The table beneath the figure gives statistics for equivalent experiments without deoxyglucose , with deoxyglucose but without Na , and with deoxyglucose and 25 mM cytoplasmic Na . DOI: http://dx . doi . org/10 . 7554/eLife . 19267 . 014 Hydrophobic fluorescent dyes have been used for decades to analyze electrogenic reactions of the Na/K pump ( Bühler et al . , 1991 ) . The interpretation of those experiments suggests that ion binding within the pump exerts long-range effects on the intramembrane electrical field that are sensed by voltage-dependent dyes some distance away . Conformational changes of the Na/K pump per se may also affect these dyes ( Fedosova et al . , 1995 ) . With this background , we tested whether conformational changes of Na/K pumps during pump current elevations might affect the currents carried by HTTP . As shown in Figure 12 , HTTP currents were indeed decreased by ~30% immediately after removing extracellular K to deactivate Na/K pumps . Furthermore , the HTTP currents routinely recovered with a time course that was similar to the recovery of pump current from inactivation . Note that the recovery of capacitance also followed this time course . In this context , we point out that Na/K pumps are present in the mouse sarcolemma at a high density . Pump density must be >2000 per μm2 to account for maximal currents of 4 pA/pF with maximal transport rates of ~300 s−1 ( Gadsby and Nakao , 1989 ) . Assuming that pumps have diameters of 3 to 4 nm , they will constitute 3 to 5% of the membrane surface area . It is therefore plausible that the conformational state of pumps could influence bulk physical properties of sarcolemma , in particular if pumps are interconnected by a protein network , such as cytoskeleton . 10 . 7554/eLife . 19267 . 015Figure 12 . C6-TPP currents decrease in response to the apparent inactivation of Na/K pumps . C6-TPP currents were repeatedly activated for 2 s , and Na/K pump currents were activated multiple times between a series of C6-TPP currents . The C6-TPP currents were decreased on average by 29% subsequent to pump activity and current decay , and the C6-TPP current recovered with a time course very similar to the recovery of pump activity and membrane capacitance . DOI: http://dx . doi . org/10 . 7554/eLife . 19267 . 015 The finding that hydrophobic cation currents are inhibited by Na/K pump activity , and recover as pump currents recover , is equivalent to results for Na/Ca exchange currents that support the occurrence of subsarcolemmal Na depletion ( Su et al . , 1998 ) . As verified in Figure 13A , and as expected for depletion of cytoplasmic Na , outward Na/Ca exchange current ( i . e . Na efflux driven by Ca influx ) is suppressed immediately after a period of high Na/K pump activity , and exchange current recovers with a time course that is similar to the recovery of pump currents . In 10 similar experiments , cytoplasmic free Ca was strongly buffered with 25 mM EGTA to 0 . 2 μM , and reverse Na/Ca exchange currents were activated repeatedly by applying extracellular Ca ( 5 mM ) for 2 to 3 s before and after activating Na/K pump currents . The extracellular solutions contained 25 mM Na to block pump current that might be activated by contaminating K . Exchange currents were decreased immediately after the pump current transient by a fractional amount that was about one-half of the fractional decay of pump current . Exchange currents then recovered with a time course that was consistently very similar to the time course with which Na/K pump current availability recovered . Based on pump currents , the apparent volume in which Na depletion would be taking place amounted to 2 . 2 ± 0 . 5 pL ( n=9 ) . 10 . 7554/eLife . 19267 . 016Figure 13 . Evidence for and against Na/K pump - Na/Ca exchange cross-talk via submembrane Na changes . ( A ) Standard cytoplasmic solution was modified to effectively buffer cytoplasmic free Ca to ~0 . 2 µM ( 25 mM EGTA + 6 mM Ca ) . Reverse Na/Ca exchange currents were activated repeatedly for 2 s before and after Na/K pump currents were activated for 15 s . Exchange currents , like C6-TPP currents in Figure 12 , are suppressed immediately after pump activity and recover over 20 s . ( B ) K-free cytoplasmic solution was buffered to contain 0 . 5 µM free Ca , thereby supporting a large , continuous inward exchange current ( i . e . Na influx and Ca extrusion ) . 50 mM EGTAi+ 35 mM Cai , 10 mM Nai , and 125 mM Nao . Inward exchange current was initially blocked , as indicated , by applying 4 mM Ni on the extracellular side . Pump current was then activated by extracellular K and exchanger block was released by removing Ni at the same time . Subsequently , Na/K pump current was activated once without and once with Na/Ca exchange block . The presence or absence of inward exchange current , before activation of pump current , does not affect the magnitude of peak pump currents . Composite results for 9 experiments are tabulated below the cartoon . ( C ) Predictions of a simple Na+ depletion model . Pump currents were simulated with a hyperbolic dependence on cytoplasmic Na+ ( K50 , 5 mM ) , and the cytoplasmic space was limited to a value of 3 pL to promote submembrane Na changes . When Na depletion during pump activity limits pump currents and suppresses reverse exchange currents , as in Panel A , forward exchange currents should very significantly enhance pump currents under the conditions of Panel B , and that is not the case . DOI: http://dx . doi . org/10 . 7554/eLife . 19267 . 016 While these results are highly consistent with Na depletion affecting Na/ Ca exchange currents , the previous results suggest that Na/K pump inactivation could affect Na/Ca exchange function by other mechanisms . Therefore , we present one more test for the existence of a restricted cytoplasmic Na space in myocytes in Figure 13B . In these experiments we tested whether Na influx by forward Na/Ca exchange can enhance Na/K pump currents , as expected for accumulation of Na in a space shared by exchangers and pumps . As described in Figure 13B , large inward Na/Ca exchange currents were generated using cytoplasmic solutions in which the free cytoplasmic Ca was heavily buffered to 0 . 5 µM ( 50 mM EGTA with 35 mM Ca ) . In this way , continuous inward exchange currents are generated and can be switched on and off by applying and removing nickel ( Ni , 4 mM ) in the extracellular solution . We note that Ni blocked isolated exchange currents within our solution switch times , and we note also that 5 mM Ni blocked equally rapidly and completely reversibly significant fractions of the Na/K pump current in mouse myocytes . This latter result is notably different from reports for myocytes from other species ( Fujioka et al . , 1998 ) . That Ni is blocking primarily inward Na/Ca exchange current in the protocols employed here is supported by the fact that Ni caused no outward current shifts under routine experimental conditions when pump activity was not activated ( 10 observations ) . As described in Figure 13B , pump currents were activated with extracellular K after prolonged ( 30 s ) inhibition of inward exchange currents and during the continued activation of inward exchange currents . The peak pump currents were in fact significantly smaller in the presence of Na influx via Na/Ca exchange than after prolonged blockade of inward exchange current . Composite results for 9 experiments are given below the pipette cartoon in Figure 13B . To highlight the contradiction that emerges , simulated results for a restricted Na space model are shown in Figure 13C . In brief , the simulation was set up conservatively , using a hyperbolic Na dependence of pump current on cytoplasmic Na with a Kd of 5 mM ( i . e . with a stong tendency to saturate at the 10 mM Nai concentration employed in experiments ) . With cytoplasmic space reduced enough to recreate pump current decay , the peak Na/K pump current should have been substantially increased in the presence of inward exchange current that amounts to 30% of the peak pump current and that moves 3 Na per elementary charges inwardly per cycle . The fundamental result of this study is that Ca influx , usually associated with cycles of internal Ca release , can strongly up-regulate cardiac Na/K pump activity by mechanisms that do not seem to involve protein kinases or redox signaling . The functional consequence will presumably be a down-regulation of basal cardiac contractility , thereby completing a negative-feedback loop in the regulation of cardiac excitation-contraction coupling . Na/K pumps are very well poised to fulfill this function: Na/K pump activity in the presence of cytoplasmic K has a steep dependence on cytoplasmic Na ( Hill coefficients > 2; Figure 2 ) . Therefore , Na/K pumps minimize changes of cytoplasmic Na whenever Na influx increases or decreases . By minimizing the influence of Na influx , and therefore of Na-conducting ion channels , Na/K pumps emerge as singularly powerful regulators of basal cardiac contractility ( Hilgemann , 2004 ) . As noted in the Introduction , previous studies already suggested that cytoplasmic Ca is in some way essential in the signaling cascade by which cAMP-dependent protein kinases enhance pump activity ( Gao et al . , 1996 ) . Our results now raise the possibility that Ca transients regulate cardiac Na/K pumps independent of cAMP , and that cAMP-dependent kinases will to a greater-or-lesser extent up-regulate Na/K pump activity secondary to their well- known actions to enhance myoycte Ca transients . In the presence of saturating cytoplasmic Na , pump currents are not increased by Ca elevations ( Figure 6B ) . Thus , attenuation of the putative inactivation reaction by Ca elevations duplicates the apparent increase of Na affinity ascribed to phospholemman phosphorylation ( Han et al . , 2010 ) . An interesting parallel to our results is the description of Na/K pump activation in skeletal muscle by a train of rapid excitations , reviewed recently ( Clausen , 2013 ) . Although extracellular K definitely accumulates in skeletal muscle T-tubules during rapid action potential firing , and thereby promotes Na/K pump activity ( DiFranco et al . , 2015 ) , this mechanism cannot explain why ouabain-sensitive rubidium uptake is strongly enhanced by skeletal muscle activity ( Buchanan et al . , 2002 ) . It seems reasonable to suggest therefore that the Ca-dependent activation of Na/K pump activity described here in cardiac myocytes is physiologically important in skeletal muscle as well . Disappointingly , our work provides no insight into the role of phospholemman phosphorylation per se by PKAs , as we observe no relevant changes in Ca-dependent modulation of pump function . At the same time , our work gives no reason to doubt that phospholemman phosphorylation increases the effective Na affinity of Na/K pump ATPase activity in isolated membranes , as described by multiple authors with multiple methods ( Bibert et al . , 2008; Manoharan et al . , 2015; Mishra et al . , 2015 ) . One potentially important regulatory factor that requires future attention is the palmitoylation of phospholemman at two cysteines , a modification that appears to be required for phospholemman to inhibit pump activity ( Tulloch et al . , 2011 ) . In this connection , it is doubtful that phospholemman is heavily palmitoylated under the conditions of our experiments ( Lin et al . , 2013 ) . We have at this time only limited insight into the pronounced inhibitory effects of extended or repeated cytoplasmic Ca elevations on Na/K pump currents ( Figures 8 and 10 ) that we observe very routinely . A Ca-dependent generation of fatty acids and lysolipids would potentially inhibit pump activity by direct actions ( Swarts et al . , 1990 ) . A combined generation of acyl CoAs and activation of PKCs has been shown to promote palmitoylation of membrane-associated proteins , including phospholemman ( Lin et al . , 2013 ) The opening of mitochondrial permeability transition pores , which also can occur in these protocols ( Hilgemann et al . , 2013; Lin et al . , 2013 ) , might promote reverse F-ATPase activity and thereby cause depletion of cytoplasmic ATP ( Bernardi et al . , 2015 ) . Biophysical analysis of cardiac myocyte ion homeostasis during patch clamp suggested that cytoplasmic Na concentrations would not deviate by more than 5% from Na concentrations employed in pipette solutions ( Mogul et al . , 1989 ) . Nevertheless , multiple groups were able to measure significant changes of cytoplasmic Na in myocytes during activation of Na/K pump currents ( Su et al . , 1998; Despa and Bers , 2003 ) , albeit on longer time scales than current decay in the present experiments . Thus , the faster components of Na/K pump current decay have remained open to other interpretations . The fact that Na/K pump activity causes a parallel decline of Na/Ca exchange activity ( Figure 13A ) is a very persuasive argument for Na depletion , although the present work suggests a plausible alternative . Given that hydrophobic cation currents decrease in parallel with Na/K pump current decay ( Figure 12 ) , it is likely that significant membrane changes are occurring and the possibility thus arises that Na pump inactivation might regulate Na/Ca exchange function by mechanisms that are independent of Na concentration changes . That Na/K pumps might physically influence Na/Ca exchangers within protein complexes was suggested already by Su et al . ( 1998 ) in this same context . This might be possible if inactivation involves unusual physical changes of pump structure , as might occur with dimerization , and that might mechanically influence the bilayer as well as interacting protein networks . Alternatively , inactivation might control an enzymatic activity that alters the membrane , similar to the long-standing ( Tian et al . , 2006 ) but still controversial ( Yosef et al . , 2016 ) hypothesis that Na/K pumps directly regulate Src kinases . The measurement of Na/Ca exchange reversal potentials in equivalent experiments with guinea pig myocytes ( Fujioka et al . , 1998 ) should have distinguished between the major two hypotheses , but the outcomes were ambivalent . Using highly Ca-buffered pipette solutions , activation of Na/K pumps by extracellular K caused a time-dependent decrease of Na/Ca exchange currents with no change of reversal potential , as expected if Na/K pumps physically regulate Na/Ca exchangers . However , Na/Ca exchange reversal potentials changed markedly during recovery from pump activity , as if subsarcolemmal Na had been depleted during pump activity and recovered after termination of pump activity . Clearly , this issue requires further work . The measurement of Na channel reversal potentials might be more decisive than Na/Ca exchange reversals because the stoichiometry of NCX1 Na/Ca exchange operation is not completely fixed and may be subject to regulatory changes ( Kang and Hilgemann , 2004 ) . While not definitive without energetic proof ( i . e . reversal potential measurements ) , all other results of this study favor the idea that Na/K pump current decay reflects primarily an inactivation mechanism: ( 1 ) Capacitance changes that arise with good certainly from Na/K pumps reverse slowly after pump activity is terminated ( Figures 2B , 3B , 5A , 6A , and 9 ) . Thus , Na/K pumps probably do not immediately return to an E2 configuration when pump activity is terminated . ( 2 ) Na/K pump current decay occurs to the same fractional extent at 23 and 37°C , although pump currents are about four-times smaller at the lower temperature ( Figure 3 ) . ( 3 ) The inclusion of cytoplasmic K in pipette solutions decreases the apparent cytoplasmic Na affinity and thereby decreases pump current densities when Na is submaximal ( Figure 2C and D ) . Nevertheless , the fractional decay of pump currents is increased ( Figure 3 ) . ( 4 ) Current decay is very strong when a high Li concentration ( 40 mM ) is used as a Na surrogate and pump currents are three-fold smaller than in our standard conditions ( Figure 3 ) . ( 5 ) We have verified a key prediction for the inactivation model , suggested in Figure 1 . Application of extracellular Na under conditions that promote accumulation of cytoplasmic ADP suppresses the transient phase of pump currents , such that pump currents activate to approximately their quasi steady state ( 15 s ) magnitude without a peak followed by current decay ( Figure 5 ) . ( 6 ) Myocytes can support outward Na/Ca exchange currents that are many times larger than pump currents ( Figure 7 ) . Within the context of a Na depletion model , one must assume that Na/Ca exchangers have access to a much larger restricted space than Na/K pumps in those myocytes , or alternatively that the restricted space and/or Na diffusion rates are strongly enhanced within a very short time by Ca influx . ( 7 ) A large , continuous Na influx via forward Na/Ca exchange fails to enhance pump currents , although the operation of pumps can depress reverse exchange currents ( Figure 13B ) . ( 8 ) The fractional decay of pump currents often increases as currents decline over time , a behavior that is opposite to the pattern expected for an ion depletion mechanism ( e . g . Figure 11 ) . Finally , as noted in Results , the rate of pump current decay does not decrease as cytoplasmic Na is increased into the saturation range for pump activity ( Verdonck et al . , 2004 ) . Apart from Na concentration changes , the circumstances in which ATP depletion , coupled with ADP and Pi accumulation , might support current decay require future attention . During a maintained average pump current of 300 pA , about 2 mmole ATP per liter cytoplasm will be cleaved every 10 s in a 20 pL cell . In the physiological setting , these nucleotide changes must be very rapidly countered by ATP regeneration via glycolysis and/or oxidative phosphorylation . On the one hand , extensive local depletion of ATP might be an immediate cause of pump current decay during metabolic stress . On the other hand , our work suggests that ADP accumulation during metabolic stress will favor occupancy of states that can inactivate . Independent of Na/K pump function , it appears that significant membrane changes are occurring rapidly in the presence of metabolic stress ( Figure 11 ) . Transient elevations of cytoplasmic Ca have two effects when cytoplasmic Na is non-saturating . Ca elevation decreases the Na/K pump current decay and increases peak pump currents ( Figures 6–10 and Table 1 ) . In both a 'Na depletion' model and an 'inactivation' model , a genuine increase of Na/K pump Na affinity will of course cause an increase of peak pump currents when extracellular K is applied . Qualitatively , a genuine increase of Na affinity can also explain changes of pump current decay that occur . In a depletion model , cytoplasmic Na must decrease more extensively to cause a fall of pump current when the Na affinity is increased . In an inactivation model , a higher Na affinity attenuates inactivation because the number of E1 pumps without a bound Na would be decreased . In this context , the existence of a labile , hydrophobic Na antagonist would go quite far to explain many results . Nevertheless , it is not necessary to evoke a genuine change of Na affinity to explain these results . One possible extension of the inactivation model is that pumps inactivate at multiple points in the pump cycle , including E2 states . If transient Ca elevation favors the recovery from multiple inactive states , not just those immediately coupled to E1 conformations , peak pump currents will increase without evoking a genuine increase of Na affinity . In a recent study employing rabbit cardiac myocytes , Na/K pump current decay during application of extracellular K was implicated to involve the glutathionylation of Na/K pump beta subunits ( Garcia et al . , 2016 ) . For unknown reasons , our equivalent results with mouse myocytes are very different , and we stress that current decay in mouse myocytes is much faster than in rabbit myocytes . As described in Table 1 , we have tested quite extensively whether Na/K pump activity is regulated by redox-dependent processes , including glutathionylation . Experimental results for our routine protocols are similar in myocytes from mice lacking Peroxyredoxin6 and in which turnover of oxidized glutathione to glutathione should be decreased ( Zhou et al . , 2013 ) . Experiments employing high concentrations of cytoplasmic glutathione ( >5 mM ) , oxidized glutathione ( >5 mM ) , superoxide dismutase , and reducing agents give no indication that redox-dependent processes are immediately involved in the effects of Ca elevations or their reversal ( Table 1 ) . Our efforts to identify specific molecular mechanisms by which Ca elevation modulate Na/K pump activity are to date unsuccessful ( Table 1 ) . These negative outcomes beg the suggestion that important Ca-dependent regulatory systems remain to be elucidated in cardiac myocytes . It may be an important 'clue' that Ca elevation and Na pump inactivation cause significant but opposite physical changes of the bulk surface membrane , as detected with hydrophobic ions . Further , it appears important that the effects of Ca elevation in intact myocytes have no clear parallels in biochemical pump studies . And finally , we stress that we have never observed comparable effects of cytoplasmic Ca on Na/K pump currents in giant excised membrane patches in which pump currents appear to be highly activated in comparison to pump currents in intact cells ( Lu et al . , 1995 ) . One reason could be that giant excised patches are highly stretched membranes which probably lack actin membrane cytoskeleton . On the other hand , pump currents in excised patches routinely show an initial 'run-up' that is consistent with the irreversible loss of an endogenous pump inhibitor ( Hilgemann , 1997 ) . With certainty , critical factors are disrupted or lost when membranes are isolated . Our data reveal three parallel changes of Na/K pump and Na/Ca exchange activities that can be tracked by a nonselective membrane probe , C6-TPP: ( 1 ) In response to a transient Ca elevation , both pump currents and exchange currents usually become facilitated ( Figure 8B ) . ( 2 ) In response to repeated transient Ca elevations , both currents become suppressed and/or are ablated ( Figure 8C and 10A ) . ( 3 ) Na/K pump current decay during application of extracellular K is associated with a substantial decline of Na/Ca exchange current ( Figure 13A ) . The fact that membrane probes sense all three events supports speculation that the three events are related . Increases of membrane capacitance that occur very reliably after Ca elevations ( Figures 6 , 9 , 10 ) indicate that membrane fusion events are occurring and/or that the membrane is becoming slightly thinner . Since maximal pump currents are not increased by transient Ca elevation ( Figure 6B ) , insertion of pumps is not a viable explanation for the activation of pump currents . However , Ca-dependent fusion events could bring into the cell surface proteins that subsequently mediate pump activation . The parallel increase of hydrophobic ion signals ( Figure 1 ) indicates that physical properties of the membrane are indeed changing . Specifically , the membrane is becoming more disordered . Ca-dependent enzymes that might mediate such changes include PLA2s ( Brown et al . , 2003 ) , diacylglycerol kinases ( Sakane and Kanoh , 1997 ) , and acyl transferases that generate acyl-phosphatidylethanolamine ( Ogura et al . , 2016 ) . An alternative is that cardiac Na transporters in intact myocytes physically regulate one another through a protein network that involves the membrane bilayer itself . Cardiac ( NCX1 ) Na/Ca exchangers have a large cytoplasmic regulatory domain with multiple Ca binding sites that effectively responds to changes of the time integral of cytoplasmic free Ca ( Matsuoka , 1993 ) . Both conformational changes per se of this domain ( John et al . , 2011 ) and its activating influence on Na/Ca exchange function ( Hilgemann et al . , 1992a ) can persist for longer than one minute after cytoplasmic Ca declines . Thus , the cytoplasmic NCX1 domain is one potential Ca sensor for the activation of Na/K pumps by Ca elevation . Conformational information would be transferred to Na/K pumps via cytoskeleton , the bilayer , and/or an enzymatic activity . The existence of physical interactions between neighboring transporters , independent of Na concentration changes , is an attractive explanation for findings that specific Na/K pump isoforms regulate cardiac excitation-contraction coupling in a specific manner ( Rindler et al . , 2013 ) . As concerns the Na depletion hypothesis , the physical basis of a restricted subsarcolemmal Na space with multi-second Na exchange remains to be established . We stress that our results do not at this time rigorously exclude its existence . New model systems , which can mimic the functions of Na/K pumps in intact myocytes , are now essential to resolve many questions opened by this study . In conclusion , we have initiated a new effort to identify physiological mechanisms that regulate Na/K pump activity in cardiac myocytes . Surprisingly , Ca elevations modulate cardiac Na/K pump activities more powerfully than any conventional signaling pathway in our experience . Transient Ca elevations attenuate Na/K pump current decay by mechanisms that remain to be elucidated . Common Ca-dependent protein kinases and phosphatases do not appear to be involved , and the underlying mechanisms result in physical changes of the membrane that can be detected by multiple nonspecific membrane probes . The Ca-dependent mechanisms that modulate Na/K pumps may also modulate the function of Na/Ca exchangers , and the involvement of membrane per se suggests that other membrane-coupled processes might also be affected . Patch clamp ( Yaradanakul et al . , 2008 ) and myocyte preparation were as described ( Lariccia et al . , 2011 ) . The UT Southwestern Medical Center Animal Care and Use Committee approved all animal studies . Highly polished pipette tips with diameters of >4 μm were employed , and access resistances during recordings ranged from 1 . 2 to 4 MΩ . All experiments presented were performed at 0 mV . Freshly isolated cardiac myocytes were incubated with 2 . 6 μM Fluo-4 AM ( Thermo Fisher Scientific ) at 23°C for 60 min . Epifluorescence imaging was performed with a Nikon Eclipse TE2000-S microscope equipped with a Photometrics Cool Snap ES2 camera and a 40 × WI objective . A 470/40 nm excitation filter was employed with a 495 dichroic and 500 LP emission filter set . Analysis was performed using Nikon NIS Elements AR 4 . 50 . 00 64-bit after background subtraction . Standard Solutions employed minimize all currents other than Na/K pump currents and Na/Ca exchange currents . The standard extracellular solution contained in mM: 110 N-methyl-D-glucamine ( NMG ) , 4 MgCl2 ± 2 CaCl2 , 0 . 5 EGTA , 20 TEA-OH , 7 NaCl or KCl , and 10 HEPES , set to pH 7 . 0 with aspartate . The standard cytoplasmic solution contained in mM: 90 KOH , 20 TEA-OH , 25 Na-OH , 15 HEPES , 0 . 5 MgCl2 , 0 . 5 EGTA , 0 . 25 CaCl2 , 1 K2PO4 , set to pH 7 . 4 with aspartate . Unless stated otherwise , 8 mM MgATP , 2 mM TrisATP , and 0 . 2 mM GTP were employed in cytoplasmic solutions , generating a free Mg2+ of 0 . 5 mM . When different monovalent cations were employed in the cytoplasmic solution , they were added as hydroxides , and NMG was used as the cation substitute in experiments varying monovalent cations . Solutions employing 20 and 50 mM EGTA were prepared by first dissolving EGTA with NMG to give a neutral pH , followed by boiling with the required amounts of CaCO3 . The final osmolarity of all solutions was 290 mosmol/L . In experiments examining the cytoplasmic Na dependence of pump currents with K , K was reduced from 90 to 70 mM in the K-containing solutions employed , and NMG was substituted for Na . NMG was substituted for cytoplasmic K in experiments without K . All salts were from Sigma-Aldrich and were the highest available grade . Reagents employed in experiments described in Table 1 were from standard chemical suppliers . PLM-deficient mice were bred from knockout mice provided by Amy L . Tucker ( U . Virginia , Charlottesville ) ( Jia et al . , 2004 ) . PLD1/2-deficient mice were bred from knockout mice provided by Michael Frohman ( Stony Brook U . , Stony Brook , NY ) ( Ali et al . , 2013 ) . DHHC5-deficient cardiac myocytes were isolated from littermates of heterozygous crosses at F2 or littermates of F2 x F2 crosses of homozygous WT or DHHC5-Gene-Trapped mice ( Li et al . , 2012 ) . PRD6-deficient mice were bred from knockout mice provided by Aron B . Fisher ( U . Pennsylvania , Philadelphia ) ( Nagy et al . , 2006 ) The simulated results were generated by assuming that peak pump currents ( Ipeak ) are proportional to the simultaneous binding of Na to three sites , two of which have the same affinity: ( 1 ) Ipeak=Imax⋅ ( Nai/ ( Nai+Kn1 ) ) 2∗Nai/ ( Nai+Kn2 ) , where Kn1 was 0 . 5 mM and Kn2 was 4 . 0 mM . In the case of a restricted space ( dotted curves in Figure 1E and F , the steady state current ( Iss ) was calculated by assuming that Na next to the membrane could decrease toward a steady state value ( Nass ) from a constant bulk Na concentration ( NaB ) in dependence on a single diffusion barrier ( Kdiff ) . Accordingly , the net Na flux through the restriction is equal to the Na flux out of the cell mediated by pumps: ( 2 ) Iss=Imax⋅ ( Nass/ ( Nass+kn1 ) ) 2⋅Nass/ ( Nass+Kn2 ) = ( NaB−Nass ) ⋅Kdiff , where NaB is the bulk cytoplasmic Na concentration , Nass is the concentration at the membrane , and diffusion into the restricted space is determined by Kdiff ( 0 . 06 pA∙pF−1∙mM−1 ) . Equation #2 was solved for Nass by a Newton procedure and the steady state current , Iss , was calculated accordingly . For the inactivation model results ( solid curves in Figure 1E and F ) , we assumed that pumps inactivate in proportion to the fraction of pumps in which the Kn2 site is not occupied , Fo=Kn2/ ( Kn2+Nai ) . Assuming that recovery from inactivation occurs at 0 . 3 s−1 and is independent of Na binding , and assuming additionally that inactivation takes place at a rate of Fo⋅1s−1 , the steady state currents ( Iss ) was calculated as: ( 3 ) Iss=Ipeak⋅0 . 3/ ( Fo+0 . 3 ) , Unless stated otherwise , error bars represent standard errors . Significance was assessed by Students T-test or , in occasional cases of inappropriate variance differences , by the Mann-Whitney Rank Sum test . Maverick data points were removed from data sets by the criterion that a data point deviated by more than two standard deviations from the data set mean .	All animal cells have pumps in their outer membrane that continuously pump out sodium ions and bring in potassium ions . As a result , the concentration of sodium ions inside cells is low in comparison to the concentration outside , and vice versa for potassium ions . These differences between inside and outside concentrations are a source of energy for cells to do a variety of important tasks , similar to using energy that is stored in a reservoir formed by a dam . When cells allow ions to move in the direction they naturally move , similar to water moving over a dam , they carry out two important roles . First , they generate electrical signals that are the basis of all fast communication in nerves and muscles . Second , they can force important molecules to be moved into or out of cells in a specific way , thereby making the inside composition of cells different from the outside . Lu et al set out to uncover how the pumps that move sodium ions out of and potassium ions into animal cells are regulated . The experiments focused on heart muscle cells from mice , and revealed that sodium-potassium pumps regulate themselves via a self-inhibitory mechanism that depends on the sodium concentration in cells . This “auto-inhibition” reaction appears to affect the activity of neighboring membrane proteins that transport sodium ions . Further experiments showed that the auto-inhibition reaction is controlled by calcium-dependent processes that change the physical properties of the cell’s membrane . The next challenges will be to determine how sodium transporters influence one another and how the calcium signals actually alter the physical properties of the surface membrane .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2016	Profound regulation of Na/K pump activity by transient elevations of cytoplasmic calcium in murine cardiac myocytes
Seeing a speaker’s face enhances speech intelligibility in adverse environments . We investigated the underlying network mechanisms by quantifying local speech representations and directed connectivity in MEG data obtained while human participants listened to speech of varying acoustic SNR and visual context . During high acoustic SNR speech encoding by temporally entrained brain activity was strong in temporal and inferior frontal cortex , while during low SNR strong entrainment emerged in premotor and superior frontal cortex . These changes in local encoding were accompanied by changes in directed connectivity along the ventral stream and the auditory-premotor axis . Importantly , the behavioral benefit arising from seeing the speaker’s face was not predicted by changes in local encoding but rather by enhanced functional connectivity between temporal and inferior frontal cortex . Our results demonstrate a role of auditory-frontal interactions in visual speech representations and suggest that functional connectivity along the ventral pathway facilitates speech comprehension in multisensory environments . When communicating in challenging acoustic environments we profit tremendously from visual cues arising from the speakers face . Movements of the lips , tongue or the eyes convey significant information that can boost speech intelligibility and facilitate the attentive tracking of individual speakers ( Ross et al . , 2007; Sumby and Pollack , 1954 ) . This multisensory benefit is strongest for continuous speech , where visual signals provide temporal markers to segment words or syllables , or provide linguistic cues ( Grant and Seitz , 1998 ) . Previous work has identified the synchronization of brain rhythms between interlocutors as a potential neural mechanism underlying the visual enhancement of intelligibility ( Hasson et al . , 2012; Park et al . , 2016; Peelle and Sommers , 2015; Pickering and Garrod , 2013; Schroeder et al . , 2008 ) . Both acoustic and visual speech signals exhibit pseudo-rhythmic temporal structures at prosodic and syllabic rates ( Chandrasekaran et al . , 2009; Schwartz and Savariaux , 2014 ) . These regular features can entrain rhythmic activity in the observer’s brain and facilitate perception by aligning neural excitability with acoustic or visual speech features ( Giraud and Poeppel , 2012; Mesgarani and Chang , 2012; Park et al . , 2016; Peelle and Davis , 2012; Schroeder and Lakatos , 2009; Schroeder et al . , 2008; van Wassenhove , 2013; Zion Golumbic et al . , 2013a ) . While this model predicts the visual enhancement of speech encoding in challenging multisensory environments , the network organization of multisensory speech encoding remains unclear . Previous work has implicated many brain regions in the visual enhancement of speech , including superior temporal ( Beauchamp et al . , 2004; Nath and Beauchamp , 2011; Riedel et al . , 2015; van Atteveldt et al . , 2004 ) , premotor and inferior frontal cortices ( Arnal et al . , 2009; Evans and Davis , 2015; Hasson et al . , 2007b; Lee and Noppeney , 2011; Meister et al . , 2007; Skipper et al . , 2009; Wright et al . , 2003 ) . Furthermore , some studies have shown that the visual facilitation of speech encoding may even commence in early auditory cortices ( Besle et al . , 2008; Chandrasekaran et al . , 2013; Ghazanfar et al . , 2005; Kayser et al . , 2010; Lakatos et al . , 2009; Zion Golumbic et al . , 2013a ) . However , it remains to be understood whether visual context shapes the encoding of speech differentially within distinct regions of the auditory pathways , or whether the visual facilitation observed within auditory regions is simply fed forward to upstream areas , perhaps without further modification . Hence , it is still unclear whether the enhancement of speech-to-brain entrainment is a general mechanism that mediates visual benefits at multiple stages along the auditory pathways . Many previous studies on this question were limited by conceptual shortcomings: first , many have focused on generic brain activations rather than directly mapping the task-relevant sensory representations ( activation mapping vs . information mapping [Kriegeskorte et al . , 2006] ) , and hence have not quantified multisensory influences on those neural representations shaping behavioral performance . Those who did focused largely on auditory cortical activity ( Zion Golumbic et al . , 2013b ) or did not perform source analysis of the underlying brain activity ( Crosse et al . , 2015 ) . Second , while many studies have correlated speech-induced local brain activity with behavioral performance , few studies have quantified directed connectivity along the auditory pathways to ask whether perceptual benefits are better explained by changes in local encoding or by changes in functional connectivity ( but see [Alho et al . , 2014] ) . And third , many studies have neglected the continuous predictive structure of speech by focusing on isolated words or syllables ( but see [Crosse et al . , 2015] ) . However , this structure may play a central role for mediating the visual benefits ( Bernstein et al . , 2004; Giraud and Poeppel , 2012; Schroeder et al . , 2008 ) . Importantly , given that the predictive visual context interacts with acoustic signal quality to increase perceptual benefits in adverse environments ( Callan et al . , 2014; Ross et al . , 2007; Schwartz et al . , 2004; Sumby and Pollack , 1954 ) , one needs to manipulate both factors to fully address this question . Fourth , most studies focused on either the encoding of acoustic speech signals in a multisensory context , or quantified brain activity induced by visual speech , but little is known about the dependencies of neural representations of the acoustic and visual components of realistic speech ( but see [Park et al . , 2016] ) . Overcoming these problems , we here capitalize on the statistical and conceptual power offered by naturalistic continuous speech to study the network mechanisms that underlie the visual facilitation of speech perception . Using source localized MEG activity we systematically investigated how local representations of acoustic and visual speech signals and task-relevant directed functional connectivity along the auditory pathways change with visual context and acoustic signal quality . Specifically , we extracted neural signatures of acoustically-driven speech representations by quantifying the mutual information ( MI ) between the MEG signal and the acoustic speech envelope . Similarly , we extracted neural signatures of visually-driven speech representations by quantifying the MI between lip movements and the MEG signal . Furthermore , we quantified directed causal connectivity between nodes in the speech network using time-lagged mutual information between MEG source signals . Using linear modelling we then asked how each of these signatures ( acoustic and visual speech encoding; connectivity ) are affected by contextual information about the speakers face , by the acoustic signal to noise ratio , and by their interaction . In addition , we used measures of information theoretic redundancy to test whether the local representations of acoustic speech are directly related to the temporal dynamics of lip movements or rather reflect visual contextual information more indirectly . And finally , we asked how local speech encoding and network connectivity relate to behavioral performance . Our results describe multiple and functionally distinct representations of acoustic and visual speech in the brain . These are differentially affected by acoustic SNR and visual context , and are not trivially explained by a simple superposition of representations of the acoustic speech and lip movement information . However , none of these local speech representations was predictive of the degree of visual enhancement of speech comprehension . Rather , this behavioral benefit was predicted only by changes in directed functional connectivity . Speech-to-brain entrainment was quantified by the mutual information ( speech MI ) between the MEG time course and the acoustic speech envelope ( not the speech + noise mixture ) in individual frequency bands ( Gross et al . , 2013; Kayser et al . , 2015 ) . At the group-level we observed widespread significant speech MI in all considered bands from 0 . 25 to 48 Hz ( FWE = 0 . 05 ) , except between 18–24 Hz ( Figure 3—figure supplement 1A ) . Consistent with previous results ( Gross et al . , 2013; Ng et al . , 2013; Park et al . , 2016 ) speech MI was higher at low frequencies and strongest below 4 Hz ( Figure 3—figure supplement 1C ) . This time scale is typically associated with syllabic boundaries or prosodic stress ( Giraud and Poeppel , 2012; Greenberg et al . , 2003 ) . Indeed , the average syllabic rate was 212 syllables per minute in the present material , corresponding to about 3 . 5 Hz . Across frequencies , significant speech MI was strongest in bilateral auditory cortex and was more extended within the right hemisphere ( Figure 3—figure supplement 1A and C ) . Indeed , peak significant MI values were significantly higher in the right compared to the left hemisphere at frequencies below 12 Hz ( paired t-tests; T ( 18 ) ≥ 3 . 1 , p≤0 . 043 Bonferroni corrected ) , and did not differ at higher frequencies ( T ( 18 ) ≤ 2 . 78 , p≥0 . 09 ) . This lateralization of speech-to-brain entrainment at frequencies below 12 Hz is consistent with previous reports ( Gross et al . , 2013 ) . Importantly , we observed significant speech-to-brain entrainment not only within temporal cortices but across multiple regions in the occipital , frontal and parietal lobes , consistent with the notion that speech information is represented also within motor and frontal regions ( Bornkessel-Schlesewsky et al . , 2015; Du et al . , 2014; Skipper et al . , 2009 ) . To determine the regions where acoustic signal quality and visual context affect the encoding of acoustic speech we modelled the condition-specific speech MI values based on effects of acoustic signal quality ( SNR ) , visual informativeness ( VIVN ) , and their interaction ( SNRxVIVN ) . Random-effects significance was tested using a permutation procedure and cluster enhancement , correcting for multiple comparisons along all relevant dimensions . Effects of experimental factors emerged in multiple regions at frequencies below 4 Hz ( Figure 3 ) . Increasing the acoustic signal quality ( SNR; Figure 3A ) resulted in stronger speech MI in the right auditory cortex ( 1–4 Hz; local peak T statistic = 4 . 46 in posterior superior temporal gyrus; pSTG-R; Table 1 ) , right parietal cortex ( local peak T = 3 . 94 in supramarginal gyrus; SMG-R ) , and right dorso-ventral frontal cortex ( IFGop-R; global peak T = 5 . 06 ) . We also observed significant positive SNR effects within the right temporo-parietal and occipital cortex at 12–18 Hz ( local peak right lingual gyrus , T = 5 . 12 ) . However , inspection of the participant-specific data suggested that this effect was not reliable ( for only 58% of participants showed a speech MI increase with SNR , as opposed to a minimum of 84% for the other SNR effects ) , possibly because the comparatively lower power of speech envelope fluctuations at higher frequencies ( c . f . Figure 1A ) ; hence this effect is not discussed further . 10 . 7554/eLife . 24763 . 005Figure 3 . Modulation of speech-to-brain entrainment by acoustic SNR and visual informativeness . Changes in speech MI with the experimental factors were quantified using a GLM for the condition-specific speech MI based on the effects of SNR ( A ) , visual informativeness VIVN ( B ) , and their interaction ( SNRxVIVN ) ( C ) . The figures display the cortical-surface projection onto the Freesurfer template ( proximity = 10 mm ) of the group-level significant statistics for each GLM effect ( FWE = 0 . 05 ) . Graphs show the average speech MI values for each condition ( mean ± SEM ) , for local and global ( red asterisk ) of the T maps . Lines indicate the across-participant average regression model and numbers indicate the group-average standardized regression coefficient for SNR in the VI and VN conditions ( >/ < 0 . 0 = positive/negative , rounded to 0 ) . ( D ) T maps illustrating the opposite SNR effects within voxels with significant SNRxVIVN effects . MI graphs for the peaks of these maps are shown in ( C ) ( IFGor-R and SFG-R = global T peaks for SNR effects in VI and VN , respectively ) . ( E ) Location of global and local seeds of GLM T maps , used for the analysis of directed connectivity . See also Tables 1 and 2 and Figure 3—figure supplements 1–2 . Deposited data: SE_meg; SE_speech; SE_miS . DOI: http://dx . doi . org/10 . 7554/eLife . 24763 . 00510 . 7554/eLife . 24763 . 006Figure 3—figure supplement 1 . Entrainment of rhythmic MEG activity to the speech envelope and lip movements . ( A ) Projection of significant speech MI maps , which quantify the entrainment of MEG source activity to the speech envelope , onto the Freesurfer template ( FWE = 0 . 05; proximity = 10 mm; surface-projected significant MI maps rescaled within volume from minimum significant MI to the 99 . 5th percentile of the surface projection ) . ( B ) Projection of significant lip MI maps . ( C ) Peak speech / lip MI values in the two hemispheres as a function of frequency ( mean ± SEM ) . Deposited data: SE_meg; SE_speech; SE_miS; LE_meg; LE_lip; LE_miL . DOI: http://dx . doi . org/10 . 7554/eLife . 24763 . 00610 . 7554/eLife . 24763 . 007Figure 3—figure supplement 2 . Information theoretic decomposition of speech entrainment . The figure shows condition-specific information terms for the relevant ROIs / bands ( c . f . Figure 3E ) . ( A ) Speech MI . ( B ) Conditional mutual information , CMI ( MEG;Speech ) , factoring out common influences between speech and lip signals . ( C ) Information theoretic redundancy between speech and lip MI . error-bars = ± SEM . See also Table 2 . Deposited data: ID_meg; ID_speech; ID_lip; ID_infoterms . DOI: http://dx . doi . org/10 . 7554/eLife . 24763 . 00710 . 7554/eLife . 24763 . 008Figure 3—figure supplement 3 . Condition-changes in the amplitude of oscillatory activity . Difference in the time-averaged Hilbert amplitude of the MEG signal between the VI and VN conditions for each ROI , as a function of frequency . Error bars bracket the 99% parametric confidence interval for the participant averaged difference . Thick bars highlight significant differences ( FWE = 0 . 05 across ROIs and frequencies ) . a . u . = arbitrary units . Deposited data: SE_meg; AMP_amp . DOI: http://dx . doi . org/10 . 7554/eLife . 24763 . 00810 . 7554/eLife . 24763 . 009Table 1 . Condition effects on speech MI . The table lists global and local peaks in the GLM T-maps . Anatomical labels and Brodmann areas are based on the AAL and Talairach atlases . β = standardized regression coefficient; SEM = standard error of the participant average . ROI-contralat . = T test for a significant difference of GLM betas between the respective ROI and its contralateral grid voxel . DOI: http://dx . doi . org/10 . 7554/eLife . 24763 . 009Anatomical labelBrodmann areaMNI coordinatesGLM effectFrequency BandT ( 18 ) β ( SEM ) T ( 18 ) ROI-contralat . HG-R4263−2111VIVN0 . 25–1 Hz4 . 750 . 39 ( 0 . 06 ) 2 . 00pSTG-R2248−308SNR1–4 Hz4 . 460 . 48 ( 0 . 08 ) 2 . 36SMG-R4057−3038SNR1–4 Hz3 . 940 . 29 ( 0 . 09 ) 0 . 22PMC-L6−54032VIVN1–4 Hz3 . 810 . 27 ( 0 . 06 ) −0 . 65IFGt-R4642332SNRxVIVN0 . 25–1 Hz3 . 620 . 29 ( 0 . 07 ) 1 . 48IFGop-R4751182SNR1–4 Hz5 . 060 . 36 ( 0 . 08 ) 6 . 03IFGor-R473026−16SNR in VI0 . 25–1 Hz5 . 070 . 44 ( 0 . 08 ) 1 . 92SFG-R6123058SNR in VN0 . 25–1 Hz−3 . 55−0 . 41 ( 0 . 09 ) −2 . 21VC-R17/1818−102-4VIVN1–4 Hz6 . 010 . 45 ( 0 . 06 ) 1 . 84denotes significant effects ( FWE = 0 . 05 corrected for multiple comparisons ) . Relevant variables in deposited data ( doi:10 . 5061/dryad . j4567 ) : SE_meg; SE_speech; SE_miS . Contrasting informative and not-informative visual contexts revealed stronger speech MI when seeing the speakers face ( VI ) at frequencies below 4 Hz in both hemispheres ( Figure 3B ) : the right temporo-parietal cortex ( 0 . 25–1 Hz; HG; T = 4 . 75; Table 1 ) , bilateral occipital cortex ( 1–4 Hz; global T peak in right visual cortex VC-R;=6 . 01 ) and left premotor cortex ( 1–4 Hz; PMC-L; local T peak = 3 . 81 ) . Interestingly , the condition-specific pattern of MI for VC-R was characterized by an increase in speech MI with decreasing SNR during the VI condition , pointing to a stronger visual enhancement during more adverse listening conditions . The same effect was seen in premotor cortex ( PMC-L ) . Since visual benefits for perception emerge mostly when acoustic signals are degraded ( Figure 2 ) ( Ross et al . , 2007; Sumby and Pollack , 1954 ) , the interaction of acoustic and visual factors provides a crucial test for detecting non-trivial audio-visual interactions . We found significant interactions in the 0 . 25–1 Hz band in the right dorso-ventral frontal lobe , which peaked in the pars triangularis ( IFGt-R; T = 3 . 62; Figure 3C; Table 1 ) . Importantly , investigating the SNR effect in the frontal cortex voxels revealed two distinct strategies for handling speech in noise dependent on visual context ( Figure 3D ) : During VI speech MI increased with SNR in ventral frontal cortex ( peak T for SNR in pars orbitalis; IFGor-R; T = 5 . 07 ) , while in dorsal frontal cortex speech MI was strongest at low SNRs during VN ( peak T in superior frontal gyrus; SFG-R; T = −3 . 55 ) . This demonstrates distinct functional roles of ventral and dorsal prefrontal regions in speech encoding and reveals a unique role of superior frontal cortex for enhancing speech representations in a poorly informative context , such as the absence of visual information in conjunction with poor acoustic signals . For further analysis we focused on these regions and frequency bands revealed by the GLM effects ( Figure 3E ) . Our results reveal significantly stronger entrainment at low frequencies ( c . f . Figure 3—figure supplement 1 ) and a prevalence of condition effects on speech MI in the right hemisphere ( c . f . Figure 3 ) . We directly tested whether these condition effects were significantly lateralized by comparing the respective GLM effects between corresponding ROIs across hemispheres ( Table 1 ) . This revealed that only the 1–4 Hz SNR effect in IFGop-R was significantly lateralized ( T ( 18 ) = 6 . 03; FWE = 0 . 05 corrected across ROIs ) , while all other GLM effects did not differ significantly between hemispheres . To complement the above analysis of speech-to-brain entrainment we also systematically analyzed the entrainment of brain activity to lip movements ( lip MI ) . This allowed us to address whether the enhancement of the encoding of acoustic speech during an informative visual context arises from a co-representation of acoustic and visual speech information in the same regions or not . As expected based on previous work , the acoustic speech envelope and the trajectory of lip movements for the present material were temporally coherent , in particular in the delta and theta bands ( Figure 1A ) ( Chandrasekaran et al . , 2009; Park et al . , 2016; Schwartz and Savariaux , 2014 ) . Lip-to-brain entrainment was quantified for the visual informative condition only , across the same frequency bands as considered for the speech MI ( Figure 3—figure supplement 1B ) . This revealed wide-spread significant lip MI in frequency bands below 8 Hz , with the strongest lip entrainment occurring in occipital cortex ( Figure 3—figure supplement 1B ) . Peak lip MI values were larger in the right hemisphere , in particular for the 4–8 Hz band ( Figure 3—figure supplement 1C ) , but this effect was not significant after correction for multiple comparisons ( T ( 18 ) ≤ 2 . 53 , p≥0 . 06 ) . We then asked whether in any regions with significant lip MI the encoding of lip information changed with SNR . No significant SNR effects were found ( FWE = 0 . 05 , corrected across voxels and 0–12 Hz frequency bands ) , demonstrating that the encoding of lip signals is invariant across acoustic conditions . We also directly compared speech MI and lip MI within the ROIs highlighted by the condition effects on speech MI ( c . f . Figure 3E ) . In most ROIs speech MI was significantly stronger than lip MI ( Table 2; T ( 18 ) HG-R , pSTG-R , IFGop-R and PMC-L ≥3 . 58; FWE = 0 . 05 corrected across ROIs ) , while lip MI was significantly stronger in VC-R ( T ( 18 ) = −3 . 35; FWE = 0 . 05 ) . 10 . 7554/eLife . 24763 . 010Table 2 . Analysis of the contribution of audio-visual signals in shaping entrainment . For each region / effect of interest ( c . f . Table 1 ) the table lists the comparison of condition-averaged speech and lip MI ( positive = greater speech MI ) ; the condition effects ( GLM ) on the conditional mutual information ( CMI ) between the MEG signal and the speech envelope , while partialling out effects of lip signals; and the condition-averaged information theoretic redundancy between speech and lip MI . DOI: http://dx . doi . org/10 . 7554/eLife . 24763 . 010Speech vs . lip MISpeech-Lip redundancySpeech CMILabelT ( 18 ) Avg ( SEM ) T ( 18 ) Avg ( SEM ) EffectT ( 18 ) β ( SEM ) HG-R4 . 2728 . 16 ( 6 . 59 ) 0 . 730 . 33 ( 0 . 44 ) VIVN4 . 370 . 35 ( 0 . 06 ) pSTG-R3 . 905 . 42 ( 1 . 39 ) 0 . 490 . 19 ( 0 . 38 ) SNR4 . 660 . 49 ( 0 . 08 ) SMG-R2 . 951 . 32 ( 0 . 45 ) 1 . 100 . 51 ( 0 . 47 ) SNR4 . 100 . 29 ( 0 . 09 ) PMC-L3 . 581 . 06 ( 0 . 30 ) 3 . 832 . 42 ( 0 . 63 ) VIVN3 . 470 . 24 ( 0 . 06 ) IFGt-R1 . 210 . 87 ( 0 . 72 ) 2 . 291 . 75 ( 0 . 77 ) SNRxVIVN4 . 070 . 31 ( 0 . 07 ) IFGopR3 . 681 . 50 ( 0 . 41 ) 4 . 691 . 56 ( 0 . 33 ) SNR4 . 700 . 35 ( 0 . 07 ) SFG-R0 . 880 . 61 ( 0 . 70 ) 4 . 132 . 37 ( 0 . 57 ) SNR in VN−3 . 62−0 . 43 ( 0 . 09 ) VC-R−3 . 35−2 . 19 ( 0 . 65 ) 2 . 370 . 68 ( 0 . 29 ) VIVN5 . 770 . 45 ( 0 . 06 ) denotes significant effects ( FWE = 0 . 05 corrected for multiple comparisons ) . Deposited data: ID_meg; ID_speech; ID_lip; ID_infoterms . Given that only the speech and not the lip representation were affected by SNR the above results suggest that both acoustic and visual speech signals are represented independently in rhythmically entrained brain activity . To address the interrelation between the representations of acoustic and visual speech signals more directly , we asked whether the condition effects on speech MI result from genuine changes in the encoding of the acoustic speech envelope , or whether they result from a superposition of local representations of the acoustic and the visual speech signals . Given that visual and acoustic speech are temporally coherent and offer temporally redundant information , it could be that the enhancement of speech MI during the VI condition simply results from a superposition of local representations of the visual and acoustic signals arising within the same brain region . Alternatively , it could be that the speech-to-brain entrainment reflects a representation of the acoustic speech signal that is informed by visual contextual information , but which is not a one to one reflection of the dynamics of lip movements . We performed two analyses to address this . First , we calculated the conditional mutual information between the MEG signal and the acoustic speech envelop while partialling out the temporal dynamics common to lip movements and the speech envelope . If the condition effects on speech MI reflect changes within genuine acoustic representations , they should persist when removing direct influences of lip movements . Indeed , we found that all of the condition effects reported in Figure 3 persisted when computed based on conditional MI ( absolute T ( 18 ) ≥ 3 . 47; compare Table 2 for CMI with Table 1 for MI; ROI-specific MI and CMI values are shown in Figure 3—figure supplement 2A , B ) . Second , we computed the information-theoretic redundancy between the local speech and lip representations . Independent representations of each speech signal would result in small redundancy values , while a common representation of lip and acoustic speech signals would reflect in a redundant representation . Across SNRs we found that these representations were significantly redundant in the ventral and dorsal frontal cortex ( T ( 18 ) ≥ 3 . 83 , for SFG-R , IFGop-R , IFGt-Rand PMC-L ) but not in the temporal lobe or early auditory and visual cortices ( FWE = 0 . 05 corrected across ROIs; Table 2; Figure 3—figure supplement 2C ) . However , the actual redundancy values were rather small ( condition-averaged values all below 3% ) . All in all , this suggests that the local representations of the acoustic speech envelope in sensory regions are informed by visual evidence but in large do not represent the same information that is provided by the dynamics of lip movements . This in particular also holds for the acoustic speech MI in visual cortex . The stronger redundancy in association cortex ( IFG , SFG , PMC ) suggests that these regions feature co-representations of acoustic speech and lip movements . The diversity of the patterns of speech entrainment in temporal , premotor and inferior frontal regions across conditions shown in Figure 3 could arise from the individual encoding properties of each region , or from changes in functional connectivity between regions with conditions . To directly test this , we quantified the directed causal connectivity between these regions of interest . To this end we used Directed Information ( DI ) , also known as Transfer Entropy , an information theoretic measure of Wiener-Granger causality ( Massey , 1990; Schreiber , 2000 ) . We took advantage of previous work that made this measure statistically robust when applied to neural data ( Besserve et al . , 2015; Ince et al . , 2017 ) . We observed significant condition-averaged DI between multiple nodes of the speech network ( FWE = 0 . 05; Figure 4A and Figure 4—figure supplement 1A ) . This included among others the feed-forward pathways of the ventral and dorsal auditory streams , such as from auditory cortex ( HG-R ) and superior temporal regions ( pSTG-R ) to premotor ( PMC-L ) and to inferior frontal regions ( IFGt-R , IFGop-R ) , from right parietal cortex ( SMG-R ) to premotor cortex ( PMC-L ) , as well as feed-back connections from premotor and inferior frontal regions to temporal regions . In addition , we also observed significant connectivity between frontal ( SFG-R ) and visual cortex ( VC ) . 10 . 7554/eLife . 24763 . 011Figure 4 . Directed causal connectivity within the speech-entrained network . Directed connectivity between seeds of interest ( c . f . Figure 3E ) was quantified using Directed Information ( DI ) . ( A ) Maximum significant condition-average DI across lags ( FWE = 0 . 05 across lags; white = no significant DI ) . ( B ) Significant condition effects ( GLM for SNR , VIVN or their interaction ) on DI ( FWE = 0 . 05 across speech/brain lags and seed/target pairs ) . Bar graphs display condition-specific DI values for each significant GLM effect along with the across-participants average regression model ( lines ) . Numbers indicate the group-average standardized betas for SNR in the VI and VN conditions , averaged across lags associated with a significant GLM effect ( >/ < 0 . 0 = positive/negative , rounded to 0 ) . Error-bars = ± SEM . See also Table 3 and Figure 4—figure supplement 1 . Deposited data: DI_meg; DI_speech; DI_di; DI_brainlag; DI_speechlag . DOI: http://dx . doi . org/10 . 7554/eLife . 24763 . 01110 . 7554/eLife . 24763 . 012Figure 4—figure supplement 1 . Directed functional connectivity within the speech-entrained network . ( A ) Significant condition-averaged directed information ( DI ) values between all seed-target pairs as a function of the speech ( τSpeech ) and brain lags ( τBrain ) . ( B ) Group-level statistical maps for the GLM effects on DI of acoustic signal quality ( SNR ) , visual informativeness ( VIVN ) and their interaction . Deposited data: DI_meg; DI_speech; DI_di; DI_brainlag; DI_speechlag . DOI: http://dx . doi . org/10 . 7554/eLife . 24763 . 012 We then asked whether and where connectivity changed with experimental conditions ( Figure 4B , Table 3 and Figure 4—figure supplement 1B ) . Within the right ventral stream feed-forward connectivity from the temporal lobe ( HG-R , pSTG-R ) to frontal cortex ( IFGt-R , IFGop-R ) was enhanced during high acoustic SNR ( FWE = 0 . 05; T ( 18 ) ≥ 3 . 1 ) . More interestingly , this connectivity was further enhanced in the presence of an informative visual context ( pSTG-R → IFGt-R , VIVN effect , T = 4 . 57 ) , demonstrating a direct influence of visual context on the propagation of information along the ventral stream . Interactions of acoustic and visual context on connectivity were also found from auditory ( HG-R ) to premotor cortex ( PMC-L , negative interaction; T = −3 . 01 ) . Here connectivity increased with increasing SNR in the absence of visual information and increased with decreasing SNR during an informative context , suggesting that visual information changes the qualitative nature of auditory-motor interactions . An opposite interaction was observed between the frontal lobe and visual cortex ( SFG-R → VC-R , T = 3 . 69 ) . Finally , feed-back connectivity along the ventral pathway was significantly stronger during high SNRs ( IFGt-R → pSTG-R; T = 4 . 56 ) . 10 . 7554/eLife . 24763 . 013Table 3 . Analysis of directed connectivity ( DI ) . The table lists connections with significant condition-averaged DI , and condition effects on DI . SEM = standard error of participant average; β = standardized regression coefficients . T ( 18 ) = maximum T statistic within significance mask . All reported effects are significant ( FWE = 0 . 05 corrected for multiple comparisons ) . Deposited data: DI_meg; DI_speech; DI_di; DI_brainlag; DI_speechlag . DOI: http://dx . doi . org/10 . 7554/eLife . 24763 . 013DICondition effects ( GLM ) SeedTargetT ( 18 ) EffectT ( 18 ) β ( SEM ) HG-RPMC-L3 . 38SNRxVIVN−3 . 01−0 . 24 ( 0 . 08 ) HG-RIFGt-R3 . 03SNR3 . 320 . 31 ( 0 . 09 ) HG-RIFGopR4 . 54SNR3 . 190 . 26 ( 0 . 07 ) pSTG-RIFGt-R3 . 39SNR3 . 910 . 32 ( 0 . 09 ) VIVN4 . 570 . 23 ( 0 . 05 ) pSTG-RIFGopR4 . 12SNR3 . 310 . 28 ( 0 . 08 ) IFGt-RIFGopR3 . 76VIVN3 . 560 . 21 ( 0 . 06 ) IFGopRpSTG-R4 . 16SNR4 . 650 . 31 ( 0 . 09 ) SFG-RVC-R4 . 40SNRxVIVN3 . 690 . 28 ( 0 . 08 ) We performed two analyses to test whether and where changes in the local representation of speech information or directed connectivity ( DI ) contribute to explaining the multisensory behavioral benefits ( c . f . Figure 2 ) . Given the main focus on the visual enhancement of perception we implemented this analysis only for speech and not for lip MI . First , we asked where speech-MI and DI relates to performance changes across all experimental conditions ( incl . changes in SNR ) . This revealed a significant correlation between condition-specific word-recognition performance and the strength of speech MI in pSTG-R and IFGt-R ( r ≥ 0 . 28; FWE = 0 . 05; Table 4 and Figure 5A ) , suggesting that stronger entrainment in the ventral stream facilitates comprehension . This hypothesis was further corroborated by a significant correlation of connectivity along the ventral stream with behavioral performance , both in feed-forward ( HG-R → IFGt-R/IFGop-R; pSTG-R → IFGt-R/IFGop-R; r ≥ 0 . 24 , Table 4 ) and feed-back directions ( IFGop-R → pSTG-R; r = 0 . 37 ) . The enhanced quality of speech perception during favorable listening conditions hence results from enhanced speech encoding and the supporting network connections along the temporal-frontal axis . 10 . 7554/eLife . 24763 . 014Figure 5 . Neuro-behavioral correlations . ( A ) Correlations between behavioral performance and condition-specific speech MI ( perform . ( r ) , and correlations between the visual enhancement of performance and the visual enhancement in MI ( vis . enhanc . ( r ) . ( B ) Same for DI . Only those ROIs or connections exhibiting significant correlations are shown . error-bars = ± SEM . See also Tables 2–3 . Deposited data: BEHAV_perf; SE_meg; DI_meg; SE_miS; DI_di; NBC_miS; NBC_di . DOI: http://dx . doi . org/10 . 7554/eLife . 24763 . 01410 . 7554/eLife . 24763 . 015Table 4 . Association of behavioral performance with speech entrainment and connectivity . Performance: T statistic and average of participant-specific correlation ( SEM ) between behavioral performance and speech MI / DI . Visual enhancement: correlation between SNR-specific behavioral benefit ( VI-VN ) and the respective difference in speech-MI or DI . DOI: http://dx . doi . org/10 . 7554/eLife . 24763 . 015Speech MIPerformanceVisual enhancementT ( 18 ) r ( SEM ) T ( 18 ) r ( SEM ) HG-R1 . 270 . 13 ( 0 . 10 ) 0 . 210 . 04 ( 0 . 15 ) pSTG-R3 . 43 0 . 30 ( 0 . 09 ) 0 . 530 . 07 ( 0 . 11 ) SMG-R2 . 350 . 23 ( 0 . 09 ) -0 . 39-0 . 07 ( 0 . 14 ) PMC-L0 . 470 . 04 ( 0 . 08 ) 0 . 130 . 03 ( 0 . 16 ) IFGt-R3 . 09 0 . 28 ( 0 . 09 ) 1 . 250 . 29 ( 0 . 18 ) IFGopR2 . 380 . 24 ( 0 . 09 ) -0 . 25-0 . 05 ( 0 . 17 ) SFG-R-0 . 47-0 . 04 ( 0 . 08 ) 1 . 610 . 35 ( 0 . 17 ) VC-R1 . 550 . 18 ( 0 . 10 ) -0 . 82-0 . 14 ( 0 . 14 ) Directed connectivityPerformanceVisual enhancementSeedTargetT ( 18 ) r ( SEM ) T ( 18 ) r ( SEM ) HG-RPMC-L0 . 900 . 06 ( 0 . 06 ) -0 . 07-0 . 01 ( 0 . 14 ) HG-RIFGt-R4 . 83 0 . 31 ( 0 . 07 ) 2 . 55 0 . 28 ( 0 . 11 ) HG-RIFGopR3 . 19 0 . 24 ( 0 . 07 ) 1 . 860 . 31 ( 0 . 17 ) pSTG-RIFGt-R4 . 28 0 . 27 ( 0 . 06 ) 1 . 280 . 16 ( 0 . 12 ) pSTG-RIFGopR3 . 59 0 . 29 ( 0 . 08 ) 1 . 820 . 32 ( 0 . 17 ) IFGt-RIFGopR1 . 110 . 08 ( 0 . 07 ) 2 . 270 . 33 ( 0 . 14 ) IFGopRpSTG-R4 . 51 0 . 37 ( 0 . 08 ) 2 . 55 0 . 37 ( 0 . 15 ) SFG-RVC-R-0 . 040 . 00 ( 0 . 08 ) 0 . 900 . 17 ( 0 . 18 ) denotes significant effects ( FWE = 0 . 05 corrected for multiple comparisons ) . Deposited data: BEHAV_perf; SE_meg; DI_meg; SE_miS; DI_di; NBC_miS; NBC_di . Second , we asked whether and where the improvement in behavioral performance with an informative visual context ( VI-VN ) correlates with an enhancement in speech encoding or connectivity . This revealed no significant correlations between the visual enhancement of local speech MI and perceptual benefits ( all T values < FWE = 0 . 05 threshold; Table 4 ) . However , changes in both feed-forward ( HG-R → IFGt-R; r = 0 . 28; Figure 5B ) and feed-back connections ( IFGop-R → pSTG-R; r = 0 . 37 ) along the ventral stream were significantly correlated with the multisensory perceptual benefit ( FWE = 0 . 05 ) . We verified that the reported condition effects on speech MI are not simply a by-product of changes in the overall oscillatory activity . To this end we calculated the condition averaged Hilbert amplitude for each ROI and performed a GLM analysis for condition effects as for speech entrainment ( FWE = 0 . 05 with correction across ROIs and frequency bands; Table 5; Figure 3—figure supplement 3 ) . This revealed a reduction of oscillatory activity during the visual informative condition in the occipital cortex across many bands ( VC-R , 4–48 Hz ) , in the inferior frontal cortex ( IFG-R and IFGop-R , 24–48 Hz ) , and in the pSTG-R at 4–8 Hz and 18–24 Hz . No significant effects of SNR or SNRxVIVN interactions were found ( FWE = 0 . 05 ) . Importantly , none of these VIVN effects overlapped with the significant changes in speech MI ( 0 . 25–4 Hz ) and only the reduction in pSTG-R power overlapped with condition effects in connectivity . All in all this suggests that the reported changes in speech encoding and functional connectivity are not systematically related to changes in the strength of oscillatory activity withy acoustic SNR or visual context . 10 . 7554/eLife . 24763 . 016Table 5 . Changes in band-limited source signal amplitude with experimental conditions . The table lists GLM T-statistics , participant averaged standardized regression coefficients ( and SEM ) for significant VIVN effects on MEG source amplitude ( FWE = 0 . 05 corrected across ROIs and frequency bands ) . . Effects of SNR and SNRxVIVN interactions were also tested but not significant Deposited data: SE_meg; AMP_amp . DOI: http://dx . doi . org/10 . 7554/eLife . 24763 . 016ROIBandT ( 18 ) β ( SEM ) pSTG-R4–8 Hz−3 . 66−0 . 38 ( 0 . 09 ) pSTG-R18–24 Hz−4 . 11−0 . 40 ( 0 . 08 ) IFGt-R24–36 Hz−3 . 91−0 . 40 ( 0 . 06 ) IFGt-R30–48 Hz−4 . 49−0 . 39 ( 0 . 08 ) IFGop-R24–36 Hz−4 . 44−0 . 40 ( 0 . 07 ) IFGop-R30–48 Hz−4 . 14−0 . 41 ( 0 . 07 ) VC-R4–8 Hz−3 . 70−0 . 55 ( 0 . 08 ) VC-R8–12 Hz−4 . 53−0 . 70 ( 0 . 05 ) VC-R12–18 Hz−5 . 20−0 . 70 ( 0 . 05 ) VC-R18–24 Hz−5 . 57−0 . 66 ( 0 . 06 ) VC-R24–36 Hz−5 . 57−0 . 55 ( 0 . 08 ) VC-R30–48 Hz−4 . 54−0 . 46 ( 0 . 10 ) Cross-frequency coupling between the phase and amplitudes of different rhythmic brain signals has been implicated in mediating neural computations and communication ( Canolty and Knight , 2010 ) . We asked whether the above results on functional connectivity are systematically related to specific patterns of phase-amplitude coupling ( PAC ) . We first searched for significant condition-average PAC between each pair of ROIs across a wide range of frequency combinations . This revealed significant PAC within VC-R , within pSTG-R and within SMG-R , as well as significant coupling of the 18–24 Hz VC-R power with the 0 . 25–1 Hz IFGop-R phase ( FWE = 0 . 05; see Table 6 ) . However , we found no significant changes in PAC with experimental conditions , suggesting that the changes in functional connectivity described above are not systematically related to specific patterns of cross-frequency coupling . 10 . 7554/eLife . 24763 . 017Table 6 . Analysis of phase-amplitude coupling ( PAC ) . The table lists the significant condition-averaged PAC values for all pairs or ROIs and frequency bands ( FWE = 0 . 05 corrected across pairs of phase and power frequencies ) . SEM = standard error of participant average . None of these changed significantly with conditions ( no GLM effects at FWE = 0 . 05 ) . Deposited data: SE_meg . DOI: http://dx . doi . org/10 . 7554/eLife . 24763 . 017Phase ROI ( band ) Power ROI ( band ) T ( 18 ) Pac ( SEM ) pSTG-R ( 1–4 Hz ) pSTG-R ( 8–12 Hz ) 3 . 260 . 22 ( 0 . 07 ) SMG-R ( 4–8 Hz ) SMG-R ( 30–48 Hz ) 3 . 580 . 27 ( 0 . 07 ) IFGop-R ( 0 . 25–1 Hz ) VC-R ( 18–24 Hz ) 3 . 080 . 22 ( 0 . 07 ) VC-R ( 4–8 Hz ) VC-R ( 8–12 Hz ) 3 . 060 . 35 ( 0 . 11 ) VC-R ( 1–4 Hz ) VC-R ( 12–18 Hz ) 3 . 440 . 48 ( 0 . 13 ) VC-R ( 4–8 Hz ) VC-R ( 24–36 Hz ) 3 . 760 . 26 ( 0 . 07 ) We observed functionally distinct patterns of speech-to-brain entrainment along the auditory pathways . Previous studies on speech entrainment have largely focused on the auditory cortex , where entrainment to the speech envelope is strongest ( Ding and Simon , 2013; Gross et al . , 2013; Keitel et al . , 2017; Mesgarani and Chang , 2012; Zion Golumbic et al . , 2013a ) , and only few studies have systematically compared speech entrainment along auditory pathways ( Zion Golumbic et al . , 2013b ) . This was in part due to the difficulty to separate distinct processes reflecting entrainment when contrasting only few experimental conditions ( e . g . forward and reversed speech [Ding and Simon , 2012; Gross et al . , 2013] ) , or based on the difficulty to separate contributions from visual ( i . e . lip movements ) and acoustic speech signals ( Park et al . , 2016 ) . Based on the susceptibility to changes in acoustic signal quality and visual context , the systematic use of region-specific temporal lags between stimulus and brain response , and the systematic analysis of both acoustic and visual speech signals , we here establish entrainment as a ubiquitous mechanism reflecting distinct acoustic and visual speech representations along auditory pathways . Entrainment to the acoustic speech envelope was reduced with decreasing acoustic SNR in temporal , parietal and ventral prefrontal cortex , directly reflecting the reduction in behavioral performance in challenging environments . In contrast , entrainment was enhanced during low SNR in superior frontal and premotor cortex . While there is strong support for a role of frontal and premotor regions in speech processing ( Du et al . , 2014; Evans and Davis , 2015; Heim et al . , 2008; Meister et al . , 2007; Morillon et al . , 2015; Rauschecker and Scott , 2009; Skipper et al . , 2009; Wild et al . , 2012 ) , most evidence comes from stimulus-evoked activity rather than signatures of neural speech encoding . We directly demonstrate the specific enhancement of frontal ( PMC , SFG ) speech representations during challenging conditions . This enhancement is not directly inherited from the temporal lobe , as temporal regions exhibited either no visual facilitation ( pSTG ) or visual facilitation without an interaction with SNR ( HG ) . We also observed significant entrainment to the temporal trajectory of lip movements in visual cortex , the temporal lobe and frontal cortex ( Figure 3—figure supplement 1 ) . This confirms a previous study , which has specifically focused on the temporal coherence between brain activity and lip movements ( Park et al . , 2016 ) . Importantly , by comparing the local encoding of both the acoustic and visual speech information , and conditioning out the visual signal from the speech MI , we found that sensory cortices and the temporal lobe provide largely independent representations of the acoustic and visual speech signals . Indeed , the information theoretic redundancy between acoustic and visual representations was small and was significant only in association regions ( SFG , IFG , PMC ) . This suggests that early sensory cortices contain largely independent representations of acoustic and visual speech information , while association regions provide a superposition of auditory and visual speech representations . However , the condition effects on the acoustic representation in any of the analyzed regions did not disappear when factoring out the representation of lip movements , suggesting that these auditory and visual representations are differentially influenced by sensory context . These findings extend previous studies by demonstrating the co-existence of visual and auditory speech representations along auditory pathways , but also reiterate the role of PMC as one candidate region that directly links neural representations of lip movements with perception ( Park et al . , 2016 ) . Visual information from the speakers’ face provides multiple cues that enhance intelligibility . In support of a behavioral multisensory benefit we found stronger entrainment to the speech envelope during an informative visual context in multiple bilateral regions . First , we replicated the visual enhancement of auditory cortical representations ( HG ) ( Besle et al . , 2008; Kayser et al . , 2010; Zion Golumbic et al . , 2013a ) . Second , visual enhancement of an acoustic speech representation was also visible in early visual areas , as suggested by prior studies ( Nath and Beauchamp , 2011; Schepers et al . , 2015 ) . Importantly , our information theoretic analysis suggests that this representation of acoustic speech is distinct from the visual representation of lip dynamics , which co-exists in the same region . The visual enhancement of acoustic speech encoding in visual cortex was strongest when SNR was low , unlike the encoding of lip movements , which was not affected by acoustic SNR . Hence this effect is most likely explained by top-down signals providing acoustic feedback to visual cortices ( Vetter et al . , 2014 ) . Third , speech representations in ventral prefrontal cortex were selectively involved during highly reliable multisensory conditions and were reduced in the absence of the speakers face . These findings are in line with suggestions that the IFG facilitates comprehension ( Alho et al . , 2014; Evans and Davis , 2015; Hasson et al . , 2007b; Hickok and Poeppel , 2007 ) and implements multisensory processes ( Callan et al . , 2014 , 2003; Lee and Noppeney , 2011 ) , possibly by providing amodal phonological , syntactic and semantic processes ( Clos et al . , 2014; Ferstl et al . , 2008; McGettigan et al . , 2012 ) . Previous studies often reported enhanced IFG response amplitudes under challenging conditions ( Guediche et al . , 2014 ) . In contrast , by quantifying the fidelity of speech representations , we here show that speech encoding is generally better during favorable SNRs . This discrepancy is not necessarily surprising , if one assumes that IFG representations are derived from those in the temporal lobe , which are also more reliable during high SNRs . Noteworthy , however , we found that speech representations within ventral IFG are selectively stronger during an informative visual context , even when discounting direct co-representations of lip movements . We thereby directly confirm the hypothesis that IFG speech encoding is enhanced by visual context . Furthermore , we demonstrate the visual enhancement of speech representations in premotor regions , which could implement the mapping of audio-visual speech features onto articulatory representations ( Meister et al . , 2007; Morillon et al . , 2015; Morís Fernández et al . , 2015; Skipper et al . , 2009; Wilson et al . , 2004 ) . We show that that this enhancement is inversely related to acoustic signal quality . While this observation is in agreement with the notion that perceptual benefits are strongest under adverse conditions ( Ross et al . , 2007; Sumby and Pollack , 1954 ) , there was no significant correlation between the visual enhancement of premotor encoding and behavioral performance . Our results thereby deviate from previous work that has suggested a driving role of premotor regions in shaping intelligibility ( Alho et al . , 2014; Osnes et al . , 2011 ) . Rather , we support a modulatory influence of auditory-motor interactions ( Alho et al . , 2014; Callan et al . , 2004; Hickok and Poeppel , 2007; Krieger-Redwood et al . , 2013; Morillon et al . , 2015 ) . In another study we recently quantified dynamic representations of lip movements , calculated when discounting influences of the acoustic speech , and reported that left premotor activity was significantly predictive of behavioral performance ( Park et al . , 2016 ) . One explanation for this discrepancy may be the presence of a memory component in the present behavioral task , which may engage other brain regions ( e . g . IFG ) more than other tasks . Another explanation could be that premotor regions contain , besides an acoustic speech representation described here , complementary information about visual speech that is not directly available in the acoustic speech contour , and is either genuinely visual or correlated with more complex acoustic properties of speech . Further work is required to disentangle the multisensory nature of speech encoding in premotor cortex . Finally , our results highlight an interesting role of the superior frontal gyrus , where entrainment was strongest when sensory information was most impoverished ( low SNR , visual not informative ) or when the speakers face was combined with clear speech ( high SNR , visual informative ) . Superior frontal cortex has been implied in high level inference processes underlying comprehension , sentence level integration or the exchange with memory ( Ferstl et al . , 2008; Hasson et al . , 2007a; Yarkoni et al . , 2008 ) and is sometimes considered part of the broader semantic network ( Binder et al . , 2009; Gow and Olson , 2016; Price , 2012 ) . Our data show that the SFG plays a critical role for speech encoding under challenging conditions , possibly by mediating sentence-level processes during low SNRs or the comparison of visual prosody with acoustic inputs in multisensory contexts . To understand whether the condition-specific patterns of local speech representations emerge within each region , or whether they are possibly established by network interactions , we investigated the directed functional connectivity between regions of interest . While many studies have assessed the connectivity between auditory regions ( e . g . [Abrams et al . , 2013; Chu et al . , 2013; Fonteneau et al . , 2015; Park et al . , 2015] ) , few have quantified the behavioral relevance of these connections ( Alho et al . , 2014 ) . We observed significant intra-hemispheric connectivity between right temporal , parietal and frontal regions , in line with the transmission of speech information from the temporal lobe along the auditory pathways ( Bornkessel-Schlesewsky et al . , 2015; Hickok , 2012; Poeppel , 2014 ) . Supporting the idea that acoustic representations are progressively transformed along these pathways we found that the condition-specific patterns of functional connectivity differed systematically along the ventral and dorsal streams . While connectivity along the ventral stream was predictive of behavioral performance and strongest during favorable listening conditions , the inter-hemispheric connectivity to left premotor cortex was strongest during adverse multisensory conditions , i . e . when seeing the speakers face at low SNR . Interestingly , this pattern of functional connectivity matches the pattern of speech entrainment in PMC , reiterating the selective and distinctive contribution of premotor regions in speech encoding during multisensory conditions ( Park et al . , 2016 ) . Our results therefore suggest that premotor representations are informed by auditory regions ( HG , pSTG ) , rather than being driven by the frontal lobe , an interpretation that is supported by previous work ( Alho et al . , 2014; Gow and Olson , 2016; Osnes et al . , 2011 ) . We also observed a non-trivial pattern of connectivity between the SFG and visual cortex . Here the condition-specific pattern of connectivity was similar to the pattern of entrainment in the SFG , suggesting that high-level inference processes or sentence-level integration of information in the SFG contribute to the feed-back transmission of predictive information to visual cortex ( Vetter et al . , 2014 ) . For example , the increase of connectivity with decreasing SNR during the visual non-informative condition could serve to minimize the influence of visual speech information when this is in apparent conflict with the acoustic information in challenging environments ( Morís Fernández et al . , 2015 ) . Across conditions behavioral performance was supported both by an enhancement of speech representations along the ventral pathway as well as enhanced functional connectivity . This enhanced functional connectivity emerged both along feed-forward and feed-back directions between temporal and inferior frontal regions , and was strongest ( in effect size ) along the feed-back route . This underlines the hypothesis that recurrent processing , rather than a simple feed-forward sweep , is central to speech intelligibility ( Bornkessel-Schlesewsky et al . , 2015; Hickok , 2012; Poeppel , 2014 ) . Central to the scope of the present study , however , we found that no single region-specific effect could explain the visual behavioral benefit . Rather , the benefit arising from seeing the speakers face was significantly correlated with the enhancement of recurrent functional connectivity along the ventral stream ( HG → IFG → pSTG ) . Our results hence point to a distributed origin of the visual enhancement of speech intelligibility . As previously proposed ( Besle et al . , 2008; Ghazanfar et al . , 2005; Ghazanfar and Schroeder , 2006; Kayser et al . , 2010; Zion Golumbic et al . , 2013a ) this visual enhancement involves early auditory regions , but as we show here , also relies on the recurrent transformation of speech representations between temporal and frontal regions . While the effects of experimental conditions on speech MI dominated in the right hemisphere we found little evidence that these effects were indeed significantly stronger in one hemisphere . Indeed , only the SNR effect in IFGop was significantly lateralized , while all other effects were comparable between hemispheres . Hence care needs to be taken when interpreting our results as evidence for a lateralization of speech encoding . At the same time we note that a potential right dominance of speech entrainment is in agreement with the hypothesis that right temporal regions extract acoustic information predominantly on the syllabic and prosodic time scales ( Giraud and Poeppel , 2012; Poeppel , 2003 ) . Further , several studies have shown that the right hemisphere becomes particularly involved in the representation of connected speech ( Alexandrou et al . , 2017; Bourguignon et al . , 2013; Fonteneau et al . , 2015; Horowitz-Kraus et al . , 2015 ) , and one previous study directly demonstrated the prevalence of speech-to-brain entrainment in delta and theta bands in the right hemisphere during continuous listening ( Gross et al . , 2013 ) . This makes it little surprising that the right hemisphere becomes strongly involved in representing continuous multisensory speech . Furthermore , we a bias towards the right hemisphere may in part also be a by-product of the use of entrainment as a n index to characterize speech encoding , given that the signal power of acoustic and visual speech is highest at low frequencies ( c . f . Figure 1 ) , and given that the right hemisphere supposedly has a preference for speech information at long time scales ( Giraud and Poeppel , 2012; Poeppel , 2003 ) . Speech perception relies on mechanisms related to predictive coding , in order to fill in acoustically masked signals and to exploit temporal regularities and cross-modal redundancies to predict when to expect what type of syllable or phoneme ( Chandrasekaran et al . , 2009; Peelle and Sommers , 2015; Tavano and Scharinger , 2015 ) . Predictions modulate auditory evoked responses in an area specific manner , involve both the ventral and dorsal pathways ( Kandylaki et al . , 2016; Sohoglu and Chait , 2016 ) , and affect both feedforward and feedback connections ( Auksztulewicz and Friston , 2016; Chennu et al . , 2016 ) . While an informative visual context facilitates the correction of predictions about expected speech using incoming multisensory evidence , we can only speculate about a direct link between the reported effects and predictive processes . Previous studies have implied delta band activity and the dorsal auditory stream in mediating temporal predictions ( Arnal and Giraud , 2012; Arnal et al . , 2011; Kandylaki et al . , 2016 ) . Hence , the changes in delta speech entrainment across conditions seen here may well reflect changes related to the prevision of temporal predictions . Several computational candidate mechanisms have been proposed for how multisensory information could be integrated at the level of neural populations ( Ohshiro et al . , 2011; Pouget et al . , 2002; van Atteveldt et al . , 2014 ) . The focus on rhythmic activity in the present study lends itself to suggest a key role of the phase resetting of oscillatory process , as proposed previously ( Schroeder et al . , 2008; Thorne and Debener , 2014; van Atteveldt et al . , 2014 ) . However , given the indirect nature of the neuroimaging signals the present study can’t rule in or out the involvement of specific neural processes . Our results provide a network view on the dynamic speech representations in multisensory environments . While premotor and superior frontal regions are specifically engaged in the most challenging environments , the visual enhancement of comprehension at intermediate SNRs seems to be mediated by interactions within the core speech network along the ventral pathway . Such a distributed neural origin of multisensory benefits is in line with the notion of a hierarchical organization of multisensory processing , and the idea that comprehension is shaped by network connectivity more than the engagement of particular brain regions . The stimulus material consisted of audio-visual recordings based on text transcripts taken from publicly available TED talks also used in a previous study ( Kayser et al . , 2015 ) ( Figure 1A; see also [Park et al . , 2016] ) . Acoustic ( 44 . 1 kHz sampling rate ) and video recordings ( 25 Hz frame rate , 1920 by 1080 pixels ) were obtained while a trained male native English speaker narrated these texts . The root mean square ( RMS ) intensity of each audio recording was normalized using 6 s sliding windows to ensure a constant average intensity . Across the eight texts the average speech rate was 160 words ( range 138–177 ) per minute , and the syllabic rate was 212 syllables ( range 192–226 ) per minute . We presented each of the eight texts as continuous 6 min sample , while manipulating the acoustic quality and the visual relevance in a block design within each text ( Figure 1B ) . The visual relevance was manipulated by either presenting the video matching the respective speech ( visual informative , VI ) or presenting a 3 s babble sequence that was repeated continuously ( visual not informative , VN ) , and which started and ended with the mouth closed to avoid transients . The signal to noise ratio ( SNR ) of the acoustic speech was manipulated by presenting the speech on background cacophony of natural sounds and scaling the relative intensity of the speech while keeping the intensity of the background fixed . We used relative SNR values of +8 , +6 , +4 and +2 dB RMS intensity levels . The acoustic background consisted of a cacophony of naturalistic sounds , created by randomly superimposing various naturalistic sounds from a larger database ( using about 40 sounds at each moment in time ) ( Kayser et al . , 2016 ) . This resulted in a total of 8 conditions ( four SNR levels; visual informative or irrelevant ) that were introduced in a block design ( Figure 1B ) . The SNR changed from minute to minute in a pseudo-random manner ( 12 one minute blocks per SNR level ) . Visual relevance was manipulated within 3 min sub-blocks . Texts were presented with self-paced pauses . The stimulus presentation was controlled using the Psychophysics toolbox in Matlab ( Brainard , 1997 ) . Acoustic stimuli were presented using an Etymotic ER-30 tubephone ( tube length = 4 m ) at 44 . 1 kHz sampling rate and an average intensity of 65 dB RMS level , calibrated separately for each ear . Visual stimuli were presented in grey-scale and projected onto a translucent screen at 1280 × 720 pixels at 25 fps covering a field of view of 25 × 19 degrees . Subjects performed a delayed comprehension tasks after each block , whereby they had to indicate whether a specific word ( noun ) was mentioned in the previous text ( six words per text ) or not ( six words per text ) in a two alternative forced choice task . The words chosen from the presented text were randomly selected and covered all eight conditions . The average performance across all trials was 73 ± 2% correct ( mean and SEM across subjects ) , showing that subjects indeed paid attention to the stimulus . Behavioral performance for the words contained in the presented text was averaged within each condition , and analyzed using a repeated measures ANOVA , with SNR and VIVN as within-subject factors . By experimental design , the false alarm rate , i . e . the number of mistaken recognitions of words that were not part of the stimulus , was constant across experimental conditions . As a consequence , condition-specific d’ measures of word recall were strongly correlated with condition-specific word-recall performance ( mean correlation and SEM across subjects = 0 . 97 ± 0 . 06; T ( 18 ) for significant group-average Fisher-Z transformed correlation = 32 . 57 , p<0 . 001 ) . We extracted the envelope of the speech signal ( not the speech plus background mixture ) by computing the wide-band envelope at 150 Hz temporal resolution as in previous work ( Chandrasekaran et al . , 2009; Kayser et al . , 2015 ) . The speech signal was filtered ( fourth order Butterworth filter; forward and reverse ) into six frequency bands ( 100 Hz - 4 kHz ) spaced to cover equal widths on the cochlear map . The wide-band envelope was defined as the average of the Hilbert envelopes of these band-limited signals ( Figure 1A ) . The temporal trajectory of the lip contour was extracted by first identifying the lips based on their hue and then detecting the area of mouth-opening between the lips ( Park et al . , 2016 ) . For each video frame , the mouth aperture was subsequently estimated as the area covered by an ellipsoid fit to the detected lip contours , which was then resampled to 150 Hz for further analysis ( Figure 1A ) . We estimated the coherence between the speech envelope and lip contour using spectral analysis ( Figure 1A ) . MEG recordings were acquired with a 248-magnetometers whole-head MEG system ( MAGNES 3600 WH , 4-D Neuroimaging ) at a sampling rate of 1017 . 25 Hz . Participants were seated upright . The position of five coils , marking fiducial landmarks on the head of the participants , was acquired at the beginning and at the end of each block . Across blocks , and participants , the maximum change in their position was 3 . 6 mm , on average ( STD = 1 . 2 mm ) . Analyses were carried out in Matlab using the Fieldtrip toolbox ( Oostenveld et al . , 2011 ) , SPM12 , and code for the computation of information-theoretic measures ( Ince et al . , 2017 ) . Block-specific data were pre-processed separately . Infrequent SQUID jumps ( observed in 1 . 5% of the channels , on average ) were repaired using piecewise cubic polynomial interpolation . Environmental magnetic noise was removed using regression based on principal components of reference channels . Both the MEG and reference data were filtered using a forward-reverse 70 Hz FIR low-pass ( −40 dB at 72 . 5 Hz ) ; a 0 . 2 Hz elliptic high-pass ( −40 dB at 0 . 1 Hz ) ; and a 50 Hz FIR notch filter ( −40 dB at 50 ± 1 Hz ) . Across participants and blocks , 7 MEG channels were discarded as they exhibited a frequency spectrum deviating consistently from the median spectrum ( shared variance <25% ) . For analysis signals were resampled to 150 Hz and once more high-pass filtered at 0 . 2 Hz ( forward-reverse elliptic filter ) . ECG and EOG artefacts were subsequently removed using ICA in fieldtrip ( runica , 40 principal components ) , and were identified based on the time course and topography of IC components ( Hipp and Siegel , 2013 ) . High resolution anatomical MRI scans were acquired for each participant ( voxel size = 1 mm3 ) and co-registered to the MEG data using a semi-automated procedure . Anatomicals were segmented into grey and white matter and cerebro-spinal fluid ( Ashburner and Friston , 2005 ) . The parameters for the affine registration of the anatomical to the MNI template were estimated , and used to normalize the grey matter probability maps of each individual to the MNI space . A group MNI source-projection grid with a resolution of 3 mm was prepared including only voxels associated with a group-average grey-matter probability of at least 0 . 25 . The projection grid excluded various subcortical structures , identified using the AAL atlas ( e . g . , vermis , caudate , putamen and the cerebellum ) . Leadfields were computed based on a single shell conductor model . Time-domain projections were obtained on a block-by-block basis using LCMV spatial filters ( regularization = 5% ) . A different LCMV filter was used for each frequency band by computing the sensor covariance for the band-pass filtered sensor signals . Further analyses focused on the maximum-variance orientation of each dipole . Motivated by previous work ( Gross et al . , 2013; Ng et al . , 2013 ) , we considered eight partly overlapping frequency bands ( 0 . 25–1 Hz , 1–4 Hz , 4–8 Hz , 8–12 Hz , 12–18 Hz , 18–24 Hz , 24–36 Hz , and 30–48 Hz ) , and isolated these from the full-spectrum MEG signals , the speech envelope and the lip trajectory in each band using a forward-reverse fourth order Butterworth filter ( magnitude of frequency response at band limits = −6 dB ) . Entrainment was quantified using the mutual information ( MI ) between the filtered MEG and speech envelope or lip time courses: ( 1 ) MI_speech=MI ( MEG;speech ) andMI_lip=MI ( MEG;lip ) The MI was calculated using a bin-less approach based on statistical copulas , which provides greater sensitivity than methods based on binned signals ( Ince et al . , 2017 ) . To quantify the entrainment of brain activity to the speech envelope / lip movement we first determined the optimal time lag between MEG signals and the stimulus for individual bands and source voxels using a permutation-based RFX estimate . Lag estimates were obtained based on a quadratic fit , excluding lags with insignificant MI ( permutation-based FDR = 0 . 01 ) . Voxels without an estimate were assigned the median estimate within the same frequency band , and volumetric maps of the optimal lags were smoothed with a Gaussian ( FWHM = 10 mm ) . Speech / lip MI were then estimated for each band and voxel using the optimal lag . The significance of group-level MI values was assessed within a permutation-based RFX framework that relied on MI values corrected for bias at the single-subject level , and on cluster mass enhancement of the test statistics corrected for multiple comparisons at the second level ( Maris and Oostenveld , 2007 ) . At the single-subject level , null distributions were obtained by shuffling the assignment of stimulus and MEG , independently for each participant , that is , by permuting the six speech segments within each of the eight experimental conditions ( using the same permutation across bands ) . Participant-specific bias-corrected MI values were then defined as the actual MI minus the median MI across all 720 possible null permutations . Group-level RFX testing relied on T-statistics for the null-hypothesis that the participant-averaged bias-corrected MI was significantly larger than zero . To this end we generated 10 , 000 samples of the group-averaged MI from the participant-specific null distributions , used cluster-mass enhancement across voxels and frequencies ( cluster-forming threshold T ( 18 ) = 2 . 1 ) to extract the maximum cluster T across frequency bands and voxels , and considered as significant a cluster-enhanced T statistic higher than the 95th percentile of the permutation distribution ( corresponding to FWE = 0 . 05 ) . Significant speech MI was determined across all conditions , whereas significant lip MI was derived only for the VI condition . To determine whether and where speech / lip entrainment was modulated by the experimental factors we used a permutation-based RFX GLM framework ( Winkler et al . , 2014 ) . For each participant individually we considered the condition-specific bias-corrected MI averaged across repetitions and estimated the coefficients of a GLM for predicting MI based on SNR ( 2 , 4 , 6 , 8 dB ) , VIVN ( 1 = Visual Informative; −1 = Visual Not informative ) , and their interaction; for lip MI we only considered the SNR effect in the VI condition . We computed a group-level T-statistic for assessing the hypothesis that the across-participant average GLM coefficient was significantly different than zero , using cluster-mass enhancement across voxels and frequencies . Permutation testing relied on the Freedman-Lane procedure ( Freedman and Lane , 1983 ) . Independently for each participant and GLM effect , we estimated the parameters of a reduced GLM that includes all of the effects but the one to be tested and extracted the residuals of the prediction . We then permuted the condition-specific residuals and extracted the GLM coefficient for the effect of interest estimated for these reshuffled residuals . We obtained a permutation T statistic for the group-average GLM coefficient of interest using the max-statistics . We considered as significant T values whose absolute value was higher than the 95th percentile of the absolute value of 10 , 000 permutation samples , correcting for multiple comparisons across voxels / bands ( FWE = 0 . 05 ) . We only considered significant GLM effects in conjunction with a significant condition-average entrainment . To quantify directed functional connectivity we relied on the concept of Wiener-Granger causality and its information theoretic implementation known as Transfer Entropy or directed information ( DI ) ( Massey , 1990; Schreiber , 2000; Vicente et al . , 2011; Wibral et al . , 2011 ) . Directed information in its original formulation ( Massey , 1990 ) ( termed DI* here ) quantifies causal connectivity by measuring the degree to which the past of a seed predicts the future of a target signal , conditional on the past of the target , defined at a specific lag ( τBrain ) : ( 2 ) DI∗ ( τBrain ) =I ( Targett;Seedt−τ\|Targett−τ ) While DI* provides a measure of the overall directed influence from seed to target , it can be susceptible to statistical biases arising from limited sampling , common inputs or signal auto-correlations ( Besserve et al . , 2015 , 2010; Ince et al . , 2017; Panzeri et al . , 2007 ) . We regularized and made this measure more conservative by subtracting out values of DI computed at fixed values of speech envelope . This subtraction removes terms – including the statistical biases described above – that cannot possibly carry speech information ( because they are computed at fixed speech envelope ) . This results in an estimate that is more robust and more directly related to changes in the sensory input than classical transfer entropy ( the same measure was termed directed feature information in [Ince et al . , 2017 , Ince et al . , 2015] ) . DI was defined here as ( 3 ) DI ( τBrain , τSpeech ) =DI∗ ( τBrain ) −DI∗ ( τBrain ) \|Speech ( τSpeech ) where DI\|Speech denotes the DI conditioned on the speech envelope . Positive values of DI indicate directed functional connectivity between seed and target at a specific brain ( τBrain ) and speech lag ( τSpeech ) . The actual DI values were furthermore Z-scored against random effects for added robustness , which facilitates statistical comparisons between conditions across subjects ( Besserve et al . , 2015 ) . To this end DI , as estimated for each participant and connection from Equation 3 , was Z-scored against the distribution of DI values obtained from condition-shuffled estimates ( using the same randomization procedure as for MI ) . DI was computed for speech lags between 0 and 500 ms and brain lags between 0 and 250 ms , at steps of one sample ( 1/150 Hz ) . We estimated DI on the frequency range of 0 . 25–8 Hz ( forward-reverse fourth order Butterworth filter ) , which spans all the frequencies relevant for the condition effects on speech MI ( Figure 3 ) . The use of a single frequency band for the connectivity analysis greatly reduced the computational burden and statistical testing compared to the use of multiple bands , while the use of a larger bandwidth here also allowed for greater robustness of underlying estimators ( Besserve et al . , 2010 ) . Furthermore , we computed DI by considering the bivariate MEG response defined by the band-passed source signal and its first-order difference , as this offers additional statistical robustness ( Ince et al . , 2017 , 2016 ) . Seeds for the DI analysis were the global and local peaks of the GLM-T maps quantifying the SNR , VIVN and SNRxVIVN modulation of entrainment , and the SFG-R voxel characterized by the peak negative effect of SNR in the visual informative condition , for a total of 8 seeds ( Table 1 and Figure 3E ) . To test for the significance of condition-average DI we used the same permutation-based RFX approach as for speech MI , testing the hypothesis that bias-corrected DI > 0 . We used 2D cluster-mass enhancement of the T statistics within speech/brain lag dimensions correcting for multiple comparisons across speech and brain lags ( FWE = 0 . 05 ) . To test for significant DI effects with experimental conditions we relied on the same GLM strategy as for MI effects , again with the same differences pertaining to cluster enhancement and comparison correction ( FWE = 0 . 05 across lags and seed/target pairs ) . We only considered DI modulations in conjunction with a significant condition-average DI . We used a permutation-based RFX approach to assess ( 1 ) whether an increase in condition-specific speech-MI or DI was associated with an increase in behavioral performance , and ( 2 ) whether the visual enhancement ( VI-VN ) of speech MI or DI was associated with stronger behavioral gains . We focused on the eight regions used as seeds for the DI analysis ( c . f . Figure 3E ) . For speech MI we initially tested whether the participant-average Fisher Z-transformed correlation between condition-specific performance and speech-MI was significantly larger than zero . Uncorrected p-values were computed using the percentile method , where FWE = 0 . 05 p-values corrected across regions were computed using maximum statistics . We subsequently tested the positive correlation between SNR-specific visual gains ( VI-VN ) in speech-MI and behavioral performance using the same approach , but considered only those regions characterized by a significant condition-specific MI/performance association . For DI , we focused on those lags characterized by a significant SNR , VIVN , or SNRxVIVN DI modulation . Significance testing proceeded as for speech MI , except that Z-transformed correlations were computed independently for each lag and then averaged across lags ( FWE = 0 . 05 corrected across all seed/target pairs ) . We tested for a significant lateralization of the GLM effects on speech MI reported in Figure 3 . To this end we extracted participant specific GLM betas for each effect in the respective ROI and band . We then extracted the same GLM coefficient for the contralateral voxel and computed the between-hemispheric difference . This was tested for significance using a two-sided RFX test based on a sign-permutation of the across-participant T value ( 10 , 000 permutations ) , with maximum-statistic multiple comparison correction across ROIs ( FWE = 0 . 05; Table 1 ) . To test whether the condition modulation of speech MI could be attributed to a co-representation of visual lip information in the same ROI we calculated the conditional information between the MEG and the speech envelope , factoring out the encoding of temporal dynamics common to the speech and lip signals . With MI_speech&lip defined as MI ( MEG;speech , lip ) , the CMI was defined as follows ( 4 ) CMI ( MEG;speech\|lip ) =MI_speech&lip−MI_lip where the first term on the right-hand side denotes the information carried by the local MEG signal about both the acoustic and visual speech , and the second term the MI about only the visual speech . The respective CMI values were then tested for significant condition effects ( Table 2 ) . To further test whether the local representations of acoustic and visual speech in each ROI were independent or possibly strongly redundant ( hence capturing the same aspect of sensory information ) , we computed a measure of normalized information theoretic redundancy during the VI condition as follows ( Belitski et al . , 2010; Pola et al . , 2003; et al . , 2003 ) : ( 5 ) Red= ( MI_speech+MI_lip−MI_speech&lip ) / ( MI_speech+MI_lip ) ∗100 This expresses redundancy as percentage of the total information that there would be in its absence of any redundancy . For these analysis both speech and lip signals were extracted at their respective optimal lag for each ROI/band and a common segment to each stimulus and the MEG activity was used for the calculation ( segment duration = 60 s – 320 ms ) . Statistical tests contrasting condition-averaged information terms relied on the same RFX permutation framework and correction across all relevant dimensions as in all other analyses ( FWE = 0 . 05 ) . We compared condition-averaged MI_speech with MI_lip values using a two-sided test , contrasted condition-averaged redundancy values with their statistical bias ( null-distribution ) , and tested for condition effects ( GLM ) on the CMI values . The amplitude within specific bands was defined as the absolute value of the instantaneous Hilbert-transformed band-pass MEG signal beamformed to each of the ROIs ( c . f . Figure 3E ) . For each participant and experimental condition , we averaged the amplitude of the MEG time courses across time and repetitions of the same condition . Significance testing of condition changes in amplitude relied on the same RFX permutation-based approach as for the other modulation analyses , with maximum statistic multiple comparisons correction across ROIs and frequency bands ( FWE = 0 . 05 ) . We computed a measure of phase-amplitude coupling ( PAC ) between the oscillatory activity in different bands and regions . PAC was defined as ( 6 ) PAC=∑t=1NAFH ∗ eiθFL/N where AFH and θFL denote the instantaneous Hilbert amplitude and phase angle of the high- and low-frequency MEG pass-band signal , respectively , and N is the number of time samples of the pass-band MEG signal in a specific condition . Low-frequency phase was extracted for the 0 . 25–1 , 1–4 , and 4–8 Hz bands . High-frequency amplitude was extracted for the 8–12 , 12–18 , 18–24 , 24–36 and 30–48 Hz bands . We tested for both a significant condition-average PAC and for a significant modulation of PAC with conditions . Significance testing relied on the same RFX permutation-based approach as for the other modulation analyses , with maximum statistic correction for multiple comparisons across pairs of phase/power frequency pairs for the significance of condition averaged PAC , and also across pairs of phase/power ROIs for the GLM modulation ( FWE = 0 . 05 ) . The data analyzed for the ROI results presented in Figures 2–5 , in the figure supplements and in Tables 1–5 , as well as the speech and lip time courses analyzed in Figure 1 , have been deposited on Dryad ( doi:10 . 5061/dryad . j4567 ) .	When listening to someone in a noisy environment , such as a cocktail party , we can understand the speaker more easily if we can also see his or her face . Movements of the lips and tongue convey additional information that helps the listener’s brain separate out syllables , words and sentences . However , exactly where in the brain this effect occurs and how it works remain unclear . To find out , Giordano et al . scanned the brains of healthy volunteers as they watched clips of people speaking . The clarity of the speech varied between clips . Furthermore , in some of the clips the lip movements of the speaker corresponded to the speech in question , whereas in others the lip movements were nonsense babble . As expected , the volunteers performed better on a word recognition task when the speech was clear and when the lips movements agreed with the spoken dialogue . Watching the video clips stimulated rhythmic activity in multiple regions of the volunteers’ brains , including areas that process sound and areas that plan movements . Speech is itself rhythmic , and the volunteers’ brain activity synchronized with the rhythms of the speech they were listening to . Seeing the speaker’s face increased this degree of synchrony . However , it also made it easier for sound-processing regions within the listeners’ brains to transfer information to one other . Notably , only the latter effect predicted improved performance on the word recognition task . This suggests that seeing a person’s face makes it easier to understand his or her speech by boosting communication between brain regions , rather than through effects on individual areas . Further work is required to determine where and how the brain encodes lip movements and speech sounds . The next challenge will be to identify where these two sets of information interact , and how the brain merges them together to generate the impression of specific words .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Contributions of local speech encoding and functional connectivity to audio-visual speech perception
Lab organisms are valuable in part because of large-scale experiments like screens , but performing such experiments over long time periods by hand is arduous and error-prone . Organism-handling robots could revolutionize large-scale experiments in the way that liquid-handling robots accelerated molecular biology . We developed a modular automated platform for large-scale experiments ( MAPLE ) , an organism-handling robot capable of conducting lab tasks and experiments , and then deployed it to conduct common experiments in Saccharomyces cerevisiae , Caenorhabditis elegans , Physarum polycephalum , Bombus impatiens , and Drosophila melanogaster . Focusing on fruit flies , we developed a suite of experimental modules that permitted the automated collection of virgin females and execution of an intricate and laborious social behavior experiment . We discovered that ( 1 ) pairs of flies exhibit persistent idiosyncrasies in social behavior , which ( 2 ) require olfaction and vision , and ( 3 ) social interaction network structure is stable over days . These diverse examples demonstrate MAPLE’s versatility for automating experimental biology . Genetic model organisms are used to advance our biological understanding in numerous areas including disease and its treatment , basic cell biology , neuroscience and behavior . Species like Saccharomyces cerevisiae , Caenorhabditis elegans , and Drosophila melanogaster are desirable lab model organisms due to their rapid reproduction , ease of rearing , and especially their deep genetic toolkits comprising strains with varying genotypes and transgenic alterations that permit rapid , mechanistic inquiries . To take advantage of these toolkits , screen experiments quantify the phenotypes of hundreds ( Vitaterna et al . , 1994 ) , thousands ( Kain et al . , 2012 ) , tens of thousands ( Ayroles et al . , 2015; Buchanan et al . , 2015; Churgin et al . , 2017 ) or even hundreds of thousands of individual animals ( Robie et al . , 2017 ) . With the ongoing improvement and widespread adoption of high-performance machine vision phenotyping ( Branson et al . , 2009; Dankert et al . , 2009; Kabra et al . , 2013; Kimura et al . , 2014 ) , the time needed to manually handle experimental animals remains the bottleneck limiting data collection . The advent of liquid-handling robots has radically changed the face of molecular biology , enabling techniques and experiments that had long been imagined , but were too complex , lengthy , or tedious to have been previously realized . But there are no comparable systems for handling of experimental organisms that are larger than ~1 mm or cannot be suspended in liquid . There are large-scale systems that handle adult flies in vials , in the form of ‘fly flipping’ robots . However , these take up a whole room , generally in a core facility , and cost hundreds of thousands of dollars to purchase and maintain , and are therefore inaccessible to most labs . And while there are many examples of high-throughput phenotypic assays in Drosophila ( Branson et al . , 2009; Kabra et al . , 2013; Kain et al . , 2012; Buchanan et al . , 2015; Geissmann et al . , 2017 ) , the systems that are not single-purpose still require human intervention to load and unload individual flies . Some researchers have used flies' natural tendency to climb up ( negative gravitaxis ) to isolate individuals for behavioral analysis ( von Reyn et al . , 2014 ) , imaging ( Medici et al . , 2015 ) , or microsurgery ( Savall et al . , 2015 ) . But dependence on this particular behavior fundamentally caps throughput , inadvertently selects for a subset of a population , and limits eligible genotypes . An alternative approach , actively conveying flies with airflow ( MacMillan and Hughson , 2014 ) permits moving animals on demand , and opens the door for increased throughput . The dearth of instruments for automating Drosophila experiments is representative of the situation for many other lab organisms , such as yeast and C . elegans , where there is no standard platform for automated handling of the organisms themselves . Here , we present an automated platform that is high-throughput and flexible enough to assist in conducting diverse experimental protocols in Drosophila and other species . Due to its modular design , the system can automate diverse assays in a wide variety of organisms ( including yeast , C . elegans , the slime mold Physarum polycephalum , and the bumblebee Bombus impatiens , an assortment chosen to demonstrate the platform’s versatility ) . For fruit flies , where we developed significant capabilities , this instrument can conduct numerous protocols , including loading of individual fruit flies for circadian rhythm ( Pfeiffenberger et al . , 2010; Geissmann et al . , 2017 ) experiments , or aiding with lab chores like collecting virgin female flies for genetic crosses or passaging individual flies in controlled culture conditions for longevity assays . The physical platform of this instrument integrates animal husbandry and phenotyping , permitting end-to-end experimental protocols . Its low cost ( ~$3 , 500 ) , programmability , and scalability permit large-scale experiments that take advantage of the many benefits lab model organisms offer , such as huge fly genetic libraries containing thousands of lines ( Jenett et al . , 2012; Thibault et al . , 2004 ) . After demonstrating MAPLE’s breadth of utility across species , we highlight the depth of its capabilities with two particularly time- and manual labor-intensive fruit fly tasks: rapidly collecting virgin females , and large-scale longitudinal measurement of fly social networks and behavior . The latter experiment reveals previously unknown stability of Drosophila social interactions , and confirms that both olfaction and vision are required for dyad-specific social interactions in this context . With the high level goals of modularity , scalability , and automatability in mind , we designed the MAPLE system with the following design constraints: ( 1 ) It features a large , flat experimental workspace with room for multiple flexibly-configurable experimental modules . ( 2 ) This workspace is physically open for user convenience , and transparent on the top and bottom for in situ optical phenotyping . ( 3 ) Multiple end-effectors can move throughout the workspace to handle organisms , capture images , and manipulate experimental modules . ( 4 ) It features failsafe mechanisms so that users can leave it unattended without worrying that it would damage itself or experimental modules . ( 5 ) It is relatively inexpensive and scalable . MAPLE ( Figure 1 , Figure 1—figure supplement 1A ) was built using extruded aluminum rails to support x- , y- , and z-carriages mounted on linear rails in a Cartesian configuration . We employed the CoreXY system ( Moyer , 2012 ) , which reduces the mass of the moving part of the X/Y gantry by fixing the stepper motors on the frame . ( However , for the speeds at which we run MAPLE , which are roughly 80% as fast as human hands conducting experiments ( Video 1 ) , mounting the y-axis stepper motor on the x-axis carriage would likely not reduce performance . ) With respect to our design constraints: ( 1 ) the accessible experimental workspace measures 100 × 28 . 2×7 . 5 cm on the x- , y- , and z-axes , respectively . Its floor is clear acrylic with cable/tubing pass-throughs . Locating brackets ( laser cut out of 6 mm acrylic ) were affixed to an interchangeable acrylic surface with the same footprint as the floor ( a ‘workspace plate’ ) , allowing experimental modules to be precisely and repeatably positioned within and removed from the workspace ( Figure 1—figure supplement 2 ) . Interchanging workspace plates allows rapid reconfiguration of the workspace for different experimental procedures . ( 2 ) The sides , top , and bottom of MAPLE are open or made of clear acrylic , permitting the optical phenotyping of flies in experimental modules at all time points other than when the end-effector carriages are above the modules . ( 3 ) The end-effector assembly comprises three independent z-axes , each featuring a single tool ( Figure 1A; Video 2 ) : an object manipulator for picking up experimental module components like plastic lids using vacuum; a USB digital camera with LED illumination for acquiring high-resolution images for machine-vision; and an organism manipulator for handling small individual animals using vacuum , or , in the case of our yeast experiments , wooden applicators ( Figure 1—figure supplement 3 ) . ( 4 ) All motion axes have physical limit switches and/or software limits preventing overtravel . The organism manipulator end effector , which is rigid and must align precisely with experimental modules at different heights , is equipped with a collision-detection switch to halt z-motion before the robot is damaged . This sensor can also be used to detect the height of rigid module components . It is safe to leave MAPLE unattended ( Video 3 ) . ( 5 ) MAPLE components cost approximately $3500 . Its bill of materials ( Supplementary file 1 , assembly instructions ( Supplementary file 2 ) , and code libraries ( see Materials and methods for links ) have been made public under open source licenses . MAPLE measures 43 . 5 cm in the y-dimension permitting mounting in standardized 19’ rack systems , so multiple robots can be arranged compactly . We first established that MAPLE can be used to automate experiments in a wide variety of lab organisms . Specifically , we implemented experimental MAPLE protocols for baker’s yeast S . cerevisiae , the nematode C . elegans , the slime mold P . polycephalum , and the fruit fly D . melanogaster ( Figure 1B–E ) . For yeast , we programmed MAPLE to transfer yeast cells from a single colony on a source plate to target plates and streak the cells out to grow new colonies from single cells ( Figure 1B , Video 4 ) . For C . elegans , we used MAPLE’s camera effector to record 1 Hz movies of worms foraging on a lawn of OP50 bacteria ( normal C . elegans culture conditions; Figure 1C ) . We tracked these worms offline to produce Video 5 , demonstrating that MAPLE’s camera is of high enough fidelity to capture worm behavior on this spatial scale . Collecting such movies from multiple plates serially would allow MAPLE to conduct behavioral screens . Next , for the slime mold Physarum , we plated nine plasmodia on nine plates of 2% agar ( Figure 1D ) . We programmed MAPLE to take a photo of each plate once per minute and recorded plasmodia movement over the next 12 hr ( Video 6 ) . From these images , we compiled a combined time-lapse ( Video 7 ) of plasmodial outgrowth and motion , which contrasts the exploratory behaviors of the different individuals . Lastly , for fruit flies , we developed a variety of experimental modules that can be flexibly reconfigured to conduct numerous experiments . These are detailed in the rest of the paper , along with new scientific results obtained with them . Automated fruit fly experiments generally exploit MAPLE’s organism manipulator to move individual flies between compartments where the flies are grown or housed and phenotyping compartments where experimental data are automatically collected ( Figure 1E ) . MAPLE’s modularity , open-source design , and hierarchical software architecture also facilitates hardware modifications that expand its multi-species capabilities . We adapted the basic MAPLE design for high-throughput imaging of uniquely identified workers within colonies of the Common Eastern bumblebee ( Bombus impatiens ) . In the Bee Experimental Ethology Colony Hardware ( BEECH ) system ( Figure 1F , G ) , up to 12 colonies of ~50 bumblebees each are housed in acrylic colony boxes featuring a dark nest chamber ( Video 8 ) and a circadian-lit foraging chamber , where bees are supplied with nectar and pollen . This MAPLE-derived two-dimensional Cartesian robot moves a multi-camera end-effector from colony to colony , recording high-resolution video of the nest and foraging chambers ( Videos 8 and 9 ) , which permits automated tracking of individual tagged bees using machine vision ( Crall et al . , 2015 ) for up to several weeks . While replacing MAPLE’s z-axis assembly with multiple cameras and extending all three dimensions of MAPLE’s frame to accommodate the 12-colony box modules , BEECH employs identical construction and motion-control , demonstrating the versatility of the MAPLE design . An important step in adopting a new automated approach is confirming that the procedure generates similar data to prior manual experiments . To do this , we focused on fruit flies , and set out to confirm that handling by MAPLE did not damage flies or introduce discrepancies compared to experiments conducted manually . Specifically , we examined the locomotor performance of flies handled by MAPLE and unhandled flies . First , we manually aspirated 96 anesthetized flies into a FlyPlate ( a modified 96-well plate for single fly storage . Detailed description below ) . After 2 hr of rest , half the flies ( 48 ) were subjected to repeated manual removal and replacement back into their well using the MAPLE organism manipulator , while the remaining flies ( 48 ) were left unhandled ( Figure 2A , B ) . After this handling procedure , the entire plate was imaged in a backlit motion tracking rig ( Buchanan et al . , 2015 ) . Handling and imaging were then repeated an additional four times to test for cumulative effects . Flies handled by MAPLE were statistically indistinguishable from unhandled flies both in the fraction that were active across imaging sessions ( Figure 2C; b = −0 . 02 , t ( 477 ) = −0 . 098 , p=0 . 92 by multinomial logistic regression ) and in their mean walking speed across imaging sessions ( Figure 2D; F ( 4 , 376 ) =0 . 51 , p=0 . 73 by mixed-effects ANOVA ) . Second , we measured the locomotor bias of individual flies in Y-shaped mazes ( Ayroles et al . , 2015; Buchanan et al . , 2015 ) configured with multi-position loading ports . Awake flies were loaded into these mazes from FlyPlates by either manual aspiration or MAPLE-handling ( Figure 2E , F ) . The across-individual distribution of walking speeds was statistically indistinguishable between the manual and MAPLE-handled group ( Figure 2G; p=0 . 61 by KS-test ) . Likewise , the across-individual distribution of turning bias ( the tendency of individuals to turn left or right at the choice point in the center of the Y-maze ) was indistinguishable between handling treatments ( Figure 2H; p=0 . 74 by KS-test ) . The same was true for comparisons of manual and MAPLE-handled behavior in semi-circular arenas ( Figure 2—figure supplement 1 ) . Having confirmed that MAPLE does not obviously distort fly behavioral data , we set out to create an ecosystem of reconfigurable modules that could be used combinatorially in the MAPLE workspace to conduct a large number of experimental protocols . We fabricated and deployed a number of modules , which fall into three categories ( Figure 3 ) : fly source , fly sink , and phenotyping modules . Fly source modules are repositories from which flies can be removed in a controlled fashion and transferred into downstream modules . These include the Fly Dispenser ( FlySorter LLC ) , a small device that outputs single flies , on demand , from a standard plastic vial pre-loaded with many flies . It can be triggered to dispense a fly via a serial command sent over USB . A standard CO2 pad with a porous polyethylene surface , used to anesthetize flies manually at the start of a MAPLE session , serves as a source of flies . Using machine vision , MAPLE is capable of recognizing flies’ positions on the pad . FlyPlates ( FlySorter LLC ) are modified 96-well plates in which the floor has been replaced with a metal mesh , allowing flies stored in the wells to feed on fly media below the plate . The lid features a nylon mesh with X-shaped slits cut above each well ( Figure 2B ) , which allow an aspirator tip or the organism manipulator end-effector to enter the well , retrieve or deposit a fly , and leave the well without permitting the fly to escape . Because flies can be deposited in this module , it also falls in the category of fly sink modules . Fly sink modules are destinations into which flies that have been handled by MAPLE can be deposited . In the case of the FlyPlate , deposited flies can be later removed . Other fly sink modules are one-way , including a morgue , a dish of slightly soapy water or 70% ethanol covered with a nylon mesh lid in the style of the FlyPlate . Standard fly culture media vials with nylon mesh loading adapters can be used to collect many flies after MAPLE handling for long-term storage . The last category of modules is phenotyping modules , which produce experimental data . We have created a number of behavior phenotyping modules ( Figure 3—figure supplement 1 ) , including arrays of circular open field arenas , arrays of Y-shaped mazes for measuring locomotor handedness ( Buchanan et al . , 2015 ) , and social arena arrays , in which pairwise social interactions can be monitored . Phenotyping modules are not limited to collecting behavioral data . For example , flies can be loaded into standard 96-well plates outfitted with nylon mesh lids . From there , they are ready for molecular protocols , including use by liquid-handling robots . Flies in FlyPlates , which have access to food , can be used for circadian assays ( Tataroglu and Emery , 2014 ) , longevity ( Stearns et al . , 2000 ) , or pharmacological experiments ( Gasque et al . , 2013 ) . Modules such as the Fly Dispenser can be situated outside MAPLE’s frame , receiving and delivering flies via air-flow tubing . 3D-printed adapter blocks connect these tubes into the MAPLE workspace and are situated with locating brackets . The modularity of MAPLE’s hardware is reflected in its software as well ( Figure 3—figure supplement 2 ) . Each experimental procedure is associated with a Python experimental script file . These files call functions that ( 1 ) implement common multi-step robot actions ( like retrieving a fly from a behavioral arena ) , ( 2 ) implement low-level elemental robot actions , ( 3 ) are specifically associated with the modules used in a particular experiment , and ( 4 ) mediate the remote control of MAPLE over the internet . This software architecture permits the rapid scripting of new experimental protocols at a high level ( Figure 3—figure supplement 3 ) . To demonstrate the experimental flexibility of this ecosystem of modules and software , we implemented a number of experimental procedures ( Figure 3B ) . These include: ( 1 ) collecting virgin flies for genetic crosses by dispensing flies as they eclose and then distributing them into individual wells of FlyPlates where their isolation preserves their virgin status indefinitely; ( 2 ) loading flies into Y-shaped arenas to measure their locomotor biases , the time-consuming step of a routine assay in our lab; ( 3 ) loading flies from the Fly Dispenser into the FlyPlate wells for long-term culturing and circadian phenotyping; and ( 4 ) loading flies into social arenas to measure their pairwise interactions . Below we describe results from procedures ( 1 ) and ( 4 ) in detail , as well as the scientific findings obtained using the latter procedure . We tasked MAPLE with performing a tedious task that consumes great amounts of time in essentially all Drosophila labs: collecting virgin females for genetic crosses . Female D . melanogaster will not mate with males for approximately 6 hr after they eclose from the pupal case . In the first portion of this interval , they have morphological characteristics ( puffy abdomens , translucent cuticle , and visible meconium in the gut ) that correlate with their young age and are reliable , but conservative , indicators of virginity . In traditional manual virgin-picking , only females with these morphological correlates are collected . This means that many virgin females lacking morphological correlates in the latter portions of the 6 hr no-mating window are discarded , unless practitioners collect virgins from a stock vial or bottle at regular intervals at least three or four times a day . MAPLE has the potential to recover 100% of females as virgins by isolating them quickly after they eclose , and storing them individually to preclude mating . To implement this procedure , we devised a simple custom fly media vial in which the lower portion containing fly food is detachable . Parental generation flies lay eggs in this food , or food from another vial containing larval flies can be transferred into the custom vial . When flies in this experimental generation climb onto the walls of the vial and pupate , the food-containing portion is manually detached , and replaced with an empty vial-bottom ( Video 10 ) . This pupae-containing , foodless vial is then placed in the Fly Dispenser . Every 30 min , all newly eclosed flies are dispensed into MAPLE , which distributes them individually into the wells of a FlyPlate ( Figure 4A ) . This process was not without error: at rates of ~3% a well was left empty , or ~8% a well was loaded with two flies . A representative outcome is shown in Figure 4—figure supplement 1 . To recover only the virgin females , fully loaded FlyPlates were removed from their media base , brought to traditional CO2 fly-pushing pads at dissecting microscopes , the flies were anesthetized through the wire mesh floor , and then the plate and pad assembly was inverted , leaving the flies on the CO2 pad in their respective positions from the wells . The sex of each fly was determined by eye , and females that had been stored either alone , or with no males were collected . In a head-to-head comparison of typical manual virgin-collecting ( picking at the beginning and end of the work day from five vials ) versus MAPLE virgin-collecting , virgin females were procured at a rate of 1 . 5/min ( including the time needed to bring vials from the incubator , anesthetization , etc . ) and 4 . 9/min using MAPLE ( including the time to set up the pupa vial , load the dispenser , manually sex and sort the flies , etc . ) ( Figure 4B ) . In the future , we anticipate MAPLE will be able to sex flies without human intervention using the high-resolution imaging module , further decreasing manual work required . Lastly , we set out to see if MAPLE could conduct experiments that would be difficult using traditional manual methods . Specifically , we set out to measure social interaction networks ( SINs; Schneider et al . , 2012; Pasquaretta et al . , 2016 ) between pairs of flies , and then determine if measures of pair-wise affiliation are preserved on the timescale of days ( Figure 5 ) . It is known from group behavioral experiments that individuals spend more time interacting with specific other individuals ( Simon et al . , 2012 ) , and such interaction-based SINs remain stable over tens of minutes in fruit flies ( Schneider et al . , 2012 ) . In other species , such as the forked fungus beetle Bolitotherus cornutus , these dyadic interactions are stable for days ( Formica et al . , 2017 ) . Because it is challenging to perfectly maintain individual identity through experiments in which multiple individuals are in the same compartment and subsequently retrieved , stored and retested , it is unknown if dyad-specific affiliative measures are stable over longer periods of time in flies . To assess this we devised a new high-throughput assay to measure affiliative behavior between pairs of flies ( Figure 5A , B ) . This consisted of a 9 × 9 array of adjacent , approximately semi-circular arenas separated by an interchangeable barrier . MAPLE can load individual flies into each arena-half using a multi-position loading port ( Video 11 ) . We made four versions of the interchangeable barrier: an open-clear barrier was made of clear acrylic with grooves to connect the half-arenas , permitting flies to see and smell each other ( Video 12 ) ; a solid-clear barrier permits flies to see each other but impedes airflow; an open-black barrier permits airflow but blocks visual cues; and a solid-black barrier blocks both airflow and visual cues . Flies in this assay are thus presented with the choice of interacting with a designated partner across the barrier , or not . As a measure of social affiliation , we determined an ‘interactivity index , ’ defined as the Pearson correlation coefficient between the distances of each fly to the barrier over time ( Figure 5C , D ) . Either significantly positive or negative values of this index indicate social interaction . We found that male-female and female-female , although not male-male , dyads produced the same distribution of interactivity indices ( Figure 5—figure supplement 1A ) . Most social experiments were performed with virgin female-female dyads , although for some control groups we also included male-female dyads . We programmed MAPLE to load pairs of flies from the FlyPlate into social arenas according to three different dyad schemes ( Figure 5E ) . These varied from high-throughput ( testing 162 flies at a time ) , with each fly tested in just a single dyad , to low-throughput ( 10 flies total ) but saturated with respect to all possible dyadic pairings . Using the high-throughput scheme , we first measured the interactivity indices of wild type ( Canton-S ) flies separated by open-clear barriers . The observed distribution was significantly different compared to shuffled controls that scramble dyadic pairings ( Figure 5—figure supplement 1B; p<0 . 0001 by KS-test ) , indicating that interactions are dyad-specific . Dyads separated by barriers that limit sensory cues exhibited significantly lower mean absolute interactivity indices ( Figure 5F ) , indicating that both olfactory and especially visual cues drive differences in dyadic interactions . This is consistent with reports that visual cues can mediate social modulation of behavior ( Kim et al . , 2012 ) . Consistently , anosmic Orco mutant flies ( Vosshall and Hansson , 2011 ) showed a significant 65% reduced mean absolute interactivity index , even with open-clear barriers ( p=0 . 00048 by t-test ) . Blind NorpA mutant flies ( Kim et al . , 1995 ) did not differ significantly from wild type flies in mean absolute interactivity index ( p=0 . 26 by t-test ) . Flies with mutations in the white gene ( w1118 ) have reduced visual acuity ( Markow and Scavarda , 1977 ) but showed a 50% increase in absolute interactivity index which was statistically significant ( p=0 . 013 by t-test ) , suggesting that social partner visual recognition does not require fine acuity ( Justice et al . , 2012 ) . The increased inter-dyad variability seen in w1118 animals may be consistent with increased inter-individual variability in phototactic preference exhibited by this genotype ( Kain et al . , 2012 ) . In a control experiment , w1118 flies separated by solid-black barriers had interactivity indices indistinguishable from Canton-S flies separated by solid-black barriers ( p=0 . 48 by t-test ) . Lastly , we set out to determine if social behavior is stable over long periods of time by measuring dyadic interactivity indices across days . Using our high-throughput dyad scheme , we measured interactivity indices from 91 dyads on each of 6 consecutive days . Interactivity indices across Canton-S dyads were statistically significantly correlated ( 0 . 31 < r < 0 . 38 , 0 . 0001 < p < 0 . 04 ) between day one vs . day 2 , day 2 vs . day 3 and day 3 vs . day 4 ( Figure 5G ) , but not days 4 vs . 5 or 5 vs . 6 . Thus , dyad-specific affiliative behavior appears to be stable over days-long timescales , though only for the first days of experiments . In control experiments , Canton-S flies with solid-black barriers and NorpA and Orco mutant flies exhibited no significant correlation in dyad interactivity indices across days . Significant correlation in interactivity index across dyads was observed in our two SIN dyad schemes as well ( Figure 5H–J ) suggesting that SIN persistence can be measured in a variety of experimental formats . Other metrics of social interaction yielded qualitatively similar results to the interactivity index ( Figure 5—figure supplement 2; Supplementary methods ) . We developed MAPLE , a modular , automated platform for large-scale experiments , to expand automated experimental capabilities for lab model organisms ( Figure 1 ) . By design , MAPLE is versatile , scalable , and relatively inexpensive , with all components of the core fly-handling robot costing roughly $3500 . We displayed the versatility of this system by adapting the MAPLE platform to detect , place , and streak out individual S . cerevisiae colonies ( Figure 1B ) , autonomously record C . elegans behavior ( Figure 1C ) and P . polycephalum movement ( Figure 1D ) , as well as automatically monitor the behavior of multiple bumblebee colonies over long periods ( Figure 1F , G ) . We have made the design and code for MAPLE open access in the hope that it , or descendant approaches , will be adopted by our field . After demonstrating that MAPLE did not discernibly alter the baseline behavior of animals compared to manual handling ( Figure 2 ) , we developed a procedure by which MAPLE could help conduct a repetitive lab chore that nearly all Drosophila biologists are familiar with — collecting virgin females for genetic crosses . Using a Fly Dispenser , MAPLE collected individual animals as they eclosed from the pupal case and kept them in isolation , ensuring their virginity ( Figure 4 ) . To arrive at a final collection of virgin females , these singly-housed individuals were sexed by hand . Even with this manual step , virgins were procured at a much higher ( ~3 x ) rate with MAPLE’s assistance than without it ( measured in terms of human labor time per virgin female ) . MAPLE’s capacity to handle individual flies in combinatorially complex experimental designs ( Figure 3 ) , allowed us to demonstrate that dyadic fly social behavior relies on both visual and olfactory cues . Dependence of social behavior on visual ( Mery et al . , 2009; Simon et al . , 2012 ) and olfactory cues ( Schneider et al . , 2012; Billeter and Levine , 2015 ) is known from experiments observing many flies simultaneously . We further found that individual differences in dyadic social behavior — as well as derived social interaction network topography — remain stable over days ( Figure 5 ) . Stability of social networks on short timescales ( Schneider et al . , 2012 ) is known , but long-term stability of social networks is a novel finding . This illustrates that MAPLE can ( 1 ) replicate known results without introducing significant behavioral confounds and ( 2 ) extend our knowledge by conducting complex , long-term experiments that are challenging for human experimentalists . Technologies that automate experiments will particularly facilitate the phenotyping of individual animals , an approach that reveals underappreciated intragenotypic variability . Our group has shown that individual flies generally exhibit very different behaviors from one another even if they are reared in the same environment and have the same genotype , and that these differences persist across days ( Kain et al . , 2012; Buchanan et al . , 2015; Ayroles et al . , 2015; Kain et al . , 2015; Todd et al . , 2017 ) . MAPLE allowed us to show that this extends to dyad-specific social interactions in flies ( Figure 5H–J ) . To our knowledge , this is the first demonstration that dyad-specific social interactions are persistent on days-long timescales in flies . As the behavioral assay used here relies on physically separating individual flies , it is limited to examining visual and olfactory aspects of social interactions . To describe persistence of dyad-specific social interactions in flies more completely , further investigation including somatosensory and gustatory information exchange is necessary ( Krstic et al . , 2009; Schneider et al . , 2012; Ramdya et al . , 2015 ) . Moreover , this assay presents flies with a choice to interact with a particular partner , or not . Assays with groups of flies , where the alternative to interacting with a particular fly includes interacting with potentially many other flies , may produce different results ( Ramdya et al . , 2017 ) , although interaction networks in a group context are at least partially stable over the course of experiments ( Schneider et al . , 2012 ) . MAPLE is designed to be a general purpose instrument , therefore it has some disadvantages compared to purpose-built devices that perform a single task . For example , while MAPLE’s camera can be used to acquire time-lapse movies of yeast colony growth , Physarum plasmodium movement , or C . elegans locomotion , purpose-built worm imaging systems can provide comparable data with higher throughput ( Churgin et al . , 2017 ) or higher resolution ( Stern et al . , 2017 ) . Likewise , purpose-built yeast colony picking robots ( e . g . BioMatrix , S and P Robotics ) achieve higher performance on specific tasks like transferring colonies , although typically at a higher price . MAPLE’s open , modular design is meant to facilitate customization and prototyping , so it may be a good platform for the development of new organism-handling hardware , for example , for the dissection and transfer of Physarum syncytium or individual worms . These and other experimental applications might rely on liquid-handling . MAPLE currently has no liquid-handling capabilities , but there are open source liquid-handling robotic systems such as Opentrons ( http://opentrons . com ) which could be integrated with MAPLE . Specifically , liquid-handling effectors could be installed on MAPLE z-axes , replacing some or all of the current effectors . In its current configuration , MAPLE has some drawbacks . Its effectors move slightly slower than the hands of a trained experimentalist , so its advantages derive from it not growing tired or losing focus . MAPLE does not execute every action with 100% success . For example , attempts to remove an awake fly from a behavioral arena , using vacuum through a loading port , succeed approximately 90% of the time . Given this error rate , the experimental procedures that include removing flies from arenas use the camera to inspect the arena after each attempt to confirm if the fly has been successfully removed . Similarly , our procedure to transfer yeast colonies used the collision-detector to confirm that the object manipulator had successfully removed petri dish lids . These checks reduce the efficiency of MAPLE-conducted experiments ( particularly compared to manual experiments ) , but mean that the operator can walk away from the device confident that MAPLE will eventually get it right . Given its redundancies and fail-safes , MAPLE can work autonomously for long periods ( e . g . 12 hr for the Physarum experiment ) without getting stuck . A potential advantage of using MAPLE for fly experiments specifically is the option to avoid anaesthesia . The Fly Dispenser releases awake animals into the MAPLE system , and these can then be moved between the FlyPlate and behavioral arenas using the multi-position loading ports we developed . Avoiding anesthesia has multiple benefits , including reducing the distortion of behavioral ( Bartholomew et al . , 2015 ) and physiological measurements ( Colinet and Renault , 2012 ) , which , depending on the form of anaesthesia , can last for hours or days ( MacMillan et al . , 2017 ) . Even seemingly benign manual aspiration can disrupt the expression of sensitive phenotypes ( Trannoy et al . , 2015 ) . Automated , anesthesia-free animal-handling thus has the potential to standardize handling effects . MAPLE’s modularity means that the platform's versatility and capabilities can be expanded in the future . As examples , we are developing fly sink modules for culturing thousands of animals in the style of population cages , facilitating selection experiments and experimental evolution . We are developing a high-resolution imaging source/sink module into which flies could be deposited , imaged in dorsal and ventral views , and released back into experimental workflows . This device is inspired by the Fly Catwalk system ( Medici et al . , 2015 ) but can be loaded with a fly on demand , rather than relying on flies to enter the imaging chamber on their own . The images from this module will allow machine-learning-based classification of morphological phenotypes , like sex and eye color , as well as quantification of fluorescent proteins; key genetic markers in flies . In addition to conducting tedious or complex experiments , MAPLE’s integrated format affords the opportunity to feed back the results of phenotypic assays into fly handling tasks . For example , MAPLE could be used to perform artificial selection , identifying individuals with a specific behavior or phenotype , and placing males and females together in wells of a FlyPlate , vials , or population cage . Likewise , multigeneration genetic crosses will be possible with our high resolution imaging module . Our current behavioral phenotyping relies on simple motion tracking , but behavioral assays employing thermogenetic or optogenetic stimulation , or sophisticated stimulus control could be implemented through custom modules . Unsupervised learning algorithms have been used to automatically phenotype flies ( Berman et al . , 2014; Todd et al . , 2017 ) , and with the capacity to store and access large numbers of flies individually , MAPLE could identify and isolate outliers within a population . MAPLE is compatible with such modern , automated approaches to fly experimentation , and brings automated animal-handling one step closer to the potential achieved by liquid-handling robots for molecular research . MAPLE’s frame is a rectangular prism constructed from extruded aluminum struts ( Misumi HFS-5 series in various sizes and lengths ) and brackets ( Misumi five series ) . The principal axes of the robot are Cartesian , that is to say linear and mutually orthogonal . The longest axis - designated the X axis - comprises two supported linear rails ( IGUS Drylin AWUM-12 ) , each with two housed bearings ( IGUS drylin OJUM-6–12 ) . The total of four housed bearings support a single , wide linear rail ( IGUS WS-10–120 ) that is the Y axis . A carriage made of two aluminum plates sandwiching four more bearing blocks ( WJ200UM-01–10 ) slides along the Y axis . Suspended from this carriage is an assembly that houses the three independent Z axis slides ( IGUS SLN-D740679-2 ) . We deliberately chose sliding bearings ( as opposed to ball bearing slides ) for the three axes to avoid noise and vibrations that might confound behavioral experiments . The first of the three end-effectors is an object manipulator . Made from an off-the-shelf vacuum cup connected to an air manifold , it can pick up and deposit lids or other small components from the workspace . A high-resolution digital camera and lens ( The Imaging Source DFM 72BUC02-ML and TBL 9 . 6-2 C 3MP ) is mounted to the second Z axis . Moving this Z carriage up and down focuses the camera . The third motorized Z slide holds a custom aspirator to move flies . Several short lengths of small diameter tubing ( sections of needle tips , McMaster 75165A553 and 75165A682 ) are fixed approximately 4 mm inside a blunt , Luer-lock needle ( McMaster 6710A61 ) , forming a barrier to flies but allowing air to flow . A custom-molded silicone rubber boot surrounds the fly vacuum tip for the object manipulator , forming a better seal against apertures in modules and improving fly handling ( Figure 1—figure supplement 3 ) . Stepper motors ( 1 . 5 A NEMA 17 60 mm bipolar stepper motor , MNEMA17-60 from RobotDigg . com ) mounted to the main frame drive two 6 mm-wide GT-2 belts arranged in a CoreXY configuration ( Moyer , 2012 ) , and these belts drive the motion in the X and Y directions . Each Z axis slide is driven by its own stepper motor , integral to the slide assembly . The maximum speed for the X and Y axes is 200 mm/s . The Z axes top out at 83 . 3 mm/s . Limit switches ( Omron SS-5 ) are mounted to the frame for each axis , providing repeatable end stops for homing . Motion control , as well as the control of auxiliary devices such as the solenoid valves and the LED illumination , is handled by a Smoothieboard v1 PCBA , running custom Smoothieware firmware ( included in our Github repository ) . G-code commands are sent from a PC connected via USB , interpreted on the Smoothieboard , and translated into electrical signals sent to each stepper motor . Scripts containing each experimental protocol , along with a common set of frequently used subroutines , were written in Python 2 . 7 ( or Matlab 2016a for BEECH ) and executed on PCs running Windows 7 . Every module has an associated Python 2 . 7 module class file ( all MAPLE control software , including module class files , are available at https://github . com/FlySorterLLC/MAPLEControlSoftware; copy archived at https://github . com/elifesciences-publications/MAPLEControlSoftware ) . These files provide the 3D coordinates of key points on each module , including the sites of any ports or adapters through which flies are conveyed , as well as the z-clearance above the module needed so that the end-effectors do not collide with the modules in the workspace . These classes can be instantiated in Python files that represent workspace configurations ( Examples/ExampleWorkspace1 . py ) . Beyond the module class files , there is a master MAPLE class file ( robotutil . py ) , which 1 ) establishes the communications connections between the experiment-coordinating computer and the motion control card and camera in MAPLE and 2 ) contains functions for all low-level robot operations , like returning to the 0 , 0 , 0 home position , moving to an arbitrary position , opening or closing the solenoid valves that control the vacuum flow in the end effectors , or acquiring a photo from the end-effector camera . There is also a file ( commonFlyTasks . py ) that contains subroutines for common usage tasks , like the combination of end effector movements , vacuum and air valve engagement , and Fly Dispenser serial commands required to retrieve a fly from the Fly Dispenser adapter on the workspace . Files of an additional type , experimental scripts , implement the actual experimental procedures ( e . g . , Figure 3—figure supplement 3; Video 4 ) . Each of these scripts load the class files for the modules used in their respective experiments , and procedurally calls the low- and mid-level functions of robotutil . py to implement each procedure . See Figure 3—figure supplement 2 for a schematic of the software architecture . Lastly , to increase the convenience of using MAPLE , we implemented a remote control system in which representations of the current status of experiments ( e . g . text reports and digital images ) are posted to a dedicated email account . This account can also receive MAPLE commands by email to remotely trigger experimental procedures . The workspace refers to a 100 cm x 28 . 2 cm x 7 . 5 cm volume that can be accessed by all end-effectors . As specified , the workspace is attached to the same frame as the axis carriages , but could be mounted separately to isolate experimental modules from the vibrations of MAPLE’s motors . The bottom of this volume is a clear acrylic floor with 5 mm holes organized in a grid ( 10 cm apart ) and cable pass-throughs for easy organization of individual modules . The 5 mm holes can be used to affix an interchangeable acrylic plate ( a ‘workplate’ ) to the acrylic workspace floor using nylon thumb screws . Workplates have locating brackets which define the positions of modules in a particular experimental configuration . Thus , every experimental script is associated with a physical workplate that locates the modules used in its experiment . Modules were made by a variety of fabrication techniques , including laser-cutting of acrylic . Vector outlines of module components were made in Autodesk Inventor or Adobe Illustrator and laser cut by way of CorelDraw . Acrylic components were joined together using Plastruct plastic weld . Phenotyping modules used to investigate affiliative behavior and locomotor handedness , that is , social arena and y-maze arrays , measured 30 cm x 30 cm ( Figure 3 ) and were fabricated from sheet acrylic using a laser cutter . Fly source modules were custom made ( CO2 pad ) or purchased from FlySorter LLC . Fly Dispenser dimensions are 22 cm x 15 cm , CO2 pad dimensions are 25 cm x 15 cm , and FlyPlate dimensions are 16 cm x 10 cm . The Fly Dispenser isolates and outputs individual flies from an attached vial . Repeated knocking motion ( which mimics the tapping gesture that people use to knock flies down in a vial ) causes flies to fall from the vial into a funnel . At the bottom of the funnel , a pair of motorized , soft foam wheels acts as a valve , and a photo interrupter detects when one fly has passed by . The wheels are stopped , preventing other flies from passing through the valve , and an air pump transports the isolated fly out of a tube . The dispensing process can be remotely triggered and monitored by Python scripts via a USB serial interface . MAPLE interfaces with the Fly Dispenser through the dispenser adaptor logistics module . The dispenser adaptor is a 4 cm x 1 . 5 cm 3D-printed ABS block with two 5 mm diameter plastic Luer lock tube sockets attached on opposite sides . The Dispenser handpiece connects to the bottom side of the dispenser adaptor , while the MAPLE fly manipulator end effector aligns to the opening on the top of the adapter . The FlyPlate is a modified 96-well plate positioned on a food tray . Each well in the FlyPlate follows the 96-well plate standard for bottomless wells ( 7 mm in diameter and 10 . 9 mm deep ) . Wells have a stainless steel mesh floor that allows feeding but prevents escape . The plate lid has x-shaped laser-cut openings over each well , cut into a flexible nylon mesh , that allow MAPLE’s individual fly manipulator ( or a handheld aspirator ) to penetrate to remove or deposit flies . The openings close back up once the aspirator/manipulator tip has been removed , keeping flies securely housed . Flies had free access to standard cornmeal diet on the food tray placed below . Food was replaced every 2 days to maintain adequate moisture and freshness and remove eggs and first instar larvae . The morgue fly sink module is a 10 cm diameter x 5 cm deep laser-cut acrylic cylinder covered by a detachable lid that allows quick disposal of its contents and is equipped with an nylon mesh adapter ( in the style of the well coverings of the FlyPlate ) that allows MAPLE to deposit flies into soapy water or ethanol that traps and euthanizes them . The fly food vial is a standard 2 . 6 cm x 10 cm vial equipped with a detachable lid that facilitates MAPLE fly depositing . A standard fly culture media vial can be placed into the fly food vial . Behavior phenotyping modules can receive flies in one of two different ways , depending on whether the flies are anesthetized or not . In a traditional experimental style ( Ayroles et al . , 2015; Buchanan et al . , 2015 ) , MAPLE can pick up anesthetized flies from the CO2 pad module with the organism manipulator , pick up the plastic lid covering a behavioral arena with the object manipulator , drop the fly in the arena , and replace the lid . In a MAPLE-optimized experimental style , awake flies are retrieved from the Fly Dispenser or FlyPlate and then transferred directly into the behavioral arena by sliding a multi-position loading port , a slidable clear lid with a 3 . 5 mm diameter opening through which flies can be deposited and removed , into place above the arena , dropping the fly , and then sliding the port so it is inaccessible to the fly ( Figure 3—figure supplement 1 and examples below ) . Arenas used for affiliative behavior ( Figure 5 ) and circling bias experiments ( Figure 2—figure supplement 1 ) are circular in shape with a 30 mm diameter and a height of 3 mm . Arenas are covered by a multi-position loading port . Two equal-sized semicircular compartments are formed by a 1 . 5-mm-thick interaction barrier . Interaction barriers refer to individually laser-cut blocks that can be placed into corresponding openings in the middle of the circular arena , allowing separation of flies into individual compartments . Barriers were laser-cut from either clear or black acrylic and were designed to be either solid or open . Solid barriers were flat on the bottom . Open barriers had 4 ~ 0 . 25 mm horizontal channels connecting the two compartments ( Figure 5B; Video 12 ) . Open barriers presumably facilitate the exchange of odor cues between compartments . Barriers could be made of clear acrylic , facilitating visual cues , or black acrylic . A social arena array comprises 81 circular arenas , with 162 semicircular arenas in total . Locomotor handedness was assessed using y-maze arenas ( Buchanan et al . , 2015 ) . Individual arms of the symmetrical Y-shaped mazes are 15 . 5 mm long and 120° apart . Arm ends are circular ( 5 . 2 mm ) in shape , making it easier for flies to turn around and permitting loading and unloading flies via multi-position loading ports ( Figures 1 ) . Arenas are covered with identical lids as those in the social arena array . A y-maze array comprises 81 y-maze arenas arranged equidistantly in a nine-by-nine grid . All parts described were manufactured from either clear or black acrylic and cut into shape using a laser cutter . Yeast expressing GFP in mitochondria ( genotype can1-100 leu2-3 , 112 his3-11 , 15 ura3-1 BUD4-S288C RAD5 TRP , mdh1::KAN ) were streaked from frozen stocks onto complete supplement mixture media lacking adenine ( CME-Ade ) and cultured at 30°C for 1–2 days . Cells of this genotype express GFP in mitochondria , and are thus fluorescent under the dissecting scope . MAPLE touched single colonies on the CME-Ade plate and streaked this material onto empty yeast extract-peptone-dextrose ( YPD ) plates . These target plates were allowed to incubate overnight at 30°C prior to imaging . We imaged N2 worms of mixed sexes and ages on standard growth media plates consisting of nematode growth medium in 1 . 7% agar with an OP50 E . coli lawn . Locomotion was recorded at 1 Hz at room temperature ( 21°C ) . We inoculated petri dishes containing 2% agar in water with ~3 mm diameter excised pieces of an oatmeal-fed , actively growing Physarum plasmodium . These were allowed to recover from excision and plating for ~12 hr at 21°C prior to the collection of the 12 hr 0 . 017 Hz time-lapse movie in MAPLE , which was also conducted at 21°C . All experiments were performed using Canton-S ( wild type ) , Orco , NorpA , or w1118 lines . Mutant lines were homozygous . We raised flies on CalTech formula cornmeal mediaunder 12 hr/12 hr light and dark cycle in an incubator at 25°C and 70% humidity . Flies were anesthetized using carbon dioxide ( CO2 ) and housed in vials of 15 to 20 flies , unless otherwise specified . Five days post-eclosion , flies were aspirated into individual wells in the FlyPlate using CO2 and used for experimentation after at least 2 hr of recovery . All experiments were conducted between 9AM and 9PM ( ZT0-ZT12 ) . Flies were loaded into individual arenas by MAPLE; arenas that remained empty after two iterations were loaded manually using an aspirator . Flies were later removed from their arenas in an identical fashion . FlyPlates , social arenas , and y-maze arrays were filled with 96 , 162 , and 81 flies , respectively . Flies were assayed using diffused white LED backlighting ( Buchanan et al . , 2015 ) in a temperature ( 23°C ) and humidity ( 41% ) controlled behavioral observation room . Fly movement was tracked for 1 hr . Fly tracks were analyzed using a custom MATLAB script . For longitudinal assaying , flies were moved back into the FlyPlate after phenotyping and allowed to feed and rest overnight or for 1 hr at minimum . Fly movement was tracked at 29 . 9 fps using a custom real-time MATLAB script interfacing with a Firefly MV FMUV-13S2C USB-camera . Tracks were analyzed in MATLAB . In total , 3% of the data were discarded because flies were immobile as determined by mean speed thresholds . Unless otherwise specified , all confidence intervals were computed by bootstrapping the data associated with individual flies or individual fly dyads ( in the case of social interaction measurements ) 1000 times , using custom MATLAB scripts . One standard deviation of the bootstrap estimates was our estimate of the standard error of the estimate . P-values reported were adjusted for multiple comparisons using the Bonferroni correction where applicable , and asterisks reflect post-correction significance .	Biological research can , at times , be mind-numbingly tedious: scientists often have to do the same experiment over and over on many different samples . When working with animals such as fruit flies , this means researchers have to physically handle large numbers of specimens , selecting certain individuals or moving them from one container to another to perform the study . This represents a serious bottleneck that slows down discovery . Automation represents an obvious solution to this issue . In fact , it has already revolutionized fields like molecular biology , where robots can handle the liquids required for the experiments . Yet , it is not so easy to automate tasks that involve animals larger than a millimeter . To fill that gap , Alisch et al . have developed a robotic system called Modular Automated Platform for Large-scale Experiments ( MAPLE ) that can manipulate fruit flies and other small organisms . Using gentle vacuum , MAPLE can pick up individual flies to move them from one compartment to another . These areas could be places where the insects grow or where experimental measurements are automatically gathered . Putting the robot to work , Alisch et al . used MAPLE to collect virgin female flies for genetic experiments , a common task in fruit flies laboratories . The system was also configured to load flies into arenas where their behavior could be measured . Finally , MAPLE assisted with an experiment that involved tracking the interactions of known individuals to examine if the flies exhibited social networks , and if those networks were stable . This logistically complicated experiment would have been difficult to run without the help of an automated system . Alisch et al . also show that the robot can be adjusted to work with various species often used for research , such as nematode worms , yeast , slime mould and even bumblebees . This allows the system to be useful in a range of research fields . As MAPLE fits on a table top and is fairly affordable , the hope is that it could help many scientists do their experiments faster and with greater consistency , freeing up time for creative thinking and new ideas . Ultimately , this tool could help to speed up scientific progress .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "tools", "and", "resources", "neuroscience" ]	2018	MAPLE (modular automated platform for large-scale experiments), a robot for integrated organism-handling and phenotyping
The roles of long non-coding RNAs ( lncRNAs ) in regulating cancer and stem cells are being increasingly appreciated . Its diverse mechanisms provide the regulatory network with a bigger repertoire to increase complexity . Here we report a novel LncRNA , Lnc34a , that is enriched in colon cancer stem cells ( CCSCs ) and initiates asymmetric division by directly targeting the microRNA miR-34a to cause its spatial imbalance . Lnc34a recruits Dnmt3a via PHB2 and HDAC1 to methylate and deacetylate the miR-34a promoter simultaneously , hence epigenetically silencing miR-34a expression independent of its upstream regulator , p53 . Lnc34a levels affect CCSC self-renewal and colorectal cancer ( CRC ) growth in xenograft models . Lnc34a is upregulated in late-stage CRCs , contributing to epigenetic miR-34a silencing and CRC proliferation . The fact that lncRNA targets microRNA highlights the regulatory complexity of non-coding RNAs ( ncRNAs ) , which occupy the bulk of the genome . A downstream target of p53 , the microRNA miR-34a is a well-known tumor suppressor in various types of cancers ( Chang et al . , 2007; He et al . , 2007 ) . Among its many functions , miR-34a has been shown to limit self-renewal of cancer stem cells ( Bu et al . , 2013; Liu et al . , 2011 ) . miR-34a mimics such as MRX34 are among the first microRNA mimics to reach clinical trial for cancer therapy ( Bader , 2012; Bouchie , 2013 ) . Besides cancer , miR-34a has been shown to regulate stem cell differentiation , somatic stem cell reprogramming , cardiac ageing , neurodegeneration , ciliogenesis , bone resorption , and metabolism ( Aranha et al . , 2011; Boon et al . , 2013; Choi et al . , 2011; Krzeszinski et al . , 2014; Liu et al . , 2012; Song et al . , 2014; Xu et al . , 2015 ) . Loss of p53 function can lead to downregulation of miR-34a; however , miR-34a expression also tends to be silenced due to aberrant CpG methylation of its promoter in many types of cancer , including breast , prostate , lung , colon , kidney , bladder , pancreatic , and ovarian cancer ( Corney et al . , 2010; Kong et al . , 2012; Lodygin et al . , 2008 ) . Methylation of the miR-34a promoter is positively correlated with and miR-34a expression is inversely correlated with progression of colorectal cancer ( CRC ) ( Siemens et al . , 2013 ) . However it is completely unclear how miR-34a was silenced by epigenetic modification . Normal stem cells often divide asymmetrically to produce one daughter cell like itself for self-renewal and another daughter cell unlike itself to go down a path of differentiation ( Neumuller and Knoblich , 2009 ) . Asymmetric division allows stem cells to maintain self-renewal while generating a heterogeneous population for cellular diversity ( Reya et al . , 2001 ) . Tumor cells are usually heterogeneous and have a wide range of potential for tumorigenesis , proliferation , and metastasis . Recent studies have reported that cancer cells , including colorectal , glioma , lung and breast cancer cells , could also divide asymmetrically , generating progenies with different proliferation capabilities ( Bu et al . , 2013; Dey-Guha et al . , 2011; Lathia et al . , 2011; O'Brien et al . , 2012; Pece et al . , 2010; Pine et al . , 2010; Sugiarto et al . , 2011 ) . The frequencies of symmetric vs . asymmetric divisions are associated with cancer proliferation and progression . Disruption to asymmetric division in favor of symmetric self-renewal alters the balance between self-renewal and differentiation , which has been linked to neoplastic transformation and tumor growth ( Cicalese et al . , 2009; Sugiarto et al . , 2011 ) . Here , we discovered that a novel lncRNA , Lnc34a , directly targets the miR-34a promoter for epigenetic silencing by recruiting the DNA methyltransferase Dnmt3a via Prohibitin-2 ( PHB2 ) and Histone Deacetylase 1 ( HDAC1 ) . Asymmetric distribution of Lnc34a during colon cancer stem cell ( CCSC ) division leads to asymmetric daughter cell fate . Its suppression leads to differentiation while its abundance leads to CCSC proliferation via symmetric self-renewal . Lnc34a tends to be upregulated in late-stage CRC , associated with miR-34a silencing . The ability of lncRNA to target microRNA provides RNA circuitry more ways to increases the complexity of the regulatory network . We performed RT-PCR with 10 pairs of primers to scan for potential transcripts overlapping the miR-34a promoter and its downstream sequence . A 293 base pair ( bp ) transcript fragment was amplified . Rapid amplification of cDNA ends ( RACE ) further identified a full-length , 693 bp transcript ( Figure 1—figure supplement 1A , B ) . Northern blot confirmed the existence and size of the transcript in seven CRC cell lines and two colon cancer stem cell ( CCSC ) lines ( Figure 1A , B , Figure 1—figure supplement 1D ) . The CCSCs were isolated from two early-stage CRC specimen , and were functionally validated by serial sphere formation , tumor initiation , and marker staining ( Bu et al . , 2013 ) . The original frozen stocks from the first passage were used in the study . The transcript is composed of two exons , spanning nearly 15 . 3 kilobases ( kb ) , and does not contain a valid Kozak sequence . The full-length transcript has no protein coding potential according to the Coding Potential Calculator ( CPC ) and Coding Potential Assessment Tool ( CPAT ) ( Kong et al . , 2007; Wang et al . , 2013 ) . We named the transcript Lnc34a . 10 . 7554/eLife . 14620 . 003Figure 1 . Characterization of Lnc34a . ( A ) Schematic illustration of Lnc34a ( shown in black ) and miR-34a ( shown in blue ) gene structure . Lnc34a and miR-34a contain two exons and are transcribed in different directions . P , probe for Northern blot in ( B ) . ( B ) Northern blot detection of Lnc34a with the probe shown in ( A ) , quantified by Image J . ( C ) RT-qPCR detection of Lnc34a expression in colon cancer stem cells ( CCSC1 and CCSC2 ) and well-established colon cancer cell lines ( HT29 and Caco-2 ) . ( D , E ) RT-qPCR detection of Lnc34A level in cellular fractions from CCSC1 ( D ) and CCSC2 ( E ) sphere cells . U6 and actin are the nuclear and cytoplasm controls , respectively . ( F ) Lnc34a expression in CCSC sphere cells detected by RNA-FISH . Scale bar , 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 14620 . 00310 . 7554/eLife . 14620 . 004Figure 1—source data 1 . Information of CRC patients . Tissue specimens collected from the listed CRC patients were used for analyses of Lnc34a and miR-34a expression and miR-34a promoter methylation . DOI: http://dx . doi . org/10 . 7554/eLife . 14620 . 00410 . 7554/eLife . 14620 . 005Figure 1—figure supplement 1 . Identification of Lnc34a . ( A ) Schematic illustration of Lnc34a and miR-34a gene structures . Primers for RT-PCR and rapid amplification of cDNA ends ( RACE ) were shown . ( B ) RT-qPCR and RACE that amplified Lnc34a . ( C ) RT-qPCR detection of Lnc34a expression in CRC cell lines . ( D ) Northern blot detection of Lnc34a in CRC cell lines and CCSCs . The quantification of each band was carried out using Image J . DOI: http://dx . doi . org/10 . 7554/eLife . 14620 . 00510 . 7554/eLife . 14620 . 006Figure 1—figure supplement 2 . RNA FISH specificity and Lnc34a knockdown efficiency . ( A ) Knockdown of Lnc34a abolished RNA FISH signals . ( B ) RT-qPCR showing Lnc34a knockdown efficiencies by shLnc34a1 and shLnc34a2 . ( C ) RT-qPCR detection of Lnc34a level in cellular fractions from CCSC1with Lnc34a ectopic expression . DOI: http://dx . doi . org/10 . 7554/eLife . 14620 . 006 To analyze Lnc34a expression in CRC cells , RT-qPCR was performed in 9 commonly used CRC cell and the two CCSC lines . Consistent with the Northern blot measurement , Lnc34a levels were significantly higher in the CCSC sphere cells ( Figure 1C , Figure 1—figure supplement 1C ) . Cellular fractionation assays show enrichment of Lnc34a in the nuclear fraction ( Figure 1D , E ) , and RNA fluorescence in situ hybridization ( RNA FISH ) indicates that Lnc34a is mainly in the nucleus ( Figure 1F ) . RNA FISH specificity was validated when the same RNA-FISH probe did not detect Lnc34a after Lnc34a was knocked down by lentiviral short-hairpin RNA ( shRNA ) vectors in CCSC spheres ( Figure 1—figure supplement 2A ) . Notably , RNA-FISH showed that a small population among the CCSC sphere cells did not express Lnc34a , although the majority did ( Figure 1F ) . We then separated the sphere cells into two populations based on the expression levels of ALDH1 , a CCSC marker ( Huang et al . , 2009 ) . Flow analysis confirmed that ALDH1+ cells also express high levels of CD133 , another CCSC marker ( Figure 2—figure supplement 1A ) . RT-qPCR showed that , in both sphere cultures ( CCSC1 and CCSC2 ) , ALDH1+ cells have much higher Lnc34a expression levels than the ALDH1- cells ( Figure 2A , B ) . We then performed the pair-cell assay by plating single cells and allowing them to progress through one cell division ( Bultje et al . , 2009 ) . α-tubulin staining was used to identify dividing cells ( Figure 2C ) . Co-staining revealed that Lnc34a was asymmetrically distributed and enriched in the ALDH1+ ( CCSC ) daughter cells , which were also CD133+ ( Figure 2C , D , Figure 2—figure supplement 1B , C ) . 10 . 7554/eLife . 14620 . 007Figure 2 . Lnc34a Asymmetry in CCSC division . ( A , B ) RT-qPCR detection of lnc34a in ALDH1+ and ALDH1- populations isolated from spheres of two independent patient-derived lines , CCSC1 ( A ) and CCSC2 ( B ) . Lnc34a is high in ALDH1+ ( CCSC ) but low in ALDH1- ( non-CCSC ) cells . ( C ) Representative images of Lnc34a distribution in dividing pairs . α-tubulin staining is consistent with the telophase ( final phase of mitosis ) configuration of microtubules – the midbody at the division plane during cytokinesis and asters at the poles . ALDH1 identifies the CCSC daughter . ( D ) Quantification of Lnc34a/ALDH1 co-expression ( C . E . ) in daughter compartments of dividing pairs as shown in ( C ) . ( E ) Representative images of Lnc34a asymmetry in dividing pairs in xenograft tumors derived from CCSC1 and CCSC2 . Dividing pairs are identified by tubulin staining . ( F ) Percentage of Lnc34a asymmetry in dividing pairs in CCSC xenografts as shown in ( E ) . ( G ) Representative images of asymmetric and symmetric Lnc34a distribution in dividing pairs in early- and late-stage human CRC specimens . ( H ) Percentage of Lnc34a asymmetry in dividing pairs in human CRC specimens . ( I ) Effect of Lnc34a knockdown on mode of division based on ALDH1 staining of dividing cell pairs . Lnc34a knockdown decreased asymmetric ( ALDH1+/ALDH1- ) division and symmetric self-renewal ( ALDH1+/ALDH1+ ) , and increased differentiation ( ALDH1-/ALDH- ) . ( J ) Effect of ectopic Lnc34a expression on mode of division . Ectopic Lnc34a increased symmetric self-renewal ( ALDH1+/ALDH+ ) , and reduced asymmetric division ( ALDH1+/ALDH1- ) and differentiation ( ALDH1-/ALDH1- ) . The effect of ectopic Lnc34a expression was abrogated by ectopic miR-34a expression . ( K ) Pair-cell BrdU incorporation assay showing asymmetric proliferative potential . Left , schematic representation of the experimental approach . Single sphere cells were allowed to divide once ( 1st division ) . Cells were then treated with BrdU for 3 hr to label cells that were re-entering the 2nd division . Right , representative images showing that the Lnc34a+ cells were more proliferative and incorporated BrdU . Scale bar , 8 μm . Error bars denote s . d . of triplicates . p<0 . 05; p<0 . 01; p<0 . 001 . p-value was calculated based on Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 14620 . 00710 . 7554/eLife . 14620 . 008Figure 2—figure supplement 1 . CCSCs co-express ALDH1 and CD133 . ( A ) FACS showing ALDH1+ sphere cells are CD133+ . ( B ) Co-immunofluorescence of ALDH1 and CD133 showing ALDH1 and CD133 are expressed in the same daughter cell during CCSC division . ( C ) Percentages of CCSC divisions wherein miR-34a and ALDH1 are coexpressed ( C . E . ) or mutually exclusive ( M . E . ) . ( D , E ) Effect of Lnc34a knockdown ( D ) and ectopic Lnc34a and miR-34a expression ( E ) on mode of division based on CD133 staining of dividing cell pairs . DOI: http://dx . doi . org/10 . 7554/eLife . 14620 . 008 Lnc34a asymmetry in dividing cell pairs was confirmed in vivo by RNA-FISH and tubulin staining of xenograft tumors derived from subcutaneously injected CCSCs ( Figure 2E , F ) . We investigated Lnc34a asymmetry in 23 early-stage ( stage I/II ) and 22 late-stage ( stage III/IV ) human CRC specimens ( Figure 1—source data 1 ) . Lnc34a asymmetry in dividing cell pairs is more strongly associated with early-stage CRC , while late-stage CRC mostly has symmetric Lnc34a levels in dividing pairs ( Figure 2G , H ) . To investigate whether Lnc34a regulates CCSC division symmetry , we first knocked down Lnc34a using lentiviral shRNAs , which have been reported to knock down certain nuclear lncRNAs efficiently ( Castel and Martienssen , 2013; Di Ruscio et al . , 2013; Wang et al . , 2015; Xing et al . , 2014 ) . Among the five tested shRNAs against Lnc34a , two showed efficient suppression of Lnc34a ( shLnc34a1 and shLnc34a2; Figure 1—figure supplement 2B ) . Lnc34a knockdown decreased asymmetric division while increasing symmetric , ALDH1-/ALDH1- division ( Figure 2I ) . We then ectopically expressed Lnc34a using lentiviral vectors . Higher level of ectopic Lnc34a was detected in the nucleus than in the cytoplasm ( Figure 1—figure supplement 2C ) . Ectopic Lnc34a expression also decreased asymmetric division , but increased symmetric , ALDH1+/ALDH1+ division instead ( Figure 2J ) . The phenotype was rescued by ectopic miR-34a expression , suggesting that Lnc34a regulates symmetry through miR-34a ( Figure 2J ) . The same trend was observed with CD133 staining ( Figure 2—figure supplement 1C , D ) . Therefore , ectopic Lnc34a seems to promote symmetric CCSC self-renewal , while Lnc34a silencing promotes differentiation . Pair-cell BrdU incorporation assay showed that , when cultured in proliferative medium ( DMEM with 10% FBS ) , the Lnc34a+ daughter cell starts incorporating BrdU and enters into the next division immediately , whereas the Lnc34a- daughter cells does not incorporate BrdU ( Figure 2K ) . Therefore , the Lnc34a+ daughter cell has higher proliferative capacity . Serial sphere propagation assays were performed to evaluate the effect of Lnc34a on CCSC self-renewal . CCSCs containing a control vector exhibited stable sphere formation capability through 3 generations of sphere propagation . Lnc34a knockdown strongly suppressed sphere formation capability , which was completely lost after 3 generations of passage ( Figure 3A , B ) . In contrast , ectopic Lnc34a expression increased sphere numbers and sizes significantly . Ectopic miR-34a abrogated the effect of Lnc34a on sphere formation regulation , suggesting that Lnc34a promotes CCSC self-renewal by targeting miR-34a ( Figure 3A , C ) . 10 . 7554/eLife . 14620 . 009Figure 3 . Lnc34a promotes CCSC self-renewal and tumor formation . ( A ) Representative images of CCSC spheres with Lnc34a knockdown ( shLnc34a1 and shLnc34a2 ) , ectopic Lnc34a expression ( Lnc34a ) , and ectopic Lnc34a/miR-34a expression . ( B , C ) Sphere formation during serial passages after Lnc34a knockdown ( B ) and ectopic Lnc34a and miR-34a expression ( C ) . Equal number of cells was passaged for 3 generations to form spheres . ( D , E ) Knockdown of Lnc34a ( shLnc34a1 and shLnc34a2 ) reduced tumorigenicity , shown by images ( D ) and weights of xenograft tumors ( E ) . ( F , G ) Ectopic Lnc34a expression ( Lnc34a ) enhances tumorigenicity , which can be abrogated by ectopic miR-34a expression . ( H , I ) FACS plots identifying ALDH1+ ( CCSC ) populations in xenograft tumors with Lnc34a knockdown ( H ) or ectopic Lnc34a expression ( I ) . Scale bar , 50 μm . Error bars denote s . d . of triplicates . p<0 . 01; *p<0 . 001 . p-value was calculated based on Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 14620 . 009 Next , we used the mouse xenograft model to examine whether Lnc34a influences tumor growth . All five mice in the control group ( injected with sphere cells containing the control vector ) developed tumors . However , only three mice injected with sphere cells expressing shLnc34a1 and two mice injected with sphere cells expressing shLnc34a2 formed tumors , which are smaller than those of the control group ( Figure 3D , E ) . All 5 mice injected with sphere cells ectopically expressing Lnc34a developed tumors , which are notably bigger than those in the control group . Ectopic miR-34a expression abrogates the effect of ectopic Lnc34a on tumor growth , resulting in similar tumor sizes as the control group ( Figure 3F , G ) . Furthermore , we performed FACS on disassociated xenograft tumor cells . Lnc34a knockdown decreased the ALDH1+ CCSC population in the xenograft tumors ( Figure 3H ) , while ectopic Lnc34a enriched the ALDH1+ CCSC population in the tumors ( Figure 3I ) . Taken together , Lnc34a contributes to CCSC self-renewal and tumorigenesis . Opposite to Lnc34a , miR-34a is downregulated in ALDH1+ CCSCs and upregulated in ALDH1- non-CCSCs ( Figure 4A , B ) . Knockdown of Lnc34a significantly increased miR-34a expression levels , while ectopic Lnc34a expression decreased miR-34a levels ( Figure 4C , D ) . Therefore , Lnc34a suppresses miR-34a expression . RNA FISH showed that Lnc34a and miR-34a are mutually exclusive in the same daughter compartment and are present in opposite daughter compartments in more than 70% of CCSC1 and around 80% of CCSC2 dividing pairs ( Figure 4E , F ) . On the other hand , we only observed symmetric distribution of p53 , the other miR-34a upstream regulator ( Figure 4—figure supplement 1 ) . Therefore , Lnc34a provides a potential mechanism that accounts for asymmetric miR-34a levels in daughter pairs . 10 . 7554/eLife . 14620 . 010Figure 4 . Lnc34a epigenetically silences miR-34a promoter . ( A , B ) RT-qPCR of miR-34a levels in CCSC1 ( A ) and CCSC2 ( B ) . ( C , D ) RT-qPCR of miR-34a levels in CCSC1 ( C ) and CCSC2 ( D ) spheres with Lnc34a knockdown ( shLnc34a1 and shLnc34a2 ) or ectopic expression ( Lnc34a ) . NC is the control vector . ( E ) Representative images of Lnc34a and miR-34a asymmetry in CCSC1 and CCSC2 sphere cells . ( F ) Quantification of ( E ) . Lnc34a and miR-34a distributions are mutually exclusive ( M . E . ) during most CCSC divisions . ( G ) Bisulfite sequencing analysis showing miR-34a promoter methylation status in ALDH1+ ( CCSC ) and ALDH1- ( non-CCSC ) cells isolated from sphere cells . PCR products amplified from bisulfite-treated genomic DNA were cloned and sequenced to reveal the methylation status of individual CpG sites . Percentages of the methylated CpG sites ( filled circles ) among all scored sites are indicated . ( H ) Lnc34a knockdown decreased miR-34a promoter methylation in sphere cells . ( I ) Ectopic Lnc34a expression increased miR-34a promoter methylation in sphere cells . ( J , K ) ChIP-qPCR with antibodies against acetylated histones H3 and H4 . Lnc34a knockdown decreased miR-34a promoter acetylation ( J ) , while ectopic Lnc34a expression increased acetylation ( K ) . ( L ) RT-qPCR measurements of Lnc34a expression in early- and late-stage CRC specimens . ( M ) RT-qPCR measurements of miR-34a expression in early- and late-stage CRC specimens . ( N ) Bisulfite sequencing analysis of miR-34a promoter methylation status in early- and late-stage CRC specimens . Scale bar , 8 μm . Error bars denote s . d . of triplicates . p<0 . 01; *p<0 . 001 . p-value was calculated based on Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 14620 . 01010 . 7554/eLife . 14620 . 011Figure 4—figure supplement 1 . p53 symmetry . ( A ) ( A ) Representative immunofluorescence images showing symmetric distribution of p53 during CCSC division . ( B ) Percentage of p53 division type . ( C ) Representative immunofluorescence images showing symmetric distribution of p21 during CCSC division . ( D ) Percentage of p21 division type . Sym , symmetric segregation; Asym , asymmetric segregation; Am , ambiguous . DOI: http://dx . doi . org/10 . 7554/eLife . 14620 . 01110 . 7554/eLife . 14620 . 012Figure 4—figure supplement 2 . Lnc34a epigenetically silences miR-34a promoters in Caco-2 and HT29 cells . ( A , B ) RT-qPCR of miR-34a levels in CRC lines Caco-2 ( A ) and HT29 ( B ) . Ectopic Lnc34a expression suppressed miR-34a expression . ( C , D ) Bisulfite sequencing analysis showing ectopic Lnc34a expression increased miR-34a promoter methylation in Caco-2 ( C ) and HT29 ( D ) . ( E , F ) ChIP-qPCR with antibodies against acetylated histones H3 and H4 . Ectopic Lnc34a expression decreased miR-34a promoter acetylation in Caco-2 ( E ) and HT29 ( F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14620 . 01210 . 7554/eLife . 14620 . 013Figure 4—figure supplement 3 . Lnc34a , miR-34a , and promoter methylation levels in CRC specimens . ( A ) RT-qPCR showing Lnc34a and miR-34a expression in individual CRC specimens . Levels are normalized to corresponding actin levels . ( B ) Bisulfite sequencing analysis of miR-34a promoter methylation in the same CRC specimens shown in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14620 . 013 Bisulfite sequencing was then performed to evaluate miR-34a promoter methylation in ALDH1+ CCSCs and ALDH1- non-CCSCs isolated from spheres . 93 . 3% of tested CpG islands were methylated in CCSCs; in contrast , methylation rate was as low as 2 . 2% in non-CCSCs ( Figure 4G ) . Knockdown of Lnc34a diminished overall miR-34a promoter methylation in sphere cells ( Figure 4H ) , whereas ectopic Lnc34a expression significantly enhanced miR-34a promoter methylation , compared with the control vector ( Figure 4I ) . Besides methylation , ChIP-qPCR showed that Lnc34a knockdown increases acetylated histones H3 and H4 associated with the miR-34a promoter ( Figure 4J ) , whereas ectopic Lnc34a expression decreased acetylated histones H3 and H4 ( Figure 4K ) . Taken together , the data suggests that Lnc34a silences miR-34a expression in CCSCs by promoting methylation and histone deacetylation of the miR-34a promoter . The effect of ectopic Lnc34a suggests that Lnc34a might act both in cis and in trans , as have been observed for various lncRNAs such as Evf-2 and the capacity of ectopically supplied cis-acting lncRNAs to act in trans ( Di Ruscio et al . , 2013; Feng et al . , 2006; Gomez et al . , 2013; Jeon and Lee , 2011; Martianov et al . , 2007; Rinn and Chang , 2012; Schmitz et al . , 2010 ) . Lnc34a also silences miR-34a in common CRC cell lines . Ectopic Lnc34a expression suppressed miR-34a expression , and promoted methylation and deacetylation of the miR-34a promoter in CRC cell lines Caco-2 and HT29 ( Figure 4—figure supplement 2 ) . RT-qPCR performed in 23 early-stage ( stage I/II ) and 22 late-stage ( stage III/IV ) CRC specimens showed that Lnc34a expression is correlated with CRC progression . Overall , Lnc34a expression is lower in early-stage CRC and increases in late-stage CRC ( Figure 4L , Figure 4—figure supplement 3A ) . miR-34a expression follows a reverse trend ( Figure 4M , Figure 4—figure supplement 3A ) . Consistent with Lnc34a methylation of the miR-34a promoter , bisulfite sequencing revealed that the miR-34a promoter is more methylated in late-stage CRC than in early-stage CRC ( Figure 4N , Figure 4—figure supplement 3B ) . To understand the mechanisms via which Lnc34a regulates miR-34a expression , we performed an RNA pull-down assay with biotin-labeled Lnc34a , followed by mass spectrometry ( MS ) , to search for potential Lnc34a-associated proteins . The DNA methyltransferase Dnmt3a , Histone Deacetylase 1 ( HDAC1 ) , and Prohibitin 2 ( PHB2 ) were identified to be associated with Lnc34a ( Figure 5A and Figure 5—source data 1 ) . RNA immunoprecipitation ( RIP ) using specific antibodies against Dnmt3a , HDAC1 and PHB2 further confirmed the interactions ( Figure 5B ) . In contrast , RNA pulldown and RIP did not detect any interaction between Lnc34a and Dnmt1 , an enzyme that plays important roles in maintaining methylation during DNA replication ( data not shown ) . 10 . 7554/eLife . 14620 . 014Figure 5 . Lnc34a recruits epigenetic regulators . ( A ) Western blot following RNA-pull down showing Lnc34a interaction with PHB2 , Dnmt3a and HDAC1 in CCSC1 ( left ) and CCSC2 ( right ) sphere cells . RNA-pull down was performed using CCSC lysates with biotin-labeled Lnc34a , antisense and tRNA . Actin was used for input control . ( B ) RNA immunoprecipitation ( RIP ) showing Lnc34a interaction with PHB2 , Dnmt3a and HDAC1 in CCSC1 ( left ) and CCSC2 ( right ) sphere cells . ( C ) RIP showing PHB2 knockdown disrupts Lnc34a interaction with Dnmt3a , but has no effect on Lnc34a interaction with HDAC1 . ( D ) RIP showing Dnmt3a knockdown does not affect Lnc34a interaction with PHB2 or HDAC1 . ( E ) RIP showing HDAC1 knockdown has limited effect on Lnc34a interaction with PHB2 or Dnmt3a . ( F ) Mapping PHB2 and HDAC1 interaction domains on Lnc34a . Upper panel , schematic illustration of full-length Lnc34a and the truncated fragments for RNA put-down . Lower panel , Western blot of PHB2 and HDAC1 from RNA put-down of the fragments . ( G ) EMSA showing Lnc34a/PHB2 ( left ) and Lnc34a/HDAC1 ( right ) interactions . ( H ) RT-qPCR of miR-34a levels after expressing full-length or truncated fragments of Lnc34a . ( I ) In vitro interaction assay binding of the truncated fragment ( 267–560 bp ) to the DNA containing the miR-34a promoter sequence . ( J ) Schematic illustration of Lnc34a interaction with PHB2 , Dnmt3a and HDAC1 . ( K , L , M ) RT-qPCR showing knockdown of Dnmt3a ( K ) , HDAC1 ( L ) , and PHB2 ( M ) increased miR-34a expression in sphere cells . ( N , O ) RT-qPCR showing treatments with HDAC inhibitor SAHA ( N ) or TSA ( O ) increased miR-34a expression in sphere cells . Error bars denote s . d . of triplicates . *p<0 . 001 . p-value was calculated based on Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 14620 . 01410 . 7554/eLife . 14620 . 015Figure 5—source data 1 . Potential Lnc34a-associated proteins identified by biotinylated Lnc34a pull-down and mass spectrometry . DOI: http://dx . doi . org/10 . 7554/eLife . 14620 . 015 To investigate how Lnc34a interacts with Dnmt3a , HDAC1 and PHB2 , we performed RIP while knocking down each of the proteins . Knockdown of PHB2 abolished the interaction between Lnc34a and Dnmt3a , but had no effect on the interaction between Lnc34a and HDAC1 ( Figure 5C ) . Knockdown of Dnmt3a did not affect the interaction of Lnc34a with either PHB2 or HDAC1 ( Figure 5D ) . Knockdown of HDAC1 did not interrupt Lnc34a and Dnmt3a interaction , and only had limited effect on Lnc34a and PHB2 interaction ( Figure 5E ) . These data suggest that Lnc34a interacts with PHB2 and HDAC1 , and recruits Dnmt3a through PHB2 . We then serially truncated Lnc34a and performed RNA pull-down assays to map HDAC1 and PHB2 binding to Lnc34a . The 1–267 bp fragment is sufficient to bind HDAC1 , and the 560–693 bp fragment is sufficient to bind PHB2 ( Figure 5F ) . Interaction between the fragments and their cognate proteins were further validated by the electrophoretic mobility shift assay ( EMSA ) . Incubation of labeled RNA probes Lnc34a:1–267 bp with recombinant HDAC1 and Lnc34a:560–690 bp with recombinant PHB2 resulted in specific gel retardation , while unlabeled RNA probes of the same fragments competitively disrupted those binding ( Figure 5G ) . All three fragments are needed for full suppression of miR-34a expression ( Figure 5H ) . Although the 267–560 bp fragment does not interact with either HDAC1 or PHB2 , the in vitro interaction assay shows that it directly binds to the miR-34a promoter ( Figure 5I ) . Therefore , Lnc34a binds to the miR-34a promoter via the 267–560 bp sequence , and recruits HDAC1 and Dnmt3a/PHB2 via the two flanking ( 1–267 bp and 560–690 ) sequences ( Figure 5J ) . We then knocked down PHB2 , Dnmt3a and HDAC1 respectively , followed by RT-qPCR measurements of miR-34a expression . Knockdown of PHB2 , Dnmt3a or HDAC1 upregulated miR-34a expression ( Figure 5I–K ) . Inhibition of HDAC activity by SAHA and TSA also increased miR-34a expression ( Figure 5L , M ) . The data suggest that these epigenetic regulators influence miR-34a expression levels . The abundance of lncRNA in the human genome is being increasingly appreciated , but our understanding of their diverse functions is still lagging ( Mercer et al . , 2009; Rinn and Chang , 2012 ) . We demonstrate that a lncRNA , Lnc34a , can initiate CCSC asymmetric division by targeting miR-34a . Previously , lncRNAs like HOTAIR and Xist have been shown to cause histone H3 lysine 27 methylation or lysine 4 demethylation ( Gupta et al . , 2010; Tsai et al . , 2010; Zhao et al . , 2008 ) . Here , Lnc34a binds to the miR-34a promoter via its middle fragment , and recruits PHB2/Dnmt3a and HDAC1 via its flanking sequences to methylate and deacetylate the promoter , silencing miR-34a expression . This process reminds us of the ordered steps of protein-mediated DNA methylation—a DNA binding protein first interacts with the promoter , via which DNA methyltransferases are further recruited ( Chu et al . , 2015; Serra et al . , 2014; Wajapeyee et al . , 2013 ) . Lnc34a promotes CCSC self-renewal , and Lnc34a asymmetry leads to cell fate asymmetry in CCSC division . This effect is mediated by miR-34a , which has been shown to target factors of Notch and Wnt signaling pathways , both of which are essential for CCSC self-renewal ( Bu et al . , 2013; Chen et al . , 2015; Vermeulen et al . , 2010 ) . In late-stage CRC , Lnc34a expression and miR-34a promoter methylation is upregulated , while miR-34a expression is downregulated . Lnc34a demonstrates that lncRNA can target microRNA for cellular control . Given that lncRNAs occupy the majority of the genome ( Mattick and Rinn , 2015 ) , lncRNA/microRNA circuitry can potentially increase the complexity of regulatory networks . p53 is a well-known upstream regulator of miR-34a , and loss of p53 function certainly downregulates miR-34a . However , the discovery of Lnc34a demonstrates an alternative , epigenetic mechanism that cancer cells can utilize to silence miR-34a without having to mutate p53 . Although p53 knockout has been reported to reduce asymmetric division in mammary stem cells ( Cicalese et al . , 2009 ) , p53 is not known to be a major regulator of differentiation and is symmetric during CCSC division . Lnc34a provides normal and cancer cells a way to decouple mir-34a mediated cell fate decisions from p53 , which may be present in both undifferentiated and differentiated cells . Human CRC cell lines Colo205 , SW480 , HT29 , SW620 , LS174T , DLD1 , Caco-2 were purchased from ATCC and cultured in RPMI-1640 medium . No mycoplasma contamination was detected . Human CCSCs were isolated and cultured as described previously ( Bu et al . , 2013 ) . Briefly , CCSCs were isolated from patient tumors by FACS based on markers CD44 , CD133 and ALDH1 and functionally validated by serial sphere formation , tumor initiation , and self-renewal assays . For this study , original frozen stocks for the first passage were used . The CCSCs have not been authenticated by STR profiling . No mycoplasma contamination was detected . CCSCs were cultured as spheres in ultralow-attachment flasks ( Corning , Tewksgury , MA ) in DMEM/F12 ( Invitrogen , Pittsburgh , PA ) , supplemented with nonessential amino acids ( Fisher , Pittsburgh , PA ) , sodium pyruvate ( Fisher ) , Penicillin-streptomycin ( Fisher ) , N2 supplement ( Invitrogen ) , B27 supplement ( Invitrogen ) , 4 μg/mL heparin ( Sigma , Mendota Heights , MN ) , 40 ng/mL epidermal growth factor ( Invitrogen ) , and 20 ng/mL basic fibroblast growth factor ( Invitrogen ) at 37°C and 5% CO2 . To measure tumor sphere formation , single CCSCs were plated in 24-well ultra-low attachment plates ( Corning ) at 1 , 000 cells per well . Tumor spheres were counted after 2 weeks in culture by an inverted microscope ( Olympus ) . 45 frozen CRC specimens of different clinical stages were acquired from Weill Cornell Medical College ( WCMC ) Colon Cancer Biobank . The CRC stage was determined according to the TNM staging system . The clinical data for the patients are summarized in Figure 1—source data 1 . The studies followed informed consent and approval of the IRB committee at Weill Cornell Medical College . Pair-cell assay for CCSC division were performed as described previously ( Bultje et al . , 2009 ) . Briefly , spheres were dissociated and the single cells were plated on an uncoated glass culture slide ( Corning ) and allowed to divide once . After being fixed and blocked , the cells were incubated with anti-ALDH1 ( clone H-4 , 1:100 , Santa Cruz , Dallas , TX ) , anti-CD133 ( 1:200 , Abcam , Cambridge , MA ) and anti-α-tubulin ( 1:500 , Abcam ) antibodies overnight at 4°C . For the BrdU incorporation assay , sphere cells were cultured in proliferative medium ( DMEM with 10% FBS ) for 24 hr . Single cells were then plated and allowed to divide once in proliferative medium ( 1st division ) . After treatment with BrdU ( Sigma ) for 3 hr , the cells were fixed in cold 70% ethanol , incubated in 2 M HCl for 1 hr , washed , and switched to 100 mM Na2B4O7 for 2 min . After being blocked in 10% normal goat serum for 1 hr , the cells were then incubated with anti-BrdU ( 1:200 , Sigma ) antibody at 4°C overnight . The cells were then incubated with fluorescence-conjugated secondary antibody or streptavidin ( Invitrogen ) for 1 hr at room temperature . After counterstained with DAPI ( Invitrogen ) , the slides were observed under a fluorescent microscope ( Olympus , Jupiter , FL ) . RNA FISH was performed as described previously ( Lu and Tsourkas , 2009 ) . In this study , Digoxigenin ( DIG ) –labeled locked nucleic acid ( LNA ) probe ( Exiqon , Woburn , MA ) against miR-34a or Biotin-labeled LNA probe against Lnc34a ( Exiqon ) were used for RNA FISH . RNA expression was detected by Rhodamine Red labeled secondary antibody or Alexa Fluor 488 conjugated streptavidin ( Invitrogen ) . Anti-α-tubulin was used to identify dividing cells and DAPI ( Invitrogen ) was used for nucleic counterstaining . A 293 bp fragment was amplified using primers: 5`-GGTGGAGGAGATGCCGC-3` and 5`-ACCTGGGTGCATGCTGGGACG-3` . To identify the full length of Lnc34a , 3`RACE and 5`RACE was performed using kit with the primers: 5`- GCAGGACTCCCGCAAAATCTC-3` and 5`- CTCAGTCCGTGCGAAAGTTTG-5` respectively . The full length of Lnc34a was then amplified using the primers: 5`-TTAACCAGTCGGCCTTCCTCGCC-3` and 5`-TGAGATTAACCGACTTTCCCAAG-3` , then cloned into pGEM-T ( Promega , Durham , NC ) for sequencing . The full length of Lnc34a was cloned into pMSCV PIG vector ( Addgene , Cambridge , MA ) for ectopic Lnc34a expression study . shRNAs against Lnc34a were designed using Invitrogen online tool and cloned in pMSCV PIG vector . shRNAs against PHB2 , Dnmt2a and HDAC1 were purchased from Sigma . The knockdown efficiency was validated by RT-qPCR . Northern blot was performed using NorthernMax Kit ( Invitrogen ) according to the manufacturer’s instructions . The probes were generated using PCR DIG Probe Synthesis Kit ( Roche , Indianapolis , IN ) with the primers: 5`- TAGCCGAGCAAAACCCC-3` and 5`- ATGTGGGACACGGATGAGA-3` . Bisulfite sequencing was performed using EZ DNA methylation kit ( Zymo , Irvine , CA ) . 9 sequencing runs were carried out for each condition . Flow cytometry were performed as described previously ( 4 ) . CD133 expression was detected using anti-CD133 ( clone C24B9 , 1:50 , Cell Signaling , Beverly , MA ) and ALDH1 levels were analyzed using the Aldeflour kit . The samples were analyzed using a BD LSR II flow cytometer . The raw FACS data were analyzed with the FlowJo software to gate cells according to their forward ( FSC ) and side ( SSC ) scatter profiles . Total RNA was extracted from the cells using the TRIzol Reagent ( Invitrogen ) . cDNA was synthesized using the High Capacity cDNA Archive Kit ( Applied Biosystems , Foster city , CA ) . Quantitative PCR was carried out using the TaqMan MicroRNA Assay ( Applied Biosystems ) to detect miR-34a levels and the SYBR Green System ( Applied Biosystems ) to detect other gene expression . The miR-34a primer and U6 primer were purchased from Applied Biosystems . Other primer sequences include: Lnc34a , 5'-GGAGGCTACACAATTGAACAGG-3' and 5`-AGTCCGTGCGAAAGTTTGC-3`; actin , 5`-CGCGAGAAGATGACCCAGAT-3` and 5`-ACAGCCTGGATAGCAACGTACAT-3`; . The expression of each gene was defined from the threshold cycle ( Ct ) , and the relative expression levels were calculated using the 2-△△Ct method after normalization to the actin expression level . Full length of Lnc34a cDNA and it truncations were cloned into pGEM-3ZF ( + ) . Biotin-labeled RNAs were transcribed from the linearized pGEM-3ZF plasmid in vitrousing a biotin labeling mix ( Roche ) and T7 polymerase ( Promega ) . The biotinylated RNA was heated to 90°C for 2 min , incubated on ice for 2 min , and then shifted to RT for 20 min with RNA renature buffer ( 10 mM tris-HCL pH7 . 0 , 0 . 1M KCL , 10 mM MgCl2 to allow proper secondary structure formation . The cell lysates were freshly prepared using RIPA buffer ( Millipore , Billerica , MA ) with proteinase inhibitor ( Roche ) . After precleared using Dynabeads M-270 streptavidin ( Invitrogen ) , the cell lysates were diluted in binding buffer and incubated with the folded RNA for 2 hr at 4°C . Dynabeads M-270 streptavidin were then added into the mixture and incubated for 1 hr at 4°C . After washing , the RNA-binding protein complexes were released from the Dynabeads . The retrieved proteins were collected for Mass Spec and Western blotting validation . RNA-EMSA was performed using a LightShift Chemiluminescent RNA EMSA Kit ( Thermo Scientific , Pittsbrugh , PA ) according to the manufacturer’s instructions . RIP assays were performed using a RIP RNA-binding protein immunoprecipitation kit ( Millipore ) according to the manufacturer’s instructions . Antibodies against PHB2 ( Bethl , Montgomery , TX ) , HDAC1 ( Bethl ) and Dnmt3a ( Abcam ) were added into the cell lysates . Lnc34a was retrieved from the complexes and evaluated by RT-qPCR . ChIP was performed using a ChIP assay kit ( Millipore ) as described previously ( 4 ) . Antibodies against acetylated histones H3 and H4 ( Millipore ) were used to evaluate histone modifications associated with the miR-34a promoter . Enrichment of miR-34a promoter fragments was quantified by RT-qPCR with the primers: 5'-CACCTGGTCCTCTTTCCTTT-3' and 5'- TCCTCCTTCCTGCTCGT -3' . Cells were lysed in RIPA lysis buffer supplemented with cocktail protease inhibitor ( Roche ) . Proteins were separated by SDS-PAGE and transferred onto a Hybond membrane ( Amersham ) . The membranes were incubated with primary antibodies either anti-PHB2 ( 1:1000 , Bethl ) , anti-Dnmt3a ( 1:500 , Abcam ) , anti-HDAC1 ( 1:1000 , Bethl ) or anti-Actin ( 1:1000 , Abcam ) in 5% milk/TBST buffer ( 25 mM Tris pH 7 . 4 , 150 mM NaCl , 2 . 5 mM KCl , 0 . 1% Triton-X100 ) overnight , followed by incubation with horseradish peroxidase ( HRP ) -conjugated anti-mouse or anti-rabbit IgG ( Santa Cruz ) for 1 hr . The target proteins were detected on membrane by enhanced chemiluminescence ( Pierce , ) . Data were expressed as mean ± standard deviation of three biological repeats . Student t-tests were used for comparisons , with p<0 . 05 considered significant . This work was supported by NIGMS R01GM95990 , R01GM114254 , NSF 1350659 , R01 Ca098626 , NSF 1137269 , DARPA 19–1091726 , and NYSTEM C029543 .	Tumors are made of millions of cells that are not all the same . A type of cancer cell known as cancer stem cells ( or CSCs for short ) are often better at dividing to produce new cells and moving to new sites in the body than other types of cancer cell . Very small molecules called micro ribonucleic acids ( or microRNAs for short ) can influence how CSCs grow and divide by regulating the activity of specific genes . For example , a microRNA molecule called miR-34a suppresses the activity of several genes – which slows the growth of various tumors , including lung and bowel cancers . This miR-34a is often missing from some types of cells in advanced tumors . Genes encode the instructions to produce RNAs , and Wang , Bu et al . wanted to find out what stops miR-34a being produced in certain bowel cancer cells . The experiments revealed a new , very long RNA molecule – named long non-coding RNA 34 ( or Lnc34a ) – that binds to the gene that encodes miR-34a . Lnc34a recruits proteins that modify the gene and switch off the production of miR-34a . Furthermore , microscopy experiments revealed that when colon cancer cells divide , Lnc34a is distributed unevenly so that it blocks the production of miR-34a in one daughter cell but not the other . Lastly , Wang , Bu et al . confirmed that Lnc34a is found in higher levels in CSCs than in other cancer cells , which helps them to grow and divide more rapidly . Future experiments will try to find out what controls the production of Lnc34a and search for drugs that can block this process in cancer cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine", "cancer", "biology" ]	2016	A long non-coding RNA targets microRNA miR-34a to regulate colon cancer stem cell asymmetric division
Feedback inhibitory motifs are thought to be important for pattern separation across species . How feedback circuits may implement pattern separation of biologically plausible , temporally structured input in mammals is , however , poorly understood . We have quantitatively determined key properties of netfeedback inhibition in the mouse dentate gyrus , a region critically involved in pattern separation . Feedback inhibition is recruited steeply with a low dynamic range ( 0% to 4% of active GCs ) , and with a non-uniform spatial profile . Additionally , net feedback inhibition shows frequency-dependent facilitation , driven by strongly facilitating mossy fiber inputs . Computational analyses show a significant contribution of the feedback circuit to pattern separation of theta modulated inputs , even within individual theta cycles . Moreover , pattern separation was selectively boosted at gamma frequencies , in particular for highly similar inputs . This effect was highly robust , suggesting that frequency-dependent pattern separation is a key feature of the feedback inhibitory microcircuit . Efficiently discriminating similar percepts or experiences is a central capability common to invertebrate and vertebrate species . In general terms , such discrimination can be achieved by decreasing the overlap in representations by neuronal ensembles between input and output patterns , a process termed ‘pattern separation’ ( Cayco-Gajic and Silver , 2019; Marr , 1971; McNaughton and Morris , 1987; Rolls , 2013 ) . Numerous studies have proposed cellular and circuit mechanisms that support this computation . For instance , sparse divergent inputs , specialized intrinsic properties and feedforward inhibition are thought to generally contribute ( Cayco-Gajic et al . , 2017; Cayco-Gajic and Silver , 2019; Krueppel et al . , 2011; Mircheva et al . , 2019 ) . Another common feature of most of these models and experimental studies is a critical role of feedback inhibition ( Cayco-Gajic et al . , 2017; Rolls , 2013 ) . Feedback circuits differ from the above mechanisms in that they can i ) implement direct competition between active cells through lateral inhibition and ii ) integrate information about the actual global activity level in a population allowing efficient normalization ( Braganza and Beck , 2018; Wick et al . , 2010; Wiechert et al . , 2010 ) . Indeed , in the insect olfactory system a critical role of such a circuit has been causally demonstrated ( Lin et al . , 2014; Papadopoulou et al . , 2011 ) . In mammals , substantial evidence points toward a role of the hippocampal dentate gyrus ( DG ) for pattern separation during memory formation and spatial discrimination ( Bakker et al . , 2008; Berron et al . , 2016; Gilbert et al . , 2001; Leal and Yassa , 2018; Leutgeb et al . , 2007; McHugh et al . , 2007; Neunuebel and Knierim , 2014; Stefanelli et al . , 2016; van Dijk and Fenton , 2018 ) . The DG is thought to subserve this task by converting different types of inputs to sparse , non-overlapping activity patterns of granule cells ( GCs ) . However , in contrast to the insect olfactory system , the DG feedback circuit is extremely complex , comprising numerous interconnected interneuron types ( Supplementary Table 1 ) ( Bartos et al . , 2002; Dasgupta and Sikdar , 2015; Espinoza et al . , 2018; Ewell and Jones , 2010; Freund and Buzsáki , 1996; Geiger et al . , 1997; Harney and Jones , 2002; Hefft and Jonas , 2005; Kraushaar and Jonas , 2000; Larimer and Strowbridge , 2008; Lee et al . , 2016; Liu et al . , 2014; Lysetskiy et al . , 2005; Sambandan et al . , 2010; Savanthrapadian et al . , 2014; Sik et al . , 1997; Yu et al . , 2015; Yuan et al . , 2017; Zhang et al . , 2009 ) . For instance , interneurons subserving feedback inhibition are also incorporated into circuits mediating feedforward inhibition ( Ewell and Jones , 2010; Hsu et al . , 2016; Lee et al . , 2016 ) and disinhibition ( Savanthrapadian et al . , 2014; Yuan et al . , 2017 ) . This makes it difficult to predict the net inhibition arising from GC activity . We reasoned that to assess if feedback inhibition is indeed suitable for the purpose of pattern separation in the DG , it is necessary to determine how efficiently the activity of sparse GC ensembles recruits net inhibition , that is the dynamic range and gain of the feedback inhibitory microcircuit . It is furthermore necessary to quantify the spatial and temporal properties of the elicited inhibition , in order to investigate its impact on biologically plausible , temporally structured input . For instance , the DG shows prominent theta oscillations during exploration and distinctive slow-gamma activity during associative memory encoding ( Hsiao et al . , 2016; Lasztóczi and Klausberger , 2017; Pernía-Andrade and Jonas , 2014; Sasaki et al . , 2018; Trimper et al . , 2017 ) . Importantly , both sparsity and temporal oscillations will critically affect a proposed pattern separation function . For instance , feedback inhibition must by definition occur with a delay , a property frequently abstracted away in computational models ( Myers and Scharfman , 2009; Rolls , 2016 ) , but potentially critical during oscillatory activity . Here , we combine patch-clamp recordings , multiphoton imaging and optogenetics to provide a first quantitative , empirical description of the net input-output function of a feedback inhibitory microcircuit . This includes the spatiotemporal organization of net feedback inhibition elicited by a spatially restricted GC population and the net short-term dynamics within the feedback microcircuit . Finally , we integrate our data into a biophysically realistic computational model and probe its ability to perform pattern separation . We find a moderate feedback inhibition mediated pattern separation effect during theta modulated input but a substantial separation , particularly of highly similar inputs , during gamma oscillations . We reasoned that the ultimately relevant parameter for the putative pattern separation effect of feedback inhibition is the net inhibition arriving at GCs . We therefore treated the feedback microcircuit as a black-box striving to relate only its net input ( fraction of GCs active ) to its net output ( feedback inhibition in GCs ) . To this end , we antidromically recruited feedback inhibitory circuits , while simultaneously recording GC inhibition and population activity ( see schematic in Figure 1A ) . Electrical stimulation reliably evoked graded IPSCs in dentate GCs , that increased with stimulation strength ( maximal amplitude of 324 . 1 ± 99 . 2 pA , n = 8; Figure 1B ) . Feedback IPSCs were completely blocked by 10 µM GABAzine ( to 1 . 5 ± 0 . 9% , n = 7 cells , P ( df = 6 , t = 117 . 4 ) <0 . 001 , one-sided t-test ) , as expected ( Figure 1C ) . To ascertain that IPSCs were mediated by synaptically activated interneurons rather than interneurons directly recruited by electrical stimulation , we only included slices where inhibition was successfully blocked by glutamatergic antagonists ( 25 µM CNQX and 50 µM D-APV , 8 of 21 experiments , Figure 1C ) . We also tested if inhibition of glutamate release from mossy fibers , which can be specifically achieved via mGluR2/3 activation by DCG-IV ( Doherty and Dingledine , 1998; Toth et al . , 2000 ) , reduces feedback IPSCs . Indeed , we found that IPSCs were reduced to 16 . 3 ± 6 . 1% by 0 . 5 µM DCG-IV ( n = 4 cells , P ( df = 3 , t = 13 . 73 ) <0 . 001 , one-sided t-test , Figure 1C ) . In order to relate the measured IPSCs to the fraction of GCs activated by a given stimulation strength , we used population Ca2+ imaging with multibeam two-photon microscopy ( Figure 1A , see Materials and methods ) . After bolus-loading GCs with the Ca2+ indicator OGB-1-AM ( see Materials and methods ) , antidromic stimulation caused action potential associated Ca2+ elevations in a subset of GCs ( Figure 1D , transients indicated by * ) . Before quantifying population activity , we verified the reliable detection of single action potentials under our conditions using simultaneous cell-attached recordings from dentate GCs ( Figure 1E; Figure 1—figure supplement 1 ) . Briefly , cells were differentiated into true responders or non-responders on the basis of cell-attached recordings ( Figure 1E , F; responders green , non-responders grey ) . A histogram of the peak ΔF/F of non-responders upon a single stimulus was fitted with a Gaussian ( Figure 1F right , grey dots , grey bars , n = 33 ) and the threshold set to the quadruple standard deviation of this fit ( 0 . 94% ΔF/F , dashed line in Figure 1F ) . We estimated that this threshold would yield approximately equal numbers of false positives and false negatives ( Figure 1—figure supplement 1F ) . We additionally controlled for possible errors through variable dye loading and the overestimation of the active cell-fraction through accidental detection of adjacent active cells ( Figure 1—figure supplement 1G , H , respectively ) . Orientation of hippocampal slices may be a critical feature in determining the extent of feedback connectivity . We therefore systematically assessed the magnitude of feedback activation of GCs using imaging in slices obtained from different dorso-ventral levels of the hippocampus ( see inset of Figure 1G ) . We found a clear connectivity maximum within horizontal slices obtained at a distance of ~1750 µm from the temporal pole ( Figure 1G , H; Bischofberger et al . , 2006 ) . In these and all following experiments we therefore used exclusively slices obtained at 1400–2100 µm from the temporal pole , where the orientation of hippocampal slices matches the orientation of mossy fibers . Combining the IPSC recordings with population Ca2+ imaging allowed us to probe the input-output relationship of the feedback inhibitory microcircuit . Inhibition was recorded in a GC within or immediately adjacent to the imaging field , and stimulation strength was increased gradually ( Figure 1I ) . The IPSC saturated at 300µA stimulation strength , where the mean active cell fraction was 2 . 2 ± 0 . 7% and the mean IPSC reached 93 . 1 ± 3 . 4% of the maximal IPSC ( Figure 1I , J , n = 20 for imaging , n = 8 for IPSCs including six slices with both ) . Plotting the IPSC magnitude vs . the cell fraction showed that the magnitude of feedback inhibition rises steeply , reaching ~90% with less than 3% of GCs active and complete saturation at 3 . 7 ± 1 . 7% of cells ( Figure 1K ) . These experiments yielded a first quantitative estimate of the input-output relation of the feedback-inhibitory microcircuit in the DG . We then decided to verify these findings using an alternative method , which allowed spatially controlled and less synchronous GC activation . Mice selectively expressing ChR2 ( H134R ) -eYFP in GCs were created by crossing Prox1-Cre mice with Ai32-mice ( Figure 2A , see Materials and methods ) . Focal optogenetic stimulation was achieved through a laser coupled into the microscope light path , yielding an 8 µm stimulation spot ( Figure 2B ) . Brief ( 20 ms , 473 nm ) light pulses within the molecular layer approximately 40 µm from the dentate GC layer elicited reliable IPSCs in GCs ( Figure 2C ) . Increasing the light intensity evoked larger IPSCs that showed clear saturation ( Figure 2C , D , Power = 7 AU corresponding to 1 . 7 mW , see Materials and methods ) . Inhibition was completely blocked by combined application of 40 µM CNQX and 50 µM D-APV ( Figure 2E , n = 9 ) , confirming that it is recruited via glutamatergic collaterals . The maximal IPSC amplitude obtained optically vs . electrically in experiments in which both stimulations were performed were similar ( Figure 2F , paired t-test , P ( df = 3 , t = 1 . 568 ) =0 . 2148 , n = 4 ) , indicating that similar maximal inhibition is recruited despite the differences in the activated GC population ( distributed vs local; synchronous vs . less synchronous ) . In order to relate feedback inhibition to the underlying GC activity levels , we performed systematic cell attached recordings of GCs in the same slices in which inhibition was recorded ( ~2 cells per slice , Figure 2—figure supplement 1 ) . Briefly , we recorded the spatial firing probability distribution in response to focal stimulation for each laser power . We then estimated the mean firing probability of GCs throughout the section , which is equivalent to the expected active GC fraction , by incorporating measurements of the light intensity distribution throughout the slice ( Figure 2G , black ) . We additionally estimated an upper and lower bound by assuming either no decay of firing probability with slice depth or isometric decay ( Figure 2G , grey dashed lines ) . Combining the input-output relations of IPSCs ( Figure 2D ) and the estimated active cell fraction ( Figure 2G ) again revealed that inhibition is recruited steeply , saturating when approximately 4% of GC are active ( Figure 2H ) . Importantly , the resulting recruitment function of inhibition is unlikely to be affected by voltage escape errors ( Figure 2—figure supplement 2 ) . This is because such errors scale linearly with synaptic conductance and will thus affect the absolute but not the relative amplitude of the somatically measured IPSC . Next , we compared the focal light activation with global activation via a light fiber positioned over the surface of the slice ( with powers up to 50 mW , Figure 2I ) . Under global stimulation all cells tested fired APs with 100% reliability and independent of location , even though focal stimulation in direct proximity to the cell led to much lower maximal firing probabilities ( Figure 2I , middle , 100 . 0 ± 0 . 0 versus 31 . 2 ± 7 . 1% respectively , paired t-test , P ( df = 7 , t = 9 . 74 ) <0 . 001 , n = 8 ) . At the same time , the maximal IPSC amplitude did not increase further upon global stimulation ( Figure 2I , right , 356 . 9 ± 76 . 2 versus 344 . 3 ± 77 . 5 pA , paired t-test , P ( df = 9 , t = 1 . 112 ) =0 . 29 , n = 10 ) . This implies that additional activation of remote GCs cannot recruit interneurons beyond those activated by local GC populations . Thus , the recruitment of feedback inhibition in the DG is steep , with a dynamic range tuned to sparse populations of GCs ( up to 3–4% of cells ) . Previous work has addressed the lower limit of the recruitment of feedback inhibition in various cortical areas ( Jouhanneau et al . , 2018; Kapfer et al . , 2007; Miles , 1990; Silberberg and Markram , 2007 ) . The authors report the ability of even a single principal cell to activate feedback inhibitory interneurons and a supralinear increase of inhibition as the second and third principal cells are co-activated ( Kapfer et al . , 2007 ) . Given our findings so far we asked whether single GCs might also suffice to elicit feedback inhibition in the DG . To this end , we performed dual patch clamp recordings and elicited short trains of 10 action potentials at 100 Hz in one cell while monitoring inhibition in the other ( Figure 2—figure supplement 3 , n = 15 ) . However , in contrast to the neocortex ( Kapfer et al . , 2007; Silberberg and Markram , 2007 ) and area CA3 ( Miles , 1990 ) , we did not find single GC-induced feedback inhibition in any of these experiments , consistent with a recent large scale study reporting that such connections are extremely sparse ( 0 . 124% ) ( Espinoza et al . , 2018 ) . Recent evidence indicates that inhibition by individual PV+ fast spiking hilar border interneurons is non-uniformly distributed over space , with decreasing connectivity and inhibition at greater distances from the interneuron ( Espinoza et al . , 2018; Strüber et al . , 2015 ) . To test whether feedback inhibition by the entire ensemble of feedback inhibitory interneurons also displays a spatial gradient , we activated cell populations at 100 µm intervals along the GC layer while recording inhibition in individual GCs ( Figure 3A ) . Spatial profiles were recorded for increasing laser powers in cells in the superior as well as inferior blade of the DG ( Figure 3B , C respectively; n = 8 cells for each blade ) . IPSC amplitudes across locations and powers were normalized to the maximal IPSC amplitude in each respective cell . This maximal amplitude did not differ between cells in different blades ( 366 ± 40 vs 390 ± 84 pA for superior and inferior blades , respectively; t-test , P ( df = 14 , t = 0 . 258 ) =0 . 0686 ) . Next , we investigated the spatial organization of feedback inhibition at stimulation powers at which inhibition had saturated ( Figure 3D , E ) . In all GCs tested , the inhibition was greatest when stimulating in the direct vicinity of the recorded cell . Activating cells at increasing distances led to monotonically decreasing IPSC amplitudes for both blades . Importantly , the term distance here refers to the functional distance along the GC layer and not to Euclidean distance . However , inhibition was observed even at the most remote stimulation sites , indicating that even the most remote cells from the contralateral blade can contribute to the activation of feedback inhibition in a given GC . In order to statistically compare the relation of local versus remote inhibition between blades , we defined a remote location in the contralateral blade at 800 µm from the recorded cell ( measured along the GC layer and equidistant in all slices; Figure 3D , E; grey lines ) and compared it to the local IPSC ( black lines ) . Remote inhibition was significantly smaller than local inhibition while no difference between blades or significant interaction was observed ( Figure 3F; two-way RM ANOVA; Distance: F ( 1 , 14 ) =3 . 341 , p<0 . 001; Blade: F ( 1 , 14 ) =2 . 615 , p=0 . 128; Interaction: F ( 1 , 14 ) =3 . 341 , p=0 . 089 ) . Posttests suggested inhibition of inferior GCs by superior activation might be greater than vice versa . However , the difference was not significant ( Sidak’s multiple comparison corrected posttest , P ( df = 28 ) =0 . 932 , P ( df = 28 ) =0 . 051 for local and remote , respectively ) . Next , we investigated whether there are differences in the steepness of recruitment of local versus remote inhibition between blades ( black and grey , respectively; Figure 3G , H ) . To this end , we calculated the active cell fraction which produces half-maximal inhibition during local or remote stimulation for each individual slice . Comparison of the recruitment between the four groups revealed no differences between blades ( Figure 3I , two-way RM ANOVA; Distance: F ( 1 , 14 ) =7 . 889 , p=0 . 014; Blade: F ( 1 , 14 ) =0 . 5506 , p=0 . 470; Interaction: F ( 1 , 14 ) =0 . 0976 , p=0 . 759 ) . However , local inhibition was significantly more steeply recruited than remote inhibition ( 1 . 99 ± 0 . 22% vs . 3 . 17 ± 0 . 57% active cells for half-maximal inhibition ) . Next , we tested if IPSCs elicited by increasing active GC populations differed between local and remote activation with respect to their kinetic properties . Since all previous data showed no indication of blade specific differences the analysis of the kinetics of feedback IPSCs were performed on the pooled data for both blades . Interestingly , local and remote inhibition differed in all tested respects ( Figure 3J–M , two-way RM ANOVAs with dfDistance = 1 , 183 , dfcell fraction = 6 , 183 and dfinteraction = 6 , 183 ) . Local IPSCs occurred with shorter latency and lower jitter than remote IPSCs ( Figure 3J , K; Latency: p<0 . 001 , <0 . 001 and =0 . 031 for distance , cell fraction and interaction , respectively; Jitter: p<0 . 001 , =0 . 037 and =0 . 707 for distance , cell fraction and interaction , respectively ) . Furthermore , both latency and jitter decreased as larger populations were activated . IPSCs were also significantly slower in remote versus local inhibition . IPSC rise time was slightly shorter in the larger local IPSCs but did not correlate with the active cell fraction ( Figure 3L: p=0 . 010 , =0 . 633 and =0 . 388 for distance , cell fraction and interaction , respectively ) . Similarly , decay times were significantly shorter in local versus remote inhibition while they progressively increased with increasing stimulation power ( Figure 3M; p<0 . 001 , <0 . 001 and =0 . 124 for distance , cell fraction and interaction , respectively ) . These data demonstrate that remote inhibition shows greater delay , greater jitter and slower kinetics than local inhibition . Different connections within the feedback inhibitory microcircuit have been shown to variably facilitate or depress during trains of activity ( Savanthrapadian et al . , 2014 ) ( Tabular overview provided in supplementary file 1 ) . This makes it difficult to predict the net effect on the short-term dynamics of GC feedback inhibition . We therefore characterized the frequency-dependence of net feedback inhibition using antidromic electrical stimulation as described above ( Figure 4A–C ) . In marked contrast to the CA1 region of the hippocampus ( Pothmann et al . , 2014 ) , feedback IPSCs showed strong frequency-dependent facilitation ( Figure 4C , n = 10 cells , one-way RM ANOVA; Frequency: F ( 2 . 69 , 29 . 54 ) =13 . 99 , p<0 . 001; Wilcoxon signed rank tests for deviation from unity at each frequency with Bonferroni corrected p-values; p>0 . 99 , p=0 . 004 , p=0 . 002 and p=0 . 002 for 1 , 10 , 30 and 50 Hz , respectively ) . Furthermore , the facilitation indices significantly increased with increasing stimulation frequency ( 1 Hz: 0 . 99 ± 0 . 07; 10 Hz: 1 . 41 ± 0 . 11; 30 Hz: 1 . 83 ± 0 . 16; 50 Hz: 2 . 09 ± 0 . 19; posttest for linear trend: p<0 . 0001 , R²=0 . 436 ) . We found no evidence for a spatial gradient of net feedback inhibitory short-term dynamics ( Figure 4—figure supplement 1 ) . Because this unusual degree of facilitation may be important in allowing sparse activity of GCs to recruit significant inhibition over time , we further examined the underlying circuit mechanisms . Interestingly , dentate interneuron inputs to GCs appear to be generally depressing ( Supplementary file 1 , blue rows ) , rendering our finding of pronounced facilitation at the circuit level even more striking . We reasoned that a facilitating excitatory synapse driving feedback interneurons could underlie circuit level facilitation . We therefore measured feedback excitation of hilar neurons by stimulating mossy fiber axons as described above ( Figure 4D–L ) . Mossy cells and interneurons were classified according to their morpho-functional properties ( Larimer and Strowbridge , 2008 ) ( Figure 4D , E , G , H , J , K ) . Cell classification was confirmed using unbiased k-means clustering ( Figure 4K ) . We found that feedback excitation of hilar cells displayed marked facilitation , which was similar for both INs and MCs ( Figure 4F , I , L; n = 9 , 12 respectively , two-way RM-ANOVA , Frequency: F ( 3 , 57 ) =6 . 642 , p<0 . 001; Cell type: F ( 1 , 19 ) =0 . 0075 , p=0 . 932; Interaction: F ( 3 , 57 ) =0 . 743 , p=0 . 531 ) . Facilitation indices of hilar cells significantly deviated from one for all frequencies tested ( Figure 4E , F; n = 23 cells; Wilcoxon signed rank tests with Bonferroni corrected p-values; p<0 . 001 for all frequencies ) . These data demonstrate a pronounced frequency-dependent net facilitation of the feedback inhibitory microcircuit , which is supported by strongly facilitating mossy fiber inputs to hilar cells . Together , these data indicate that the dentate feedback circuit is able to deliver strong , spatially graded inhibition with a high gain and the ability for temporal integration . To probe how these quantitative properties of the feedback circuit affect the pattern separation capability of the DG , we incorporated them into a biophysically realistic model of the lamellar microcircuit ( Figure 5 ) based on Santhakumar et al . ( 2005 ) ; Yim et al . ( 2015 ) , making use of their carefully experimentally constrained DG cell-types ( Figure 5A; Figure 5—figure supplement 1A ) . To maximize our models inferential value we clearly separated a tuning phase , in which we constrained the model by our experimental data , and an experimental phase , in which pattern separation was tested without further changes to the model . In the tuning phase , we first scaled up the model four-fold to contain 400 perforant path afferents ( PPs ) , 2000 GCs , 24 basket cells ( BCs ) , 24 hilar perforant path associated cells ( HC ) and 60 MCs ( Figure 5A , B ) . BCs , HCs and MCs comprise the feedback inhibitory circuit and BCs receive direct PP input thereby additionally mediating feedforward inhibition ( Ewell and Jones , 2010 ) . We then adapted the spatial extent of the target pools of BC and HC outputs to produce local and global inhibition , respectively , reproducing the experimentally determined spatial tuning of net feedback inhibition ( Figure 5C ) . We further adjusted synaptic decay time constants and weights in order to reproduce the measured PSCs of hilar neurons and GCs and the empirical recruitment curves ( Figure 5D , Figure 5—figure supplement 1 ) . Finally , we incorporated facilitation of the experimentally determined magnitude into feedback excitatory mossy-fiber outputs , leading to GC IPSC facilitation in the experimentally observed range ( Figure 5E , Figure 5—figure supplement 1B ) . Together , these minimal adaptations resulted in a model with remarkably similar properties to our experimental findings ( Figure 5C–E ) . We therefore concluded the tuning phase of the model and proceeded to an in silico pattern separation experiment without further changes to the model . To investigate the implications for pattern separation , we probed the ability of this model to separate PP input patterns with behaviorally relevant temporal structure and varying degrees of overlap ( Myers and Scharfman , 2009; Yim et al . , 2015 ) . Specifically , we created input trains with constant mean rate , but with either theta ( 10 Hz ) or slow-gamma ( 30 Hz ) modulation ( Figure 5—figure supplement 1C ) , which are prominent during exploration and novelty exposure , respectively ( Sasaki et al . , 2018; Trimper et al . , 2017 ) . To model rapid pattern separation in a behaviorally relevant timescale we chose an input duration of approximately five theta cycles ( 600 ms , corresponding to the approximate duration of place cell spiking during traversal of its place field ) . To obtain a range of input similarities , we generated input patterns in which 24 of 400 PP afferents were activated ( Figure 6A ) and compared pairs of such patterns ranging from no overlap ( two separate sets of afferents ) to complete overlap ( identical trains in the same 24 afferents in both patterns ) . Each model network was run with 25 input patterns leading to a total of 325 comparisons ( data points in Figure 6C ) . To quantify pattern separation we compared input correlation ( Rin ) to output correlation ( Rout; Figure 6B ) both measured as Pearson’s R between the population rate vectors over the full 600 ms time window ( Leutgeb et al . , 2007; Wiechert et al . , 2010 ) . Our full , tuned model reliably decreased the population vector correlations for similar patterns ( 0 < Rin < 1 ) thereby demonstrating robust pattern separation over the whole range of input similarities ( Rout <Rin; Figure 6C , left ) . Next , we isolated the contribution of feedback inhibition to pattern separation by rerunning the same input pattern combinations on the network in which mossy fiber outputs to interneurons were removed ( Figure 6C , middle ) . As expected this manipulation decreased interneuron activity and GC sparsity ( Figure 6—figure supplement 1C , D ) leading to impaired pattern separation ( Figure 6D , noFB ) . Note that removing mossy fiber outputs also eliminates BC activity through cooperative activation of summating feedforward and feedback inputs ( Ewell and Jones , 2010 ) . Removal of all inhibitory outputs led to a further decrease in pattern separation , demonstrating the effect of additionally removing feedforward inhibition ( Figure 6C , right ) . As expected , each of these manipulations increased both the fraction of active GCs and the activity per GC ( Figure 6—figure supplement 1C , D ) . In order to quantify the respective pattern separation effects over the full range of input similarity , we computed the bin wise mean Rout ( Figure 6C , Rin bin-width: 0 . 1 , dashed line ) and measured the area to the identity line ( Figure 6C , black lines ) . The resulting mean ΔRout was calculated for seven separate random networks , each challenged with theta as well as slow-gamma modulated inputs in each of the three conditions . Both the frequency of the input modulation as well as network manipulations significantly affected pattern separation ( Figure 6D; two-way RM ANOVA with both factors matching , condition: F ( 2 , 12 ) =145 . 1 , p<0 . 001; frequency: F ( 1 , 6 ) =31 . 48 , p=0 . 001; interaction: F ( 2 , 12 ) =11 . 77 , p=0 . 002; n = 7 random network seeds for these and all subsequent analyses ) . Specifically , both feedback and feedforward inhibition significantly contributed to pattern separation ( Sidak’s multiple comparison posttest , P ( df = 12 , t = 11 . 33 ) <0 . 001 and P ( df = 12 , t = 5 . 36 ) <0 . 001 , respectively ) . These results are consistent with the standard account , by which any source of inhibition supports pattern separation by decreasing GC activity ( Figure 6—figure supplement 1C , D ) . Notably , the effect of inhibition on GC sparseness was more pronounced during gamma than theta modulated activity , translating to improved pattern separation in the sparser gamma regime ( Figure 6D , Figure 6—figure supplement 1C , D ) . Remarkably , this increased sparsity in the gamma domain was achieved despite the same excitatory drive from perforant path ( Figure 6—figure supplement 1A , B ) , and with less interneuron activity ( Figure 6—figure supplement 1C , D ) . Next , we more closely investigated the isolated pattern separation effects of feedback and feedforward inhibition . To this end , we computed the difference in Rout between the respective conditions for each individual comparison ( i . e . data point in Figure 6C ) . For instance , the individual comparison shown in Figure 6A , will lead to a single Rout value in the network with MF inputs to interneurons ( full model ) , which is subtracted from the corresponding Rout value in the same network without this input ( no FB ) . This procedure isolates the effect of interest ( ΔRout ) for each individual comparison , controlling for other sources of variability . A single pattern separation measure was then obtained as before , as the area under the curve of bin-wise means of these ΔRout values ( Figure 6E , bottom ) . We found a significant effect of both inhibitory motif and frequency domain ( Figure 6E; two-way RM ANOVA with both factors matching , Motif: F ( 1 , 6 ) =15 . 58 , p=0 . 008; Frequency: F ( 1 , 6 ) =9 . 91 , p=0 . 020; Interaction: F ( 1 , 6 ) =76 . 37 , p<0 . 001 ) . Posttests revealed that the frequency dependence of pattern separation was driven by feedback inhibition ( Sidak’s multiple comparison posttest: FB: P ( df = 6 , t = 13 . 68 ) <0 . 001; FF: P ( df = 6 , t = 1 . 33 ) =0 . 412 . Interestingly , this frequency dependence of feedback inhibition mediated pattern separation was particularly pronounced for highly similar input patterns ( 0 . 9 < Rin < 1; Figure 6E , right; Motif: F ( 1 , 6 ) =261 . 7 , p<0 . 001; Frequency: F ( 1 , 6 ) =108 . 1 , p<0 . 001; Interaction: F ( 1 , 6 ) =109 . 5 , p<0 . 001; Sidak’s multiple comparison posttest: FB: P ( df = 6 , t = 15 . 78 ) <0 . 001; FF: P ( df = 6 , t = 0 . 98 ) =0 . 595 ) . Indeed , feedback inhibitory pattern separation for highly similar input at 30 Hz compared to 10 Hz was more than doubled ( from 0 . 04 ± 0 . 01 to 0 . 09 ± 0 . 01 , mean ± SD , Cohen’s d = 4 . 1 , Figure 6E , right ) . This again demonstrates feedback inhibitory pattern separation effects beyond those explainable by decreases in GC activity , since comparisons for highly similar inputs are computed on the exact same model runs as comparisons for less similar inputs and thus by definition have the same GC activity levels ( also see Figure 6F , G , arrows ) . It has recently been emphasized , that the assessment of pattern separation can depend critically on the similarity measure used ( Madar et al . , 2019; Wick et al . , 2010 ) . Therefore , we tested the robustness of this result for two alternative similarity measures , namely normalized dot product ( NDP , also known as cosine similarity ) and pattern overlap ( # of coactive/ # of totally active cells; Figure 6—figure supplement 2 ) . The frequency dependence of feedback inhibition-mediated pattern separation , especially for highly similar inputs , proved robust for all three similarity measures . Next , we investigated the specific effects of two interesting empirical findings of the present study , 1 ) the spatial tuning and 2 ) the facilitation of the feedback circuit ( Figure 6—figure supplement 3 ) . To this end , we undertook two targeted , minimal manipulations of the full tuned network . To probe the effect of spatially graded inhibition , we redistributed BC output synapses to a global target pool ( the whole GC population ) , leading to spatially uniform inhibition ( global FB; Figure 6—figure supplement 3B , E ) . To probe the effect of facilitation , we removed facilitation from mossy fiber outputs ( Figure 6—figure supplement 3C , E ) . We isolated the effects of these manipulations by pairwise comparison to the corresponding full tuned networks as described above ( Figure 6—figure supplement 3F–I ) . The results showed a small but significant contribution of facilitation ( ~20% of the isolated FB effect for both frequency paradigms ) , but not spatial tuning to pattern separation ( Figure 6—figure supplement 3G , left; Wilcoxon signed rank test for deviation from 0 , n = 7 , Bonferroni corrected p-values: p=0 . 031 and p=1 respectively for 10 Hz; p=0 . 031 and p=1 respectively for 30 Hz ) . We noted that while spatial tuning did not affect mean pattern separation , it appeared to reduce its variability ( CoV ) for a given input similarity , although the effect was again small ( Figure 6—figure supplement 3G , right; Wilcoxon signed rank test for deviation from 0 , n = 7 , Bonferroni corrected p-values: p=0 . 031 and p=0 . 750 for tuning and facilitation respectively at 10 Hz; p=0 . 438 and p>0 . 999 respectively at 30 Hz ) . So far , all pattern separation analyses were conducted on the population rate vectors during a 600 ms time window . However , many neural computations are likely to occur on shorter timescales , such as within individual theta ( ~100 ms ) and gamma ( ~10–33 ms ) cycles ( Buzsáki , 2010; van Dijk and Fenton , 2018 ) . Indeed , the time window in which correlation is recorded can nontrivially affect the resulting correlation , depending on the timing of spikes within it ( Madar et al . , 2019 ) . We therefore first computed the networks pattern separation ability within 100 ms time windows , revealing i ) the pattern separation ability within such short timescales and ii ) the temporal evolution of pattern separation throughout a 600 ms stimulus presentation ( Figure 6F , G ) . We find that pattern separation occurs even within a single theta cycle , including a contribution of feedback inhibition in both frequency paradigms ( mean ΔRout > 0 within the first 100 ms bin , Wilcoxon signed rank test with Bonferroni corrected p-values: p=0 . 031 , =0 . 031 for full and FB effect respectively in both paradigms ) . While mean ΔRout did not differ between frequency paradigms within this first time window , it was significantly elevated in the 30 Hz paradigm in all subsequent time windows ( full model effect , two-way RM ANOVA , p<0 . 001 , <0 . 001 and=0 . 004 for time-bin , frequency and interaction respectively , Sidak’s posttest p=0 . 234 for 1st bin and p<0 . 001 for all subsequent bins ) . Again , the selective increase during slow-gamma modulated inputs was driven by feedback inhibition ( isolated FB effect , two-way RM ANOVA , p=0 . 007 , <0 . 001 and=0 . 041 for time-bin , frequency and interaction respectively , Sidak’s posttest p=0 . 708 for 1st bin and p<0 . 002 for all subsequent bins ) , including a contribution from MF facilitation ( Figure 6—figure supplement 3 ) . As above , the effect was predominantly driven by the separation of highly similar input patterns ( isolated FB effect , Rin >0 . 5; two-way RM ANOVA on last time-bin , p<0 . 001 , =0 . 010 and<0 . 001 for Rin-bin , frequency and interaction respectively , Sidak’s posttest on differences between frequency paradigms for each input similarity: p=1 for Rin <0 . 6 and p=0 . 032 to p<0 . 001 for Rin = 0 . 6 to 0 . 9 ) . These results were robust when analysis time windows were decreased even further ( to the duration of a slow gamma cycle , 33 ms , Figure 6—figure supplement 4 ) . This 33 ms resolved analysis additionally reveals that the pattern separation effect , particularly of feedback inhibition , ramps up within a 100 ms window , becoming effective only at the end of a theta cycle ( Figure 6—figure supplement 4A ) . Next , we asked if the frequency dependence of feedback inhibitory pattern separation was sensitive to variations of the inhibitory decay time constants and if there might be an interaction between these decay time constants and the frequency range at which pattern separation is most effective ( Figure 6—figure supplement 5 ) . Remarkably , we found the differential effect between 10 and 30 Hz to be highly robust across a range of different decay time-constants ( 0 . 5x to 5x of the experimentally matched decay , Figure 6—figure supplement 5A–C , Supplementary file 2 ) . Furthermore , the selective enhancement of feedback inhibitory pattern separation of highly similar inputs was robust over the entire gamma range ( up to 100 Hz , Figure 6—figure supplement 5D , E ) . Next , we tested if our main results were robust to alterations in the relative strengths of feedforward vs . feedback inhibition . Since , our model is closely constrained with respect to the recruitment and functional properties of the feedback circuit , we are confident about the resulting computational inferences concerning this circuit . However , the model does not allow strong inferences about the relative roles of feedback and feedforward inhibition , and it is thus necessary to probe if extremely powerful feedforward inhibition might occlude the effects described here . We therefore selectively enhanced the feedforward inhibitory circuit in our model by increasing the PP to BC circuit 2x ( Figure 6—figure supplement 6 ) . This robustly increased the feedforward inhibitory contribution to pattern separation above that of feedback inhibition ( Figure 6—figure supplement 6B ) . However , it did not affect the frequency dependence of the feedback inhibitory effect . Indeed , for highly similar input patterns , the feedback inhibitory effect was so prominently enhanced during gamma input , as to again dominate the feedforward inhibitory effect ( Figure 6—figure supplement 6C ) . Finally , we probed the robustness of our findings for various perforant path input strengths ( Figure 6—figure supplement 7 ) . We found that frequency-dependent pattern separation by the feedback circuit occurred over a large range of PP-input strengths and resulting mean sparsities of the GC population ( Figure 6—figure supplement 7B–D ) . These data additionally suggest that for highly similar input patterns , the more efficient sparsification of the GC population at 30 Hz did not fully account for the gains in pattern separation ( Figure 6—figure supplement 7F ) . Specifically , selecting a PP-input strength at 10 Hz that produced the same sparsity as during 30 Hz did not allow to reach similar pattern separation ( Figure 6—figure supplement 7F ) . This result suggests that the feedback circuit mediates direct assembly competition , allowing pattern separation beyond a pure sparsification effect . Together these results suggest that frequency dependence is a key feature of the feedback inhibitory microcircuits and predict that feedback inhibition selectively boosts the separation of highly similar input patterns during gamma oscillations . We have therefore quantitatively described the recruitment of net feedback inhibition by defined GC population sizes in space and time in the hippocampal DG , a structure in which sparse activity and inhibition are thought to critically contribute to the function of pattern separation ( Gilbert et al . , 2001; Hunsaker et al . , 2008; Leal and Yassa , 2018; McHugh et al . , 2007; Stefanelli et al . , 2016 ) . The proposed role of the feedback inhibitory circuit depends critically on its dynamic range , that is the relation between the number of active principal cells and the resulting feedback inhibition . This property of the feedback circuit is determined by complex , mainly hilar cellular connectivity patterns including the synaptic and intrinsic properties of all participating cells ( see e . g . Espinoza et al . , 2018; Savanthrapadian et al . , 2014 ) , tabular overview in Supplementary file 1 ) . While delving into detailed cell-cell connectivities is clearly important , such studies do not allow the quantitative determination of the gain and dynamic range of net feedback inhibition ( Kapfer et al . , 2007; Silberberg and Markram , 2007 ) . Using two complementary experimental approaches , we found that net feedback inhibition is steeply recruited by sparse populations of GCs ( <4% ) . This is in good agreement with the sparse range of GC activity reported in vivo ( Diamantaki et al . , 2016; Hainmueller and Bartos , 2018; Pernía-Andrade and Jonas , 2014; Pilz et al . , 2016; Schmidt et al . , 2012 ) . In these studies , different time windows were used to define active vs . non-active granule cell populations ( one to tens of minutes for electrophysiological , imaging or immediate-early gene studies ) . The relevant window for assembly competition , however , is much shorter . If we assume random Poisson firing , the electrophysiologically determined rates by Pernía-Andrade and Jonas ( 2014 ) and Diamantaki et al . ( 2016 ) , suggest active GCs fractions of <2% for realistic assembly competition time windows of <100 ms . Accordingly , the gain and sensitivity of the circuit are well suited to strongly modulate feedback inhibition within the range of GC activity reported in vivo . In addition to steep recruitment , we have described the temporal and spatial distribution of net inhibition delivered by feedback circuits in the DG . How do these properties influence the pattern separation capability of the dentate gyrus ? To address this question , we adapted an established biophysically realistic computational model of the DG circuitry ( Santhakumar et al . , 2005; Yim et al . , 2015 ) . We first carefully constrained the model to match the spatial and temporal properties of net feedback inhibition as assessed in our physiological data . We then fixed all model parameters , and proceeded to probe the ability of this circuit to perform pattern separation on temporally complex oscillatory inputs . The major , highly robust , result of this computational study was that the impact of feedback inhibition on pattern separation is frequency-dependent . Specifically , we find that the separation of input patterns during gamma oscillations > 30 Hz is powerfully and selectively enhanced by the feedback circuit . Remarkably , this mechanism involved decreased interneuron activity and was particularly efficient for very similar input patterns . Such an effect has not been discovered in earlier modeling studies , because most models have discretized time , calculating the pre-inhibition population activity , the resulting inhibition , and the inhibition-corrected population activity in a single time step , sometimes assuming an average corrected population rate within this time step ( Myers and Scharfman , 2009; Rolls and Treves , 1998; Trappenberg , 2010 ) . Thus , they do not capture temporal features of feedback circuits . On the other hand , a number of spike based , temporally resolved models have considered only temporally unstructured ( Poisson ) inputs ( Chavlis et al . , 2017; Hendrickson et al . , 2015; Hummos et al . , 2014; Yim et al . , 2015 ) . We suggest that the precise spatiotemporal organization of the feedback circuit , together with the temporal structure of DG inputs is a crucial determinant of pattern separation . Indeed , the DG and its inputs have a strong , behaviorally relevant , temporal structure ( Lasztóczi and Klausberger , 2017; Mizuseki et al . , 2009; Pernía-Andrade and Jonas , 2014; Skaggs et al . , 1996 ) . Novelty experience can induce increased gamma and beta range activity ( Berke et al . , 2008; Rangel et al . , 2015; Trimper et al . , 2017 ) , and explorative activity with rearing is also associated with increased gamma oscillations ( Barth et al . , 2018 ) . A previous model has addressed how fast , rhythmic gamma-frequency feedback inhibition may implement a type of ‘k-winners-take-all’ operation , a basic computational component of pattern separation models ( de Almeida et al . , 2009 ) , although this model relies on faster synaptic timescales than we observed in our compound IPSCs . Perhaps most interestingly , the occurrence of oscillations in the slow-gamma range has recently been reported to be causally related to associative memory formation ( Sasaki et al . , 2018; Trimper et al . , 2017 ) , a process thought to require pattern separation . Consistent with this finding , Hsiao et al . ( 2016 ) report DG driven gamma entrainment of CA3 , the presumed primary storage location of associative memories . Together , this suggests that the dentate pattern separator may be optimized to rapidly detect subtle degrees of difference within the environment in gamma-dominated exploratory brain states , a capability likely to support successful memory encoding of novel environmental features , and potentially aiding in rapid discrimination during recall . Importantly , the frequency-dependency of pattern separation was driven by the feedback circuit . This effect was highly robust when varying the decay time constants of the inhibitory synaptic conductances , the time windows of analysis , the similarity measure , or the PP input strength . By contrast , feedforward inhibition and anatomical pattern separation was robustly independent of frequency modulation . Together this suggests that frequency-dependent pattern separation is a key property of the local inhibitory feedback circuit . Importantly , this does not preclude that additional , long range projections may add further complexity ( Szabo et al . , 2017 ) . Also note that in addition to the instantaneous pattern separation mechanisms investigated here , potentially complementary mechanisms at much longer time scales have been proposed involving ongoing neurogenesis ( Aimone et al . , 2011; Clelland et al . , 2009; Li et al . , 2017; Sahay et al . , 2011; Severa et al . , 2017; Temprana et al . , 2015 ) . The model also allowed us to examine the impact of the spatiotemporal organization of inhibition on pattern separation . Facilitation of feedback circuits produced a small but robust enhancement of pattern separation , while spatial tuning of feedback inhibition did not . The facilitation of feedback inhibition is a remarkable feature of the DG , which we to our knowledge have described for the first time . It is in marked contrast to area CA1 , where somatically measured feedback inhibition shows strong depression ( Pothmann et al . , 2014; Pouille and Scanziani , 2004 ) and is particularly surprising given the prevalence of depression in the literature on pairwise connections ( Supplementary file 1 ) . Our experimental and modeling data suggest that the strong facilitation of the mossy fiber input to the feedback circuit is the principal mediator of this net facilitation . Physiologically , facilitation may aid sparse GC spiking to efficiently recruit inhibition , particularly during burst-like activity ( Pernía-Andrade and Jonas , 2014 ) . In our model , spatial tuning of feedback inhibition had no effects on pattern separation . This may derive from the fact that PP inputs were spatially broad and random , as suggested by anatomical studies ( Tamamaki , 1997; Tamamaki and Nojyo , 1993 ) . In general , the effect of localized inhibition could be more relevant if synchronously activated populations of GCs are locally clustered ( Feldt Muldoon et al . , 2013 ) . For instance , GCs in the inferior and superior blades of the DG are known to be differentially active ( Alme et al . , 2010; Chawla et al . , 2005 ) . Accordingly , localized inhibition might be important for independent processing between the two blades . An alternative function of spatially graded inhibition has been proposed by Strüber et al . ( 2015 ) , who suggest that it is more effective in promoting synchronous gamma oscillations . Accordingly , spatial tuning may play a role in creating the oscillatory dynamics , found here to critically impact the feedback inhibitory pattern separation performance . In conclusion , this study provides the first comprehensive , quantitative description of the spatiotemporal properties of the DG feedback inhibitory microcircuit , and predicts that these properties will selectively enhance the separation of highly similar input patterns during learning-related gamma oscillations . This mechanism may be relevant for understanding disease states in which there is a coincidence of dentate gyrus-centered pathology with abnormal oscillatory activity , and memory and pattern separation deficits such as temporal lobe epilepsy , Alzheimer’s disease or schizophrenia ( Andrews-Zwilling et al . , 2012; Gillespie et al . , 2016; Leal and Yassa , 2018; Verret et al . , 2012 ) . All experimental procedures were conducted in accordance to federal law of the state of North Rhine-Westphalia ( Aktenzeichen 84–02 . 04 . 2014 . A254 ) , minimizing unnecessary pain and discomfort . Experiments were performed on horizontal hippocampal slices of 21- to 97-day-old mice . Ca2+ imaging and a subset of dual recording experiments were performed in C57/Bl6 mice obtained from Charles River Laboratories ( Wilmington , MA ) . Optogenetic experiments and the remaining dual recording experiments were performed on double transgenic offspring of Tg ( Prox1-cre ) SJ39Gsat/Mmucd , MMRRC Cat# 036632-UCD , RRID: MMRRC_036632-UCD ) obtained as cryopreserved sperm and rederived in the local facility ( Gong et al . , 2007; Gong et al . , 2003 ) and Ai32-mice ( B6;129S-Gt ( ROSA ) 26Sortm32 ( CAG-COP4H134R/EYFP ) Hze/J , IMSR Cat# JAX:012569 , RRID: IMSR_JAX:012569 ) . For preparation the animals were deeply anesthetized with Isoflurane ( Abbott Laboratories , Abbot Park , USA ) and decapitated . The head was instantaneously submerged in ice-cold carbogen saturated artificial cerebrospinal fluid ( containing in mM: NaCl , 60; sucrose , 100; KCl , 2 . 5; NaH2PO4 , 1 . 25; NaHCO3 , 26; CaCl2 , 1; MgCl2 , 5; glucose , 20 ) and the brain removed . Horizontal 350 µm thick sections were cut with a vibratome ( VT1200 S , Leica , Wetzlar , Germany , 300 µm sections for hilar recordings ) . To obtain maximum-connectivity-plane slices the brain was glued to its dorsal surface ( compare Bischofberger et al . , 2006 ) . The slicing depth at which the temporal pole of the hippocampus first became visible was noted ( depth = 0 µm ) . From here the first four sections were discarded ( up to a depth of 1400 µm ) . The following two to three sections were secured such that one further section before the beginning of the dorsal hippocampus ( approximately 2400 µm ) could be discarded . Slices were incubated at 35°C for 20 to 40 min and then stored in normal ACSF ( containing in mM: NaCl , 125; KCl , 3 . 5; NaH2PO4 , 1 . 25; NaHCO3 , 26; CaCl2 , 2 . 0; MgCl2 , 2 . 0; glucose , 15 ) at room temperature . Recordings were performed in a submerged recording chamber at 33–35°C under constant superfusion with carbogen saturated ACSF ( 3 ml/min ) . Experiments were performed in the superior blade unless otherwise indicated . Hippocampal dentate GCs were visually identified using infrared oblique illumination contrast microscopy in a 20x or 60x water immersion objective ( Olympus , XLumPlanFl , NA0 . 95W or Nikon , N60X-NIR Apo , NA1 . 0W ) on an upright microscope ( TriMScope , LaVision Biotech , Bielefeld , Germany or Nikon Eclipse FN1 , Tokyo , Japan ) . For IPSC measurements the whole-cell patch-clamp configuration was established with a low chloride cesium-methane-sulfonate based intracellular solution ( intracellular solution containing in mM: CH3O3SCs , 140; 4- ( 2-hydroxyethyl ) −1-piperazineethanesulfonic acid ( HEPES-acid ) , 5; ethylene glycol tetraacetic acid ( EGTA ) , 0 . 16; MgCl2 , 0 . 5; sodium phosphocreatine , 5; glucose , 10 ) . For GC current clamp experiments a low-chloride solution ( CC-intracellular solution containing in mM: K-gluconate , 140; 4- ( 2-hydroxyethyl ) −1-piperazineethanesulfonic acid ( HEPES-acid ) , 5; ethylene glycol tetraacetic acid ( EGTA ) , 0 . 16; MgCl2 , 0 . 5; sodium phosphocreatine , 5 ) was used . GCs with input resistances greater than 300 MΩ were discarded in order to exclude immature GCs ( Schmidt-Hieber et al . , 2004 ) . Hilar cells were recorded with intracellular solution containing in mM: K-gluconate , 140; KCL , 5; HEPES-acid , 10; EGTA , 0 . 16; Mg-ATP , 2; Na2-ATP , 2; pH adjusted to 7 . 25; 277 mmol/kg without biocytin . 0 . 3% biocytin ( Sigma-Aldrich , B4261 ) . In all imaging experiments and a subset of optogenetic experiments , the intracellular solution additionally contained 100 µM Alexa 594 hydrazide sodium salt ( Life Technologies , Carlsbad , USA ) . The identity of visually and electrophysiologically identified mature GC was confirmed by their dendritic morphology after dye filling in every case tested . Pipette resistance of the patch pipettes was 3–7 MΩ . Voltage-clamp recordings were performed with a Multiclamp 700B ( Molecular Devices , Sunnyvale ) or a BVC-700A amplifier ( Dagan Corporation , Minneapolis ) . Current-clamp recordings were performed with a Multiclamp 700B . Voltage or current signals were digitized with a Digidata 1322A ( Molecular Devices ) or ( Instrutech ITC-16 , Heka Electronics , Ludwigshafen , Germany ) at 10 or 50 kHz and recorded using Clampex 10 . 2 ( Molecular Devices ) or Igor Pro 6 ( Wavemetrics , Lake Oswego ) on a PC running Windows XP . All electrophysiological recordings were obtained at least in triplicate , then averaged and counted as a single biological replicate . For IPSC measurements , cells were held at 0 mV including liquid-junction potential correction ( estimated at 16 mV ) . To aid the voltage clamp throughout the cell , this depolarized membrane potential was slowly approached during a 15 min pre-equilibration period , during which Cs+ entered the cell . For CC-recordings liquid junction potential was not corrected . IPSCs were normalized to the maximally elicited IPSC over space and power for each respective cell . Importantly , this normalization does not require prespecification of the location or power at which a respective cell’s maximum occurs . Note , that due to this procedure all normalized IPSC values are by definition below 100% . Chemicals for electrophysiological experiments were obtained from Sigma-Aldrich ( St . Louis ) . All drugs were purchased from Tocris Bioscience ( Bristol , UK ) . Two GCs within 100 µm of each other were recorded . To test for single GC-induced feedback inhibition 10 to 15 trains of 10 APs at 100 Hz were elicited by brief ( 3 ms ) current injections in one cell . Inhibition was monitored either in VC , while holding the cell at 0 mV to allow the detection of small IPSCs ( Figure 2—figure supplement 3 , n = 7 cell pairs , seven directions ) or current clamp while holding the cell at −60 mV , allowing to probe for inhibition in both directions ( not shown , n = 4 cell pairs , eight directions ) . Dye loading was modified from Garaschuk et al . ( 2006 ) and performed in the submerged chamber at 35°C under constant superfusion . Briefly , a dye solution containing: 1 mM Oregon Green 488 BAPTA-1 acetoxy-methyl ester ( OGB-1 AM ) ; 2% pluronic F-127; 150 mM; 2 . 5 mM KCl; 10 mM HEPES ) . The dye was injected into the slice along the superior blade of the GC layer using standard patch pipettes ( 4–5 locations , 100 µm intervals , 30 µm depth , 3 min at 500 mbar per location ) . Recordings were started at least 45 min after the staining procedure . Population Ca2+ Imaging was performed using a multibeam two-photon fluorescence microscope ( TriMScope , LaVision Biotech , Bielefeld , Germany ) with excitation light at 810 nm . Images were acquired with a digital CMOS camera ( ORCA-Flash , Hamamatsu ) through a high numerical aperture 20x water immersion Objective ( XLumPlanFl , NA-0 . 95 , Olympus ) . This allowed imaging of a large field of view ( 320 × 240 µm ) with high spatial and temporal resolution ( 1920 × 1440 pixels , 20 Hz ) at acceptable signal to noise ratios . Time series were processed with ImageJ 1 . 48o and IGOR Pro 6 . 3 in a semiautomatic manner . Regions of interest were manually placed onto all well loaded cells which remained visible throughout the experiment . Ca2+ fluorescence increase normalized to baseline ( ΔF/F ) traces of individual cells were calculated without background subtraction . The fraction of responders for each time series was extracted by automatic thresholding at ΔF/F = 0 . 94% . The threshold was determined by combined cell-attached and Ca2+ imaging experiments . Note , that for these experiments the stimulation electrode was placed into the hilus in order to obtain a sufficient number of true positive responders . The imaged cell population comprised on average 46 ± 18 ( standard deviation ) cells ( n = 23 slices ) . The active cell fraction corresponds to the fraction of responders normalized to the dye-loaded population within each section . To assess the spatial distribution of cell activation in imaging experiments , ΔF/F projections were created by averaging and smoothing four frames during the transient and four frames at baseline fluorescence and then calculating the pixel wise ΔF/F . Antidromic electrical stimulation was achieved using a bipolar cluster microelectrode ( FHC , Bowdoin ) connected to a digital stimulus isolator ( AM-systems , Sequim ) , placed into stratum lucidum in the CA3 region . IPSCs at individual powers were elicited 5 to 13 times at 0 . 1 Hz and averaged ( 0 . 1 ms pulse time ) . The amplitude beyond which the stimulus isolator could not pass the full current , determined the maximal stimulation amplitude for each experiment . In order to obtain the input-output relationships of the feedback inhibitory circuit data , each variable was averaged over slices by power . This was necessary since only a small subset of experiments in which inhibition was completely blocked could also be successfully imaged ( 6 of 8 sections ) . Due to the small numbers of active cells within individual slices with sufficient dye loading ( n = 23 slices ) analysis of only these six slices leads to a very piecemeal recruitment curve . A more accurate estimation of the recruitment of feedback inhibition was obtained by averaging the cell activation and inhibition over all appropriate slices and relating them by power , respectively . Note that while the fraction of activated cells in non-MCP sections ( not included in the quantitative analysis ) was mostly zero , IPSCs were almost always present ( in 28 of 29 cells in non-MCP sections ) . Focal optogenetic stimulation was achieved through a galvanometer driven spot illumination device coupled to a 473 nm DPSS Laser ( UGA-40 , DL-473 , Rapp Optoelectronics , Hamburg , Germany ) on an upright microscope ( Nikon Eclipse FN1 , Tokyo , Japan ) . The width of the resulting stimulation spot at the focal plane was 8 . 36 ± 0 . 04 µm ( full width at half max; Nikon 10X Plan Fluor , NA 0 . 3 Laser powers are given in arbitrary units from 1 to 7 corresponding to 15 ± 1 µW , 107 ± 14 µW , 292 ± 42 µW , 762 ± 105 µW , 1433 ± 49 µW , 1729 ± 165 µW and 1660 ± 163 µW at the objective ( n = 5 measurements ) . All illumination spots were placed at approximately 40 µm into the ML at the slice surface . Stimulation pulses were of 20 ms duration . To measure the light intensity distribution throughout a slice the setup was modified to image the slice from below while the laser beam was focused to its surface ( Figure 2—figure supplement 1C–F ) . This was achieved by focusing a surgical Microscope with 36x magnification ( M695 , Leica Microsystems , Wetzlar , Germany ) to the lower slice surface . Images were taken with a CCD camera ( Nikon D60 ) . Acute sections of 100 , 150 , 200 , 250 , 300 and 350 µm thickness were cut from Prox1-ChR-eYPF mice as described above . The laser was focused to the surface of the slice in the molecular layer and an image was taken at every laser power ( p=1 to 7 AU ) . The stage was moved for every image to avoid bleaching or phototoxicity . Linear profiles of the resulting isometric light distribution were measured in several directions and averaged to obtain an x profile per section . The x-profiles of slices of different thickness were then stacked to obtain the xz-profile . Values below 100 µm depth were obtained through fitting a Gaussian function in x-direction at 100 µm depth and an exponential function in z-direction . Complete three-dimensional intensity profiles of three different locations of two slices within the dentate molecular layer were averaged . To assess the active fraction of GCs , approximately two GCs were recorded in cell-attached mode in each slice in which an IPSC was recorded . Illumination spots were placed along the GC layer at 100 µm intervals ( Figure 2—figure supplement 1 ) . The entire profile was probed in triplicate with 1 s intervals between individual locations . When the stimulation spot was in sufficient proximity to the recorded cell clear APs were generally visible ( in 25 of 26 cells ) , and otherwise could be induced through simultaneous cell attached depolarization . Cell-attached spikes were detected by automatic thresholding at 6x standard deviation of the baseline . The spatial profile of firing probabilities , centered on the recorded cells , was averaged within each section . To test if cell activation properties differed between blades the maximum firing probabilities ( at p=7 ) as well as the slopes ( increase in firing probability from p=1 to 7 ) when simply averaging over all location of a given cell were compared by t-test ( n = 7 sections per blade , p=0 . 490 and 0 . 684 for max . AP probability and slope , respectively ) . Since no difference was observed a single firing probability distribution as a function of the distance along the GC layer ( x – distance ) was calculated for each power ( Figure 2—figure supplement 1B , n = 14 sections , seven per blade ) . However , the firing probability of cells in the vicinity of the illumination spot is likely to increase not only as a function of the laser power and spread at the surface , but also of the penetration depth of the light cone . In order to calculate the firing probabilities throughout the slice , the firing probability distribution at the surface was related to the measured light intensity distribution throughout the slice ( Figure 2—figure supplement 1C–F; see above ) utilizing a ‘virtual distance’ measure . Since cells were measured at random distances from the molecular layer border , the light intensity distribution , like the firing probabilities were collapsed to two dimensions , x-distance along the GC layer and z-distance with increasing slice depth . The ‘virtual distance’ was calculated as the mean distance from a given slice-surface pixel to all other pixels of the light intensity distribution weighted by the intensity within those pixels ( Figure 2—figure supplement 1G ) . Assigning the firing probabilities of pixels at the slice surface to their respective virtual distance yields the firing probability distribution as a function of virtual distance , which was well approximated by a gaussian fit ( Figure 2—figure supplement 1H ) . This fit was used to also calculate the firing probabilities of pixels/cells deeper in the slice using the measured light intensity distribution as input . The active cell fraction then corresponds simply to the mean firing probability throughout the slice . This calculation is independent of the size and number of GC and was performed for every power individually . We noted that a large fraction of the recorded spikes occurred with larger latency than the typical IPSC following the beginning of the 20 ms stimulation pulse ( Figure 2—figure supplement 1I , example from a single slice ) . Since only APs preceding the IPSC can participate in its recruitment , we calculated the fraction of total spikes which preceded mean IPSC latency for every power , and fitted the resulting relation with an exponential function ( Figure 2—figure supplement 1J ) . All active cell fractions were corrected by this factor ( Figure 2—figure supplement 1J , bottom ) . Note that this does not take account of the disynaptic delay between mossy fiber output and GC input , thereby potentially slightly overestimating the true recruiting population . For comparison , the active cell fraction was also computed with alternative assumptions about the decay of the firing probability with increasing slice depth . If no firing probability decay with increasing depth is assumed , the active cell fraction throughout the slice is given simply by the average of the measured firing probabilities at the slice surface ( Figure 2—figure supplement 1K , upper grey dashed line ) . Alternatively , the firing probability decay with depth was assumed to be identical to the measured decay along the slice surface ( isometric firing probability distribution; Figure 2—figure supplement 1K , lower grey dashed line ) . In this case , Gaussian functions were fit to the probability distributions at the surface and these Gaussian functions were then assumed to extend also in the z-dimension . The active GC fraction was then calculated by numerical integration under the two dimensional Gaussian ( with the bounds from 0 to 350 µm in z and −888 to 888 µm in x , which corresponds to the mean GC layer length ) normalized to the same area with a uniform firing probability of one . The best estimate of the active GC fraction , incorporating light intensity measurements ( Figure 2—figure supplement 1K , black line ) , was within these upper and lower bound estimates . To globally activate the GC population a multimode light fiber ( BF-22 , Thorlabs , New Jersey ) coupled to a 473 nm laser ( Omicron Phoxx , Rodgau-Dudenhofen , Germany ) was placed above the slice surface , non-specifically illuminating the entire hippocampus . Analogous to focal stimulations , the activated cell fraction was calculated as the firing probability of individual cells following 20 ms pulses . Here , no spatial normalization is necessary since cells were sampled from random locations with respect to the light fiber . Firing probabilities for the focal stimulation in these sections was calculated as the simple average of all stimulation locations . The same stimulation paradigm which was used to assess cell activation was used to assess the spatial distribution of feedback inhibition . For individual cells , IPSCs at each location and power were averaged . The entire profile was normalized to the largest measured IPSC of that cell , independent of the power and stimulation location at which it occurred . For analysis , all IPSC profiles were spatially aligned to the recorded cells . The mean distance to apex ± one standard deviation was 356 ± 163 µm and 322 ± 97 µm for cells from the superior and inferior blade , respectively ( n = 8 cells in each blade ) . In order to test whether there were any distinct effects of the apex , such as a steep decay of inhibition , which would be masked by alignment to the recorded cells , we also aligned the profiles to the apex ( not shown ) . However , no such effects were visible . To analyze the saturated IPSC profiles , normalized IPSC amplitudes from p=5 to 7 were averaged for each cell . In order to analyze the effects of local versus remote stimulation for each blade a distance was chosen such that each remote location was still within the DG but in the other blade ( 800 µm from the recorded cell ) . Normalized IPSCs of the three locations surrounding the recorded cell or this remote location were averaged within each power to obtain the IPSC amplitudes for further analysis . The cell fraction required for the activation of a half-maximal IPSC in each section was assessed for each cell by linear interpolation between the measured values . Since no differences were found between superior and inferior inhibition , recordings of both blades were pooled to analyze the kinetic properties of IPSCs . All parameters were calculated on the multiple trials of individual cells . The latency was measured as the time from the beginning of the pulse to when the IPSC superseded six fold standard deviation of the baseline . The jitter was calculated as the standard deviation of these latencies for individual cells . The rise time was calculated as the mean 20 to 80 rise time of each cell and the decay time constant was obtained from an exponential fit to the decaying phase of the compound IPSC . Intrinsic properties of hilar cells were quantified based on 4 . 6 s long depolarizing current steps or 500 ms hyperpolarizing current steps . AP threshold and fast AHP amplitude were measured from the first AP in the first current step in which an AP occurred within the first 10 ms . Clustering fraction and mean AP time were calculated from the current injection that elicited the maximum average AP frequency . The Clustering fraction represents the fraction of APs that occur within 60 ms before or after another AP ( Larimer and Strowbridge , 2008 ) . Mean AP time was calculated as the mean AP time point normalized to the duration of the current injection ( 4 . 6 s ) . Input resistance was calculated as the slope of the IO curve from the hyperpolarizing current ladder . Cells were manually classified as mossy cells or interneurons based on these intrinsic properties . To objectively confirm classification , we performed unsupervised k-means clustering using scikit-learn ( Pedregosa et al . , 2011 ) . For clustering all six measures were normalized by mean and variance . Two cells with conflicting classification were not included in further analysis . After recording , slices were fixed for 1 hr in 4% PFA and stored overnight in 0 . 25% PBS-T at room temperature . The following day they were transferred to PBS for short term storage or immediately stained . For biocytin staining , sections were washed with PBS and incubated with Streptavidin-Alexa-Fluor-555 Conjugate ( Invitrogen , S32355 ) , 1:1000 in 0 . 25% PBS-T overnight at 4°C . The following day they were co-stained with DAPI 1:1000 in PBS for 30 min and mounted with Aqua-Poly/Mount . Cells were imaged with the Leica SP8 Confocal Microscope of the Microscopy Core Facility at the University Clinic Bonn using a 40x water immersion objective . Short-term dynamics of GCs and hilar cells were assessed using antidromic electrical or optogenetic stimulation at minimal power ( the smallest stimulation power that yielded reliable responses ) . Trains of 10 pulses at 1 , 10 , 30 , 50 Hz were delivered in triplicate and averaged ( excluding sweeps with action currents for hilar cells ) . In all GCs and a subset of hilar cells we confirmed that PSCs could be blocked by at least 90% with 40 µM CNQX + 50 µM D-APV ( n = 12 , 23 for GCs and hilar cells respectively ) . Facilitation indices were obtained by normalizing the average of the last three PSC peaks to the first . To test for differential dynamics between local and remote inhibition analogous trains of optogenetic 20 ms pulses at powers below saturation ( usually p=2 for local inhibition and p=3 for remote inhibition ) were delivered . For each power and frequency , five repeats were recorded and averaged . AP probabilities were assessed by cell-attached recordings with the stimulation site close to the recorded cell . Cell-attached spikes were detected by automatic thresholding as above . A simple multicompartmental passive ‘ball and stick’ model with number of segments following the d_lambda rule ( Carnevale and Hines , 2006 ) and passive properties Ra = 181 Ωcm , Cm = 1 uFcm−2 and a leak conductance = 0 . 0002 Scm−2 , which gave an Rin of 165 MΩ , were adopted from Carnevale and Hines ( 2006 ) and Krueppel et al . ( 2011 ) . A soma ( 20 µm diameter ) contained one dendrite ( 3 µm diameter , 200 µm length ) with an alpha synapse point mechanism ( Erev −90 mV ) placed at 180 µm from the soma . The range of synaptic conductances ( 0 . 1–50 nS; adopted from Williams and Mitchell , 2008 ) elicited IPSC amplitudes in the model , which covered the range of somatic IPSC amplitudes that were experimentally measured ( 3 pA – 1nA ) . Voltage clamp experiments were simulated using a single electrode point mechanism at the soma ( Rs 5 MOhms , to model a Rs of 15 MΩ compensated 70% ) with a holding potential of 0 mV . The transfer ( Zc ) and input impedance ( Zn ) were determined from the model and used to calculate the actual peak IPSC amplitude at the soma for a given synaptic conductance . Simulations were run in the Neuron 7 . 5 simulation environment . Simulations were run in python 2 . 7 with NEURON 7 . 4 ( Carnevale and Hines , 2006 ) on Windows 7/10 . We created a generic python-NEURON interface ( https://github . com/danielmk/ouropy; copy archived at https://github . com/elifesciences-publications/ouropy ) which wraps NEURON’s python module , into which we ported the conductance based DG model by Santhakumar et al . ( 2005 ) . Model code is available at https://github . com/danielmk/pyDentateeLife2020 ( copy archived at https://github . com/elifesciences-publications/pyDentateeLife2020 ) . We first tuned the original model to capture our experimentally determined properties in the most parsimonious way . During tuning we also updated some model properties to better reflect current data and our experimental paradigm in an individual DG lamella: We introduced a T-type Ca2+ channel mechanism into MCs to more realistically reflect the depolarizing envelope at the onset of a positive current step observed in real MCs . Furthermore , while the original model placed the perforant path input at the distal dendrite of GCs , we moved all perforant path synapses to the middle compartment of the dendrite . In order to be able to capture the results of convergent and divergent synaptic inputs in sufficient resolution to produce the empirically observed activity gradations , we up-scaled cell numbers by a factor of four . To model space , we assumed all cell types to be spread out on a 2 mm DG lamella . Since MCs project to GCs primarily outside the lamellar plane , we removed the MC to GC connection . To allow patterned PP input we adapted PP input specifications from Yim et al . ( 2015 ) . We then proceeded in a first phase of model adjustment , and adapted several parameters to reproduce our in vitro findings regarding spatial and temporal feedback inhibition ( Supplementary file 2 ) . To model frequency-dependent facilitation on mossy fiber outputs , we implemented a simple frequency-dependent synapse model ( tmgsyn ) ( Tsodyks et al . , 1998 ) , and matched the facilitation time constant as well as the decay time constants of individual PSCs to our experimental observations . As in the original model , each cell gives rise to a fixed number of synaptic connections which are spatially restricted to a target pool of adjacent cells . We tuned the size and spatial extent of this target pool to reproduce our spatial data . To provide local inhibition we implemented a ‘local’ interneuron type ( BC ) , whose inputs and outputs were spatially restricted to an ~600 µm area ( as described by Strüber et al . , 2015 ) . To provide global inhibition we implemented a second class of inhibitory interneurons ( HC ) whose inputs and outputs connect to GCs independent of space . This simple formulation allowed us to reproduce the recruitment curves seen for local , remote and global GC activation paradigms . To achieve plausible activity levels , we further adapted synaptic weights similar to Yim et al . ( 2015 ) . We call the network incorporating both spatially restricted BC synapses and mossy fiber facilitation the full tuned network . To isolate the contribution of intrinsic GC properties to pattern separation , we created a disinhibited network by setting the synaptic weight from all interneurons to zero . We also isolated feedforward inhibition by decreasing the mossy fiber to interneuron synaptic weight to zero . To evaluate the effect of spatially constrained inhibition , we created a global network , where the target pool of all interneuron was the entire GC population . To evaluate the effect of mossy fiber facilitation , we set the facilitation time constant to zero , effectively eliminating facilitation . Details on the model parameters are summarized in Supplementary file 2 ) . To study pattern separation , we generated 400 PP inputs . Each PP synapsed onto 100 randomly chosen GCs with the spatial connection probability being governed by a gaussian probability distribution with standard deviation 1 mm and random peak position , modeling a full , nearly uniform input connectivity of individual afferents ( Tamamaki and Nojyo , 1993 ) . To generate theta modulated spike patterns , we used the inhomogeneous Poisson generator from Elephant 0 . 5 . 0-Electrophysiology-Analysis-Toolkit with a 10 Hz ( theta ) sinusoidal rate profile with a peak of 100 Hz , a minimum of 0 Hz and a duration of 600 ms . To generate input patterns with varying overlap from PP afferents i = 1 to 400 , we activated afferents i to i+23 in increments of i = 1 per run . We performed 25 runs for each condition resulting in 300 unique comparisons , excluding self-comparisons . The random seed was held constant between different runs of the same condition , resulting in differing input patterns being fed into the same network . All randomness was generated with the python module numpy . random . To quantify pattern similarity , we used Pearson’s product moment correlation coefficient R of the population rate vectors for input and output patterns . The population rate vector refers to the vector of the mean firing rates of all cells in the population within the entire 600 ms simulation , or 100 or 33 ms time windows for the time resolved analyses . All statistical analyses of the model were performed with n = 7 different random network seeds . During Model development ( tuning phase ) , we first ported the model by Santhakumar et al . ( 2005 ) with closely constrained DG cell-types , and further constrained it to reproduce our physiological data . We then locked the model and proceeded to an ( in silico ) experimental phase , in which pattern separation was investigated . To compute full pattern separation effects ( Figure 6D ) , we calculated the mean Rout within Rin bins of 0 . 1 and measured the area to the unity line ( computed as the mean of the binwise Rin – Rout differences ) . To compute isolated pattern separation effects of specific manipulations we subtracted the respective Rout values with and without the manipulation , thereby obtaining a ΔRout value for each individual Rin . We then again computed the bin-wise mean and quantified the area under the curve , yielding the mean ΔRout analogous to the full effects . Note , that the sequence of averaging and subtracting is irrelevant , and was inverted only to match the figure panels . Data are displayed as mean ± SEM for each Rin bin ( Figure 6E–G ) . The coefficient of variance ( CoV ) was calculated by normalizing the standard deviation of ΔRout within each bin by the mean of that bin , and then averaging over bins , analogous to the previous analyses . However , only bins within 0 . 2 < Rin < 0 . 8 were included , since at the borders very small means led to unreliable results . ΔCoV represents the difference between the mean CoV of the global ( or nonfacilitating ) and the tuned network models . For the temporally resolved pattern separation analysis , all measures were computed as above , but on population vector correlations within 100 or 33 ms time bins . Analyses were performed using ImageJ , Microsoft Excel , Python and Igor Pro . Fits were performed using Igor Pro . Statistical analyses were performed using GraphPad Prism six or Igor Pro . Comparisons were two-tailed whenever applicable . Replicates refer to cells unless otherwise indicated ( slices for imaging experiments and network seeds for modeling data ) . Given the lack of previous information on effect sizes , sample sizes were chosen according to field norms , such that only large effects can be detected ( e . g . Cohen’s d > 1 for paired tests ) . A single outlier facilitation index ( Figure 5E ) during model tuning was removed , as it was outside the triple standard deviation ( due to a very small initial IPSC ) . Group allocation was achieved by alternating acquisition between groups . Statistical significance in Analysis of Variance ( ANOVA ) is indicated by § . F-values and degrees of freedom are given as F ( DFn , DFd ) . When ANOVAs were followed by specific comparisons these are indicated by asterisks , where p<0 . 05 , p<0 . 01 and *p<0 . 001 . Bargraphs and XY plots show means where error bars indicate standard error of the mean . In boxplots error bars represent the data range and boxes the upper and lower quartiles and the median .	You can probably recall where you left your car this morning without too much trouble . But assuming you use the same busy parking lot every day , can you remember which space you parked in yesterday ? Or the day before that ? Most people find this difficult not because they cannot remember what happened two or three days ago , but because it requires distinguishing between very similar memories . The car , the parking lot , and the time of day were the same on each occasion . So how do you remember where you parked this morning ? This ability to distinguish between memories of similar events depends on a brain region called the hippocampus . A subregion of the hippocampus called the dentate gyrus generates different patterns of activity in response to events that are similar but distinct . This process is called pattern separation , and it helps ensure that you do not look for your car in yesterday’s parking space . Pattern separation in the dentate gyrus is thought to involve a form of negative feedback called feedback inhibition , a phenomenon where the output of a process acts to limit or stop the same process . To test this idea , Braganza et al . studied feedback inhibition in the dentate gyrus of mice , before building a computer model simulating the inhibition process and supplying the model with two types of realistic input . The first consisted of low-frequency theta brainwaves , which occur , for instance , in the dentate gyrus when animals explore their environment . The second consisted of higher frequency gamma brainwaves , which occur , for example , when animals experience something new . Testing the model showed that feedback inhibition contributes to pattern separation with both theta and gamma inputs . However , pattern separation is stronger with gamma input . This suggests that high frequency brainwaves in the hippocampus could help animals distinguish new events from old ones by promoting pattern separation . Various brain disorders , including Alzheimer’s disease , schizophrenia and epilepsy , involve changes in the dentate gyrus and altered brain rhythms . The current findings could help reveal how these changes contribute to memory impairments and to a reduced ability to distinguish similar experiences .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2020	Quantitative properties of a feedback circuit predict frequency-dependent pattern separation
The large GTPase dynamin catalyzes membrane fission in eukaryotic cells , but despite three decades of experimental work , competing and partially conflicting models persist regarding some of its most basic actions . Here we investigate the mechanical and functional consequences of dynamin scaffold shape changes and disassembly with the help of a geometrically and elastically realistic simulation model of helical dynamin-membrane complexes . Beyond changes of radius and pitch , we emphasize the crucial role of a third functional motion: an effective rotation of the filament around its longitudinal axis , which reflects alternate tilting of dynamin’s PH binding domains and creates a membrane torque . We also show that helix elongation impedes fission , hemifission is reached via a small transient pore , and coat disassembly assists fission . Our results have several testable structural consequences and help to reconcile mutual conflicting aspects between the two main present models of dynamin fission—the two-stage and the constrictase model . Of the three 80% homologous mammalian isoforms of dynamin , Dyn1 , highly expressed in neurons , has been studied best ( Hinshaw , 2000; Praefcke and McMahon , 2004 ) . Crystallographic analysis ( Chen et al . , 2004; Faelber et al . , 2011; Ferguson et al . , 1994; Ford et al . , 2011; Zhang and Hinshaw , 2001; Chappie et al . , 2010 ) reveals five distinct domains: an N-terminal GTPase or G-domain; a ‘stalk’ region consisting of a four helix bundle; the signaling BSE domain that links G-domain and stalk; a pleckstrin homology ( PH ) domain binding phosphatidylinositol-4 , 5-biphosphate ( PIP2 ) lipids; and a proline-rich C-terminal domain ( PRD ) that mediates interactions with membrane scaffolding proteins . Dynamin monomers oligomerize via their stalks in a criss-cross fashion , resulting in stable dimers ( Cocucci et al . , 2014 ) or tetramers ( Ramachandran et al . , 2007 ) in solution . Continuing this assembly path yields helical filaments ( Carr and Hinshaw , 1997 ) , which have an outer diameter of about 50 nm and a helical pitch ( i . e . , height increment during one complete helical turn ) between 10 nm and 20 nm in the absence of GTP ( Chen et al . , 2004 ) . Specific interactions among dynamin subunits are not conserved throughout the large dynamin superfamily , yet all members share similar architectural assembly properties ( Praefcke and McMahon , 2004 ) . Dynamin helices are narrower in the presence of GTP; the strongest constriction has been seen with the point mutant K44ADyn1: it forms a stable superconstricted 2-start helix , tightening membrane tubules to a water-filled inner lumen with a diameter as narrow as 3 . 7 nm , but not cutting them ( Sundborger et al . , 2014 ) . Since K44ADyn1 is very inefficient in GTP hydrolysis , which however is necessary for remodeling ( Zhang and Hinshaw , 2001; Chappie et al . , 2011 ) , this mutant is believed to trap the system in a pre-fission state . This leaves open the role of GTP activity , with presently two major competing scenarios: the two-stage and the constrictase model . In the two-stage model ( Bashkirov et al . , 2008; Mattila et al . , 2015 ) , nucleotide-loaded dynamin assembles into highly constricted helices , whose entrapped lipid tubules spontaneously undergo hemifission; GTP hydrolysis induces subsequent detachment of the dynamin scaffold from the membrane , allowing the hemifission state to proceed to complete fission . However , neither the lack of dynamin’s cooperativity with GTP during hydrolysis ( Tuma and Collins , 1994 ) , nor the short lifetime of the post-hydrolysis GDP + Pi state ( Antonny et al . , 2016 ) are compatible with key requirements of the two-stage model . It also remains unclear why the experimentally observed superconstricted state does not reach hemifission . In the constrictase model ( Chappie et al . , 2011 ) , several cycles of GTP hydrolysis induce adjacent turns ( or ‘rungs’ ) to stepwise slide past one another , actively constricting the helix and its enclosed membrane tube until the latter fissions; disassembly is then a consequence of the vanishing lipid substrate . The problems here are that the cryo-EM density of stalks in a constricted ( albeit: 2-start ) helix does not match the X-ray structure of dimers , that it is not known how G-G links across rungs would unbind and step cooperatively , and that the mechanics of the final disassembly is less clear ( Antonny et al . , 2016 ) . To refine these two partially conflicting scenarios into a consistent and realistic model requires information about fast changes at the nanometer-scale . This is a challenge for all structural techniques capable of reaching the necessary resolution , and it often requires the use of mutants that might introduce artifacts . However , there is also a second problem: it is by no means obvious how a highly curved bilayer responds to the forces and torques imposed by such complex geometric constraints . Theoretical calculations have provided constraints on energetics and morphology ( Kozlov , 1999; Kozlovsky and Kozlov , 2003; Bashkirov et al . , 2008; McDargh et al . , 2016; McDargh and Deserno , 2018 ) , but these assume high symmetries and ignore fluctuations . Recent coarse-grained ( CG ) simulations , in which the dynamin helix is represented as a pair of rings , have elucidated the relevance of local torques ( Fuhrmans and Müller , 2015 ) and the possibility of long-lived hemifission intermediates ( Mattila et al . , 2015; Zhang and Müller , 2017 ) , but the implied mirror symmetry differs from the actual helicoidal one . Here we go beyond these studies and investigate the constriction of fluctuating lipid bilayer tubes by geometrically realistic dynamin helices . We employ an implicit-solvent CG membrane model ( Cooke et al . , 2005; Cooke and Deserno , 2005 ) ( see Materials and methods ) in which lipids are represented by three consecutive beads , each with a diameter of σ∼0 . 8⁢nm ( Figure 1a ) , that assemble via tail attraction and form bilayers with a bending rigidity of κ≈13kBT ( Cooke et al . , 2005; Cooke and Deserno , 2005 ) , where the thermal energy kBT=4 . 1×10−21J≈0 . 6kcal/mol provides the natural energy scale . The CG time scale maps to approximately τ≃15⁢ns , based on lipid self-diffusion , but we stress that this needs to be interpreted cautiously , since coarse-graining might speed up different dynamic processes differently . The emergent CG membranes capture not only a wide range of mesoscopic phenomena ( Deserno , 2009 ) , but also very local bilayer properties that likely play a role during leaflet-breaking fission events , such as a well-placed pivotal plane ( Wang and Deserno , 2015 ) , a correct magnitude and correlation length for lipid tilt ( Wang and Deserno , 2016 ) , and a pore-opening scenario ( Cooke et al . , 2005; Deserno , 2009 ) in agreement with previous simulations and continuum theory ( Farago , 2003; Tolpekina et al . , 2004 ) . The dynamin filament is composed of similar CG beads , arranged and elastically connected to capture filament radius r , helical radius R and pitch 2⁢π⁢p , and the location of an effective PH binding strip ( Figure 1b and Figure 5 , Materials and methods ) . In the present study we forgo the dynamin assembly process and start with a pre-formed helix winding around the neck of an elongated vesicle comprising about 10 , 000 lipids ( Figure 1c ) . We change the scaffold geometry by tuning the equilibrium distances between its CG beads , that is by imposing internal stresses that trigger a global elastic shape relaxation , and we do this slowly enough to avoid viscous stresses in the bilayer ( Figure 6 , Materials and methods ) . All simulations were performed using the ESPResSo package ( Limbach et al . , 2006 ) and run in triplicate , leading to consistent results . Visualization was done with VMD ( Humphrey et al . , 1996 ) . We begin by exploring a process in which the filament constricts its helical radius R but maintains both the pitch and the orientation of the PH domain with respect to the substrate ( Figure 2c ) . At a confinement approximately corresponding to the superconstricted state , the enclosed membrane tube remains stable . The hemifission intermediate only forms at constriction levels that eliminate the inner lumen ( Figure 2f , blue curve , and Video 1 ) . This seems to contradict the findings by Kozlovsky and Kozlov ( 2003 ) that an inner lumen diameter smaller than approximately 2 nm should spontaneously proceed towards hemifission , but this is not so: in their study an approximately catenoidal neck fissions by the application of bending moments at the upper and lower edge of the neck , while in our case the neck is created by external constriction—a mechanically different scenario . It is of course conceivable that our bilayer tube is merely kinetically stable , but nothing analytical is known about the barrier towards the topologically distinct hemifission state . Indeed , the dynamin-coated superconstricted lipid tubule is stable in experiment ( Sundborger et al . , 2014 ) . Some experiments have shown that GTP hydrolysis may increase the filament’s pitch at fixed radius ( Figure 2d ) , which has led to the suggestion that dynamin might act as an extension spring ( ‘poppase model’ ) ( Stowell et al . , 1999 ) . However , since dynamin changes its shape slowly compared to a membrane’s viscoelastic time scale , believed to be shorter than about 10 ms ( Camley and Brown , 2011 ) , this putative mechanism cannot rely on viscous stresses but instead only on the reversible work being performed during elongation of the neck . To test this , we simultaneously decrease the radius and increase the pitch ( Materials and methods , Figure 6 ) . Contrary to expectation , helical elongation impedes fission: at the same helical scaffold radius R , a larger pitch results in a larger effective radius Rg of the enclosed membrane tubule , requiring further constriction to achieve hemifission . This happens because a large pitch prevents the helical filament from symmetrically confining the membrane , allowing it to ‘bulge out’ at the open groove ( Figure 2f , dark blue curve ) . In nature this could be prevented by accessory proteins , or—if the two-start helix is physiologically relevant—by the second interlocking filament . Compared to the pure constriction case , which in all three runs transitioned into the hemifission step at the same constriction step , we here observe a slight scatter of the precise transition points between two consecutive steps—see Figure 2—figure supplement 1d . We attribute this to the reduced confinement of a membrane inside a helical scaffold with a widening groove , permitting more fluctuations that render the transition point less definite . Beyond changes of radius and pitch , the set of macroscopic motions available to a helical filament contains a third possibility: a ‘twirling’ motion around the local filament axis ( Figure 1b and Figure 2e ) . Its relevance for dynamin derives from a growing number of studies that provide experimental evidence for the tilting of dynamin’s PH domains . Mehrotra et al . ( 2014 ) have shown that a change in PH domain orientation may regulate fission . Sundborger et al . ( 2014 ) fitted dynamin’s crystallographic structure to EM density maps and noticed that one of the two PH domains per dimer tilts out of the membrane , thereby breaking the symmetry of the dimer . And a very recent 3 . 75 Å resolution Cryo-EM reconstruction of human Dyn1 in the GMPPCP-bound state ( Kong et al . , 2018 ) shows that the bundle signaling element ( BSE ) is asymmetrically bent , presumably due to forces generated from the GTPase dimer interaction . These forces are further transferred across the stalk to the PH domain and onto the lipid membrane . On mesoscopic scales , this local rearrangement of the tertiary structure can effectively be described as a net rotation of the helical scaffold around its local longitudinal axis , even if the core of the protein filament does not actually co-rotate . If the PH domains of each monomer were to tilt in the same fashion , then due to dynamin’s criss-cross assembly half of the PH domains would tilt ‘up’ while the other half would tilt ‘down’ , which on average does not displace the filament’s binding surface . However , if only one of these two sets of PH domains tilts , this breaks the up-down symmetry and—at the level of our CG model—effectively rotates the average position of the adhesive strip away from the substrate . This creates a tangential torque on the membrane , because adhesion pins the filament’s local material frame to its Darboux frame on the surface ( Guven et al . , 2014 ) , and therefore it has appropriately been called ‘Darboux torque’ ( Fierling et al . , 2016 ) —see Figure 2g and Box 1 . Considering that the PH domain is connected to the stalk rather flexibly , it is not obvious that it can actually transmit such a torque . However , flexibility does not preclude the transmission of stresses , provided other contacts are in place that help acting as a pivot , and these might simply be steric interactions with other parts of the assembled scaffold . We recall that Kong et al . ( 2018 ) provide structural evidence that forces are transmitted all the way from dynamin’s G-domain to the PH domain and onto the membrane , even though this is indirectly deduced by aligning the coordinates of dynamin with an unbent BSE to the structure of a bent dynamin in the DynGMPPCP map at the GTPase domain . Still , neither this new structure , nor the previously observed tilting of the PH domain ( Sundborger et al . , 2014 ) unambiguously proves the existence of torques . But while elucidating the mechanical nature of this process will have to await more detailed molecular modeling , the existence of a Darboux torque as been posited and exploited in another model for dynamin-driven fission , which simplifies the complex scaffold geometry to two counter-rotating apposing rings ( Fuhrmans and Müller , 2015; Shnyrova et al . , 2013 ) , a geometry that remote-pinches the membrane between the rings . Notice , though , that both geometry and symmetry are different in the latter case: counter-rotating rings have a mirror symmetry and hence exert oppositely acting Darboux torques that constructively interfere in the middle of the scaffold . Since a helix is one contiguous filament , it will rotate everywhere in the same sense , rendering torque amplification much less straightforward . In our own simulations , which account for the full helical geometry , we find that rotation is necessary for a constricted membrane tube with a remaining aqueous lumen to transition into the hemifission intermediate . Consistent across all three simulation runs , the constriction plus rotation sequence ( Figure 2f , cyan curve , Figure 2—Figure Supplement 1a , and Video 2 ) transitions at R=Rc≈10 . 5σ≈8 . 4nm , when the inner lumen is small but has not yet vanished; we will subsequently refer to Rc as the ‘critical constriction radius’ . Since rotation gradually diminishes binding between membrane and filament , a non-rotating scaffold maintains adhesion while constricting , which could hold the membrane tube open and thereby prevent it from transitioning into the hemifission case . To test this , we considered a constricted but not rotated filament at R=10σ ( which is even below the critical radius ) and artificially switched off the adhesion between scaffold and membrane . As shown in Figure 3 , the neck did not progress towards hemifission; in fact , the response to this change is a reduction in Rg which , considering the error bars ( determined via blocking ( Flyvbjerg and Petersen , 1989 ) ) is so small that it is not statistically significant . This suggests that the functionally important consequence of rotation is not merely a reduction of binding energy , but active stresses , likely due to the above-mentioned Darboux torque . If in addition to constriction and rotation we also elongate the helix , hemifission is still reached before the inner lumen disappears , but this requires a further filament radius reduction ( Figure 2f , dark cyan curve ) . Consistent with our observations on constriction plus elongation , this case with additional rotation also shows a greater variability across our three runs , transitioning over three consecutive constriction steps , which however fall between the bounds of constriction plus rotation and constriction only ( see Figure 2—figure supplement 1b ) . These findings again endorse the view that rotation supports hemifission , while elongation opposes it . In the two-stage model the constricted state assembles from GTP-loaded dynamin , while subsequent hydrolysis-triggered depolymerization of the filament induces fission ( Bashkirov et al . , 2008; Mattila et al . , 2015 ) . Two pathways are conceivable: pre-hemifission coat disassembly would need to drive the membrane tube towards ( at least ) hemifission , while post-hemifission disassembly would only need to destabilize this hemifission state . To test the first pathway , we start with a filament at R=Rc ( induced only via constriction ) and then cut it into approximately dimer-sized fragments . Irrespective of whether these retain their adhesion capability to the membrane ( Video 3 ) or lose it and hence immediately unbind ( Video 4 ) , the membrane neck rapidly widens following scaffold rupture ( see also Figure 3 ) . Of note , even binding-capable fragments are individually not strong enough to impose their curvature on a substrate whose geometry becomes progressively unfavorable , and they ultimately detach from it . This also documents that our scaffold only binds weakly , and that rotation-driven hemifission is not the trivial result of massive forces resulting from large adhesion , or even of pulling lipids out of the membrane . Observe that our simulation setup does not exert external membrane tension , since our implicit solvent enforces no volume constraint and permits the membrane—even though closed and constricted—to relax the area per lipid . Nevertheless , it is unlikely that a nonzero tension would prevent the widening of the neck: the value that would be needed to maintain a cylindrical membrane at the critical radius Rc is given by κ/2Rc2 , or a few mN/m ( Deserno , 2015; Bukman et al . , 1996; Hochmuth et al . , 1996 ) , which is very high . Even disregarding the question how such a large tension would arise in a physiological context , the mere fact that this approaches a membrane’s rupture stress and would hence place the system dangerously close to unspecific bilayer failure renders a recourse to such a high tension implausible . Recent experiments have observed that fission of dynamin-coated membrane tubules occurs within the coated region ( Dar et al . , 2015 ) , although earlier studies had suggested that fission occurs at the edge between the dynamin helix and the uncoated membrane ( Morlot et al . , 2012 ) . We always observe that an initial hemifission seed appears in the middle of the neck for short filaments ( up to 1 . 5 turns ) , or two seeds appear near the edges for longer ones ( more than two turns ) , see Figure 2h . This suggests that two regions of highly localized stress appear slightly inwards of either edge , which overlap and may mutually amplify for sufficiently short filaments . Furthermore , since longer filaments trigger hemi-fission only at an even smaller scaffold radius ( Rc , long≈10σ ) , both findings indicate that short filaments induce hemifission more efficiently than long ones . That hemifission is seeded by a small pore is unexpected , not only because fission is believed to be non-leaky ( Bashkirov et al . , 2008 ) , but also because the existence of a hemifission intermediate is generally taken to exclude the need for ( or even possibility of ) pores . However , a tube’s inner leaflet must change topology , posing the question how a filament only in contact with the outer leaflet could promote this transition . Our simulations suggest an intriguing non-axisymmetric pathway: the filament seeds a small pore that widens around the circumference of the neck , while its top and bottom edges pull radially inward and fuse with the apposing intact membrane . This creates two defect lines at which three bilayers meet , and contracting them results in the top and bottom point singularity of the hemifission state’s cylindrical micelle . This sequence of events never involves a large pore , provided the filament stays short enough , a condition that imposes constraints on the constriction process itself . If dynamin constricts due to GTP hydrolysis , as posited in the constrictase model , the initially unconstricted helix has to be long enough for the two rungs to meet , so that apposing G-domains can cross-catalyze hydrolysis . Constricting the radius by a factor of two then necessarily doubles the number of turns if the helical length stays fixed . We find that such long scaffolds induce rather sizable pores ( Figure 2h ) that would likely result in leakage . But the ends of a filament that in the constricted state only takes one turn would not be able to meet up in the preceding unconstricted state , and GTP driven constriction could not commence . This dilemma could be resolved if constriction and disassembly happen in short succession: when the two ends of the initially relaxed helical filament first meet , they hydrolyze GTP and trigger a first constriction step . This brings a new pair of GTP loaded G-domains in contact that drive the next constriction step , and so forth until full constriction . But since GTP hydrolysis enhances disassembly , a filament end comprising ever longer stretches of spent dynamin can shed its monomers concurrently with the ongoing constriction , especially if its PH domain retracted from the membrane . The helix would tighten but never extend much beyond one turn , and hence never induce large pores ( Video 5 ) . This hypothesis is compatible with the experimental evidence that , upon GTP hydrolysis , the scaffold seems to adjust to a length optimal for fission ( about one full turn ) ( Shnyrova et al . , 2013 ) . Without concurrent filament shortening , our hemifission state resembles a short cylindrical micelle ( Figure 2f , background image ) , a transition state explored in two other recent investigations ( Mattila et al . , 2015; Zhang and Müller , 2017 ) . These studies find that this micelle is remarkably stable , suggesting that completing fission faces a very sizable second free energy barrier . Additional tension might reduce its height , and indeed in vitro experiments by Roux et al . ( 2006 ) and Morlot et al . ( 2012 ) have found that longitudinal tension assists in producing fission ( though it is interesting to note that our simulations present a tension-free pathway to fission ) . However , Zhang and Müller ( 2017 ) estimate that the barrier might remain as large as 30kBT even at the largest biologically justifiable tensions ( close to uncontrolled rupture ) , which leads them to speculate that other factors help to complete fission , such as the high curvature in the region where the micelle merges with the bilayer . Here we propose that this second barrier might indeed not be biologically relevant , since it is a consequence of the simulation setup in these studies , which prevents the micelle from shrinking and instead requires it to rupture mid-length . In our simulations we also never observe the micelle to break while being enclosed by the dynamin scaffold ( from which it has already unbound ) , but we argue that this is due to the scaffold preventing the two point defects at the end of the micelle from merging . Once the scaffold disassembles ( whether gradually or abruptly ) , they are being pulled together by a force that can be estimated to be F≃π⁢κ/2⁢z0∼𝒪⁢ ( 100⁢pN ) . This is the energy per unit length of creating a cylindrical micelle of curvature radius z0 ( the pivotal plane distance ) out of lipids that have a monolayer leaflet rigidity of κ/2 . This force pulls the two daughter membranes together , whereupon bilayer contact catalyzes micellar fission ( see Video 6 ) . This hypothesis , even though derived from a fairly coarse-grained model , is nevertheless plausible , because splitting the cylindrical micelle somewhere along its length would further raise the free energy by creating two spherical caps of even more unfavorable packing geometry ( Zhang and Müller , 2017 ) , while the contact between the daughter membranes annihilates the two already existing point defects at the micellar ends . The latter scenario seems to conflict with the widely held belief that highly curved vesicles are fusogenic . However , recent experiments by François-Martin et al . ( 2017 ) paint a more nuanced picture . These authors measured the free energy barrier towards fusion between ∼60⁢nm diameter POPC ( 16:0–18:1 PC ) and DOPC [18:1 ( Δ9–Cis ) PC] vesicles at 37°C . On the one hand , the observed free energy barrier of ΔF∼30kBT is at the lowest end of theoretical estimates , indicating that protein-free fusion between such vesicles is easier than previously thought . On the other hand , even when incubating such vesicles at the unusually high concentrations of 18 mM PC , only 2% had completed fusion after 30 min , which does not support the notion of spontaneous fusion—as these authors indeed conclude . Nevertheless , the free energy pathways for events that change membrane topology remain challenging , for at least four unrelated reasons: first , the choice of boundary conditions and constraints in theory or simulations can strongly affect the outcome ( for instance , whether cylindrical micelles can shrink , hemifusion diaphragms can expand , or tensions can relax ) ; second , the key physics happens at the nanometer scale , and hence the results of highly coarse-grained models such as ours or continuum theory must be interpreted cautiously; third , a topological barrier to fusion exists , whose height depends on the Gaussian curvature modulus and is thus not well known ( see the contribution of Deserno in Bassereau et al . ( 2018 ) ) ; and fourth , additional complexity due to accessory proteins or lipid mixtures may qualitatively change the energetics , for instance by lipid sorting in very high leaflet curvature gradients ( Cooke and Deserno , 2006; Tian and Baumgart , 2009 ) . For the pathway that naturally arises in our simulations—a shrinking of the hemifission micelle post-disassembly , followed by autocatalytic cutting—the free energy barriers are not obvious , especially for solvent free models like ours ( or those in the above-mentioned studies by Mattila et al . ( 2015 ) and Zhang and Müller ( 2017 ) ) , and hence this subject warrants more dedicated future studies . Taken together , our studies support a number of conclusions . To begin with , the poppase model ( Stowell et al . , 1999 ) faces not only kinetic but also equilibrium obstacles: extension of the helical scaffold disfavors fission , because a widening groove offsets constriction . In contrast , hemifission is promoted by tilting one of the two symmetry subsets of PH domains ( Mehrotra et al . , 2014; Sundborger et al . , 2014; Kong et al . , 2018 ) , which effectively rotates the filament’s adhesive strip away from normal contact . The resulting Darboux torque ( Fierling et al . , 2016; Guven et al . , 2014 ) has been previously invoked to explain fission ( Fuhrmans and Müller , 2015; Shnyrova et al . , 2013 ) , but these earlier models pictured the dynamin scaffold as two rings , whose counter-rotation-induced torque remote-pinches the enclosed membrane tube . The effect is particularly intuitive under mirror symmetry , a geometry at odds with the actual helical one . Surprisingly , this difference appears not to be central . We show that pre-hemifission scaffold breakage aborts fission by allowing the enclosed membrane tube to re-expand . But disassembly is still essential to complete fission , as argued in the two-stage model , because even if the severed bilayer is energetically preferable to the hemifission state ( Kozlovsky and Kozlov , 2002 ) , a substantial energy barrier is required to explicitly break the hemifission micelle ( Mattila et al . , 2015; Zhang and Müller , 2017 ) . Our simulations lead us to suggest that this can be circumvented by merging and annihilating the micelle’s two endpoint defects , a process that commences once the two daughter membranes are pulled into close proximity and hence critically depends on the scaffold getting out of the way by disassembling . Notice that this mechanism is not scaffold specific and may hence apply to any other fission machinery that leads to a hemifission state involving a small stretch of a cylindrical micelle . How hemifission is induced is experimentally less clear ( Antonny et al . , 2016 ) . In our studies constriction alone , down to the smallest experimentally observed luminal radii , does not suffice . We believe this to be independent of whether the constricted state is reached passively ( two-stage model ) or actively ( constrictase model ) , since our dynamic protocol could simply be viewed as a means to adiabatically prepare a highly constricted state . PH domain tilting thus emerges as a way to catalyze hemifission , and since applying a Darboux torque costs energy , we posit that the energy of GTP hydrolysis is at least in part used to drive this conformational change . This is in accord with experimental findings by Dar and Pucadyil ( 2017 ) that replacing the PH domain by a simple binding motif strongly slows down the fission rate . It is also supported by the recent Cryo-EM reconstruction by Kong et al . ( 2018 ) , who suggest that forces generated from the GTPase dimer interaction are transferred across the stalk to the PH domain and from there onto the membrane . It is worth noting that in their reconstruction Dyn1 was bound to the nonhydrolyzable guanosine triphosphate analogue GMPPCP , and so the question whether GTP hydrolysis would create additional forces or torques that then could be transduced to the membrane remained open . To better understand the mechanical basis and viability of such a force transmission , it will be essential to structurally resolve the connection between the PH domain to the stalk domain . Finally , we consistently observe that hemifission is initiated by a transient pore puncturing the constricted neck ( see arrows in Figure 2h , the time sequence in Figure 4 , and Video 2 ) . This is reminiscent of likewise non-axisymmetric pathways seen in fusion simulations ( Noguchi and Takasu , 2001; Noguchi and Takasu , 2002; Müller et al . , 2002; Müller et al . , 2003 ) , whose implied transient leakage currents were soon after confirmed to be common in electrophysiological measurements on hemagglutinin-mediated fusion ( Frolov et al . , 2003 ) . We estimate that our pores open to a diameter of no more than 2 nm , which under physiological ionic strength equips them with a conductivity on the order of 1 nS , while their lifetime is a few microseconds ( mapped very roughly from our coarse-grained model via lipid self-diffusion ) . This would not be trivial to detect experimentally , but similarly fast transients ( few microseconds ) have been observed in the gating currents of Shaker potassium channels embedded in Xenopus oocytes , using an eight-pole Bessel filter and achieving a bandwidth of about 200 kHz ( Bezanilla , 2018; Sigg and Bezanilla , 2003 ) . Increasing the ionic strength in reconstituted systems to 1 M , a further gain in sensitivity by up to an order of magnitude could be obtained , provided this does not interfere with dynamin’s operation . But notice also that our leakage pores are partially covered by the dynamin scaffold ( see again Figure 2h ) and , under physiological conditions , possibly by additional proteins proposed to assist in dynamin-driven fission , such as BAR domains ( Takei et al . , 1999; Farsad et al . , 2001; Itoh et al . , 2005; Mim et al . , 2012 ) . Hence , a pore’s effective conductance might be significantly lower than that of a freely accessible membrane pore or channel . Our simulations show that pores are larger—and hence potential leakage more severe—if hemifission is triggered at larger radii , or if the filament is longer , suggesting that a close coordination of constriction , rotation , and possibly concomitant disassembly renders fission not only more efficient but also more tight . We hence expect mutants disrupting this coordination ( for instance the ΔPH mutant of Dar and Pucadyil ( 2017 ) ) to have larger pores . This would exacerbate leakage problems and could be experimentally observed . While we have explicitly focused on the case of classical dynamin , several of our findings have implications beyond this particular protein and offer lessons for membrane topology remodeling that go beyond Dyn1 . For instance , all members of the dynamin superfamily are believed to oligomerize ( Praefcke and McMahon , 2004 ) , likely into helices , for which our generic basic model holds . Furthermore , the mismatch between the normal curvature direction of such a helical filament and the direction of its adhesive domain ( measured by φ in our case ) is a generic degree of freedom for such a structure . Whether directly present during assembly or only later triggered by a ( possibly GTP-dependent ) conformational change , it will give rise to Darboux torques on the membrane . We have explicitly shown that these support the transition into the hemifission state , and hence they provide a means to trigger a topological transition that is different from the notions of constriction or elongation . This matters because other topology remodeling proteins exist that are unlikely to work by either . Consider in particular the reverse-geometry scission driven by the ESCRT-III complex ( Wollert et al . , 2009 ) , which involves the adsorption and polymerization of a helical filament on the inside of the neck to be cut . Present models ( for a recent review , see Schöneberg et al . ( 2017 ) ) rely on adhesion and/or geometric changes of the spiraling filaments , which result in either direct forces or boundary-induced stresses at the contact site ( arising from , for instance , a contact curvature condition ( Deserno et al . , 2007 ) ) . But geometric rearrangements of a curved elastic filament with a finite twist rigidity , whose material frame is pinned to the membrane ( Guven et al . , 2014 ) , almost invariably result in additional Darboux torques , as well as twist-induced analogs ( Fierling et al . , 2016 ) , the relevance of which is only beginning to emerge ( Quint et al . , 2016 ) . In this work we have explicitly demonstrated , for the particular example of Dyn1 , that this geometrically elementary mechanism is remarkably effective , which suggest that it might be more common than so far realized: traces of the essential action could date back to the earliest bacterial FtsZ ancestor , which shares many of the key geometric ( Erickson , 2000 ) and biochemical ( Lu et al . , 2000 ) features . This might hence suggest novel functional models also for other membrane remodelers ( such as ESCRT-III ) , for which the mode of operation is much less well understood than for classical dynamin . Our investigation focuses on the mesoscopic effects of a constricting helical dynamin filament on a tubular lipid membrane , expressed primarily by the interplay between ( i ) filament geometry and binding affinity and ( i⁢i ) membrane and filament elasticity . A long-standing aim in the dynamin field has been to develop an explanatory model for dynamin’s membrane fission mechanism within the framework of such mesoscopic emergent properties . By creating a coarse-grained ( CG ) model of dynamin fission that captures precisely these mesoscopic aspects , our goal is to explore the consequences of this interplay , while either disregarding much of the microscopic or chemical detail , or implicitly accounting for it in terms of effective interactions . The fundamental degrees of freedom in our model are mesoscopic ‘beads’ with a size ( diameter ) σ≃0 . 8⁢nm , which hence correspond to 𝒪 ( 10 ) heavy atoms . Physics below this scale leaves its trace at larger dimensions through the collective action of effective potentials—much like the quantum mechanics of correlated electron clouds re-emerges classically as effective van der Waals interactions , or the configurational distributions of charged moieties in a molecule yield effective dipole moments and polarizabilities . A rich literature exists outlining the technique of coarse graining in a soft-matter and biophysics context ( Deserno , 2009; Ingólfsson et al . , 2014; Izvekov and Voth , 2005a; Izvekov and Voth , 2005b; Brini et al . , 2013; Müller-Plathe , 2002; Murtola et al . , 2009; Noid , 2013; Noid et al . , 2008; Peter and Kremer , 2009; Riniker et al . , 2012; Saunders and Voth , 2013; Voth , 2008 ) . In the terminology of Noid ( 2013 ) , we employ a ‘top down’ modeling approach . Since we are not concerned with hydrodynamic effects , we account for the embedding water implicitly via effective attractive interactions between hydrophobic species ( such as CG lipid tail beads ) . The reduction of the number of degrees of freedom allows not only for a computational speed-up and a better statistical sampling; it also offers insight into the nature of the original problem: physical effects that are correctly represented in a CG model prove to be largely independent of microscopic specifics , rendering them important players in an emergent mesoscopic explanatory model . All simulations were performed using the ESPResSo package ( Limbach et al . , 2006 ) , using an integration time step Δt=0 . 01τ ( where τ is the simulations CG time unit ) . A Langevin thermostat ( Grest and Kremer , 1986 ) with friction constant Γ=1 . 0τ−1 was used to keep the temperature constant . All simulations have been performed in an N⁢V⁢T ensemble . We use an implicit solvent lipid model ( Cooke et al . , 2005; Cooke and Deserno , 2005 ) in which a single lipid molecule is replaced by three consecutive beads , one for the hydrophilic head and two for the hydrophobic tails , which are not individually resolved ( Figure 1a ) . In the absence of solvent , the hydrophobic effect , and hence aggregation of lipids into bilayer membranes , is driven by attractive interactions between the CG tail beads of depth ε ( the simulations’ energy unit ) and range wc ( the precise potential forms are detailed by Cooke et al . ( 2005 ) ) . The magnitude of wc and the temperature T determine whether lipids aggregate into fluid membranes , and if so , what their material properties are . We work at the frequently employed state point wc=1 . 6σ and kBT=1 . 1ε ( which also sets the energy scale ) . Under these conditions lipids aggregate into fluid membranes with an area per lipid of about aℓ=1 . 2σ2≈0 . 77nm2 , an average head-bead distance from the midplane of about dH=2 . 2σ≈1 . 8nm , and a separation between the half-maximum density points ( a possible proxy for the Luzzati thickness ( Luzzati and Husson , 1962 ) ) of about d1/2=5 . 6σ≈4 . 5nm . Considering the limitations of the underlying microscopic basis of nanometer-sized CG beads , these numbers are reasonably close to typical biologically relevant lipids , such as POPC ( aℓ≈0 . 643nm2 and d1/2≈3 . 91nm at 30°C ( Kučerka et al . , 2011 ) ; still , sub-nanometer resolution should not be taken too literally in a model like this . The membrane has a bending rigidity of κ≈12 . 8kBT ( Cooke et al . , 2005; Cooke and Deserno , 2005; Harmandaris and Deserno , 2006; Hu et al . , 2013 ) , purposefully chosen smaller than a typically value of 20kBT ( Kučerka et al . , 2005 ) , since this affords a significantly entropy-deprived coarse-grained system an alternative opportunity to undergo fluctuations . Wang and Deserno ( 2016 ) , recently also measured the tilt modulus of this model , finding it to be κt≈7kBT/nm2 . This results in the characteristic tilt decay length ℓt=κ/κt≈1 . 3nm , in good agreement with experimental values ( Jablin et al . , 2014 ) , indicating that the model does not merely capture overall fluidity and curvature elasticity , but also local lipid reorientation physics , which matters for the intermediate states of fission and fusion ( Kozlovsky and Kozlov , 2002 ) . We model the right-handed helical dynamin scaffold by CG beads of the same diameter σ as the lipid CG beads ( represented by a purely repulsive Weeks-Chandler-Andersen potential that only acts between filament and lipid beads ) , placing them such as to represent the shape of a CG filament ( Figure 1b ) . This first requires setting up a coordinate system that embodies helical symmetry—see Figure 5 for the following discussion . The arc-length ( s ) parametrization of a right-handed helix of radius R and pitch 2⁢π⁢p , whose axis coincides with the z-axis of the coordinate system , is given by ( 1 ) X ( s ) = ( Rcos⁡ ( s/R2+p2 ) Rsin⁡ ( s/R2+p2 ) ps/R2+p2 ) . This has a length 2⁢π⁢R2+p2 per turn and a local ( Frenet ) curvature of R/ ( R2+p2 ) . The local unit tangent vector T ( s ) =X′ ( s ) is then ( 2a ) T ( s ) =1R2+p2 ( −Rsin⁡ ( s/R2+p2 ) −Rcos⁡ ( s/R2+p2 ) p ) . We need to define a finite-thickness filament for which X ( s ) is the central axis . Moreover , since we have to specify the location of an adhesive strip relative to a central cylinder that is being wrapped by the helix , it is convenient to extend the tangent vector T along the helical curve into a local basis by defining two additional vectors N and L as follows: both have unit length and are perpendicular to T , the normal vector N coincides with the local surface normal of the inscribed cylinder , and the remaining co-normal vector L is given by L=T×N: ( 2b ) N ( s ) = ( −cos⁡ ( s/R2+p2 ) −sin⁡ ( s/R2+p2 ) 0 ) ( 2c ) L ( s ) =1R2+p2 ( −psin⁡ ( s/R2+p2 ) −pcos⁡ ( s/R2+p2 ) R ) . The thus defined triplet {T , N , L} of vectors constitutes a right-handed orthonormal basis at every point along the filament . It is called the ‘Darboux frame’ ( with respect to the underlying cylindrical surface on which the filament rests ) . On the {N , L} plane perpendicular to the filament we now place CG beads that will form one of the cross-sectional circular discs from which we build the filament slice by slice . Each disk consists of a central bead placed directly onto the filament axis , a first ring of 6 beads and radius R6 , and a second ring of 12 beads and radius R12 . The coordinates of the beads sitting on the 1- , 6- and 12-ring can be parametrized as ( 3a ) X1 ( s ) =X ( s ) ( 3b ) X6 ( s , n ) =X ( s ) +R6[N ( s ) cos⁡φn ( 6 ) +L ( s ) sin⁡φn ( 6 ) ] ( 3c ) X12 ( s , n ) =X ( s ) +R12[N ( s ) cos⁡φn ( 12 ) +L ( s ) sin⁡φn ( 12 ) ]where n∈{0 , … , 5} for the 6-ring and n∈{0 , … , 11} for the 12-ring . The angles are given by ( 4 ) φn ( 6 ) =2⁢π⁢n6+φ , φn ( 12 ) =2⁢π⁢n12+φ , and the overall phase shift φ denotes the extent to which bead 0 on the 6- and 12-ring is rotated around the local filament axis , see Figure 1b . Since we will subsequently equip bead 0 on the 12-ring with an additional adhesion towards lipid head groups , representing the inner binding region of the dynamin filament due to the PH domains , this phase angle φ describes a rotation of the CG filament with which we effectively capture the asymmetric tilting of half of the PH domains . Notice that due to the way the Darboux frame is set up , φ=0 corresponds to an adhesive strip that exactly sits on the enclosed cylindrical surface ( and hence does not exert any Darboux torque ) . We typically consider filaments consisting of 9 dynamin dimers , each comprising five discs , for a total of 45 discs . We take R6=2σ and R12=4σ , making for a filament diameter of about 8σ≈6 . 4nm . The spacing Δ⁢s between discs along the central helix is set so that a full turn in the unconstricted state ( R=20σ and 2πp=11σ ) comprises 40 discs , leading to Δs=2πR2+p2/40=3 . 1536σ . The location of every bead in the filament at a given shape triplet {R , p , φ} is now completely specified . The shape is elastically fixed by introducing harmonic bonds between all beads within a cutoff distance of rcut=5σ , whose rest length equals the equilibrium distance between the beads according to the chosen shape triplet {R , p , φ} , and whose spring constant is K=200ε/σ2 , which is similar to the choice for other elastic networks for coarse-grained protein models ( Periole et al . , 2009 ) . This renders the filament sufficiently stiff so that it can impose its shape on an underlying membrane tube , and not vice versa . To change the filament’s shape , we adjust the rest lengths of all elastic springs so that they represent distances in a filament at a new shape triplet , {R , p , φ}→{R′ , p′ , φ′} . This introduces local stresses which the filament relaxes by transforming into the new equilibrium shape . We consider parameter ranges within R/σ∈[6 , 18] , 2⁢π⁢p/σ∈[11 , 20] and φ∈[0∘ , 80∘] . Figure 6 illustrates the time series of these parameter changes , which we use individually or in combination , as outlined in the main text . For instance , during constriction we reduce the helical radius by 1σ every 300τ , corresponding to about 0 . 18nm/μs . Albeit much faster than in reality , this is still effectively quasistatic . Beads on the red adhesive strip ( accounting for the PH domains ) additionally experience an adhesion of strength ε towards lipid head beads , represented by a standard Lennard-Jones potential that is truncated and shifted to zero at r=2 . 5σ . For the beads on the two immediately adjacent neighboring strips ( i . e . , numbers 1 and 11 on the 12-ring ) we turn off the hard core repulsion with respect to lipid head ( but not tail ) beads , in order to permit the adhesive domain to embed into the head group region . The free energy of binding for dynamin dimers or larger fragments depends not only on the local chemistry , but also on the curvature of both dynamin and the membrane . We are not aware of measurements that are precise enough to help parametrizing this interaction , but in order to avoid driving fission by overly strong interactions ( which could , for instance , exert unrealistically large torques or even pull lipids out of the membrane ) , we have opted for a lower-bound scenario , in which we made the interaction between dimer-equivalents ( blocks of 5 discs ) and a highly curved membrane neck just strong enough to trigger binding; but when substrate curvature decreases due to membrane tube widening , the interplay between curvature energy and adhesion increasingly disfavors binding ( McDargh et al . , 2016 ) , and fragments detach ( Video 3 ) . Several possibilities exist to quantify the extent of tubular membrane constriction , but the two most obvious ones have significant drawbacks: the midplane radius cannot be defined once the membrane is in the hemifission state , and the radius of the inner lumen cannot distinguish between a completely closed bilayer tube and a hemifission micelle . We hence use as a metric for constriction the cross-sectional gyration radius Rg of beads in the vicinity of the constriction point , defined as follows: ( 5 ) Rg2=1M∑i=1M[ ( ri−r0 ) xy]2 , where the sum extends over all lipids that are at most a radial distance of 20σ and an axial distance of ±1σ away from the filament’s center of mass , r0 is the center of mass of these selected lipids , and the subscript 'x⁢y' indicates that we first take the projection of ri−r0 into the x⁢y-plane . If Rm is the midplane radius of a cylindrical lipid tube and w the monolayer width , then in continuum approximation we get ( 6 ) Rg2=∫Rm−wRm+wdr2πrr2∫Rm−wRm+wdr2πr=Rm2+w2 , showing that Rg=Rm[1+12 ( w/Rm ) 2+⋯] is close to Rm , with quadratic higher order corrections in w/Rm . Moreover , a membrane cylinder of vanishing inner lumen has Rm=w and hence Rg=2w , while the hemifission micelle has Rg=w/2 . Hence , the jump in Rg is w/2=12×5 . 6⁢σ/2≈2⁢σ , using the Luzzati width of our CG membrane . This is very close to our observed jump distance in the constriction only case ( Figure 2f , blue curve ) , which indeed only transitions when the inner lumen disappears . For the other cases we investigate the jump tends to be higher , because hemifission occurs while the membrane tube is still water-filled .	When cells take up material from their surroundings , they must first transport this cargo across their outer membrane , a flexible sheet of tightly organized fat molecules that act as a barrier to the environment . Cells can achieve this by letting their membrane surround the object , pulling it inwards until it is contained in a pouch that bulges into the cell . This bag is then corded up so it splits off from the outer membrane . The ‘cord’ is a protein called dynamin , which is thought to form a tight spiral around the bag’s neck , closing it over and pinching it away . The structure of dynamin is fairly well known , and yet several theories compete to explain how it may snap the bag off the outer membrane . Here , Pannuzzo et al . have created a computer simulation that faithfully replicates the geometry and the elasticity of the membrane and of dynamin , and used it to test different ways the protein could work . The first test featured simple constriction , where the dynamin spiral contracts around the membrane to pinch it; this only separated the bag from the membrane after implausibly tight constriction . The second test added elongation , with the spiral lengthening as well as reducing its diameter , but this further reduced the ability for the protein to snap off the membrane . The final test combined constriction and rotation , whereby dynamin ‘twirls’ as it presses on the neck of the bag: this succeeded in efficiently severing the membrane once the dynamin spiral disassembled . Indeed , the simulations suggested that dynamin might start to dismantle while it constricts , without compromising its role . In fact , getting rid of excess length as the protein contracts helps to dissolve any remnants of a membrane connection . Defects in dynamin are associated with conditions such as centronuclear myopathy and Charcot‐Marie‐Tooth peripheral neuropathy . Recent research also indicates that the protein is involved in a much wider range of neurological disorders that include Alzheimer's , Parkinson's , Huntington's , and amyotrophic lateral sclerosis . The models created by Pannuzzo et al . are useful tools to understand how dynamin and similar proteins work and sometimes fail .	[ "Abstract", "Introduction", "Results", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2018	The role of scaffold reshaping and disassembly in dynamin driven membrane fission
The observation of animal orofacial and behavioral reactions has played a fundamental role in research on reward but is seldom assessed in humans . Healthy volunteers ( N = 131 ) received 400 mg of the dopaminergic antagonist amisulpride , 50 mg of the opioidergic antagonist naltrexone , or placebo . Subjective ratings , physical effort , and facial reactions to matched primary social ( affective touch ) and nonsocial ( food ) rewards were assessed . Both drugs resulted in lower physical effort and greater negative facial reactions during reward anticipation , especially of food rewards . Only opioidergic manipulation through naltrexone led to a reduction in positive facial reactions to liked rewards during reward consumption . Subjective ratings of wanting and liking were not modulated by either drug . Results suggest that facial reactions during anticipated and experienced pleasure rely on partly different neurochemical systems , and also that the neurochemical bases for food and touch rewards are not identical . Rewards are salient stimuli , objects , events , and situations that induce approach and consummatory behavior by their intrinsic relevance for survival or because experience has taught us that they are pleasurable ( Schultz , 2015 ) . Today , our understanding of the neurochemical basis of reward processing rests on 30 years of animal research , and on preliminary confirmatory findings in humans , which led to the identification of two distinct components: wanting , that is , the motivation to mobilize effort to obtain a reward , and liking , that is , the hedonic experience evoked by its consumption ( Berridge , 1996; Berridge , 2018; Berridge and Kringelbach , 2015; Berridge and Robinson , 1998 ) . This conceptual division is paralleled in cognitive theories of economic decision making ( Kahneman et al . , 1997; Berridge and O’Doherty , 2014 ) that similarly distinguish between decision utility ( how much the value attached to an outcome determines its choice or pursuit ) and experienced utility ( referring to the hedonic experience generated by an outcome ) . In animal research , the ‘taste reactivity test’ , a method to assess eating-related pleasure by observing facial and bodily reactions of animals and human infants to palatable and aversive tastes , has played a fundamental role in the identification of discrete reward systems in the brain ( Barbano and Cador , 2007; Berridge , 2000; Dolensek et al . , 2020; Grill and Norgren , 1978; Steiner et al . , 2001 ) . Indeed , it has been shown that neither pharmacological disruption nor extensive lesion of dopaminergic neurons affects facial liking reactions ( e . g . relaxed facial muscles and licking of the lips ) to the consumption of sweet foods in rats ( Berridge et al . , 1989; Treit and Berridge , 1990 ) , and that greater mesolimbic dopamine release induced by electric stimulation of the hypothalamus results in greater food intake without modulating hedonic reactions ( Berridge and Valenstein , 1991 ) . On the other hand , ( facial ) hedonic reactions to sensory pleasure are amplified by opioid , orexin , and endocannabinoid stimulation of various ‘hedonic hotspots’ of the brain , including the nucleus accumbens ( NAc ) shell and limbic areas such as the insula and the orbitofrontal cortex ( Berridge and Kringelbach , 2015 ) . These stimulations not only increase liking but also result in food approach and feeding behavior ( Peciña and Smith , 2010; Taha , 2010 ) . Evidence of similar neurochemical parsing of reward processing in humans is mainly derived from research in clinical populations and a handful of recent pharmacological studies in healthy volunteers . For example , stimulation of D2/D3 receptors through dopamine agonists can induce compulsive medicament use , gambling , shopping , hypersexuality , and other addictive activities in some patients with Parkinson's disease , often without corresponding changes in subjective liking ( Callesen et al . , 2013; Evans et al . , 2006; Weintraub et al . , 2010; but see Meyer et al . , 2019 for an account of the complexity of compulsive disorders in Parkinson's disease ) . Evidence for disrupted motivation to gain immediate rewards has been observed after dopamine D2/D3 receptors blockade in healthy volunteers , in both a pavlovian-instrumental-transfer task and a delay discounting task ( Weber et al . , 2016 ) . Administration of µ-opioid receptor agonists in healthy individuals has been associated with changes in subjective feelings and motivational responses to different types of rewards , as indicated , for example , by higher pleasantness of the highest-calorie ( but least palatable ) food option available ( Eikemo et al . , 2016 ) , greater effort to view , and liking of the most attractive opposite-sex faces ( Chelnokova et al . , 2014 ) , stronger preference for stimuli with high-reward probability ( Eikemo et al . , 2017 ) , and enhanced emotional ratings to positive and negative images ( Atlas et al . , 2014 ) . Furthermore , administration of the non-selective opioid receptor antagonist naloxone to healthy men decreased subjective pleasure associated with viewing erotic pictures and reduced the activation of reward related brain regions such as the ventral striatum ( Buchel et al . , 2018 ) . Despite the progress made , the animal research is only partly informative to comprehend reward processing in humans . While animal research allows to investigate the activity of neurons and neurotransmitters in a much more targeted way , it is also limited to certain measures of liking ( i . e . behavior and facial expressions , while humans can also provide subjective reports ) , and has mainly focused on food rewards in the past . Moreover , human and animal research about the neurochemical regulation of reward processing remain difficult to compare , as human pharmacological studies have struggled to adopt translational paradigms and an operationalization of reward that resembles the one used in animal research , that is , measuring decision utility and experienced utility in the same task , providing primary rewards on a trial-by-trial basis , and/or using objective hedonic reactions to consumed rewards , in addition to relying on subjective verbal report ( Der-Avakian et al . , 2016; Pool et al . , 2016 ) . Including the recording of hedonic facial reactions , the ‘gold standard’ in animal research on the neurochemical regulation of reward processing in human studies seems to be a promising avenue in this regard . Recently , the use of facial electromyography ( fEMG ) has gained increased attention in the context of human reward processing . Results suggested that human adults relax the corrugator muscle ( involved in frowning ) , and to a lesser extent activate the zygomaticus muscle ( involved in smiling ) , during both anticipation and consumption of different types of pleasurable stimuli , although differences between types of rewards exist ( Bershad et al . , 2019; Franzen and Brinkmann , 2016; Korb et al . , 2020; Mayo et al . , 2018; Pawling et al . , 2017; Rasch et al . , 2015; Ree et al . , 2019; Sato et al . , 2020; Wu et al . , 2015 ) . Notably , to the best of our knowledge , no study has yet investigated implicit hedonic facial reactions to different types of rewards after pharmacological drug challenge in humans . To fill this knowledge gap , we pharmacologically manipulated the dopaminergic and opioidergic systems in humans via oral administration of the highly selective D2/D3 dopamine receptor antagonist amisulpride ( 400 mg ) , the non-selective opioid receptor antagonist naltrexone ( 50 mg ) , or placebo , in a randomized , double-blind , between-subject design in 131 healthy volunteers ( group sizes were 42 , 44 , and 45 , respectively , for amisulpride , naltrexone , and placebo ) , and investigated the effects with a recently developed experimental paradigm ( Korb et al . , 2020 ) , in which reward processing is operationalized similarly to animal research . Explicit subjective ratings of wanting and liking , physical effort ( squeezing of an individually thresholded hand-dynamometer to obtain rewards ) and implicit hedonic reactions ( fEMG ) during anticipation and consumption of primary social and nonsocial rewards of similar magnitude were obtained on a trial-by-trial basis ( Figure 1 ) . Sweet milk with different concentrations of chocolate flavour served as nonsocial food rewards . Gentle caresses to the forearm , delivered by a same-sex experimenter at different speeds , resulting in different levels of pleasantness ( Ackerley et al . , 2014; Löken et al . , 2009; McGlone et al . , 2014 ) , served as non-sexual social rewards . By adopting a translational approach , which makes human research comparable to animal research ( e . g . measuring both real effort and hedonic facial reactions to primary rewards ) , we investigated two fundamental yet unresolved research questions: ( 1 ) to what extent do motivational and hedonic implicit and explicit responses during the anticipation and consumption of rewards rely on the dopaminergic and opioidergic systems in humans , and ( 2 ) do food and touch rewards share the same neurochemical basis in humans . We made the following hypotheses based on the literature . First , because liking relies heavily on the opioidergic but not the dopaminergic system ( Berridge and Kringelbach , 2015 ) , subjective ratings of liking , and hedonic facial reactions during reward consumption , were expected to be lower after administration of the opioid antagonist naltrexone , compared to placebo , but not after administration of the dopamine antagonist amisulpride , particularly for the most preferred rewards ( Eikemo et al . , 2016; Smith and Berridge , 2007 ) . Second , because wanting is believed to be regulated by the dopaminergic and opioidergic systems ( Peciña and Berridge , 2013 ) , we expected subjective ratings of wanting , and physical effort applied to obtain the preferred announced reward , to be lower after administration of both naltrexone and the D2/D3 receptor antagonist amisulpride . Third , because facial responses during reward anticipation – previously shown to occur to learned cues for rewards in rats ( Delamater et al . , 1986 ) , and humans ( Korb et al . , 2020 ) – may reflect anticipated pleasure during a period commonly associated with wanting , they were expected to be affected by naltrexone , as well as by amisulpride , compared to placebo . Finally , based on fEMG results showing similar hedonic facial reactions to food and touch rewards , such as relaxation of the corrugator supercilii muscle and in some cases activation of the zygomaticus major muscle ( Bershad et al . , 2019; Korb et al . , 2020; Mayo et al . , 2018; Pawling et al . , 2017; Ree et al . , 2019; Sato et al . , 2020 ) , and on evidence from neuroimaging studies that supports the ‘common currency hypothesis’ of reward processing ( Berridge and Kringelbach , 2015; Ruff and Fehr , 2014 ) , we expected the same pattern of results for both types of rewards . In order to rule out eventual group differences that were not of interest , we conducted a series of statistical tests to verify the matching of the three groups . The three drug groups did not differ significantly in rankings of rewards before the main task , as shown by the absence of a significant Drug X Reward Level interaction for both food and touch rewards ( Figure 2 ) . Only a significant main effect of Reward Level was found for food ( X2 ( 2 ) =78 . 1 , p<0 . 001 ) and for touch rewards ( X2 ( 2 ) =115 . 71 , p<0 . 001 ) , confirming the expected pattern of preferred food rewards ( milk with greater chocolate content being preferred to milk with lower chocolate content ) , and of touch rewards ( slower caresses being preferred to faster caresses ) . In the main task , the level of reward ( high , low , verylow ) received in each trial depended on both the announcement cue at the beginning ( high or low ) and the force exerted to obtain it ( verylow rewards were only obtained when participants exerted low effort , which linearly converted into low probability to obtain the announced reward ) . The number of trials in which high , low , and verylow rewards were obtained did not differ significantly across groups . Only a significant main effect of Reward Level was found ( F ( 2 , 763 ) =27 . 84 , p<0 . 001 ) , due to a greater number of high ( M = 33 . 07 , SD = 4 . 97 ) than low ( M = 29 . 85 , SD = 6 . 03 ) and verylow ( M = 17 . 14 , SD = 8 . 99 ) trials received , across all three drug groups and both reward types . As expected , ratings of wanting and liking , as well as effort exerted to obtained the cued rewards , decreased over the course of the experiment due to habituation and fatigue ( Figure 2—figure supplement 1 ) . This decrease was similar across drug groups . Nevertheless , the covariate Block ( recoded into first and second block separately by Reward Type ) was included in the analyses of behavioral and fEMG measures ( see below ) , to control for habituation and fatigue . The three groups of participants did not differ significantly in their maximum voluntary contraction ( MVC ) of the hand dynamometer , which was measured right before the main task and at the end of the main task , nor in their positive and negative mood measured with the PANAS at time of pill intake and 3 hr later ( all b < 0 . 6 , all t < 0 . 8 , all p>0 . 4; see Table 1 ) . Finally , the three groups of participants did not differ significantly in terms of possible side effects , which were self-reported at time of pill intake and 3 hr later ( see Table 1 for nausea scores ) . Drug effects were investigated on ratings of wanting provided at the beginning of each trial , on physical effort to obtain an announced reward , and on ratings of liking provided after having obtained a reward . Interactions with the factor Drug were only found for physical effort . Behavioral analyses on ratings of wanting ( Figure 3—figure supplement 1 , A–B ) resulted in an expected significant main effect of Reward Level ( F ( 1 , 128 . 0 ) =119 . 28 , p<0 . 001 ) , due to higher ratings of wanting for high reward ( M = 4 . 83 , SD = 4 . 31 ) compared to low reward ( M = 1 . 14 , SD = 4 . 46 ) , and a significant main effect of Block ( F ( 1 , 9653 . 7 ) =54 . 30 , p<0 . 001 ) due to decreasing wanting from the first ( M = 3 . 19 , SD = 4 . 70 ) to the second block ( M = 2 . 85 , SD = 4 . 83 ) . All other effects were not significant ( all F < 2 . 3 , all p>0 . 11 ) . To verify the lack of drug effects on ratings of wanting , we ran the same LMM using the full Bayesian method with the brms package ( Bürkner , 2017 ) . A normal prior with M = 0 and SD = 5 was defined for population-level ( fixed ) effects , and a half student-t prior with 3 degrees of freedom , M = 0 , and scaling parameter = 4 . 7 , was set for the standard deviation of subject-specific ( random ) effects . Results showed that neither amisulpride ( βmean = 1 , 95% Bayesian credible interval [−0 . 03 , 2 . 07] ) , nor naltrexone ( βmean = 0 . 07 , 95% Bayesian credible interval [−0 . 94 , 1 . 09] ) had credible main effects on ratings of wanting ( based on their respective 95% Bayesian credible interval crossing zero ) , nor did they interact with Reward Level or Reward Type ( all βmean < 0 . 3 , all 95% Bayesian credible interval crossing zero ) . In addition , a second Bayesian LMM was fitted without the main and interaction effects for the predictor Drug . The two models were compared using a Bayesian leave-one-out cross-validation ( LOO-CV; Vehtari et al . , 2017 ) . This revealed a greater predictive ability for the null model ( weight = 0 . 67 , averaging via stacking of predictive distributions ) than the full model ( weight = 0 . 33 ) , meaning that the null model is expected to be two times more accurate in predicting new data . One can thus conclude , that taking into account the drug administered does not improve the ability to predict participants’ ratings of wanting . The LMM on effort ( Figure 3—figure supplement 1 , C–D ) resulted in the expected significant main effect of Reward Level ( F ( 1 , 128 . 5 ) =54 . 41 , p<0 . 001 ) , due to stronger force applied for high ( M = 80 . 49 , SD = 22 . 35 ) than low rewards ( M = 71 . 74 , SD = 25 . 42 ) ; a significant main effect of Block ( F ( 1 , 7527 . 4 ) =175 . 49 , p<0 . 001 ) due to decreasing effort from the first ( M = 78 . 27 , SD = 23 . 79 ) to the second block ( M = 74 . 02 , SD = 24 . 65 ) ; and a significant Reward Type X Drug interaction ( F ( 2 , 128 . 4 ) =4 . 71 , p=0 . 01; Figure 3A ) reflecting lower effort for food in the amisulpride ( M = 74 . 98 , SD = 26 . 57 ) and naltrexone ( M = 73 . 51 , SD = 24 . 43 ) groups compared to the placebo ( M = 80 . 20 , SD = 22 . 41 ) group , but similar force across drug groups for touch ( amisulpride: M = 78 . 34 , SD = 25 . 14; naltrexone: M = 73 . 78 , SD = 23 . 15; placebo: M = 76 . 11 , SD = 23 . 51 ) . The Reward Level X Drug interaction ( Figure 3B ) reflected lower effort for low rewards in the amisulpride ( M = 71 . 67 , SD = 27 . 60 ) and naltrexone ( M = 67 . 90 , SD = 24 . 45 ) groups compared to Placebo ( M = 75 . 65 , SD = 23 . 60 ) , but failed to reach significance ( F ( 2 , 128 . 5 ) =2 . 95 , p=0 . 056 ) . All other effects were not significant ( all F < 0 . 9 , all p>0 . 4 ) . The same LMM on ratings of liking ( Figure 3—figure supplement 1 , E–F ) resulted in the main effect of Reward Level ( F ( 1 , 126 . 4 ) =150 . 55 , p<0 . 001 ) , with greatest liking of high rewards ( M = 5 . 02 , SD = 4 . 10 ) , followed by low rewards ( M = 1 . 79 , SD = 4 . 15 ) , and verylow rewards at the bottom ( M = −1 . 19 , SD = 3 . 89 ) . Decrease of liking over time was shown by a significant main effect of Block ( F ( 1 , 9400 . 7 ) =129 . 40 , p<0 . 001 ) , due to a decrease in liking from the first ( M = 2 . 90 , SD = 4 . 61 ) to the second block ( M = 2 . 25 , SD = 4 . 79 ) . All other effects were not significant ( all F < 1 . 1 , all p>0 . 34 ) . The lack of an effect of drug on liking was confirmed by fitting a Bayesian LMM ( the same priors were set as for the models on wanting ) . Neither amisulpride ( βmean = 0 . 78 , 95% Bayesian credible interval [−0 . 20 , 1 . 77] ) , nor naltrexone ( βmean = 0 . 25 , 95% Bayesian credible interval [−0 . 71 , 1 . 18] ) had main effects on ratings of liking , nor did they interact with Reward Level or Reward Type ( all βmean < 0 . 30 , all 95% Bayesian credible interval crossing zero ) . The full model had lower predictive ability ( weight = 0 . 002 ) than the model without main and interaction effects of the predictor Drug ( weight = 0 . 998 ) , as shown with LOO-CV . These Bayesian analyses strengthen the view , already conveyed by the frequentist LMMs , that neither drug affected explicit wanting or liking of both types of rewards in this study . To investigate drug effects on reward anticipation and reward consumption , facial EMG was analyzed in relation to trial-by-trial subjective ratings and effort ( as continuous predictors ) in four periods of interest ( see Figure 1 ) . In short , the following results were found . In the Pre-Effort anticipation period ( Figure 4 ) the corrugator was , as expected , relaxed for greater wanting and effort , and was more activated to food in the amisulpride and naltrexone groups compared to the placebo group . In the same time window , the zygomaticus muscle showed , as expected , stronger activation for greater wanting , however only in food trials . In the Post-Effort anticipation period , a non-significantly greater zygomaticus activation for greater wanting was found . In the Delivery phase , a Liking X Drug interaction was found in the zygomaticus muscle ( Figure 5 ) , reflecting the expected zygomaticus activation for greater liking in the placebo and ( to a lesser extent ) amisulpride group , while the opposite pattern of lower zygomaticus contraction for greater liking was found in the naltrexone group . Importantly , this interaction did not survive FDR correction ( p=0 . 09 ) but seems credible based on a Bayesian LMM . Finally , in the Relax window immediately following reward administration , the corrugator significantly relaxed for the most liked food rewards , but not touch rewards . For the corrugator muscle by Wanting , significant main effects of Reward Type ( F ( 1 , 267 . 9 ) =10 . 31 , p=0 . 01 ) , Wanting ( F ( 1 , 238 . 5 ) =7 . 75 , p=0 . 01 ) , and Block ( F ( 1 , 7867 . 4 ) =7 . 76 , p=0 . 01 ) were found . Activation of the corrugator was greater for food ( M = 116 . 35 , SD = 84 . 72 ) than touch ( M = 110 . 21 , SD = 60 . 63 ) and decreased , as expected , with increasing ratings of wanting ( slope b = −2 . 80; Figure 3A ) . A significant Drug X Reward Type interaction ( F ( 2 , 267 . 8 ) =4 . 08 , p=0 . 04 ) reflected ( Figure 4D ) greater corrugator activation to food than touch in the amisulpride group ( p=0 . 006; food: M = 119 . 12 , SD = 95 . 82; touch: M = 109 . 46 , SD = 53 . 44 ) and naltrexone group ( p=0 . 001; food: M = 120 . 00 , SD = 101 . 29; touch: M = 109 . 66 , SD = 67 . 25 ) , while the placebo group had similar activations across both reward types ( p=0 . 66; food: M = 110 . 30 , SD = 48 . 29; touch: M = 111 . 44 , SD = 59 . 87 ) . Corrugator activation to food was also significantly greater in the amisulpride and naltrexone groups compared to the placebo group ( p=0 . 03 and . 01 ) . For the corrugator muscle by Effort , we found a significant main effect of Reward Type ( F ( 1 , 277 . 2 ) =11 . 04 , p=0 . 008 ) , with greater activation for food ( M = 116 . 35 , SD = 84 . 72 ) than touch ( M = 110 . 21 , SD = 60 . 63 ) , a significant main effect of Effort ( F ( 1 , 207 . 9 ) =6 . 38 , p=0 . 04 ) due to greater corrugator relaxation with increasing levels of Effort ( b = -2 . 59; Figure 4B ) a significant main effect of Block ( F ( 1 , 7662 . 4 ) =5 . 72 , p=0 . 04 ) due to greater corrugator activation in the second ( M = 115 . 55 , SD = 87 . 5 ) compared to the first block ( M = 110 . 98 , SD = 56 . 4 ) , and a significant Drug X Reward Type interaction ( F ( 2 , 277 . 2 ) =4 . 03 , p=0 . 04 ) reflecting greater corrugator activation for food than touch in the amisulpride group ( p=0 . 007; food: M = 119 . 12 , SD = 95 . 82; touch: M = 109 . 46 , SD = 53 . 44 ) and naltrexone group ( p<0 . 001; food: M = 120 . 00 , SD = 101 . 29; touch: M = 109 . 66 , SD = 67 . 25 ) , while the placebo group had similar activations across both reward types ( p=0 . 72; food: M = 110 . 30 , SD = 48 . 29; touch: M = 111 . 44 , SD = 59 . 87 ) . Corrugator activation for food was also significantly greater in the naltrexone group compared to the placebo group ( p=0 . 03 ) . For the zygomaticus muscle by Wanting ( random slopes for the Reward Type X Effort interaction were removed to allow model convergence ) , a significant main effect of Reward Type ( F ( 1 , 125 . 9 ) =13 . 78 , p=0 . 001 ) was found , reflecting greater zygomaticus activation for food ( M = 138 . 79 , SD = 145 . 48 ) than touch ( M = 122 . 98 , SD = 130 . 67 ) . Moreover , greater wanting predicted zygomaticus contraction in food trials ( b = 5 . 6 ) but not in the touch trials ( b = −2 . 73 ) , as shown by a significant Reward Type X Wanting interaction ( F ( 1 , 2925 . 7 ) =6 . 62 , p=0 . 04; Figure 4C ) . All other effects were not significant ( all F < 0 . 8 , all p>0 . 68 ) . For the zygomaticus muscle by Effort ( random slopes for the Reward Type X Effort interaction were removed to allow model convergence ) , only a significant main effect of Reward Type was found ( F ( 1 , 129 . 2 ) =14 . 70 , p=0 . 001 ) , with greater zygomaticus activation to food ( M = 138 . 79 , SD = 145 . 48 ) compared to touch ( M = 122 . 98 , SD = 130 . 67 ) . All other effects were not significant ( all F < 1 . 2 , all p>0 . 81 ) . No significant effects were found for the corrugator muscle , neither by Wanting nor by Effort ( all F < 1 . 7 , all p>0 . 78 ) . For the Zygomaticus , a greater contraction for increasing levels of Wanting ( b = 6 . 82 ) was observed , but the effect fell short of significance ( F ( 1 , 186 . 9 ) =6 . 55 , p=0 . 08 ) . All other effects were not significant ( all F < 1 . 5 , all p>0 . 51 ) . Analysis of the corrugator resulted in a significant main effect of Reward Type ( F ( 1 , 121 . 9 ) = 8 . 21 , p=0 . 04 ) , due to greater corrugator activation in response to food ( M = 153 . 25 , SD = 216 . 96 ) than touch ( M = 117 . 23 , SD = 298 . 61 ) . All other effects were not significant ( all F < 2 . 4 , all p>0 . 48 ) . For the zygomaticus a significant main effect of Reward Type ( F ( 1 , 126 . 1 ) =77 . 97 , p<0 . 001 ) , was observed , due to greater zygomaticus activation in response to food ( M = 264 . 83 , SD = 233 . 77 ) than touch ( M = 125 . 21 , SD = 211 . 39 ) . A Liking X Drug interaction , falling short of significance after FDR correction ( F ( 2 , 132 . 2 ) =4 . 22 , p=0 . 09 , Figure 5 ) , was also found . Greater liking resulted , as expected , in greater zygomaticus activation in the placebo group ( b = 5 . 74 ) , and to a lesser extent also in the amisulpride group ( b = 0 . 56 ) . By contrast , the opposite pattern was found in the naltrexone group , with a negative slope ( b = −7 . 86 ) that was significantly different from the placebo group ( p=0 . 006 ) but not from the amisulpride group ( p=0 . 09 ) . The amisulpride and placebo group did not differ significantly between each other ( p=0 . 3 ) . To further probe the Liking X Drug interaction , we also ran a Bayesian LMM with the same predictors ( dummy coding was applied to the drug groups amisulpride and naltrexone , to compare them to placebo ) . A normal , and a half student-t prior were chosen for , respectively , population-level ( fixed ) , and group-level ( random ) effects . Results confirmed a credible difference , compared to placebo , in the effect of liking on the zygomaticus activation in the naltrexone group ( βmean = 13 . 62 , 95% Bayesian credible interval [−23 . 37 , –3 . 68] ) , but not in the amisulpride group ( βmean = -4 . 92 , 95% Bayesian credible interval [−14 . 99 , 5 . 09] ) . For the corrugator by Liking ( the random slope for the Reward Type X Liking interaction was removed to allow model convergence ) , significant main effects of Reward Type ( F ( 1 , 155 . 0 ) =20 . 36 , p<0 . 001 ) and Liking ( F ( 1 , 184 . 6 ) =12 . 41 , p=0 . 001 ) , and a significant Reward Type X Liking ( F ( 1 , 9231 . 8 ) =7 . 66 , p=0 . 01 ) interaction were found . The interaction reflected a significant corrugator decrease with greater liking for food rewards ( b = −23 . 2 ) but not for touch rewards ( b = −5 . 6 ) . For the zygomaticus by Liking ( the random slope for the Reward Type X Liking interaction was removed to allow model convergence ) , a significant main effect of Reward Type was found ( F ( 1 , 126 . 7 ) =144 . 25 , p<0 . 001 ) , reflecting greater zygomaticus contraction to food ( M = 211 . 17 , SD = 162 . 63 ) than touch ( M = 132 . 05 , SD = 208 . 32 ) . Zygomaticus contraction was overall greater in the second ( M = 166 . 22 , SD = 148 . 85 ) compared to the first block ( M = 175 . 58 , SD = 225 . 67 ) , as indicated by a significant main effect of Block ( F ( 1 , 9472 . 0 ) =7 . 44 , p=0 . 03 ) . No significant main or interaction effects for the factor Drug were found in the CS and ZM data ( all F < 1 . 7 , all p>0 . 19 ) . In line with animal models and recent human pharmacological studies , indicating that both the dopaminergic and the opioidergic systems underlie the motivation to obtain rewards ( Chelnokova et al . , 2014; Peciña and Smith , 2010; Weber et al . , 2016 ) , we observed a similar effect of the D2/D3 antagonist amisulpride and the non-specific opioid receptor antagonist naltrexone on the effort produced to obtain the announced reward , resulting in a reduction of applied force . Notably , and differently from our original hypothesis , the effect was most pronounced for the second-preferred ( low ) rewards , as indicated by a Drug x Reward Level interaction ( which however was not significant , p=0 . 056 ) . A possible explanation for this finding is that our food stimuli did not vary in caloric content ( i . e . the three reward levels were matched for fat and sugar ) . Therefore , individual preferences were derived from other mechanisms than energy value , possibly leading to a different effect of the drug on food reward processing ( Barbano et al . , 2009; Salamone et al . , 2007 ) . Another possible explanation is that high rewards are less susceptible to changes in their incentive salience when more options are available . Indeed , the majority of the studies in animals and humans have used only two reward levels , and have found that interference with dopaminergic or opioidergic transmission alters the outcome of cost/benefit analyses involving work-related response costs for the most valuable option ( Salamone et al . , 2007 ) . Our finding suggests a similar shift in cost/benefit that is possibly sensitive to a different experimental set-up ( Barbano et al . , 2009 ) . A similar effect of both amisulpride and naltrexone during food anticipation was observed in the implicit measure of fEMG: corrugator activation was greater in both drug groups , compared to placebo . Because frowning typically reflects a more negative ( or less positive ) reaction to rewards and emotional stimuli ( Fernández-Dols and Russell , 2017; Heller et al . , 2011; Lang et al . , 1993; Larsen et al . , 2003 ) , the observation that dopamine and opioid antagonists led to greater frowning might be interpreted as a reflection of less anticipated pleasure , independently of reward level , in these groups of participants . During reward consumption , only the opioidergic antagonist naltrexone had an effect on implicit hedonic facial reactions . Lower zygomaticus activation for greater liking was found in the naltrexone group ( with both frequentist and Bayesian LMMs ) , and this effect was significantly different from the placebo group , who instead showed the expected pattern of higher zygomaticus activation for greater liking . The amisulpride group showed the same pattern as the placebo group , although to a lesser extent . We interpret this finding as less smiling to liked rewards after administration of the opioid receptor antagonist naltrexone , as it parallels observations of fewer orofacial hedonic reactions to most preferred foods after opiodergic blockage in animals ( Smith and Berridge , 2007 ) . Zygomaticus activation is also known to increase for highly negative stimuli , in addition to positive stimuli ( Lang et al . , 1999; Larsen et al . , 2003 ) , which could suggest that greater zygomaticus activation with decreasing liking in the naltrexone group is due to disgust expressions , or other negative facial reactions to the least-liked rewards , rather than to a reduction in positive hedonic responses ( smiling ) . This , however , seems unlikely , as we only administered positive rewarding stimuli , and because drug groups did not differ in ( 1 ) initial reward preferences ( Figure 2 ) ; ( 2 ) the number of high , low , and verylow rewards received; ( 3 ) the ratings of wanting and liking of these rewards; ( 4 ) the change of ratings of liking over time ( Figure 2—figure supplement 1 ) ; and ( 5 ) amounts of nausea or other side-effects ( Table 1 ) . Moreover , this finding is unlikely to be explained by mouth movements related to food ingestion , because participants were instructed to swallow the food after the Delivery window , and no interaction with the factor Reward Type was found . Taken together , and partially in line with the behavioral results ( where the effects of the drugs were only observable for effort , but not for subjective wanting and/or liking ratings ) , fEMG data suggest a differential action of dopaminergic and opioidergic drug manipulation during the anticipation and consumption of rewards , with an effect of both drugs during the anticipation of rewards , but only of naltrexone during subsequent reward consumption . Interestingly , while the corrugator showed a general increase due to drug administration , zygomaticus activity was only affected in its relationship to subjective pleasure , and not in terms of overall activation . This differential pattern suggests that the two muscles , despite both tracking changes in hedonic value ( Lang et al . , 1993; Larsen et al . , 2003 ) , do not necessarily behave in a complementary , but rather in an independent way . This might also explain the heterogeneity of findings in previous studies , which often report an effect of reward valence on only one of the two muscles . Interestingly , drug effects were found on effort levels and fEMG , but not on subjective ratings of wanting and liking . While this may come as a surprise , it is in line with several previous studies , which have reported either null or weak effects of pharmacological interventions on pleasantness likings of affective touch ( Case et al . , 2016; Ellingsen et al . , 2014; Løseth et al . , 2019; Trotter et al . , 2016 ) . However , several studies have reported significant effects of naltrexone/morphine on food liking/consumption ( Bertino et al . , 1991; Eikemo et al . , 2016; Yeomans and Gray , 1996; Yeomans and Gray , 1997; but see Hetherington et al . , 1991 for a null effect ) . One possibly relevant difference between some of the previous work , and our study , is that we kept calory intake constant across food stimuli , and thus across participants . More research will be needed to clarify if drug-induced changes in reward pleasantness can be reliably assessed with explicit measures ( ratings ) for some types of rewards ( food ) , but instead require more implicit measures ( facial responses , effort ) for other types of rewards ( affective touch ) . Inclusion of both food and touch rewards allowed us to indirectly address the yet unresolved question ( Ruff and Fehr , 2014 ) , whether different types of rewards are processed by the same neurobiological systems ( as proposed by the ‘common currency hypothesis’ ) , or if representations coding for different rewards occur in distinct neural circuits , albeit on a common scale ( Grabenhorst and Rolls , 2011 ) . In particular social rewards , like affective touch , may constitute a separate class of stimuli , with a dedicated neural circuitry ( Rademacher et al . , 2010 ) , which can be specifically impaired , for example in people with autism spectrum disorders ( Chevallier et al . , 2012; Haggarty et al . , 2020 ) . Although the magnitude of the two types of rewards in terms of subjective ratings and effort was carefully matched ( Korb et al . , 2020 ) , most drug effects were either stronger or restricted to food trials , as indicated by significant Drug x Reward Type interactions for measures of effort to obtain the announced reward , and for corrugator activation during Pre-Effort anticipation . This suggests that the decision utility of touch and food rewards may not rely on the same neurochemical brain systems . However , fEMG responses to food were also stronger to begin with , as indicated by significant main effects of Reward Type for both muscles during the Pre-effort anticipation , Delivery , and Relax analysis windows . This might explain why only reactions to food rewards were modulated by opioidergic and dopaminergic antagonists . Another possible explanation for the less pronounced drug effects for touch is that responses to social rewards , including touch , might also depend on oxytocin and serotonin , in addition to dopamine and opioids ( Fischer and Ullsperger , 2017; Tang et al . , 2020; Walker and McGlone , 2013 ) . This is also suggested by the finding of higher pleasantness ratings and greater zygomaticus activation to touch after administration of 3 , 4-methylenedioxymethamphetamine ( MDMA ) , a drug that modulates serotonin , dopamine , and possibly oxytocin levels ( Bershad et al . , 2019; de Wit and Bershad , 2020 ) . Of note , drug effects on the activity of the zygomaticus muscle during reward consumption were similar for touch and food rewards , as indicated by the absence of a Drug X Reward Type interaction during the Delivery and Relax periods . To the best of our knowledge , this is the second study ( after Bershad et al . , 2019 ) to report a pharmacological modulation of hedonic responses to experienced touch ( for a weaker effect see also Case et al . , 2016 ) . Major differences in our study compared to previous work ( Ellingsen et al . , 2014; Løseth et al . , 2019 ) are the delivery of affective touch with the hand instead of a brush ( Ellingsen et al . , 2014 included touch by hand but wearing a glove ) , and allowing participants to select their preferred touch speed . Regarding the way touch was delivered , it is possible that the social saliency of the touch stimuli delivered through direct skin contact was enhanced , compared to when the touch is delivered with a brush , allowing us to detect subtle effects of the drug . Regarding the selection of the preferred touch speed , even if the majority of our subjects selected the slower speed as the preferred one , inter-individual differences were observed , and the implementation of a task that could account for those , may have helped to detect the effect of the drug . Further studies should investigate how such factors modulate drug responses to perceived affective touch . The current study is characterized by a number of limitations . First , only two types of stimuli ( food and touch ) were used to define social and non-social rewards , and only two neurochemical systems ( dopaminergic and opioidergic ) were challenged . However , from the animal and human literature , we know that other systems ( e . g . endocannabinoids , orexin , benzodiazepine , etc . ) also contribute to the motivational and hedonic components of reward processing ( Berridge and Kringelbach , 2015 ) . Future studies should therefore broaden the neuropharmacological investigation of social vs . nonsocial reward processing , by using other compounds and different rewarding stimuli , to allow for the generalizability of the findings to other social and nonsocial rewards , and to better understand their neurochemical basis . Furthermore , computational approaches will most likely be useful to reveal hidden psychological states subtending motivation and experienced pleasure , allowing to refine how drug administration acts on these two components ( Meyer et al . , 2019 ) . Second , we used a relatively low dose of amisulpride ( 400 mg ) . Amisulpride can both increase dopaminergic neurotransmission by blocking presynaptic autoreceptors when given at low doses ( 50–300 mg ) , and decrease it by blocking postsynaptic D2/D3 receptors when given at higher doses of 400–1200 mg ( Racagni et al . , 2004; Schoemaker et al . , 1997 ) . The used dose of 400 mg is at the lower end of postsynaptically active high doses ( Rosenzweig et al . , 2002 ) and occupies ∼70% of D2 receptors when given for 2 weeks ( Meisenzahl et al . , 2008 ) . We decided on 400 mg of amisulpride based on previous studies in humans ( e . g . Weber et al . , 2016 ) , to obtain postsynaptic dopaminergic effects , while ensuring the safety and well-being of our participants , as well as allowing both participants and experimenters to remain in the dark about the type of compound or placebo administered ( double blinding ) . The effects we observed ( e . g . less effort ) are in line with amisulpride’s antagonistic action on postsynaptic D2/D3 receptors . Upon availability of drug compounds that more strongly modulate the dopaminergic systems with minimal side effects , future studies should , however , explore dose-dependent changes in human reward processing . Third , we used a cross-sectional design for drug/placebo administration . A within-subjects design would certainly have resulted in greater statistical power . However , this would have come with the cost of even greater habituation to the rewards . Fourth , the study suffered from a lack of power to detect small effects . We had modeled the sample size on a previous study using the same drugs and doses ( Weber et al . , 2016 ) . However , Weber et al . , 2016 only found relatively small drug effects , and several other studies have failed to show effects of a pharmacological modulation of the opioid , serotonin , or oxytocin systems on the liking of affective touch ( Ellingsen et al . , 2014; Løseth et al . , 2019; Trotter et al . , 2016 ) . This reveals the difficulty of uncovering the neurochemical basis of reward processing in humans and suggests that larger sample sizes should be used in future pharmacological studies to investigate the neurochemical bases of touch and other rewards . We report pharmacological evidence in healthy human volunteers , across several measures including the monitoring of facial expressions with fEMG , about the role of the dopaminergic system for the motivational component and of the opioidergic system for both motivational and hedonic components of reward processing . The effort to obtain a reward and valenced facial reactions during reward anticipation were both modulated by the administration of dopaminergic or opioidergic antagonists . By contrast , facial reactions during reward experience were only altered by the opioidergic antagonist , suggesting neurochemical differences underlying hedonic expressions during anticipation and experience of pleasure . Explicit ratings of reward wanting and liking were not modulated by either drug . This constitutes the first demonstration of this kind in adult humans , using an operationalization of reward closely resembling previous animal research , and it suggests that the neurochemical regulation of pleasure ( as indicated by hedonic facial reactions ) is phase-specific , depending on whether the reward is anticipated or experienced . The finding that most drug effects were either stronger for , or restricted to , food trials may indicate different neurochemical brain mechanisms for social and nonsocial rewards . This point however requires further investigation via brain imaging or more direct measures of brain activity in addition to pharmacological challenges tailored to investigate the role of different neurochemical systems in the processing of social versus nonsocial rewards . Based on previous work that had used the same compounds and doses ( Weber et al . , 2016 ) , we aimed at collecting data from 40 participants per group or more . The final study sample included 131 volunteers ( 88 females ) aged 18–35 years ( M = 23 . 3; SD = 3 . 5 ) . In the amisulpride group , blood concentrations of the drug ( measured 5 hr after intake ) were in or above the therapeutic range ( blood samples missing for six people ) . Specifically , the minimum was 212 ng/mL , and 19 participants were above 604 ng/mL . All participants reported being right-handed , to smoke less than five cigarettes daily , to have no history of current or former drug abuse , to like milk and chocolate , not to suffer from diabetes , lactose intolerance , lesions or skin disease on the left forearm , and to be free of psychiatric or neurological disorders . Participants’ average Body Mass Index ( BMI ) was 22 . 6 ( SD = 2 . 5 , range 17 . 7–29 . 3 ) . To reduce the chances that social touch would be perceived as a sexual reward , the touch stimulation was always carried out by a same-sex experimenter ( see Procedure ) , and only participants who reported to be heterosexual were included . The study was approved by the Ethical Committee of the Medical University of Vienna ( EK N . 1393/2017 ) and was performed in line with the Declaration of Helsinki ( World Medical Association , 2013 ) . Participants signed informed consent and received monetary compensation of 90€ . Three stimuli with identical fat and sugar content ( 1 . 5 g fat , 10 g of sugar per 100 g ) were used as food rewards: milk , chocolate milk , and a 4:1 mix of milk and chocolate milk . Tap water served for rinsing at the end of each trial . The initial stimulus temperature of these liquids was kept constant ( ~4°C ) across participants . Stimulus delivery was accomplished through computer-controlled pumps ( PHD Ultra pumps , Harvard Apparatus ) attached to plastic tubes ( internal ø 1 , 6 mm; external ø 3 , 2 mm; Tygon tubing , U . S . Plastic Corp . ) , which ended jointly on an adjustable mount positioned about 2 cm in front of the participant’s mouth . In each trial , 2 mL of liquid was administered for 2 s . Overall , including stimulus pretesting ( see Procedure ) , participants consumed 196 mL of liquids , composed of 98 mL of water , and 98 mL of sweet milk with different concentrations of chocolate aroma ( depending on effort , see below ) . Touch rewards consisted of gentle caresses over a previously-marked 9-cm area of the participant’s forearm ( measurement started from the wrist towards the elbow ) . Three different caressing frequencies , chosen based on the literature and pilot testing , were applied for 6 s by a same-sex experimenter: 6 cm/s , 21 cm/s , and 27 cm/s . To facilitate stroking , the stimulating experimenter received extensive training and , in each trial , heard rhythmic sounds , indicating the rhythm for stimulation , through headphones . After cleansing of the corresponding face areas with alcohol , water , and an abrasive paste , reusable Ag/AgCl electrodes with 4 mm inner and 8 mm outer diameter were attached bipolarly according to guidelines ( Fridlund and Cacioppo , 1986 ) on the left corrugator supercilii ( corrugator ) and the zygomaticus major ( zygomaticus ) muscles . A ground electrode was attached to the participants' forehead and a reference electrode on the left mastoid . The EMG data were sampled at 1200 Hz with impedances below 20 kOHM using a g . USBamp amplifier ( g . tec Medical Engineering GmbH ) and the software Matlab ( MathWorks , Inc ) . A monocentric , randomized , double-blind , placebo-controlled , three-armed study design was used . The study took place in the Department of Psychiatry and Psychotherapy at the Medical University of Vienna . Participants visited the laboratory for the first visit ( T0 ) in which they received a health screening , followed by a second visit ( T1 ) that included oral drug intake and the experiment described here . Pharmacological dosage , and length of waiting time after drug intake ( 3 hr ) , were modeled on previous work ( Weber et al . , 2016 ) , and on the drug’s pharmacodynamics . Amisulpride reaches the first peak in serum after 1 hr , and a second ( higher ) peak after approximately 4 hr . The elimination half-life is 12 hr ( Rosenzweig et al . , 2002 ) . At doses of 400 mg or higher , amisulpride acts as a postsynaptic D2/D3 receptor antagonist and thus results in lower dopaminergic action ( Racagni et al . , 2004; Schoemaker et al . , 1997 ) . Naltrexone reaches maximal concentration in plasma after 1 hr , has an elimination half-life in plasma of approximately 4 hr , and is completely cleared from plasma after 96 hr ( Meyer et al . , 2019 ) . Importantly , up to 90% of mu-opioid receptors in the brain remain blocked by naltrexone after 48 hr , and partial receptor blockade could be shown up to 168 hr after intake ( Lee et al . , 1988 ) . Participants came to T1 with an empty stomach ( it was morning , and they had been instructed not to eat in the preceding 6 hr ) , filled out the PANAS questionnaire , tested negative ( or were excluded ) on a urine drug screen sensitive to opiates , amphetamine , methamphetamine , cocaine ( among other things ) , and then received a capsule filled with either 400 mg of amisulpride ( Solian ) , 50 mg of naltrexone ( Dependex ) , or 650 mg of mannitol ( sugar ) from the study doctor . All capsules looked identical from the outside , and neither participants nor the experimenters were informed of their content . Drug intake was followed by a waiting period , EMG preparation , and task instructions . The experiment comprised two tasks following procedures described elsewhere ( Korb et al . , 2020 ) . The main task started 3 hr after pill intake . Participants were seated at a table and comfortably rested their left forearm on a pillow . A curtain blocked their view of the left forearm and the rest of the room . This was particularly relevant for touch trials , in which one of two same-sex experimenters applied the touch rewards to the participant’s left forearm . Two experimenters were always present during testing , to limit the influence of participants’ experimenter preferences , and to allow participants to better concentrate on the ( touch ) stimuli . Participants first completed a short task , in which they experienced and individually ranked three food rewards , and separately three touch rewards , presented randomly in sets of three of the same reward type . In the main task , which started 3 hr after pill intake , the previously most liked stimuli were used as ‘high’ rewards , the stimuli with medium liking as ‘low’ rewards , and the least liked stimuli were used as ‘verylow’ rewards . To calibrate the dynamometer , the MVC was established right before the short task , by asking participants to squeeze the dynamometer ( HD-BTA , Vernier Software and Technology , USA ) with their right hand as hard as possible three times , each lasting 3 s . The average MVC ( peak force in newtons across all three trials ) was 212 ( SD = 80 . 4 ) and did not differ significantly between drug groups , as tested by linear regression ( β = 1 . 6 , SE = 8 . 68 , t = 0 . 19 , p=0 . 85 ) . After calibration of the dynamometer , EMG electrodes were attached , participants received detailed instructions , and completed four practice trials ( two per reward type ) . The main task included four experimental blocks with 20 trials each . Each block contained either food or touch trials , and the blocks were interleaved ( ABAB or BABA ) in a counterbalanced order across participants . Each trial included the following steps ( Figure 1; see Figure 1—figure supplement 1 and Supporting Information for all elements of a trial ) : ( 1 ) a picture announcing the highest possible reward ( high or low , 3 s ) , ( 2 ) a continuous scale ranging from ‘not at all’ to ‘very much’ to rate ( without time limit ) wanting of the announced reward ( ratings were converted to a Likert scale ranging from −10 to +10 ) , ( 3 ) a 4 s period of physical effort , during which probability of receiving the announced reward was determined by the amount of force exerted by squeezing the dynamometer with the right hand , while receiving visual feedback ( sliding average of 1 s , as percentage of the MVC ) , ( 4 ) a picture announcing the obtained reward ( 3 s for food , 7 . 3 s for touch ) , which could be high , low , or – if insufficient effort had been exerted – verylow ( the greater participants’ effort , the higher the probability of obtaining the announced reward ) , ( 5 ) a phase of reward delivery ( 2 s for food , 6 . 5 s for touch – this difference in timing was necessary to obtain sufficiently long tactile stimulation , while keeping the overall trial duration similar across reward types ) , ( 6 ) for food trials instructions to lean back and swallow the obtained reward ( duration 3 s ) , ( 7 ) a relaxation phase ( 5 s ) , and ( 8 ) a continuous scale to rate the liking of the obtained reward . In food trials , participants then received water for mouth rinsing . In both reward types , trials ended with a blank screen for 3–4 s . The last four trials in each block did not require pressing of the dynamometer . These trials were added to the design , in case participants would never press at all , which did not happen for any participant . Trials without pressing were kept in the data , as removing them from analyses reduced power but did not change the pattern of results . After each block participants were allowed to take a short break . Both tasks were run on a desktop computer with Windows seven using MATLAB 2014b and the Cogent 2000 and Cogent Graphics toolboxes and presented on an LCD monitor with a resolution of 1280 × 1024 pixels . The positive and negative affect schedule ( PANAS; Watson et al . , 1988 ) , and a questionnaire assessing nausea and 50 other side effects , was filled out twice at the main laboratory visit: just before pill intake , and 3 hr later . Levels of amisulpride ( ng/mL ) were measured in blood samples taken 5 hr after pill intake ( after both tasks ) . Data and analysis scripts are available online ( https://osf . io/vu8dz ) . Group comparisons for age , BMI , MVC , PANAS scores , and side effects , were made with linear regressions using the lm ( ) function . Differences in the ranking of rewards across drug groups were tested with separate ordinal regressions by Reward Type ( food , touch ) , using the package ordinal . All other analyses were done with linear mixed-effects models ( LMMs ) , fitted through restricted maximum likelihood ( REML ) estimation , using the lmer ( ) function of the lmerTest package in R ( which adds p values to the lme4 output; Bates and Maechler , 2014; R Development Core Team , 2019 ) , and with Helmert contrast coding . In comparison to ANOVAs , LMMs reduce Type-I errors and allow for a better generalization of findings ( Judd et al . , 2012 ) . To control for the effect of time – possibly inducing fatigue and/or habituation ( Figure 2—figure supplement 1 ) – the four blocks were recoded to two blocks by Reward Type and entered as covariates to the LMMs . Figures ( except Figure 1 ) were created in R using the packages ggplot2 , ggpirate , and cowplot . Behavioral data were analyzed in the following manner . Outlier trials were defined as those with a rating of wanting , rating of liking , or amount of exerted force , which was greater/smaller than the subject’s mean +/- 2 times the subject’s standard deviation . This led to an average rejection of 6 . 56 trials per participant ( SD = 3 . 71 ) . The total number of excluded trials did not differ significantly between groups ( t ( 133 ) = −1 . 28 , p=0 . 20 ) . For each behavioral dependent variable ( ratings of wanting and liking , and effort ) , a LMM was fitted with the fixed effects Reward Type ( food , touch ) , Reward Level ( high , low , verylow ) , Drug ( amisulpride , naltrexone , placebo ) , their interactions , and with Block ( first , second ) as a covariate . Categorical predictors were centered through effect coding , and by-subject random intercepts and slopes for all within-subjects factors and their interactions were included as random effects ( unless the model did not converge , in which case the random-effects structure was gradually simplified , e . g . by first dropping the interaction among within-subjects factors ) . Type-III F-tests were computed with the Satterthwaite degrees of freedom approximation . We report all statistically significant ( p<0 . 05 ) effects , and non-significant effects with p<0 . 1 that are of interest because related to the main hypotheses , as Anova ( ) outputs . Model tables showing all fixed and random effects can be found in the Supporting Information . Due to technical failure , one participant lacked EMG data entirely , and another participant lacked the EMG for half of the trials . The EMG data were pre-processed in Matlab R2018a ( www . themathworks . com ) , partly using the EEGLAB toolbox ( Delorme and Makeig , 2004 ) . A 20 to 400 Hz bandpass filter was applied , then data were rectified and smoothed with a 40 Hz low-pass filter . Epochs were extracted focusing on periods of reward anticipation ( Pre-Effort and Post-Effort anticipation ) and reward consumption ( Delivery and subsequent Relax ) . EMG was averaged over time-windows of one second , with exception of the 6 . 5-seconds-long period of touch Delivery , which was averaged over five windows of 1 . 3 s each , to obtain the same number of windows as for food delivery . We excluded for each participant trials on which the average amplitude in the baseline period ( 1 s during fixation ) of the corrugator or zygomaticus muscles was lower than M−2SD , or higher than M+2SD ( M = average amplitude over all trials' baselines for the respective muscle and participant ) . On average , this led to the rejection of 7 . 7% of trials per participant ( SD = 2 . 5 ) . EMG analyses were carried out in four periods of interest: Pre-effort anticipation during reward announcement at the beginning of each trial ( 3 s ) , Post-effort anticipation during the announcement of the gained reward ( 3 s ) , Delivery ( 5 s for food and 6 . 5 s for touch , both averaged to five 1 s time windows ) , and Relax ( 5 s ) . For each trial , values in these epochs were expressed as percentage of the average amplitude during the fixation cross at the beginning of that trial . For the Pre- and Post-Effort anticipation periods , separate LMMs were fitted by muscle , with the fixed effects Drug ( amisulpride , naltrexone , placebo ) , Reward Type ( food , touch ) , and either trial-by-trial Wanting or Effort ( these were continuous predictors ) , and all interactions . During the Post-Effort anticipation period , participants could receive the information that they were going to obtain the verylow reward , to which the preceding ratings of wanting and effort did not apply . Because this may have been frustrating for participants , we also carried out analyses excluding trials , in which verylow rewards were obtained . As the results did not change , we kept all trials . For the Delivery and Relax periods , separate LMMs on all trials were fitted by muscle , with the fixed effects Drug , Reward Type , and Liking . In all LMMs Wanting , Effort , and Liking were centered and scaled by subject , and Block ( first , second ) was added as a covariate to control for the effects of fatigue or habituation . We controlled for the false discovery rate ( FDR ) associated with multiple testing of the EMG data using the Benjamini-Hochberg method ( Benjamini and Hochberg , 1995 ) . Model tables with un-corrected p-values can be found in the Supporting Information .	Studies in rats and other species have shown that two chemical messengers in the brain regulate how much an animal desires a reward , and how pleasant receiving the reward is . In this context , chemicals called opioids control both wanting and enjoying a reward , whereas a chemical called dopamine only regulates how much an animal desires it . However , since these results were obtained from research performed on animals , further studies are needed to determine if these chemicals play the same roles in the human brain . Korb et al . show that the same brain chemicals that control reward anticipation and pleasure in rats are also at work in humans . In the experiment , 131 healthy volunteers received either a drug that blocks opioid signaling in the brain , a drug that blocks dopamine signaling , or a placebo , a pill with no effect . Then , participants were given , on several occasions , either sweet milk with chocolate or a gentle caress on the forearm . Participants rated how much they wanted each of the rewards before receiving it , and how much they liked it after experiencing it . To measure their implicit wanting of the reward , participants also pressed a force-measuring device to increase their chances of receiving the reward . Additionally , small electrodes measured the movement of the volunteer’s smiling or frowning muscles to detect changes in facial expressions of pleasure . Volunteers taking either drug pressed on the device less hard than the participants taking the placebo , suggesting they did not want the rewards as much , and they frowned more as they anticipated the reward , indicating less anticipatory pleasure . However , only the volunteers taking the opioid-blocking drug smiled less when they received a reward , indicating that these participants did not get as much pleasure as others out of receiving it . These differences were most pronounced when volunteers looked at or received the sweet milk with chocolate . This experiment helps to shed light on the chemicals in the human brain that are involved in reward-seeking behaviors . In the future , the results may be useful for developing better treatments for addictions .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2020	Dopaminergic and opioidergic regulation during anticipation and consumption of social and nonsocial rewards
Charcot–Marie-Tooth disease type 2A ( CMT2A ) is an untreatable childhood peripheral neuropathy caused by mutations of the mitochondrial fusion protein , mitofusin ( MFN ) 2 . Here , pharmacological activation of endogenous normal mitofusins overcame dominant inhibitory effects of CMT2A mutants in reprogrammed human patient motor neurons , reversing hallmark mitochondrial stasis and fragmentation independent of causal MFN2 mutation . In mice expressing human MFN2 T105M , intermittent mitofusin activation with a small molecule , MiM111 , normalized CMT2A neuromuscular dysfunction , reversed pre-treatment axon and skeletal myocyte atrophy , and enhanced axon regrowth by increasing mitochondrial transport within peripheral axons and promoting in vivo mitochondrial localization to neuromuscular junctional synapses . MiM111-treated MFN2 T105M mouse neurons exhibited accelerated primary outgrowth and greater post-axotomy regrowth , linked to enhanced mitochondrial motility . MiM111 is the first pre-clinical candidate for CMT2A . Charcot–Marie-Tooth disease ( CMT ) describes a family of genetically diverse and clinically heterogeneous peripheral neuropathies ( Fridman et al . , 2015 ) . Type 2A CMT ( CMT2A ) is caused by mutations of the mitochondrial fusion protein , mitofusin 2 ( MFN2 ) ( Züchner et al . , 2004 ) , and is distinguished from other CMT subtypes by onset of neuromuscular signs in early childhood and progressive loss of neuromuscular coordination and strength in arms throughout the first two decades of life , thought to be the consequence of dying-back of long peripheral nerves ( Fridman et al . , 2015; Feely et al . , 2011; Bombelli et al . , 2014; Yaron and Schuldiner , 2016; Berciano et al . , 2017 ) . Because there are currently no disease-modifying treatments , CMT2A is managed with braces , wheelchairs , and social support . Over 100 different dominant missense MFN2 mutations are implicated in CMT2A ( Beręsewicz et al . , 2018 ) . MFN2 and related MFN1 are nuclear-encoded dynamin-family GTPases located at the mitochondrial outer membrane-cytosol interface where they promote mitochondrial fusion essential to mitochondrial respiratory function and repair ( Chan , 2012 ) . Dominant inhibition by MFN2 mutants of mitochondrial fusion ( Chen and Chan , 2006; Pareyson et al . , 2015 ) , mitophagy ( Rizzo et al . , 2016; Filadi et al . , 2018 ) , and/or neuronal mitochondrial transport ( Pareyson et al . , 2015; Baloh et al . , 2007; Crunkhorn , 2018 ) are proposed to evoke neuronal degeneration in CMT2A . Because CMT2A is an autosomal dominant genetic condition , gene editing could potentially correct causal MFN2 mutant alleles , but the large number of different causal CMT2A MFN2 mutations complicates an editing approach . Alternately , forced expression of normal mitofusins could oppose mutant MFN2 dysfunction , as demonstrated in a recent study in transgenic mice ( Zhou et al . , 2019 ) . However , MFN gene therapy would be difficult to discontinue or reverse if postulated adverse effects of MFN overactivity are encountered ( El Fissi et al . , 2018 ) . Here , we describe a therapeutic approach to CMT2A that is agnostic to MFN2 genotype and does not require genetic manipulation: intermittent or ‘burst’ activation of endogenous normal mitofusins . Pharmacological mitofusin activators improved mitochondrial morphology , fitness , and motility in human and mouse CMT2A neurons in vitro . Daily administration of a short acting mitofusin activator to mice with late stage CMT2A reversed neuromuscular dysfunction . Mechanistically , neuronal repair and regeneration were linked to enhanced mitochondrial transport to , and mitochondrial occupation within , axonal termini . Reversal of pre-existing CMT2A neuromuscular degeneration in vivo has not previously been achieved by any means , and provides a powerful rationale for advancing mitofusin activators to first in human trials . One of the central features of CMT2A is the large number of different MFN2 mutations that provoke the syndrome . Common MFN2 GTPase and coiled-coiled domain mutations induce more severe and earlier onset disease , whereas rare carboxy terminal domain mutations confer later onset and milder disease ( Feely et al . , 2011; Stuppia et al . , 2015 ) . We compared mitochondrial phenotypes in cells from four CMT2A patients , two having MFN2 mutations within the canonical dynamin/Fzo-like GTPase domain ( MFN2 T105M in the G1 motif and MFN2 R274W between the G4 and G5 motifs ) , and two with mutations in the MFN2 coiled-coiled helix bundle core ( MFN2 H361Y and R364W ) . ( Figure 1—figure supplement 1 ) . Donor patient characteristics are in Table 1 . To avoid loss of some CMT2A-associated mitochondrial phenotypes in iPSC-derived neurons ( Rizzo et al . , 2016; Saporta et al . , 2015 ) , we directly reprogrammed CMT2A patient fibroblasts into motor neurons via microRNA-mediated neuronal conversion ( Figure 1b; Abernathy et al . , 2017 ) . Reprogramming efficiency was similar between CMT2A and control patient fibroblasts:>90% neurons ( measured as β-III tubulin staining ) , and >85% motor neurons ( measured as β-III tubulin , HB9/MNX1 co-staining ) ( Figure 1—figure supplement 2 ) . Compared to neurons reprogrammed from individuals with no evident disease at the time of sampling and who had none of the tested MFN2 mutations by Sanger sequencing ( ‘normal’ ) , all four CMT2A motor neuron lines exhibited fragmented mitochondria ( decreased mitochondrial aspect ratio; length/width ) that is a consequence of impaired fusion in this context Franco et al . , 2016; accompanying mitochondrial depolarization reflected characteristic functional impairment ( Figure 1c; Crowley et al . , 2016 ) . Moreover , all four CMT2A motor neuron lines exhibited abnormal mitochondrial transport through axons , with diminished proportion and velocity of motile mitochondria ( Figure 1c ) . Mitochondrial fragmentation , respiratory dysfunction , and dysmotility observed in reprogrammed neurons are prototypical features of CMT2A ( Baloh et al . , 2007; Zhou et al . , 2019; Verhoeven et al . , 2006; Rocha et al . , 2018 ) . Dominant inhibition of normal MFN1 and MFN2 by CMT2A MFN2 mutants produces an imbalance between mitochondrial fission and fusion that underlies mitochondrial pathology in CMT2A ( Zhou et al . , 2019 ) . This dynamic imbalance can be reversed in transfected mouse cells and in vivo mouse models by forced overexpression of normal MFN1 or MFN2 ( Zhou et al . , 2019; Detmer and Chan , 2007 ) . We posited that pharmacological activation of normal endogenous human MFN1 and MFN2 would also reverse mitochondrial abnormalities in CMT2A patient motor neurons . Chimera C is one of a new class of direct mitofusin activators that promotes conformational activation of MFN1 and MFN2 , thereby stimulating endogenous mitofusins to improve mitochondrial dysmorphology and dysfunction ( Rocha et al . , 2018; Dang et al . , 2020 ) . Chimera C ( 100 nM , 48 hr ) enhanced mitochondrial fusion ( i . e . it increased aspect ratio ) and improved respiratory function ( i . e . it reversed mitochondrial depolarization ) in cells lacking either MFN1 or MFN2 , but had no effects in cells lacking both mitofusin targets ( Figure 1—figure supplement 3 ) . Chimera C ( 100 nM , 48 hr ) also improved mitochondrial aspect ratio , depolarization , and motility in all four CMT2A patient motor neuron lines ( Figure 1c ) . Children with CMT2A are typically healthy during early years , but develop signs of neuromuscular dysfunction during the mid first decade of life . Neurogenic distal limb muscular atrophy is progressive until the end of the second decade , at which time the disease stabilizes; longevity is normal , but disability is permanent ( Fridman et al . , 2015; Feely et al . , 2011 ) . No mouse models of CMT2A recapitulate all of these key clinical features in the absence of confounding developmental phenotypes ( Zhou et al . , 2019; Detmer et al . , 2008; Cartoni et al . , 2010; Bannerman et al . , 2016; Dorn , 2020 ) . Therefore , a prerequisite for proof-of-concept testing of mitofusin activation in vivo was to generate a mouse CMT2A model having greater similarity to the human condition . By combining Rosa26 <fs-MFN2 ( T105M ) > ( Bannerman et al . , 2016 ) and Mnx1-Cre ( HB9 ) ( Yang et al . , 2001 ) alleles ( Figure 2a ) we drove human MFN2 T105M expression in mouse neurons ( Figure 2b; CMT2A mouse ) . Neuromuscular functional integrity over time was assessed as the duration mice could walk on an elevated accelerating rotating cylinder without falling off ( RotaRod latency ) . RotaRod latency of CMT2A mice was normal at 10 weeks of age , progressively declined thereafter , and stabilized beyond 30 weeks ( Figure 2c ) . As in clinical CMT2A , axonal mitochondria of MFN2 T105M mice were fragmented with disorganized cristae ( Sole et al . , 2009; Figure 2d ) . Neuroelectrophysiological testing of CMT2A patients characteristically reveals reduced compound muscle action potentials ( CMAP ) with normal nerve conduction velocities ( Berciano et al . , 2017; Harding and Thomas , 1980 ) . Recapitulating this clinical finding , sciatic nerve-tibialis muscle CMAP amplitudes of 50-week-old MFN2 T105M mice were diminished with no change in signal latency , which reflects conduction velocity ( Figure 2e ) . Tibialis myofiber atrophy and loss of large axons without demyelination in the MFN2 T105M mouse ( vide infra ) also mimicked clinical CMT2A ( Verhoeven et al . , 2006; Muglia et al . , 2001; Neves and Kok , 2011 ) . To further evaluate the relevance of the MFN2 T105M mouse to human CMT2A , dorsal root ganglion ( DRG ) sensory neurons were isolated and placed in culture , the MFN2 T105M transgene induced with Adeno-Cre , and neurons assayed for the mitochondrial pathologies delineated in reprogrammed CMT2A patient motor neurons ( vide supra ) . CMT2A-associated abnormalities in axon mitochondrial aspect ratio and transport ( Figure 2f ) and polarization status ( Figure 2—figure supplement 1 ) were each mimicked in mouse CMT2A DRGs . As in reprogrammed human CMT2A motor neurons , mitofusin activation improved these abnormalities ( Figure 2f and Figure 2—figure supplement 1 , compare to Figure 1c ) . Collectively , the above results show that activating mitofusins can improve multiple mitochondrial abnormalities manifested by cultured human and mouse CMT2A neurons . To determine whether benefits of mitofusin activation in cultured neurons would translate to therapeutic effects on neuromuscular dysfunction in CMT2A we contemplated an in vivo trial in our CMT2A mouse . However , Chimera C is rapidly degraded by the liver and undergoes first-pass metabolism , making it impractical for in vivo studies ( Dang et al . , 2020 ) . We therefore evaluated in vivo efficacy of mitofusin activation in CMT2A using MiM111 , a structurally distinct compound having a mitofusin activation profile similar to Chimera C ( Figure 1—figure supplement 3 ) , but which is metabolically stable with good nervous system bioavailability ( Dang et al . , 2020 ) . We hypothesized that intermittent or ‘burst’ mitofusin activation ( a dosing schedule that reversed mitochondrial dysfunction for <12 hr each day ) ( Figure 3—figure supplement 1 ) would confer therapeutic benefits by cyclically enhancing mitochondrial fitness and transport , while minimizing the possibility of mitofusin toxicity that might occur with constant mitofusin activation ( El Fissi et al . , 2018; Meyer et al . , 2017 ) . Based on a minimum effective MiM111 plasma concentration of 30 ng/ml ( Dang et al . , 2020 ) and a plasma t1/2 of 2 . 3 hr with Cmax of 24 , 000 ng/ml after intramuscular administration of 30 mg/kg ( Figure 3—figure supplement 1 ) , we estimated that daily IM dosing would reverse mitochondrial dysmotility in CMT2A mice for ~12 hr out of every 24 hr . Indeed , Figure 3—figure supplement 1 show that mitochondrial motility in sciatic nerve axons of MFN2 T105M mice was normalized 4 hr after a single intramuscular injection of MiM111 ( 30 mg/kg ) , declined by approximately half after 12 hr , and returned to CMT2A baseline after 24 hr . If CMT2A neuron die-back is reversible then burst mitofusin activation should improve neuromuscular degeneration in MFN2 T105M mice who had progressed to the severe and stable CMT2A phenotype . To test this notion , 50-week-old MFN2 T105M mice and littermate controls were randomized to receive daily MiM111 or its vehicle . Researchers blind to genotype and treatment group performed Rotarod and neuro-electrophysiological testing after 4 and 8 weeks ( Figure 3b ) . The characteristic decreases in RotaRod latency and CMAP amplitude in MFN2 T105M mice ( see Figure 2c and e ) were reversed 4 weeks after MiM111 treatment and remained near normal after 8 weeks ( Figure 3c and d ) ; MiM111 had no effect on control mice ( Figure 3—figure supplement 2 ) . Compared to sciatic nerves of vehicle-treated MFN2 T105M CMT2A mice , MiM111 treatment reduced axon damage ( Figure 4a ) , increased axon diameter ( Figure 4b ) , and increased staining for superior cervical ganglion 10 ( SCG10; a marker of neuron regeneration ) ( Shin et al . , 2014; Figure 4—figure supplement 1 ) . These findings suggest that mitofusin activation reversed CMT2A-induced neuronal degeneration . Skeletal myocytes of CMT2A mouse tibialis muscle innervated by the sciatic nerve were abnormally small ( Figure 4c ) , reflecting neurogenic muscle atrophy ( because the MFN2 T105M transgene is directed by neuron-specific HB9-Cre ) . In agreement with muscle atrophy being a secondary effect , skeletal myocyte mitochondria of CMT2A mice appeared structurally normal ( Figure 4—figure supplement 1 ) compare to sciatic nerve axon mitochondria in Figures 2d and 4b . Therefore , normalization of tibialis myocyte diameter after mitofusin activation ( Figure 4c ) indicates restoration of neuromuscular integrity . Collectively , the above findings provide indirect support for the idea that CMT2A mice suffer from distal neuron dieback that can be reversed by activating mitofusins . Reasoning that decreased neuromuscular junction density in CMT2A mice would constitute direct evidence for dieback , we quantified neuromuscular junctional synapses containing receptors for the neurotransmitter acetylcholine ( AchR ) in tibialis muscles of CMT2A mice . Compared to normal mice , CMT2A mice had ~50% fewer synaptic junctions/myocyte , which was reversed after mitofusin activator treatment ( Figure 4—figure supplement 1 ) . Strikingly , mitochondrial occupancy of vehicle-treated CMT2A neuromuscular synaptic junctions was also reduced by ~half compared to normal mice , and was normalized by MiM111 treatment ( Figure 4d ) . Because mitochondrial transport can play a central role in neuron repair and regeneration ( Sheng , 2017 ) , the observation that MiM111 promoted mitochondrial localization within terminal neuromuscular synaptic junctions provided a plausible mechanistic link between mitofusin activation , mitochondrial motility , neuronal regrowth , and reversal of neuromuscular dysfunction in this preclinical CMT2A model . Reversal of CMT2A-induced distal neuron die back implies neuronal regrowth . Indeed , sensory DRG neurons isolated from CMT2A mice and cultured in the presence of MiM111 ( 100 nM , 48 hr ) exhibited not only enhanced mitochondrial fusion ( increased aspect ratio ) and transport ( greater mitochondrial motility and velocity ) , but axon outgrowth ( length and branching ) ( Figure 5 ) . Similar effects were seen in CMT2A DRGs treated with Chimera C ( 100 nM , 48 hr ) ( Figure 5—figure supplement 1 ) . Both MiM111 and Chimera C provoked mitochondrial redistribution to axonal termini of cultured CMT2A DRGs ( Figure 5—figure supplement 1 ) recapitulating mitochondrial occupation of neuromuscular synapses after MiM111 treatment of CMT2A mice in vivo ( see Figure 4d ) . Comparing the mitochondrial motility , aspect ratio , and neuron growth responses at different times after mitofusin activation revealed significantly increased mitochondrial trafficking within 2 hr , whereas enhanced axon outgrowth was significant after 24 and 28 hr , and mitochondrial aspect ratio ( i . e . fusion ) was significant only after 48 hr ( Figure 5—figure supplement 2 ) . Given the established role for mitochondrial transport in neuronal repair ( Sheng , 2017 ) , this temporal sequence lends credence to the idea that accelerated neuron growth is a consequence of enhanced mitochondrial function and redistribution . DRG outgrowth measures in vitro regrowth of neuronal extensions that are amputated during the cell isolation trituration procedure . We considered that a more appropriate model of regrowth after dieback in CMT2A would test intact neurons lacking only the distal axons . Because CMT2A mouse neurons grow poorly in tissue culture in the absence of mitofusin activators ( vide supra ) , this was not feasible using DRGs . Therefore , we seeded cortical neurons collected from MFN2 T105M allele mice in chambers separated from empty chambers by linear microchannels . In the absence of Cre-recombinase these ‘normal’ neurons grew axons through the microchannels that branched into the empty chambers ( Figure 6a ) . Adenoviral Cre was then used to activate the CMT2A MFN2 T105M transgene , followed after 48 hr by aspiration amputation of the branched axonal termini ( Figure 6b and c ) . Mitochondrial motility and aspect ratio were measured 1 hr before and after axotomy; axon regrowth was measured 3 days after axotomy . The aspect ratio of mitochondria in the distal linear axons of normal and CMT2A neurons was unaffected either by axotomy or by MiM111 ( Figure 6d , left panel ) . By contrast , and consistent with a previous study in normal neurons ( Zhou et al . , 2016 ) , mitochondrial motility was reduced by axotomy ( Figure 6d middle panels ) . Mitofusin activation with MiM111 after axotomy restored mitochondrial motility and neuronal outgrowth to pre-axotomy levels . Thus , the link between experimentally activating mitofusins , the subsequent increase in mitochondrial transport , and enhanced neuronal growth/repair was consistent for mouse CMT2A sciatic nerve axons in vivo , cultured CMT2A DRG neuron outgrowth , and cultured CMT2A cortical nerve regrowth after distal axotomy . These preclinical studies show that activating endogenous normal mitofusins can improve stable neuromuscular dysfunction caused by a CMT2A MFN2 mutant . Pharmacological mitofusin activation enhanced CMT2A neuron growth in vivo and in vitro by promoting mitochondrial fitness and transport , thereby reversing CMT2A-associated mitochondrial fragmentation , depolarization and stasis . We believe the key benefit that accrued from directly activating mitofusin-mediated mitochondrial fusion and motility was improved delivery of healthy mitochondria to neuromuscular junctions and axon growth buds . To our knowledge , this is the first report of any experimental intervention that fully reverses in vivo CMT2A phenotypes , demonstrating the feasibility of a clinically translatable disease-modifying therapeutic modality for this incurable condition . Our studies integrated findings from multiple complementary experimental platforms . Motor neurons directly reprogrammed from CMT2A patient fibroblasts have not previously been described , and provided a platform in which effects of therapeutic interventions could be assessed on different patients’ mutations and individual genetic backgrounds using a disease-relevant cell type . Compared to iPS cell-derived CMT2A neurons ( Rizzo et al . , 2016; Saporta et al . , 2015 ) , direct reprogramming more faithfully reproduced prototypical CMT2A mitochondrial phenotypes . Compared to the parental patient fibroblasts ( Dang et al . , in preparation ) , neurons permitted assessment of the CMT2A-associated mitochondrial motility disorders . While CMT2A has both sensory and motor neuropathy components , we reprogrammed specifically for motor neurons because motor components of this disease are the major source of patient disability . All mouse disease models have advantages and limitations . Our CMT2A mouse model expresses human MFN2 T105M using a ‘motor neuron selective’ promoter ( vide infra ) , and therefore does not exhibit sensory nerve involvement that is sometime manifested in clinical CMT2A . However , compared to other CMT2A mice , the current model more faithfully recapitulates CMT2A neuromuscular dysfunction that is the dominant cause of morbidity in the human condition ( Fridman et al . , 2015; Feely et al . , 2011; Bombelli et al . , 2014; Yaron and Schuldiner , 2016; Berciano et al . , 2017 ) . We previously used young adult mice carrying this combination of MFN2T105M and HB9-Cre alleles to evaluate effects of a topically applied prototype mitofusin activator on mitochondrial motility in sciatic nerve axons ex vivo ( Rocha et al . , 2018 ) . It was not known if , with age , these mice would develop neuromuscular signs similar to clinical CMT2A . As shown here , motor function in these mice is normal at age 10 weeks , but declines until age 30 weeks whereupon it stabilized for at least another 20 weeks . This pattern is similar to the clinical course of CMT2A , in which apparently normal children typically manifest neuromuscular signs in the mid first decade of life , exhibit progressive loss of motor function in distal extremities over the next 10–15 years , and then stabilize . Moreover , the functional ( neuroelectrophysiological ) , histological , and ultrastructural features of axonal tissue in the mice were similar to the human condition . Together with the positive response to mitofusin activation in patient neurons , the improvement in neuromuscular function and cell/organelle pathology in MiM111-treated CMT2A mice supports the approach of mitofusin activation for the clinical disease . Perhaps the most remarkable finding here is that mitofusin activation reversed the signs of CMT2A in mice with severe , stable disease . Every measured endpoint was improved , including gross neuromuscular function ( RotaRod ) , electrophysiological metrics of neuromuscular integrity ( CMAP ) , read-outs for axon degeneration , and multiple histological and ultrastructural assays of mitochondrial pathology in neurons . In vivo and in vitro results pointed to enhanced CMT2A neuron repair and regrowth as a central reason for phenotype reversal . Because it is not possible to functionally dissociate mitofusin-mediated increases in mitochondrial fusion and motility , it is unclear if one or the other of these responses preferentially underlies the neuroregenerative effects of mitofusin activation . However , it seems reasonable to postulate that mitochondrial delivery to distal neurons has greatest importance in the long peripheral nerves innervating hands , forearms , feet , and forelegs , i . e . those areas most impacted in CMT2A ( Fridman et al . , 2015; Feely et al . , 2011; Bombelli et al . , 2014; Yaron and Schuldiner , 2016; Berciano et al . , 2017 ) . In agreement with this notion , we observed a positive correlation between mitochondrial delivery to or occupancy of axonal termini and CMT2A neuron growth in vivo and in vitro . As introduced above , damaging MFN2 mutations are a straightforward cause of CMT2A , but MFN2 multifunctionality complicates delineating the underlying cellular pathology ( Filadi et al . , 2018; Dorn , 2020 ) . For this reason , the specific functional benefits accruing from mitofusin activation in CMT2A cannot unambiguously be defined . Mitochondrial fusion and motility are impaired in CMT2A ( Chen and Chan , 2006 ) and allosteric mitofusin activation corrects both of these parameters ( Franco et al . , 2016; Rocha et al . , 2018; Dang et al . , 2020 ) . Mitochondrial respiratory dysfunction , measured here as loss of inner membrane polarization , is a consequence of diminished fusion-mediated homeostatic repair ( Chen and Chan , 2006 ) , and its improvement can therefore also be explained by enhanced fusogenicity . MFN2 has a role in mitophagic mitochondrial quality and quantity control ( Chen and Dorn , 2013; Gong et al . , 2015 ) , and allosteric mitofusin activation suppressed the increase in autophagy/mitophagy induced by CMT2A mutant MFN2 T105M in cultured cells ( Rocha et al . , 2018 ) . Finally , MFN2 can mediate physical interactions and calcium signaling between mitochondria and endoplasmic reticulum that may also have a role in CMT2A ( Larrea et al . , 2019 ) , but effects of mitofusin activation on mitochondria-reticular interactions have not been described . Here , we studied two structurally distinct but functionally similar allosteric mitofusin activators , a new class of drug that is the first to directly enhance mitochondrial fusion and transport . Although the prototype compounds were found not to be ‘druggable’ , a new generation of mitofusin activators have addressed pharmaceutical limitations of the initial chemical series ( Dang et al . , 2020 ) . As described previously ( Rocha et al . , 2018; Dang et al . , 2020 ) , mitofusin activators have minimal effects in normal cells , likely because increasing the probability that mitofusins are in their open/ ‘active’ conformation is a subtle intervention that can be readily compensated for in the absence of a pre-existing imbalance between mitochondrial fusion and fission . The current in vivo studies used a short-acting compound administered once daily to evaluate the effects of intermittent , or burst , mitofusin activation on CMT2A neuromuscular dysfunction . We considered that continuous long-term activation of mitochondrial fusion and transport might possibly be deleterious ( El Fissi et al . , 2018 ) ( although it is worth noting that adverse effects of MFN1 and MFN2 overexpression in transgenic mice have not been reported ) . Moreover , we reasoned that the problem underlying CMT2A is the cumulative effects of long-term mitochondrial stasis and dysfunction on mitochondrial fitness and neuromuscular integrity . This scenario can explain why CMT2A progresses over many years in people and many weeks in mice . Our aim with burst activation was to turn back the disease clock through daily re-setting of mitochondrial function . By intermittently mobilizing healthy mitochondria to distal areas of physiological need , and simultaneously removing senescent or impaired mitochondria , neuron repair , renewal , and neuromuscular signaling were improved . A mouse is not a man and human neuroregenerative capacity declines with age ( Mattson and Magnus , 2006 ) . For this reason we do not expect that mitofusin activation can fully reverse CMT2A phenotypes in older human patients with long-term stable disease . Nevertheless , the current results suggest that pharmacological mitofusin activation could offer the first disease-altering therapy for younger CMT2A patients . An intriguing possibility is that mitofusin activation may also have a therapeutic role in some of the many other neurodegenerative conditions not directly caused by mitofusin defects wherein mitochondrial fusion or transport are defective ( Dang et al . , in preparation; Knott et al . , 2008; Burté et al . , 2015 ) . Rosa-STOP-mMFN Thr105Met ( T105M ) mice ( C57BL/6 Gt ( ROSA ) 26 Sortm1 ( CAG-MFN2T105M ) Dple/J ) from The Jackson Laboratory ( Bar Harbor , Maine , USA; Stock No: 025322 , RRID:MGI:_JAX:025322 ) were crossed to HB9-Cre mice ( B6 . 129S1-Mnx1tm4 ( cre ) Tmj/J ) from The Jackson Laboratory ( Stock No: 006600 , ( RRID:MGI:_JAX:006600 ) ) to generate neuron-targeted MFN2 T105M mice . HB9 is a motoneuron-specific promoter ( Yang et al . , 2001 ) , but JAX data indicates that this HB9-Cre line also drives expression in some sensory DRG neurons ( https://images . jax . org/webclient/img_detail/20564/ ) . All experimental procedures were approved by Washington University in St . Louis School of Medicine Animal Studies Committee; IACUC protocol number 19–0910 , Exp:12/16/2022 . Normal mouse embryonic fibroblasts ( MEFs ) were prepared by enzymatic dissociation from embryonic day E . 13 . 5–14 . 5 C57BL/6J mice ( The Jackson Laboratory Cat:# 000664 , RRID:IMSR_JAX:000664 ) . Mfn1 null and Mfn2 null Mfn1/Mfn2 double null MEFs fibroblasts were purchased from American Type Culture Collection ( ATCC Manassas , Virginia , USA ) ( CRL-2992 , RRID:CVCL_L691 and CRL-2994 , RRID:CVCL_L692 and CRL-2993 , RRID:CVCL_L693 respectively ) . Human fibroblast: Dermal fibroblast ( MFN2 T105M ) and Dermal fibroblast ( MFN2 H361Y ) from Dr . Robert H . Baloh ( Cedars Sinai ) , Dermal fibroblast ( MFN2 R274W ) from Dr . Barbara Zablocka ( Mossakowski Med Res Ctr ) , Dermal fibroblast ( MFN2 R364W ) from Dr . Michael E . Shy ( University of Iowa ) . Dermal fibroblast ( Normal ) from NINDS respectively: ND34769 , ( RRID:CVCL_EZ04 , ND36320 , RRID:CVCL_EZ16 and ND29510 , RRID:CVCL_Y813 ) . Adenovirus expressing human FLAG-hMFN2 -T105M was prepared at Vector Biolabs ( Malvern , PA , USA ) . Adenoviri expressing β-galactosidase ( Ad-CMV-β-Gal; #1080 ) , and Ad-Cre ( #1794 ) were purchased from Vector Biolabs . Adenovirus for Mito-Ds-Red2 came from Signagen ( Cat:#SL1007744 ) . Lentivirus packaging vectors: psPAX2 ( Addgene , Cat#: 12260 ) pMD2 . G ( Addgene , Cat#: 12259 ) , Lentiviral vectors with recombinant DNA: rtTA-N144 ( Addgene , Cat#: 66810 ) pTight-9–124-BclxL ( Addgene , Cat#: 60857 ) , human LHX3-N174 and human ISL1-N174 were packaged and used as described ( Abernathy et al . , 2017 ) . Mouse monoclonal anti-mitofusin 2 ( Cat # ab56889 - 1:1000 , RRID:AB_2142629 ) , anti-COX-IV ( Cat #ab16056 - 1:1000 , RRID:AB_443304 ) and anti-GAPDH ( Cat #ab8245 - 1:1000 , RRID:AB_2107448 ) were from AbCAM ( Cambridge , MA , USA ) . Rabbit polyclonal anti-Stathmin-2/Superior Cervical Ganglion 10 ( SCGN10; Cat # NBP1-4946 , RRID:AB_10011569 ) was from Novus Biologicals ( Littleton , CO , USA ) . Rabbit polyclonal FSP-1 was from Sigma Millipore ( Cat # 07–2274 , RRID:AB_10807552 ) . Anti-mouse monoclonal -MNX1was from DSHB ( 1:10 , Cat# 81 . 5C10 , RRID:AB_2145209 ) . Mouse monoclonal anti-β -tubulin III ( Cat # 801201- 1:500 , RRID:AB_2313773 ) was from Biolegend ( San Diego , CA , USA ) . Peroxidase-conjugated anti-mouse IgG ( Cat #7076S - 1:1000 , RRID:AB_330924 ) was from Cell Signaling ( Danvers , MA , USA ) . Goat anti-rabbit IgG ( Spicier reactivity Goat , Host/Isotype Rabbit/IgG; Cat #31460 , RRID:AB_228341 ) and Alexa-Fluor 488 anti-mouse ThermoFisher ( Waltham , MA , USA Cat #A11008 , RRID:AB_143165 ) . α-Bugarotoxin Alexa flour 594 was from ThermoFisher , Waltham , MA , USA Cat:# B12423 . DNA was extracted from 5 × 106 primary human fibroblasts using the DNeasy blood and tissue kit ( Qiagen , Cat#: 69506 ) according to manufacturer’s protocol . PCR used Taq Plus Master Mix 2X ( Cat#: BETAQR-L , Bulls eye ) . 50 ng of genomic DNA template , and the following primers: PCR products were purified using PureLink Quick Gel Extraction Kit ( Cat#: K21000-12 , Invitrogen ) . Sanger sequencing of PCR products was performed at GENEWIZ . Direct reprogramming of human motor neurons from patient fibroblasts used the procedure as described ( Abernathy et al . , 2017 ) . Reprogramming cocktail consisted of 1 ml concentrated lentivirus containing the reverse tetracycline-controlled transactivator ( rtTA; Addgene , Cat# 66810 ) , 1 ml virus containing pT-BclXL-9/9−124 , 125 µl virus containing motor neuron transcription factor ISL1 , and 125 µl virus containing motor neuron transcription factor LHX3 with polybrene ( 8 μg/ml; Sigma-Aldrich , Cat# H9268 ) . Human skin fibroblasts of low passage number ( P4-P7 ) were spinfected at 37°C for 30 min at 1 , 000 × G . Doxycycline ( Dox , 1 µg/ml; Sigma Aldrich , Cat# D9891 ) and antibiotics for respective vectors ( Puromycin , 3 µg/ml; Invitrogen , Cat# A11138-03; Geneticin , 400 µg/ml; Invitrogen , Cat# 10131–035 ) were added to culture medium for 4 days after viral transduction . On day 5 cells were re-plated on poly-ornithine/laminin/fibronectin ( Sigma , Cat# P4957 , # L2020 , # F4759 ) coated 18 mm glass coverslips and on the following day changed to neuronal medium supplemented with Dox ( 1 µg/ml added every other day ) , valproic acid ( 1 mM; Sigma , Cat# 676380 ) , dibutyryl cAMP ( 200 µM; Sigma , Cat# D0627 ) , BDNF , NT-3 , CNTF , GDNF ( all 10 ng/ml , Peprotech , Cat# 450–02 , #450–03 , #450–13 , #450–10 ) , retinoic Acid ( 1 µM; Sigma , Cat# R2625 ) and antibiotics . Neuronal medium was refreshed by replacing half every 4 days . Antibiotics were discontinued on day 14; Dox was discontinued on day 30 . Cells underwent studies beginning on day 35 . Motor neurons were identified after formalin fixation by labeling with mouse anti-MNX1 ( 1:10; DSHB , Cat# 81 . 5C10 ) and mouse anti-TUBB3B ( 1:2000; Biolegend , Cat#PRB-435P-100 ) . Fibroblasts were identified by labeling with rabbit anti FSP-1 ( 1:200; Sigma , Cat: # 07–2274 ) . Adult mouse dorsal root ganglion ( DRG ) neurons were prepared from ~8-week-old HB9Cre- MFN2 Thr105Met flox-stop transgenic mice as previously described ( Rocha et al . , 2018 ) . To comprehensively induce MFN2 T105M transgene expression , the DRGs were infected with Adeno-Cre ( M . O . I . of 50 ) 48 hr prior to study . DRG neurons were distinguished from non-neuronal cells by staining with anti-β-III tubulin . Mouse cortical neurons were isolated from individual embryonic day E . 18 . 5 MFN2 Thr105Met flox-stop transgenic mice by papain digestion and mechanical dispersion using a published procedure ( Sobieski et al . , 2015 ) . Briefly , mouse brain cortices were isolated under a dissecting microscope and sliced into 0 . 5–1 mm thick sections in Leibovitz’s L-15 Medium ( Gibco Cat:#11415–064 ) containing BSA ( 0 . 23 mg/ml , Sigma Cat:#A7030 ) . Papain ( 1 mg/ml , Sigma Cat #P4762 ) was added and the tissue digested for 20 min at 37°C . The papain solution was replaced and micropipettes used to triturate the solution until no more tissue was visible . Cortical neurons were plated in microfluidic neuron XonaChip chambers as described below . Static confocal imaging of cultured neurons used triple-stained with MitoTracker Green ( 200 nM; Invitrogen , Thermo Fisher Scientific Cat:# M7514 ) to visualize mitochondria , tetramethylrhodamine ethyl ester ( TMRE , 200 nM , Invitrogen Thermo Fisher Scientific Cat:# T-669 ) that labels mitochondria with normal polarization of the mitochondrial inner membrane , and Hoechst ( 10 μg/ml; Invitrogen , Thermo Fisher Scientific Cat:# H3570 ) that stains nuclei blue as described ( Franco et al . , 2016 ) . Static live cell images were acquired on a Nikon Ti Confocal microscope using either a 60 × 1 . 3 NA oil-immersion objective or 10 × 0 . 3 NA dry objective , in Krebs-Henseleit buffer ( 138 NaCl , 3 . 7 nM KCL , 1 . 2 n M KH2PO4 , 15 nM glucose , 20 nM HEPES pH: 7 . 2–7 . 5 , and 1 mM CaCl2 ) : laser excitation was 488 nm with emission at 510 nm for MitoTracker Green and Ad-Mito GFP , 549 nm with emission at 590 nm for TMRE , and 306 nm with emission 405 nm for Hoecsht and DAPI . Axon branching analysis of CMT2A mouse DRGs was performed at various times after isolation and plating , as indicated . In some studies neurons were infected with Ad-mito-RFP to label mitochondria red . Cells were fixed and labeled with anti-β-tubulin III ( 1:200 in 10% goat serum in PBS ) to identify neurons . Images were acquired using the 10x objective and excitation at 488 nm/emission 510 nm for Alexa-Flour 488 and 579 excitation/599 emission for mito-RFP . Sholl analysis of axonal branching used ImageJ ( Schneider et al . , 2012 ) and an open source plugin ( https://imagej . net/Sholl_Analysis ) . Briefly , a starting radius was set to encompass the soma of β-tubulin III-positive DRG neurons and concentric circles established at 10 micron increments , to 40 microns . Numbers of axon and radii intersections were totaled for all circles to derive intersection number , which is a measure of axonal branching . Special attention was given to ensure that there was uniform staining along all parts of the DRG soma and axons so that the plugin was able to accurately assess the number of intersections accurately . Video confocal studies of mitochondrial motility studies in neurons and sciatic nerves used time-lapse imaging ( 1 frame every 5 s ) for 121 frames ( 10 min , sciatic nerve ) or 180 frames ( 15 min , cultured neurons ) at 37°C on a Nikon A1Rsi Confocal Microscope using a 40x oil objective as described ( Rocha et al . , 2018 ) . Cultured neuron mitochondria were labeled with Adeno-mitoDsRed2 or MitoTracker Orange ( 200 nM , Invitrogen Thermo Fisher Scientific Cat:# M7510 ) excited at 561 nm , emission 585 nm . Sciatic nerve axon mitochondria were labeled with TMRE . Kymographs and quantitative data were generated using an Image-J plug-in . In vitro microfluidic studies of axon growth used primary cortical neurons isolated from embryonic day 18 . 5 MFN2 T105M flox-stop mice . 50 , 000–90 , 000 suspended cells in 20 µl of Earle’s Minimal Essential Medium ( MEM; #11090–081; Gibco ) supplemented with 5% FBS ( Gibco #16140–063 ) , 5% horse serum ( HS ) ( Gibco #26050–070 ) , 400 µM L-glutamine ( Gibco #25030–149 ) , 50 units/ml each penicillin/streptomycin ( Gibco #15070–063 ) and 0 . 3% glucose ( Sigma G 5767 ) ( 5–5 media ) was added to the left chambers of XonaChips with 450 μm microgroove barriers ( #XC450; Xona Microfluidics , Temecula , CA , USA ) coated with 0 . 5 mg/ml poly ( D ) lysine ( Sigma #P7280 ) . Ten minutes thereafter , 150 µl of 5–5 medium supplemented with 0 . 5 µl insulin-transferrin-sodium selenite ( Sigma I 1884 ) was added to each well and neurons cultured under standard conditions . After 24 hr the medium was changed to neurobasal medium ( #21103–049; Gibco , Carlsbad , CA , USA ) , 1x B27 supplement ( #17504–044 , Gibco , Carlsbad , CA , USA ) , 50 units/ml each penicillin/streptomycin ( #15070–063; Gibco , Carlsbad , CA , USA ) and 400 µM L-glutamine ( #25030–149; Gibco , Carlsbad , CA , USA ) with feeding every 2 to 3 days until axotomy ( DIV 12 ) , and infected with adeno-Cre for 48 hr to induce MFN2 Thr105Met expression . Vacuum aspiration axotomy and post-axotomy regrowth analyses were performed as described ( Zhou et al . , 2016 ) . Aspiration axotomy was followed by application of fresh neuron feeding medium containing MIM111 ( 100 nM ) or its vehicle ( Me2SO , 1:1 , 000 ) . Cells were fixed in situ; axonal outgrowth and post-axotomy regrowth were analyzed by confocal analysis of β-III tubulin positive cells . The area of βIII-tubulin signals above the same threshold within a 1024 × 1024 image that covers all axon segments extending from microgrooves was measured using ImageJ ( NIH ) and reported as pixels density of axon segments extending from an average of all microgrooves . Immunoblot analysis was performed on mouse sciatic nerve proteins size-separated on 10% SDS-PAGE gels ( Biorad Cat# 456–1036 ) and transferred to 0 . 45 µM Polyvinylidene fluoride ( PVDF ) membranes ( GE- Amersham , Freiburg , Germany Cat# 10600023 ) . Membranes were blocked with 5% non-fat milk for 30 min and incubated with primary antibody overnight at 4°C . Peroxidase-conjugated secondary antibodies and Chemiluminescence Substrate ( Thermo Scientific #32132 ) were used for signal detection . Quantification of immunoreactive proteins was performed on a LI-COR Odyssey infrared detection system ( Lincoln , NE , USA , version 1 . 0 . 17 ) . Cultured neurons were stained in situ with TMRE ( 200 nM , Invitrogen Thermo Fisher Scientific Cat:# T-669 ) for 30 min at 37°C in 5% CO2–95% air , washed twice in PBS , and released from culture substrates with 0 . 05% Trypsin-EDTA ( Gibco , cat:# 1995647 ) . After centrifugation , the DRG pellets were re-suspended in 200 µl of FACS buffer ( PBS 1X , BSA 1X , 2 Mm EDTA ) . Flow cytometry of TMRE fluorescence was performed on a Gallios instrument ( Beckman Coulter ) and analyzed using FlowJo10 software . ~3500 events were acquired for each sample . Data are presented as Mean Fluorescence Intensity per experiment . In some studies , carbonyl cyanide-p-trifluoromethoxyphenylhydrazone ( FCCP , 10 μM for 1 hr ) ( Sigma , Cat #C2759 ) was added as a positive control for mitochondrial depolarization . Rotarod studies were performed on mice initially acclimated to the RotaRod ( Ugo Basile , Gemonio , VA , Italy;# 47650 ) at a speed of 5 r . p . m . CMT2A mice underwent RotaRod evaluations weekly from 10 to 50 weeks for disease development , and 4 and 8 weeks after mitofusin activator therapy . The acceleration protocol increased from 5 to 40 r . p . m over 120 s and then maintained 40 r . p . m . indefinitely . Each mouse underwent five separate trials per testing event with 5 min rest between trials . Latency ( time to falling off ) was averaged for all trials . Neuroelectrophysiologic recordings of tibialis/gastrocnemius compound muscle action potentials ( CMAP ) were performed with a Viasys Healthcare Nicolet Biomedical instrument ( Middleton , WI , USA Cat:# OL060954 ) running Viking Quest version 11 . 2 software by an operator ( A . F . ) blinded to genotype and treatment group . Mice were anesthetized with isofluorane ( 4–5% induction , 1 . 5% maintenance ) , shaved , and the proximal sciatic nerves stimulated using a needle electrode ( Natus , Mundelein , IL , USA Cat:# F-E2-48 ) with 3 . 9 mV pulses of 0 . 002 ms duration . Ring electrodes ( Natus , Mundelein , IL , USA Cat:# 291965 ) were positioned at the mid forelimb at the belly of the tibialis anterior and gastrocnemius muscles to record CMAP . Optimal stimulating electrode position was determined as that giving the greatest CMAP amplitude; 3–4 independent events were recorded and averaged . Twelve 50-week-old CMT2A mice ( HB9-Cre + MFN2 Thr105Met flox-stop ) and six littermate controls were randomized and blinded to daily intramuscular treatment with MiM111 or vehicle for 8 weeks: under sterile conditions18 . 75 mg/ml ( 64 . 8 mM ) MIM 111 was dissolved in 10% Me2SO/90% ( 2-hydroxypropyl ) -β-cyclodextrin ( HP-BCD; Sigma , Cat: #332607 ) , sterile-filtered ( 0 . 22 µm PVDF , #SLGV033RS , Millipore , Cork , Ireland ) , and drug- or vehicle-containing syringes were assigned to individual mice by XD using a randomization table . Daily intramuscular injections ( biceps femoris muscle , alternating left and right every other day ) were performed by AF , who was blinded to both mouse genotype and drug treatment group . Rotarod and neurophysiological testing were performed before , and 4 and 8 weeks after initiation of therapy . Mice were terminated by anesthesia overdose after 8 weeks for tissue studies . Sciatic and mid tibial nerves were dissected from both legs of all mice . For histology and immunohistology the left leg nerves or muscles were fixed in PFA for 2 hr , transferred to 30% sucrose/PBS overnight at 4°C , and embedded in optimal cutting temperature ( OCT , Tissue-TEK Cat: 4583 ) medium for storage at −80°C . Immunostaining with anti-Superior Cervical Ganglion 10 ( SCG10 ) or wheat germ agglutinin labeling ( WGA , Cat:#W834 , Invitrogen ) was performed on 10 μm cryostat sections briefly ( 5 min ) brought to room temperature and then re-cooled to −20°C for 30 min . RGB rightness of the representative images was increased uniformly for presentation purposes . Mitochondrial occupancy in neuromuscular synaptic junctions was assessed in 10 μm cryostat tibialis muscle sections using anti-COXIV ( 1:200 in 10% goat serum ) to label mitochondria and anti-acetylcholine receptor with α-Bugarotoxin to label neuronal synapses . Transmission electron microscopy and toluidine blue staining used standard techniques ( Zhou et al . , 2019 ) . Data are reported as means ± SD . Two-group comparisons used Student’s t-test; multiple group comparisons used one-way ANOVA , and time-course by treatment group or genotype by treatment group comparisons used two-way ANOVA , with Tukey’s post-hoc test for individual statistical comparisons . p<0 . 05 was considered significant . Mouse treatment was randomized according to a random integer table ( even or odd ) and performed by investigators blind to both genotype and treatment status . Post terminal analysis of tissue and cell phenotypes was performed blindly . Samples size estimation: Using two-sample t-test based on the preliminary data where the coefficients of variation ( CV ) at 50 weeks were 10% and 15% for rotarod latency and CMAP amplitude , respectively , the study was initially designed to have a sample size of 15 mice/group , providing 80% power at 1-sided α = 0 . 05 . Because the therapeutic response for targeted differences was greater than anticipated , the study was completed with a reduced sample size of n = 6/group .	Charcot-Marie-Tooth disease type 2A is a rare genetic childhood disease where dying back of nerve cells leads to muscle loss in the arms and legs , causing permanent disability . There is no known treatment . In this form of CMT , mutations in a protein called mitofusin 2 damage structures inside cells known as mitochondria . Mitochondria generate most of the chemical energy to power a cell , but when mitofusin 2 is mutated , the mitochondria are less healthy and are unable to move within the cell , depriving the cells of energy . This particularly causes problems in the long nerve cells that stretch from the spinal cord to the arm and leg muscles . Now , Franco , Dang et al . wanted to see whether re-activating mitofusin 2 could correct the damage to the mitochondria and restore the nerve connections to the muscles . The researchers tested a new class of drug called a mitofusin activator on nerve cells grown in the laboratory after being taken from people suffering from CMT2A , and also from a mouse model of the disease . Mitofusin activators improved the structure , fitness and movement of mitochondria in both human and mice nerve cells . Franco , Dang et al . then tested the drug in the mice with a CMT2A mutation and found that it could also stimulate nerves to regrow and so reverse muscle loss and weakness . This is the first time scientists have succeeded to reverse the effects of CMT2A in nerve cells of mice and humans . However , these drugs will still need to go through extensive testing in clinical trials before being made widely available to patients . If approved , mitofusin activators may also be beneficial for patients suffering from other genetic conditions that damage mitochondria .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "neuroscience" ]	2020	Burst mitofusin activation reverses neuromuscular dysfunction in murine CMT2A
The Ras family of GTPases are important in cell signaling and frequently mutated in human tumors . Understanding their regulation is thus important for studying biology and human diseases . Here , we report that a novel posttranslational mechanism , reversible lysine fatty acylation , regulates R-Ras2 , a member of the Ras family . SIRT6 , a sirtuin with established tumor suppressor function , regulates the lysine fatty acylation of R-Ras2 . In mouse embryonic fibroblasts ( MEFs ) , Sirt6 knockout ( KO ) increased R-Ras2 lysine fatty acylation . Lysine fatty acylation promotes the plasma membrane localization of R-Ras2 and its interaction with phosphatidylinositol 3-kinase PI3K , leading to activated Akt and increased cell proliferation . Our study establishes lysine fatty acylation as a previously unknown mechanism that regulates the Ras family of GTPases and provides an important mechanism by which SIRT6 functions as a tumor suppressor . SIRT6 ( sirtuin 6 ) belongs to the Sir2 ( silencing information regulator 2 ) family of nicotinamide adenine dinucleotide ( NAD+ ) -dependent protein lysine deacylases . It plays important roles in a variety of biological processes , including DNA damage and repair ( Kaidi et al . , 2010; Mao et al . , 2011; Toiber et al . , 2013 ) , glucose metabolism ( Zhong et al . , 2010 ) , and cell proliferation ( Sebastián et al . , 2012 ) . Sirt6 knockout ( KO ) mice display multiple defects and die a few weeks after birth ( Mostoslavsky et al . , 2006 ) . Underlying its biological functions , SIRT6 has multiple enzymatic activities . It can deacetylate histone H3 lysine 9 ( Ac-H3K9 ) , lysine 18 ( Ac-H3K18 ) , and lysine 56 ( Ac-H3K56 ) ( Michishita et al . , 2008; Yang et al . , 2009; Michishita et al . , 2009; Tasselli et al . , 2016 ) , to suppress target gene expression of several transcription factors , including NF-κB ( Kawahara et al . , 2009 ) , HIF-1α ( Zhong et al . , 2010 ) , c-Jun ( Sundaresan et al . , 2012 ) , and c-Myc ( Sebastián et al . , 2012 ) . SIRT6 has also been reported to be an adenosine diphosphate ( ADP ) -ribosyltransferase ( Mao et al . , 2011; Liszt et al . , 2005 ) . We have recently identified SIRT6 as an efficient lysine defatty-acylase that regulates the secretion of tumor necrosis factor ( TNFα ) ( Jiang et al . , 2013 ) . Mechanistically , lysine fatty acylation promotes TNFα targeting to lysosome and thus decreases its secretion ( Jiang et al . , 2016 ) . However , it remains unclear whether SIRT6 regulates other proteins by defatty-acylation . The Ras family of proteins plays important roles in numerous biological pathways , including signal transduction , membrane trafficking , nuclear export/import , and cytoskeletal dynamics ( Wennerberg et al . , 2005 ) . Five branches of the Ras superfamily ( Ras , Rho , Rab , Arf , and Ran ) are classified according to sequence similarity . Ras proteins can exist in two conformational states: a GDP-bound inactive state and a GTP-bound active state . In the GTP-bound state , Ras proteins can recruit effector proteins and turn on specific signaling pathways ( Hancock , 2003; Karnoub and Weinberg , 2008 ) . Protein post-translational modifications ( PTMs ) play important roles in regulating the Ras family of proteins ( Ahearn et al . , 2011 ) . Ras and Rho families of GTPases are modified by prenylation ( farnesylation or geranylgeranylation ) on their C-terminal CaaX motif at the cysteine residue . Some proteins from Ras and Rho families , such as H-Ras and N-Ras , contain cysteine palmitoylation as a second lipidation . Both prenylation and palmitoylation serve as important membrane targeting signals ( Ahearn et al . , 2011; Linder and Deschenes , 2007 ) . Other proteins from Ras and Rho families , such as K-Ras4B , do not have cysteine palmitoylation and are thought to use a C-terminal polybasic sequence for membrane targeting ( Ahearn et al . , 2011; Linder and Deschenes , 2007 ) . Here , we demonstrate that a Ras family protein , R-Ras2 ( also called TC21 because it was cloned from a human teratocarcinoma cDNA library ) ( Drivas et al . , 1990 ) , which is able to transform cells and is up-regulated in several human cancers ( Lee et al . , 2011; Gutierrez-Erlandsson et al . , 2013; Erdogan et al . , 2007; Murphy et al . , 2002 ) , is regulated by a novel PTM , lysine fatty acylation . Importantly , SIRT6 is identified as the defatty-acylase of R-Ras2 . SIRT6 defatty-acylates R-Ras2 and attenuates its plasma membrane localization , therefore inhibits its ability to activate PI3K signaling pathway and cell proliferation . SIRT6 has been reported to be a tumor suppressor ( Sebastián et al . , 2012 ) . Decreased SIRT6 expression level is found in several human cancers . Enhanced glycolysis is observed in SIRT6-deficient cells and tumors , which is thought to drive tumor formation in vivo ( Sebastián et al . , 2012 ) . Interestingly , the tumor formation promoted by the loss of SIRT6 is oncogene HRAS-independent . However , the exact role of SIRT6 in cancer is still not well understood . In particular , which enzymatic activity is important for the tumor suppression function is not clear . We have recently identified a single point mutation ( G60A ) of SIRT6 that maintains the defatty-acylase activity but exhibits no detectable deacetylase activity in cells ( Zhang et al . , 2016 ) . Utilizing this mutant , we first investigated whether the defatty-acylase activity of SIRT6 contributes to tumor suppression and aimed to identify the defatty-acylation substrate protein that is important for this function . We stably expressed SIRT6 wild type ( WT ) ( exhibits both deacetylase and defatty-acylase activities ) , G60A ( exhibits only defatty-acylase activity in cells ) , or H133Y ( exhibits neither activity in cells ) into Sirt6 KO mouse embryonic fibroblasts ( MEFs ) and tested their effects on cell proliferation . Sirt6 KO MEFs expressing SIRT6 WT showed decreased cell proliferation compared to those without SIRT6 expression ( Figure 1A ) , consistent with the reported role of SIRT6 in suppressing cell proliferation . In contrast , expressing the H133Y mutant had no significant effect on cell proliferation . Interestingly , expression of the G60A mutant , which only exhibits the defatty-acylase activity in cells , also decreased cell proliferation , similar to the expression of SIRT6 WT ( Figure 1A ) . The suppression of cell proliferation by SIRT6 WT was only slightly better than that by the G60A mutant . This suggests that although both deacetylase and defatty-acylase activities of SIRT6 contribute , the defatty-acylase activity plays a major role in regulating cell proliferation . 10 . 7554/eLife . 25158 . 003Figure 1 . Identification of defatty-acylation targets of SIRT6 that contribute to its function in cell proliferation . ( A ) Cell proliferation of Sirt6 KO MEFs stably expressing pCDH empty vector , SIRT6 WT , G60A , or H133Y . Cell proliferation was assayed and quantified using crystal violet staining . Values with error bars indicate mean ± s . d . of three biological replicates . * indicates p<0 . 05 and ‘ns’ indicates no statistical significance when comparing to Sirt6 KO MEFs with pCDH . ( B ) Schematic overview of the SILAC experiment to identify SIRT6 lysine defatty-acylation targets . ( C ) Data analysis and filter of SILAC results . ( D ) Effects of R-Ras2 knockdown on the proliferation of Sirt6 WT and KO MEFs . Values with error bars indicate mean ± s . d . of three biological replicates . *** indicates p<0 . 005 for comparing to Sirt6 KO MEFs with ctrl shRNA . DOI: http://dx . doi . org/10 . 7554/eLife . 25158 . 00310 . 7554/eLife . 25158 . 004Figure 1—figure supplement 1 . Schematic overview of the SILAC experiment to compare the protein abundance in Sirt6 WT and KO MEFs . DOI: http://dx . doi . org/10 . 7554/eLife . 25158 . 004 To identify the lysine defatty-acylation targets of SIRT6 that contribute to its tumor suppressing function , we used a quantitative mass spectrometry method , stable isotope labeling with amino acids in cell culture ( SILAC ) , to identify proteins with different lysine fatty acylation levels in Sirt6 WT and KO MEFs . We used fatty acid analogs to metabolically label fatty acylated proteins and enriched these proteins by incorporating a biotin tag through the copper ( I ) -catalyzed alkyne-azide cycloaddition ( typically called click chemistry ) followed by streptavidin pull-down ( Figure 1B ) . Two fatty acid analogs were used for cross comparison: Alk12 which better mimics myristic acid and Alk14 which better mimics palmitic acid ( Charron et al . , 2009 ) . When processing the samples , we used IpaJ , a cysteine protease from Shigella flexneri that has been reported to hydrolyze the peptide bond after N-myristoylated glycine ( Burnaevskiy et al . , 2013 ) , to reduce the labeling background from N-myristoylation . We also used hydroxylamine ( NH2OH ) as a nucleophile to remove fatty acylation on cysteine residues . There were 865 proteins identified in Alk12 SILAC and 1285 proteins identified in Alk14 SILAC ( Supplementary files 1 and 2 ) . To narrow down the target lists , we filtered the data using four criteria ( Figure 1C ) : ( 1 ) Protein score ≥10 , and the number of unique peptide ≥2; ( 2 ) The target was present in both SILAC experiments , which could decrease the N-myristoylation background ( proteins present only in Alk12 SILAC ) and S-palmitoylation background ( proteins present only in Alk14 SILAC ) ; ( 3 ) Heavy/Light ratio ≥1 . 5 , which allowed us to pick proteins that exhibited higher lysine fatty acylation in Sirt6 KO MEFs than in Sirt6 WT MEFs; ( 4 ) The target had similar protein levels in Sirt6 WT and KO MEFs by comparing the data from a total protein SILAC experiment ( Figure 1—figure supplement 1 . and Supplementary file 3 ) , which was to make sure that proteins with higher Heavy/Light ratios in Alk12 and Alk14 SILAC were not due to increased protein levels . With these criteria , we narrowed down the target lists to five proteins ( Figure 1C ) . Interestingly , a Ras-related small GTPase , R-Ras2 , was one of the possible defatty-acylation targets of SIRT6 . R-Ras2 , like its cousins ( H- , N- , and K-Ras ) , is known to be highly relevant to cancer ( Erdogan et al . , 2007; Clark et al . , 1996; Rosário et al . , 1999; Rong et al . , 2002 ) . We thus hypothesized that the tumor suppressing function of SIRT6 could be through regulating R-Ras2 . To test this hypothesis , we knocked down R-Ras2 in Sirt6 WT and KO MEFs and examined the cell proliferation . Knockdown of R-Ras2 in both Sirt6 WT and KO MEFs by two different shRNAs significantly decreased cell proliferation ( Figure 1D ) , suggesting that R-Ras2 is important for cell proliferation in MEFs . We then validated whether R-Ras2 had lysine fatty acylation . We first examined R-Ras2 mRNA and proteins levels in Sirt6 WT and KO MEFs . The results showed that SIRT6 did not affect the transcription and translation of R-Ras2 ( Figure 2A ) . We then used the fatty acid analog Alk14 to metabolically label an overexpressed FLAG-tagged R-Ras2 . After FLAG immunoprecipitation , we conjugated a fluorescent dye ( 520-BODIPY-azide ) through click chemistry to allow visualization of fatty acylated R-Ras2 by in-gel fluorescence ( Figure 2—figure supplement 1 ) . We used NH2OH to remove potential cysteine fatty acylation . Thus , the in-gel fluorescence signal should be mainly attributed to fatty acylation on lysine residues . R-Ras2 exhibited increased lysine ( NH2OH-resistant ) fatty acylation in Sirt6 KO MEFs than in WT MEFs ( Figure 2B and Figure 2—figure supplement 2A ) . We also observed increased R-Ras2 lysine fatty acylation in Human Embryonic Kidney ( HEK ) 293 T cells after knocking down SIRT6 with two different Sirt6 shRNAs , but not with a control shRNA ( Figure 2C ) . We then used an alternative method to detect lysine fatty acylation on overexpressed FLAG-tagged R-Ras2 . We treated the cells with Alk14 , and then incorporated a biotin tag on fatty acylated proteins ( including R-Ras2 ) through click chemistry . We pulled down the fatty acylated proteins using streptavidin beads and then removed S-palmitoylation from the streptavidin beads by NH2OH treatment . The proteins on beads were then resolved by gel electrophoresis and the level of R-Ras2 was detected by FLAG western blot . This method also revealed more lysine fatty acylation on overexpressed R-Ras2 in Sirt6 KO MEFs than in Sirt6 WT MEFs ( Figure 2—figure supplement 2B ) . 10 . 7554/eLife . 25158 . 005Figure 2 . Validation of R-Ras2 as the defatty-acylation target of SIRT6 . ( A ) mRNA and protein levels of R-Ras2 in Sirt6 WT and KO MEFs . ( B ) In-gel fluorescence ( with NH2OH treatment ) showing that R-Ras2 has higher lysine fatty acylation level in Sirt6 KO MEFs than in Sirt6 WT MEFs . Right histogram shows the quantification of bands on the fluorescence gel . Values with error bars indicate mean ± s . d . of three biological replicates . * indicates p<0 . 05 . The full fluorescence gel is shown in Figure 2—figure supplement 2A . ( C ) Detection of R-Ras2 lysine fatty acylation levels in control and SIRT6 knockown HEK 293 T cells by in-gel fluorescence . Right histogram shows the quantification of bands on the fluorescence gel . Values with error bars indicate mean ± s . d . of three biological replicates . * indicates p<0 . 05 . ( D ) Lysine fatty acylation levels of endogenous R-Ras2 in Sirt6 WT and KO MEFs . ( E , F ) SIRT6 defatty-acylated R-Ras2 in a NAD+-dependent manner in vitro . In-gel fluorescence was used to detect R-Ras2 lysine fatty acylation ( E ) . A 32P-NAD+ assay was used to detect fatty acyl ADPR product from defatty-acylation reaction . ( F ) . ( G ) In-gel fluorescence ( with NH2OH treatment ) showing that mutation of four lysine residues at the C-terminus of R-Ras2 significantly decreased lysine fatty acylation in Sirt6 KO MEFs . Right histogram shows the quantification of bands on the fluorescence gel . Values with error bars indicate mean ± s . d . of three biological replicates . * indicates p<0 . 05 . The full fluorescence gel including R-Ras2 total fatty acylation levels ( without NH2OH treatment ) is shown in Figure 2—figure supplement 2C . ( H ) Tandem mass ( MS/MS ) spectrum of doubly charged Alk14 modified ( on K194 ) R-Ras2 peptide . The b- and y- ions are shown along with the peptide sequence . ( I ) In-gel fluorescence ( with NH2OH treatment ) showing that single mutation of four lysine residues at the C-terminus of R-Ras2 did not affect R-Ras2 lysine fatty acylation . ( J ) Confocal imaging showed that R-Ras2 WT was mainly localized in the plasma membrane in Sirt6 KO MEFs . R-Ras2 WT in Sirt6 WT MEFs as well as R-Ras2 4KR in Sirt6 WT and KO MEFs was localized in both the intracellular vesicles and plasma membrane ( n = 5 , 5 , 5 , 6 cells for each sample from left to right , respectively . The images of other cells were shown in Figure 2—figure supplement 4A . DOI: http://dx . doi . org/10 . 7554/eLife . 25158 . 00510 . 7554/eLife . 25158 . 006Figure 2—figure supplement 1 . Scheme showing the in-gel fluorescence method with Alk14 metabolic labeling to identify R-Ras2 as a lysine fatty acylated protein . FLAG-tagged R-Ras2 protein was enriched from whole cell lysates by FlAG immunoprecipitation . Alk14-labeled R-Ras2 protein was detected by in-gel fluorescence after incorporating BODIPY ( B ) -azide using click chemistry . DOI: http://dx . doi . org/10 . 7554/eLife . 25158 . 00610 . 7554/eLife . 25158 . 007Figure 2—figure supplement 2 . Validation of R-Ras2 as the defatty-acylation target of SIRT6 . ( A ) Full gel image of R-Ras2 fatty acylation level with or without NH2OH treatment in Sirt6 WT and KO MEFs . ( B ) Lysine fatty acylation levels of overexpressed R-Ras2 in Sirt6 WT and KO MEFs . ( C ) Full gel image of R-Ras2 WT and 4KR fatty acylation levels with or without NH2OH treatment in Sirt6 KO MEFs . ( D ) In-gel fluorescence showing lysine fatty acylation level of overexpressed R-Ras2 WT and C199S mutant in HEK 293T cells with SIRT6 knockdown . C , control shRNA . S6 , SIRT6 shRNA#1 . DOI: http://dx . doi . org/10 . 7554/eLife . 25158 . 00710 . 7554/eLife . 25158 . 008Figure 2—figure supplement 3 . Total ion chromatogram ( TIC ) , extracted ion chromatogram ( XIC ) and parent MS ( MS1 ) of Ak14 modified R-Ras2 peptide . DOI: http://dx . doi . org/10 . 7554/eLife . 25158 . 00810 . 7554/eLife . 25158 . 009Figure 2—figure supplement 4 . Lysine fatty acylation targets R-Ras2 to plasma membrane . ( A ) Confocal imaging showing subcellular localization of GFP-tagged R-Ras2 WT and 4KR in Sirt6 WT and KO MEFs . ( B ) Subcellular fractionation of R-Ras2 WT and 4KR in HEK 293T cells with palmitic acid or TM3 treatment . GAPDH was used as the marker of cytosol fraction and Na , K-APTase was used as the marker of plasma membrane fraction . DOI: http://dx . doi . org/10 . 7554/eLife . 25158 . 009 Using a similar method , we set out to examine whether endogenous R-Ras2 was regulated by lysine fatty acylation . Endogenous R-Ras2 labeled with Alk14 was conjugated to biotin and pulled down with streptavidin beads . More fatty-acylated endogenous R-Ras2 was pulled down from Sirt6 KO MEFs than from WT MEFs ( Figure 2D ) , suggesting that endogenous R-Ras2 contained more lysine fatty acylation in Sirt6 KO MEFs . To confirm that SIRT6 could defatty-acylate R-Ras2 directly , we overexpressed FLAG-tagged R-Ras2 in Sirt6 KO MEFs , metabolically labeled with Alk14 , and then purified R-Ras2 protein from total cell lysates . We then incubated R-Ras2 with SIRT6 in the presence of NAD+ and subsequently performed click chemistry to detect R-Ras2 lysine fatty acylation . In the presence of NAD+ , SIRT6 removed most of the lysine fatty acylation signal from R-Ras2 ( Figure 2E ) . We also used a previously established 32P-NAD+ assay ( Du et al . , 2011 ) , in which the newly generated fatty acyl adenosine diphosphate ribose ( ADPR ) product could be easily detected on thin layer chromatography ( TLC ) plate , to detect lysine fatty acylation on R-Ras2 . SIRT6 WT , but not the catalytic mutant SIRT6 H133Y , generated the fatty acyl ADPR spot in the presence of R-Ras2 isolated from Sirt6 KO MEFs and NAD+ ( Figure 2F ) . These data suggested that R-Ras2 was fatty-acylated on lysine residues and SIRT6 could directly remove the fatty acylation in vitro . Next , we set out to identify which lysine residues of R-Ras2 were fatty acylated . We noticed that there are four lysine residues in the C-terminal hypervariable region ( HVR ) of R-Ras2: K192 , K194 , K196 , and K197 ( Figure 2G ) . This region is known as a C-terminal polybasic sequence and is generally believed to help anchor Ras proteins to the membrane through electrostatic interaction ( Ahearn et al . , 2011 ) . We suspected that some of these lysine residues might be fatty acylated . We thus mutated these four lysine residues to arginine ( the 4KR mutant ) , which should maintain the positive charge and thus , should not disrupt the membrane association provided by the positive charge . We examined the lysine fatty acylation of R-Ras2 WT and 4KR in Sirt6 KO MEFs and found that the 4KR mutant significantly decreased the lysine fatty acylation signal ( Figure 2G and Figure 2—figure supplement 2C ) . We then utilized mass spectrometry ( MS ) to directly identify the modification site using FLAG-tagged R-Ras2 purified from Alk14 treated Sirt6 KO MEFs . A peptide ( residue 193–197 ) carrying Alk14 modification on K194 was detected ( Figure 2H and Figure 2—figure supplement 3 ) . Interestingly , when we mutated each of these four lysine residues to arginine and detected their Alk14 labeling by in-gel fluorescence , the hydroxylamine-resistant labeling of all the single mutants was similar to that of WT ( Figure 2I ) . This result suggested that K192 , K194 , K196 , and K197 are likely to be fatty acylated redundantly , although the MS result implied that K194 is preferentially fatty acylated on WT R-Ras2 protein . R-Ras2 is known to have palmitoylation on cysteine 199 , which is close to the fatty acylated lysine cluster . It is possible that lysine fatty acylation occurs via acyl transfer from the nearby acylated cysteine . To test this possibility , we mutated cysteine 199 to serine ( the C199S mutant ) and assayed its Alk14 labeling in HEK 293T cells . Alk14 labeling still occurred on R-Ras2 C199S mutant , which was NH2OH resistant ( Figure 2—figure supplement 2D ) , suggesting that cysteine fatty acylation of R-Ras2 is not required for the occurrence of its lysine fatty acylation . To investigate the function of R-Ras2 lysine fatty acylation , we first examined the subcellular localization of R-Ras2 WT and 4KR in both Sirt6 WT and KO MEFs by confocal imaging . In Sirt6 WT MEFs , both R-Ras2 WT and 4KR were localized in the plasma membrane and intracellular vesicles ( Figure 2J and Figure 2—figure supplement 4A ) . In contrast , in Sirt6 KO MEFs , R-Ras2 WT was mainly localized in the plasma membrane , while R-Ras2 4KR was localized in both the plasma membrane and intracellular vesicles ( Figure 2J and Figure 2—figure supplement 4A ) . Considering that R-Ras2 WT in Sirt6 KO MEFs has the highest lysine fatty acylation level ( Figure 2B and G ) , this data suggests that lysine fatty acylation helps targeting R-Ras2 to the plasma membrane . In addition , we performed subcellular fractionation of HEK 293T cells overexpressing FLAG-tagged R-Ras2 . Under normal conditions , SIRT6 was present and removed most fatty acyl group from R-Ras2 lysine residues , and thus we did not observe obvious difference in plasma membrane and cytosolic localizations between WT R-Ras2 and the 4KR mutant ( Figure 2—figure supplement 4B ) . When we treated the cells with palmitic acid ( to increase R-Ras2 lysine fatty acylation ) or SIRT6 inhibitor TM3 ( to inhibit SIRT6 defatty-acylase activity ) ( He et al . , 2014 ) , we observed decreased cytosolic localization of WT R-Ras2 , but not the 4KR mutant ( Figure 2—figure supplement 4B ) . This data also supports that lysine fatty acylation of R-Ras2 targets it to the plasma membrane . To further confirm that SIRT6 regulates R-Ras2 by defatty-acylation and not deacetylation , we expressed SIRT6 G60A , which exhibits no detectable deacetylase activity in cells , in Sirt6 KO MEFs and examined the lysine fatty acylation and subcellular localization of R-Ras2 . Expressing the G60A mutant in Sirt6 KO MEFs decreased R-Ras2 lysine fatty acylation to the same level as expressing WT SIRT6 ( Figure 3A ) . In vitro , the G60A mutant removed fatty acylation from R-Ras2 similar to WT SIRT6 ( Figure 3B ) . Confocal imaging results showed that in Sirt6 KO MEFs , expression of the G60A mutant promoted R-Ras2 dissociation from the plasma membrane , similar to the expression of WT SIRT6 ( Figure 3C and Figure 3—figure supplement 1A ) . Subcellular fractionation of endogenous R-Ras2 showed that , similar to WT SIRT6 , the G60A mutant decreased the plasma-membrane-localized endogenous R-Ras2 and increased the intracellular vesicle-localized endogenous R-Ras2 ( Figure 3—figure supplement 1B ) . These data collectively demonstrated that the defatty-acylase activity of SIRT6 is sufficient to regulate R-Ras2 . 10 . 7554/eLife . 25158 . 010Figure 3 . SIRT6 defatty-acylase activity is required for regulating R-Ras2 lysine fatty acylation and subcellular localization . ( A ) Detection of R-Ras2 lysine fatty acylation levels in Sirt6 KO MEFs expressing pCDH empty vector , SIRT6 WT , G60A or H133Y by in-gel fluorescence . ( B ) In-gel fluorescence showed that SIRT6 WT and G60A defatty-acylated R-Ras2 in a NAD+-dependent manner in vitro . ( C ) Confocal imaging showing the localization of GFP-tagged R-Ras2 WT in Sirt6 KO MEFs expressing pCDH empty vector , SIRT6 WT , G60A or H133Y ( n = 5 , 6 , 5 , 5 cells for each sample from left to right , respectively ) . The images of other cells were shown in Figure 3—figure supplement 1A . DOI: http://dx . doi . org/10 . 7554/eLife . 25158 . 01010 . 7554/eLife . 25158 . 011Figure 3—figure supplement 1 . The defatty-acylase activity of SIRT6 regulates R-Ras2 subcellular localization . ( A ) Confocal imaging showing the localization of GFP-tagged R-Ras2 in Sirt6 KO MEFs expressing pCDH empty vector , SIRT6 WT , G60A or H133Y . ( B ) Subcellular fractionation showing the localization of endogenous R-Ras2 in Sirt6 WT MEF , Sirt6 KO MEFs , and Sirt6 KO MEFs expressing SIRT6 WT , G60A or H133Y . DOI: http://dx . doi . org/10 . 7554/eLife . 25158 . 011 We next investigated how lysine fatty acylation of R-Ras2 affected its downstream signaling effect . There are two well-studied effector pathways of R-Ras2 , the PI3K/Akt pathway ( Rong et al . , 2002; Rosário et al . , 2001 ) and the Raf/MAPK pathway ( Rosário et al . , 1999 ) . We first examined whether R-Ras2 can activate the Raf/MAPK pathway in MEFs . We used Raf RBD-conjugated argarose beads to pull-down the active GTP-bound form of R-Ras2 . The endogenous total Ras ( H-Ras , N-Ras and K-Ras ) , but not the overexpressed or endogenous R-Ras2 , was pulled down by Raf RBD ( Figure 4—figure supplement 1 ) , suggesting that in MEFs , R-Ras2 was unlikely to activate the Raf/MAPK pathway . We then examined the PI3K/Akt pathway by co-immunoprecipitation ( co-IP ) . We overexpressed FLAG-tagged R-Ras2 WT and 4KR in Sirt6 WT and KO MEFs , immunoprecipitated R-Ras2 and examined p110α , one isoform of the catalytic subunit of PI3K , which has been reported to interact with R-Ras2 ( Rodriguez-Viciana et al . , 2004 ) . WT R-Ras2 in Sirt6 KO MEFs , which was shown to have the highest level of lysine fatty acylation ( Figure 2B and G ) , had more p110α interaction than the WT R-Ras2 in Sirt6 WT MEFs or the 4KR mutant in either Sirt6 WT or KO MEFs ( Figure 4A ) . This data suggested that lysine fatty acylation of R-Ras2 promotes its interaction with p110α . 10 . 7554/eLife . 25158 . 012Figure 4 . Lysine fatty acylation of R-Ras2 activates PI3K-Akt pathway and promotes cell proliferation in MEFs . ( A ) Co-IP experiment showed that R-Ras2 WT in Sirt6 KO MEFs had more p110α interaction than R-Ras2 WT in Sirt6 WT MEFs and R-Ras2 4KR in Sirt6 WT or KO MEFs . ( B ) Knockdown of R-Ras2 in Sirt6 KO MEFs decreased p-Akt Thr308 , but not p-Akt Ser473 , to a level similar to that in Sirt6 WT MEFs . Right histogram shows the quantification of bands on the Western blot membrane . Values with error bars indicate mean ± s . d . of three biological replicates . * indicates p<0 . 05 . The images of the other western blots used for quantification are shown in Figure 4—figure supplement 2A . ( C ) Expressing SIRT6 WT or SIRT6 G60A in Sirt6 KO MEFs decreased p-Akt Thr308 , but not p-Akt Ser473 . Right histogram shows the quantification of bands on the western blot membrane . Values with error bars indicate mean ± s . d . of three biological replicates . * indicates p<0 . 05 . The images of other western blots for quantification are shown in Figure 4—figure supplement 2B . ( D ) R-Ras2 WT but not R-Ras2 4KR promoted cell proliferation in Sirt6 KO MEFs . Values with error bars indicate mean ± s . d . of three biological replicates . *** indicates p<0 . 005 . DOI: http://dx . doi . org/10 . 7554/eLife . 25158 . 01210 . 7554/eLife . 25158 . 013Figure 4—figure supplement 1 . Raf-RBD binding assay showed that both overexpressed and endogenous R-Ras2 did not bind Raf RBD-conjugated argarose beads . Total endogenous Ras ( H-Ras , N-Ras and K-Ras ) was used as the positive control . DOI: http://dx . doi . org/10 . 7554/eLife . 25158 . 01310 . 7554/eLife . 25158 . 014Figure 4—figure supplement 2 . The images of other western blots for quantification . ( A ) Knockdown of R-Ras2 in Sirt6 KO MEFs decreased p-Akt Thr308 , but not p-Akt Ser473 , to a level similar to that in Sirt6 WT MEFs . ( B ) Expressing SIRT6 WT or SIRT6 G60A in Sirt6 KO MEFs decreased p-Akt Thr308 , but not p-Akt Ser473 . DOI: http://dx . doi . org/10 . 7554/eLife . 25158 . 014 Then , we examined whether the interaction of R-Ras2 with p110α could activate its downstream kinase Akt . Sirt6 KO MEFs showed higher phosphorylated Akt on Thr308 ( p-Akt Thr308 ) , but not on Ser473 ( p-Akt Ser473 ) , when compared to Sirt6 WT MEFs ( Figure 4B ) . Knockdown of R-Ras2 in Sirt6 WT MEFs did not affect p-Akt Thr308 , whereas knockdown of R-Ras2 in Sirt6 KO MEFs decreased p-Akt Thr308 to the same level as that in Sirt6 WT MEFs ( Figure 4B and Figure 4—figure supplement 2A ) . We further found that expressing WT SIRT6 or the G60A mutant in Sirt6 KO MEFs decreased p-Akt Thr308 levels ( Figure 4C and Figure 4—figure supplement 2B ) , suggesting that the defatty-acylation activity of SIRT6 is important for regulating p-Akt Thr308 . All the data combined suggested that SIRT6 defatty-acylates R-Ras2 and attenuates its interaction with p110α , resulting in decreased Akt phosphorylation on Thr308 . Finally , to confirm that lysine fatty acylation of R-Ras2 is important for its role in promoting cell proliferation , we overexpressed R-Ras2 WT and 4KR in Sirt6 WT and KO MEFs and measured the cell proliferation . Expression of R-Ras2 WT in Sirt6 KO MEFs significantly increased cell proliferation compared with vector control ( Figure 4D ) , while expressing R-Ras2 4KR in Sirt6 KO MEFs did not increase cell proliferation ( Figure 4D ) . In Sirt6 WT MEFs , neither overexpressing WT R-Ras2 nor the 4KR mutant increased cell proliferation compared with vector control ( Figure 4D ) . These data suggested that lysine fatty acylation of R-Ras2 is important for promoting MEF cell proliferation . In this study , we found that the defatty-acylase activity of SIRT6 is important for regulating cell proliferation . Using a proteomics approach , we identified a small GTPase , R-Ras2 , as a defatty-acylation target of SIRT6 . R-Ras2 has a higher lysine fatty acylation level in Sirt6 KO MEFs than that in Sirt6 WT MEFs . Lysine fatty acylation targets R-Ras2 to the plasma membrane , facilitates its interaction with p110α , and activates the Akt signaling pathway . This is important for the proliferative phenotype of Sirt6 KO MEFs because when we knocked down R-Ras2 or decreased its lysine fatty acylation in Sirt6 KO MEFs , the p-Akt Thr308 level and cell proliferation became similar to those in Sirt6 WT MEFs . Previously , the function of SIRT6 was mainly explained through transcriptional regulation , which typically occurs at the end of signal transduction pathways . Now we showed that SIRT6 can also function at the top of the signal transduction pathway by regulating a GTPase in the Ras family . SIRT6 is known to be a tumor suppressor and down-regulated in many different human cancers ( Sebastián et al . , 2012 ) . Our study suggests that Sirt6 KO increases lysine fatty acylation of R-Ras2 and activates the PI3K-Akt pathway , leading to higher cell proliferation . Thus , the defatty-acylation and suppression of R-Ras2 is a major contributor to the tumor suppressor role of SIRT6 . For the Ras proteins that do not have palmitoylated cysteine ( such as K-Ras4B ) , they are thought to use polybasic sequences for membrane targeting ( Hancock , 2003; Cox et al . , 2014 ) . Interestingly , in addition to palmitoylation on Cys199 and farnesylation on Cys201 , R-Ras2 contains several lysine residues at the C-terminus ( Lys192 , 194 , 196 , 197 ) . One question is , if these lysine residues also help to target R-Ras2 to membranes via electrostatics , why cells use lysine instead of arginine ? Our study has shed lights onto this question and identified a novel regulatory mechanism for this family of proteins , which may help to develop new strategies to pharmacologically target R-Ras2 to treat human diseases . In a recent paper , the lysine residues at the C-terminal of K-Ras4B have been shown to play important roles for phospholipid binding and K-Ras4B signal output ( Zhou et al . , 2017 ) . The lysine to arginine mutant had distinct lipid-binding capacity , suggesting that simple electrostatic interaction may not be the mechanism how polybasic sequences target to the membrane . It is possible that the lysine residues at the C-terminal of K-Ras4B are similarly regulated by fatty acylation , which could explain the distinct lipid-binding capacity of the lysine-to-arginine mutant . However , we could not completely rule out that the lysine residues at the C-terminal of R-Ras2 may also function in phospholipid binding and thus affect its signaling output , in addition to being regulated by lysine fatty acylation and SIRT6 . Many Ras family of small GTPases contain polybasic sequences at the C-termini . It is possible that lysine fatty acylation also occur on these GTPases and regulate their localization and activities . If so , study how these GTPases are regulated by lysine fatty acylation and whether sirtuins also serve as deacylases for these proteins is an ongoing direction in our laboratory . Our study also expands the biological significance of protein lysine fatty acylation . Protein lysine fatty acylation was first reported in 1992 ( Stevenson et al . , 1992 ) . Since then , only a few proteins were known to have this modification and whether this PTM has important biological functions or not was not known . We previously identified that lysine fatty acylation regulates protein secretion ( e . g . TNFα and ribosomal proteins ) ( Jiang et al . , 2013; Zhang et al . , 2016 ) . The finding that lysine fatty acylation regulates the Ras family of proteins suggest that this PTM may have very broad and important biological functions . Anti-FLAG affinity gel ( #A2220 , RRID: AB_10063035 ) and anti-FLAG antibody conjugated with horseradish peroxidase ( #A8592 , RRID: AB_439702 ) were purchased from Sigma . Human/mouse SIRT6 ( #12486 ) , p110α ( #4249 , RRID: AB_2165248 ) , p-Akt Thr308 ( #13038 , RRID: AB_2629447 ) , p-Akt Ser473 ( #4060 , RRID: AB_2315049 ) , Akt ( #4691 , RRID: AB_915783 ) , HSP90 ( #4877 , RRID: AB_2121214 ) and Na , K-ATPase ( #3010 , RRID: AB_2060983 ) antibodies were purchased from Cell Signaling Technology . β-Actin ( sc-4777 ) and GAPDH ( sc-20357 , RRID:AB_641107 ) antibodies were purchased from Santa Cruz Biotechnology . R-Ras2 antibody ( #H00022800-M01 , RRID:AB_547895 ) was purchased from Novus Biologicals . 32P-NAD+ was purchased from PerkinElmer . 3X FLAG peptide , Azide-PEG3-biotin , Tris[ ( 1-benzyl-1H-1 , 2 , 3-triazol-4-yl ) methyl]amine ( TBTA ) , Tris ( 2-carboxyethyl ) phosphine ( TCEP ) , hydroxylamine , NAD+ , and protease inhibitor cocktail were purchased from Sigma . Sequencing grade modified trypsin and FuGene six transfection reagent were purchased from Promega ( Madison , WI ) . ECL plus western blotting detection reagent and Streptavidin agarose beads were purchased from Thermo Scientific Pierce ( Rockford , lL ) . Sep-Pak C18 cartridge and polyester-backed silica plate were purchased from Waters ( Milford , MA ) . 520-BODIPY azide was purchased from Active Motif ( Carlsbad , CA ) . Ras assay kit ( Raf-1 RBD , agarose ) was purchased from EMD Millipore ( Temecula , CA ) . R-Ras2 and SIRT6 shRNA lentiviral plasmids ( pLKO . 1-puro vector ) were purchased from Sigma , the sequence of shRNA was: R-Ras2 shRNA#1: TRCN0000306170 ( ccggtaagagtcccttgaggtttagctcgagctaaacctcaagggactcttatttttg ) ; R-Ras2 shRNA#2: TRCN0000077748 ( ccggcgctagatattgactgttatactcgagtataacagtcaatatctagcgtttttg ) ; SIRT6 shRNA#1: TRCN0000378253 ( ccggcagtacgtccgagacacagtcctcgaggactgtgctcggacgtactgtttttg ) ; SIRT6 shRNA#2: TRCN0000232528 ( ccgggaagaatgtgccaagtgtaagctcgagcttacacttggcacattcttctttttg ) . Alk12 and Alk14 were synthesized according to reported procedures ( Charron et al . , 2009 ) . Plasmid of IpaJ in pET2-8b vector was a kind gift from Prof . Neal M . Alto at Department of Microbiology , University of Texas Southwestern Medical Center . IpaJ was purified according to reported procedures ( Burnaevskiy et al . , 2013 ) . Sirt6 WT and knockout ( KO ) MEFs were kindly provided by Prof . Raul Mostoslavsky at Massachusetts General Hospital Cancer Center , Harvard Medical School , which were prepared and authenticated by the authors as described previously ( Sebastián et al . , 2012 ) . Human Embryonic Kidney ( HEK ) 293T cells were purchased from ATCC ( RRID: CVCL_0063 ) . All the cell lines have been tested for mycoplasma contamination by PCR-based mycoplasma detection kit ( Sigma , MP0025 ) and showed no contamination . Sirt6 WT , Sirt6 KO MEFs and Sirt6 KO MEFs expressing SIRT6 WT , G60A or H133Y were cultured in Dulbecco's Modified Eagle Medium ( DMEM ) with 10% fetal bovine serum ( FBS ) . HEK 293T cells were cultured in DMEM medium with 10% FBS . Human SIRT6 was inserted into pET-28a vector . SIRT6 G60A and H133Y mutants were made by QuikChange . The plasmids of SIRT6 WT and mutants were transformed into E . coli BL21 ( DE3 ) cells . The proteins were purified according to reported procedures ( Jiang et al . , 2013 ) . Human R-Ras2 was inserted into lentiviral vector ( pCDH-CMV-MCS-EF1-Puro ) with N-FLAG tag . R-Ras2 4KR mutant was made by QuikChange . R-Ras2 lentivirus was generated by co-transfection of R-Ras2 , pCMV-dR8 . 2 , and pMD2 . G into HEK 293T cells . After transfection for 48 hr , the medium was collected and used for infecting Sirt6 KO MEFs or HEK 293T cells . To obtain the R-Ras2 stable overexpressed cells , the cells were treated by 1 . 5 mg/mL of puromycin 48 hr after infection and cultured for 1 week while passing cells every 2–3 days . To purify R-Ras2 from Sirt6 KO MEFs or HEK 293T cells , the cells were collected at 500 g for 5 min and then lysed in Nonidet P-40 lysis buffer ( 25 mM Tris-HCl pH 7 . 4 , 150 mM NaCl , 10% glycerol and 1% Nonidet P-40 ) with protease inhibitor cocktail ( 1:100 dilution ) . Anti-FLAG affinity gel was used to enrich the R-Ras2 protein from total cell lysates . Then R-Ras2 was eluted and purified by 3X FLAG peptide following the manual . Human SIRT6 WT was inserted into pCDH-CMV-MCS-EF1-Puro vector without tag . SIRT6 G60A and H133Y mutants were made by QuikChange . SIRT6 lentivirus was generated by co-transfection of SIRT6 , pCMV-dR8 . 2 , and pMD2 . G into HEK 293T cells . After transfection for 48 hr , the medium was collected and used for infecting Sirt6 KO MEFs . Sirt6 KO MEFs with stable expressed SIRT6 WT , G60A or H133Y was selected by 1 . 5 mg/mL of puromycin . Empty pCDH vector was used as the negative control . R-Ras2 WT or 4KR was transfected into MEFs or HEK 293T cells by FuGene six transfection reagent . After 24 hr , the cells were treated with 50 µM of Alk14 for 6 hr . The cells were collected at 500 g for 5 min and then lysed in Nonidet P-40 lysis buffer ( 25 mM Tris-HCl pH 7 . 4 , 150 mM NaCl , 10% glycerol , and 1% Nonidet P-40 ) with protease inhibitor cocktail . The total lysate was incubated with anti-FLAG affinity gel at 4°C for 1 hr . The affinity gel was then washed three times by immunoprecipitation ( IP ) washing buffer ( 25 mM Tris-HCl pH 7 . 4 , 150 mM NaCl and 0 . 2% Nonidet P-40 ) and then re-suspended in 18 µL of IP washing buffer . The click chemistry reaction was performed by adding the following reagents: 520-BODIPY azide ( 0 . 8 µL of 1 . 5 mM solution in DMF ) , TBTA ( 1 . 2 µL of 10 mM solution in DMF ) , CuSO4 ( 1 µL of 40 mM solution in H2O ) and TCEP ( 1 µL of 40 mM solution in H2O ) . The reaction was allowed to proceed at room temperature for 45 min . Then , the SDS loading buffer was added and heated at 95°C for 10 min . After centrifugation at 15 , 000 g for 2 min , the supernatant was collected and treated with 300 mM hydroxylamine at 95°C for 7 min . The samples were resolved by 12% SDS-PAGE . In-gel fluorescence signal was recorded by Typhoon 9400 Variable Mode Imager ( GE Healthcare Life Sciences ) . Sirt6 WT and KO MEFs were treated with 50 µM Alk14 for 6 hr . Cells were collected and lysed by Nonidet P-40 lysis buffer using the same method descried above . The total lysates were subjected to click chemistry reaction by adding the following reagents: Azide-PEG3-biotin ( final concentration was 100 µM ) , TBTA ( final concentration was 0 . 5 mM ) , CuSO4 ( final concentration was 1 mM ) , and TCEP ( final concentration was 1 mM ) . The reaction was allowed to proceed at room temperature for 45 min , and then the total proteins were precipitated by methanol/chloroform ( 2 . 5/1 ) and washed by ice-cold methanol . The protein pellets were re-solubilized in 1 . 5% SDS , 1% Brij97 , 100 mM NaCl and 50 mM triethanolamine . Streptavidin agarose beads were added and incubated at room temperature for 1 hr . After washing the beads three times by 0 . 2% SDS in PBS buffer , the beads were treated with 1 M hydroxylamine ( pH 7 . 4 ) at room temperature for 1 hr . Then , the beads were washed three times with 0 . 2% SDS in PBS buffer . The SDS loading buffer was added to the beads , heated at 95°C for 10 min , and then used for Western blot . R-Ras2 WT was transfected into Sirt6 KO MEFs treated with 50 µM Alk14 for 6 hr and purified using the method described above . The in vitro assay was proceeded in the assay buffer ( 50 mM Tris-HCl pH 8 . 0 , 100 mM NaCl , 2 mM MgCl2 , 1 mM DTT , 1 mM NAD+ ) with 15 μM of SIRT6 and incubated at 37°C for 2 hr . Proteins were precipitated by methanol/chloroform ( 2 . 5/1 ) and washed by ice-cold methanol . The protein pellets were re-solubilized in 4% SDS , 150 mM NaCl and 50 mM triethanolamine . Then click chemistry reaction and in-gel fluorescence were carried out as described above . R-Ras2 WT was transfected into Sirt6 KO MEFs and purified by FLAG affinity gel . Purified R-Ras2 on the FLAG affinity gel was used for 32P-NAD+ assay . 10 μL of reaction buffer containing 50 mM Tris pH 8 . 0 , 150 mM NaCl , 1 mM DTT , 5 μM SIRT6 WT or H133Y , and 0 . 1 μCi of 32P-NAD was mixed with R-Ras2 protein . The reaction was allowed to proceed at 37°C for 2 hr . 2 μL of the reaction mixture was spotted onto the polyester-backed silica plate . The plate was developed in 30:70 ( v/v ) 1 M ammonium bicarbonate/95% ethanol . Then , the plate was exposed in the phosphor imaging screen ( GE Healthcare , Piscataway , NJ ) for 4 hr . The signal was detected using Typhoon 9400 Variable Mode Imager . H3K9 myristoyl and H3K9 palmitoyl peptides were incubated with SIRT6 in the same reaction buffer as positive control for fatty acyl ADPR . H2BK12 acetyl peptide was incubated with SIRT1 in the same reaction buffer as positive control for acetyl ADPR . All the peptides were synthesized according to the reported procedures ( Zhu et al . , 2012 ) . The proteins were resolved by 12% SDS-PAGE and transferred to polyvinylidene fluoride ( PVDF ) membrane . The membrane was incubated with 5% bovine serum albumin ( BSA ) in TPBS buffer ( 0 . 1% Tween-20 in PBS solution ) at room temperature for 60 min . Then , the antibody was diluted with fresh 5% BSA in TPBS buffer and incubated with the membrane for different time points according to the manual . After washing three times by TPBS buffer , the secondary antibody was diluted with fresh 5% BSA in TPBS buffer and then incubated with the membrane at room temperature for 1 hr . The chemiluminescence signal in membrane was recorded after developing in ECL plus western blotting detection reagents using Typhoon 9400 Variable Mode Imager . Cells were seeded in 12-well plates ( 5000 cells/well ) or 24-well plates ( 2500 cells/well ) , and then maintained in DMEM medium with 10% FBS for 5 days . After washing twice with ice-cold PBS , cells were fixed by ice-cold methanol for 10 min . Then , methanol was removed and crystal violet ( 0 . 2% in 2% ethanol solution ) was added and incubated for 5 min . Cells were then washed with water until all excess dye was removed . Crystal violet dye that remained with the cells was solubilized by 0 . 5% SDS in 50% ethanol solution . The absorption of crystal violet was measured at 550 nm . Sirt6 KO MEFs were cultured in DMEM medium with [13C6 , 15N2]-L-lysine and [13C6 , 15N4]-L-arginine for five generations . Sirt6 WT MEFs were cultured in normal DMEM medium for five generations . The cells were treated with 50 µM Alk12 or Alk14 for 6 hr . After quantifying the concentration of total proteins by Bradford assay , 2 . 5 mg of total proteins from each sample were mixed . IpaJ was added ( final concentration was 150 µg/mL ) and incubated with the total lysate at 30°C for 1 hr . The click chemistry reaction was then performed by adding the following reagents: Azide-PEG3-biotin ( final concentration was 100 µM ) , TBTA ( final concentration was 0 . 5 mM ) , CuSO4 ( final concentration was 1 mM ) , and TCEP ( final concentration was 1 mM ) . The reaction was allowed to proceed at room temperature for 45 min , and then the total proteins were precipitated by methanol/chloroform ( 2 . 5/1 ) and washed by ice-cold methanol . After re-solubilize the protein pellets in 1 . 5% SDS , 1% Brij97 , 100 mM NaCl and 50 mM triethanolamine , streptavidin agarose beads were added and incubated with the lysates at room temperature for 1 hr . After washing the beads three times by 0 . 2% SDS in PBS buffer , the beads were treated with 0 . 5 M hydroxylamine ( pH 7 . 4 ) at room temperature for 1 hr . Then , the beads were washed three times with 0 . 2% SDS in PBS buffer . The beads were incubated with 6 M urea and 10 mM TCEP in PBS at 37°C for 30 min , then 400 mM iodoacetamide was added and incubated with the beads at 37°C for 30 min . After washing the beads with 2 M urea in PBS , the beads were incubated with 2 µg trypsin in 2 M urea in PBS at 37°C for 8 hr . The digestion reaction was quenched with 0 . 1% trifluoroacetic acid and the mixture was desalted using a Sep-Pak C18 cartridge . The lyophilized peptides were used for nano LC-MS/MS analysis . LC-MS/MS analysis was performed using LTQ-Orbitrap Elite mass spectrometer . The lyophilized peptides were dissolved in 2% acetonitrile containing 0 . 5% formic acid . The Orbitrap was interfaced with a Dionex UltiMate3000 MDLC system . The peptide samples were injected onto C18 RP nano column ( 5 µm , 75 µm × 50 cm , Magic C18 , Bruker ) at a flow rate of 0 . 3 µL/min . The gradient for HPLC condition was 5–38% acetonitrile containing 0 . 1% formic acid in 120 min . The Orbitrap Elite was operated in positive ion mode with spray voltage 1 . 6 kV and source temperature 275°C . Data-dependent acquisition ( DDA ) mode was used by one precursor ions MS survey scan from m/z 300 to 1800 at resolution 60 , 000 using FT mass analyzer , followed by up to 10 MS/MS scans at resolution 15 , 000 on 10 most intensive peaks . All data were acquired in Xcalibur 2 . 2 operation software . Sirt6 KO MEFs stably expressing FLAG R-Ras2 were used for detecting lysine fatty acylation on R-Ras2 . Cells were treated with 50 µM Alk14 for 6 hr , collected and lysed by Nonidet P-40 lysis buffer using the same method described above . 40 mg of total protein lysates were used for FLAG IP . After washing the FLAG resin three times with IP washing buffer , R-Ras2 protein was eluted by heating at 95°C for 10 min in buffer containing 1% SDS and 50 mM Tris-HCl pH 8 . 0 . The supernatant was treated with 300 mM NH2OH pH 7 . 4 at 95°C for 10 min . R-Ras2 protein was then precipitated by methanol/chloroform and processed for disulfide reduction , denaturing , alkylation , and neutralization using the same method described above . The processed R-Ras2 protein was digested with 1 . 5 µg of trypsin in a glass vial at 37°C for 2 hr , and then desalted using Sep-Pak C18 cartridge . For the LC-MS/MS analysis of lysine fatty acylated peptides , the same settings as SILAC experiment was applied except the LC gradient , which was 5–95% ACN with 0 . 1% trifluoroacetic acid from 0 to 140 min . All data were acquired in Xcalibur 2 . 2 operation software . RNA extraction , reverse transcription and PCR analysis of mRNA levels . Total RNAs were extracted using RNeasy Mini kit ( QIAGEN ) . Reverse transcription was performed using SuperScript III First-Strand Synthesis kit ( Invitrogen ) . PCR amplification was performed using Herculase II Fusion Enzyme with dNTPs Combo kit ( Agilent ) . Subcellular fractionation and confocal imaging were performed according to reported procedures ( Huang et al . , 2012 ) . Confocal imaging was performed on a Zeiss LSM880 confocal/multiphoton microscope . Data were expressed as mean ± s . d . ( standard deviation , shown as error bars ) . Differences were examined by two-tailed Student’s t-test between two groups; p<0 . 05 , p<0 . 01 , **p<0 . 005 .	Cancer is one of the leading causes of death worldwide . Proteins that cause and promote cancer are called oncoproteins . Other proteins , called tumor suppressors , counteract the oncoproteins but are frequently inactive or not present in cancer cells . SIRT6 is a tumor suppressor protein that has been studied in many different types of cancer . In 2013 , researchers found that SIRT6 can remove chemical groups known as fatty acyl groups from the lysine residues of proteins . However , it was unclear whether and how this activity of SIRT6 contributes to its role as a tumor suppressor . Zhang et al . – who are part of the research group who performed the 2013 study – have now compared mouse cells that lack SIRT6 with normal mouse cells to find out which proteins SIRT6 removes fatty acyl groups from . A biochemical technique that makes use of synthetic fatty acids , which get incorporated into the mouse cells , showed that SIRT6 removes fatty acyl groups from a protein called R-Ras2 . This protein is part of a large family of oncoproteins . Zhang et al . discovered that when R-Ras2 is tagged with the fatty acyl group it moves to the cell’s membrane and causes the cell to divide more rapidly . Hence , this promotes the growth and spread of cancerous tumors . SIRT6 acts as an eraser , removing the fatty acyl group , and therefore slows down the growth of cancer cells . Future experiments will aim to find out whether fatty acyl groups also control the activity of other oncoproteins that are similar to R-Ras2 . If that is the case , drugs that can regulate the removal of fatty acyl groups from oncoproteins may eventually form new cancer treatment options .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2017	SIRT6 regulates Ras-related protein R-Ras2 by lysine defatty-acylation
Purkinje cells ( PC ) , the sole output neurons of the cerebellar cortex , encode sensorimotor information , but how they do it remains a matter of debate . Here we show that PCs use a multiplexed spike code . Synchrony/spike time and firing rate encode different information in behaving monkeys during saccadic eye motion tasks . Using the local field potential ( LFP ) as a probe of local network activity , we found that infrequent pause spikes , which initiated or terminated intermittent pauses in simple spike trains , provide a temporally reliable signal for eye motion onset , with strong phase-coupling to the β/γ band LFP . Concurrently , regularly firing , non-pause spikes were weakly correlated with the LFP , but were crucial to linear encoding of eye movement kinematics by firing rate . Therefore , PC spike trains can simultaneously convey information necessary to achieve precision in both timing and continuous control of motion . Movements are often executed with high precision in timing and trajectory control . The cerebellum is heavily involved in online motor control and should process sensorimotor information with great accuracy . In particular , PCs , which deliver the final output from the cerebellar cortex , should use an appropriate coding strategy for this task , but the nature of their coding mechanism is actively debated ( De Zeeuw et al . , 2011; Heck et al . , 2013 ) . In one view , transmission of timing-sensitive information is performed by precisely timed PC spikes and their synchronized firing ( Ebner and Bloedel , 1981; Gauck and Jaeger , 2000; Shin and De Schutter , 2006; de Solages et al . , 2008; De Zeeuw et al . , 2011; Person and Raman , 2012 ) . In the other , PCs use linear firing-rate coding with weak PC-to-PC correlations to robustly control continuous movement kinematics , where high signal-to-noise ratio is achieved by averaging the rates of many PCs ( Shidara et al . , 1993; Thier et al . , 2000; Roitman et al . , 2005; Medina and Lisberger , 2007; Catz et al . , 2008; Herzfeld et al . , 2015 ) . In this study , we re-examined this controversy using a new approach to analyze PC spike trains . We classified PC spikes into specific spike categories , and correlated spike categories with the LFP , using the LFP as a proxy signal for local network activity . In particular , we focused on the role of long , infrequent interspike intervals ( ISI ) , called pauses , which abruptly interrupt the rapid and very regular firing of PCs ( Schonewille et al . , 2006; Shin and De Schutter , 2006; Shin et al . , 2007; Yartsev et al . , 2009 ) . Pauses in the PC spike train are a well-known phenomenon in many contexts , such as saccades ( Ohtsuka and Noda , 1995; Arnstein et al . , 2015; Herzfeld et al . , 2015 ) , and classical conditioning ( Rasmussen et al . , 2008 ) , etc . Spikes that initiate and terminate pauses often synchronize sharply across nearby PCs ( Shin and De Schutter , 2006 ) . This suggests that pauses can be simultaneously involved in spike coding by individual PCs and with collective encoding in a local network . Our approach was to examine the relationships among PC spikes , cerebellar LFPs , and eye motion , with a focus on how those relationships change , depending on the spike category . Using this , we demonstrate that PC spikes simultaneously contribute to precision , both in timing and control of motion by adaptive use of synchrony/spike time and rate coding scheme . From three rhesus ( Macaca mulatta ) monkeys ( E , H , and N ) , we simultaneously recorded spikes of single PCs , the slow component of the LFP below low-γ frequency ( 42 Hz ) , and eye positions during spontaneous and visually guided saccades ( Figure 1A ) . Only recordings where both simple and complex spikes could be isolated were used . In this study , we focused exclusively on simple spikes , since complex spikes comprised a very small percentage of total spikes ( 1 . 71 ± 0 . 13% with a rate of 0 . 81 ± 0 . 07 Hz , mean ± SEM ) and their overall effect on firing statistics was not significant ( Figure 1—figure supplement 1H , see also below ) . 10 . 7554/eLife . 13810 . 003Figure 1 . Cerebellar LFP and PC spikes correlate differently with saccadic eye movements . ( A ) Left: Schematics of eye motion tasks . Right: Simultaneously recorded eye speed , PC spike train , and LFP . Both the recorded ( black ) and filtered LFP below the low-γ frequency ( red ) are shown , but only the latter was analyzed . ( B ) CCFLFP-EV and CCFSpike-EV computed with EV ( Top ) and eye speed in the direction with the angle θ ( Bottom ) . Shaded regions represent the 99% confidence interval . Data are normalized as described in the Materials and methods and plotted as mean ± SEM . ( C ) θ-dependent variability of CCFLFP-EV ( x-axis ) versus CCFSpike-EV ( y-axis ) . CCFSpike-EV varied significantly more ( p<0 . 05 , t-test ) than CCFLFP-EV in 74% ( n = 25; magenta ) , and less in 15% ( n = 5; cyan ) of all recordings . No difference was found in the rest ( n = 4; black ) . Error bars are omitted for clarity . The gray line represents equal variability . The red circle denotes data in A and B . DOI: http://dx . doi . org/10 . 7554/eLife . 13810 . 00310 . 7554/eLife . 13810 . 004Figure 1—figure supplement 1 . Saccade angle and duration dependence of the onset-triggered average LFP , simple and complex spikes . ( A ) Onset-triggered average LFP with different saccade angle θ and , ( B ) L . Each data point at θ or L represents the average over [θ − 45º , θ + 45º] or [L − 4 ms , L + 4 ms] . The LFP was normalized by its standard deviation . In ( A ) the curve in the center represents the normalized peak-to-peak amplitude of the average LFP . In ( B ) white lines mark the saccade beginning and end , respectively . ( C ) The average LFP for the different saccade lengths ( top ) , and for all the saccades ( bottom ) . ( D , E , F ) The same plots as ( A–C ) for the firing rate . In ( D ) red and blue lines represent normalized maximal and minimal firing rates , respectively . The rate was computed with a Gaussian smoothing kernel of σ = 10 ms . Note that waveforms in ( A ) and ( D ) are similar to CCFs in Figure 1B . ( G ) Firing rate of complex spikes , computed with σ = 25 ms , varying with θ ( left ) and L ( right ) . The boxed insert shows spike waveforms . ( H ) Histograms of ISIs after simple ( cyan ) , complex ( black ) , and all spikes ( red ) . Note that the simple and all-spike cases are nearly identical while the complex spike case is shifted , reflecting pauses after complex spikes . Cumulative distributions ( inset ) show that the all-spike case has a small number of short ISIs < 5 ms , caused by complex spikes that immediately follow simple spikes . Data are mean ± SEM . The data are the same as in Figure 1A , B . DOI: http://dx . doi . org/10 . 7554/eLife . 13810 . 004 We first estimated the sensitivity of neural signals to eye motion by computing their cross-correlation functions to eye velocity ( EV ) , CCFLFP-EV and CCFSpike-EV for the LFP and simple spikes , respectively ( Figure 1B ) . Two sharp differences between the LFP and spikes were noticed: First , CCFLFP-EV was significant in recordings from 49 cells ( p<0 . 01 , t-test , n = 33 , 15 , 1 from E , H , and N , respectively ) , but significant CCFSpike-EV was found only in a subset of 34 cells ( p<0 . 01 , t-test , n = 21 , 12 , 1 from E , H , and N , respectively ) . These 34 cells were used in the rest of the analysis . Second , when computed to account for the eye speed component for specific directions , the CCF varied less significantly with the angles of eye movements for the LFP than for spikes , in most cases ( Figure 1B , C ) . We also computed the average LFP and firing rates for saccades with different directions and durations , and found that they shared the same properties , having similar waveforms as the CCFs ( Figure 1—figure supplement 1A–F ) . We identified eye saccade-related neurons based on CCFLFP-EV and CCFSpike-EV; this is essentially identical to the classical method used to localize eye motion-sensitive cells ( Ohtsuka and Noda , 1992 ) . Differences in how LFP and PC spikes relate to eye movements were similar to previously reported differences between multiple PCs versus single units , where the averaged activity of multiple PCs is more robust , but much less direction-dependent ( Ohtsuka and Noda , 1995; Thier et al . , 2000 ) . This supports the view that the LFP represents the average activity of many neurons with diverse angle-dependencies in the local network ( Buzsáki et al . , 2012 ) rather than reflecting the activity of a single PC or a cluster of PCs with similar directional tuning . For this reason , the LFP and spikes apparently do not encode identical information about eye motion . All recorded PCs fired rapidly ( 55 ± 18 Hz , mean ± SD ) with high regularity . To quantify this regularity , we computed the ISI asymmetry index ( AI ) and coefficient of variation ( CV2 ) ( Shin et al . , 2007 ) for each data set , and compared them with those of control spike trains . In controls , spikes were randomly generated with the same instantaneous firing rates and refractory period ( 4 ms ) as in the reference PC spike train . All of our spike trains contained a significantly higher number of regularly firing spikes ( AI ≈ 0 , or equivalently CV2 ≈ 0 ) than the control ( p<0 . 01 , one-sided z-test; Figure 2A ) . ISI distributions had long tails that followed a power law , PISI ( x ) ~ x-α ( α = 3 . 78 ± 0 . 64; tail onset = 30 . 4 ± 13 . 4 ms; p<0 . 05 , n = 32 ) ( Figure 2B ) . Furthermore , the size of the ISI after each spike quickly diverged as the AI of the spike increased ( Figure 2C ) , implying that deviation from regular firing tends to be associated with long ISIs . 10 . 7554/eLife . 13810 . 005Figure 2 . Regular and pause spikes in the PC spike train . ( A ) Left: The ISI asymmetry index ( AI ) measures local variability at each spike and is related to the local coefficient of variation ( CV2 ) . Right: Distribution of AI for PC spikes ( black ) and rate-matching control spike train ( green , mean ± SD ) . Significantly more PC spikes occurred around AI ≈ 0 . ( B ) The ISI histogram with a fitted power-law tail ( red ) . Inset: the same histogram and tail in linear scales . ( C ) ISI after each spike vs . AI . The black line represents mean ± SD in each bin ( center = [-0 . 8 , -0 . 6 , … , 0 . 8] , width = 0 . 2 ) . ( D ) Top: Three types of spiking pattern classified by their AI and associated ISI length . Bottom: Examples of pause and regular spikes in an actual PC spike train . Data are the same as in Figure 1A , B . DOI: http://dx . doi . org/10 . 7554/eLife . 13810 . 00510 . 7554/eLife . 13810 . 006Figure 2—figure supplement 1 . Complex and simple spike pauses . ( A ) ISI after each spike vs . AI for all spikes . Orange dots are complex spikes . Light blue dots are pause-initiating spikes selected based on the AI value and ISI size ( see next section and Materials and methods for details ) . Lines represent correlation coefficients R . Here the fraction of selected pauses caused by complex spikes is 1 . 75% . ( B ) ISIs after spikes vs . ISIs before spikes . Red , cyan , and black dots are pause-initiating , -terminating , and regular spikes , respectively . The rest of the spikes are gray dots . Contours represent log10 ( density ) . Note that the density becomes stretched along x- and y-axes and develops two tails as the ISI becomes larger . Data are the same as in Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 13810 . 006 Our observations are consistent with previous studies describing the PC spike train as typically composed of prolonged periods of fast and highly regular firing , occasionally interrupted by longer ISIs , called pauses ( Schonewille et al . , 2006; Shin and De Schutter , 2006; Shin et al . , 2007; Yartsev et al . , 2009 ) ( Figure 2D ) . Some previous studies discussed extremely long ( >200 ms ) pauses and related them to membrane bistability of the PC ( Loewenstein et al . , 2005; Schonewille et al . , 2006 ) . However , the majority of longer ISIs , or pauses , were much shorter ( <100 ms ) , even when they were associated with a moderately large AI ( Figure 2C ) . Note that these were therefore comparable or larger in duration to pauses triggered by complex spikes ( Latham and Paul , 1970 ) ( see also Figure 1—figure supplement 1H ) and also those following simple spike bursts induced by optogenetic excitation ( Lee et al . , 2015 ) . Although complex spikes triggered pauses , their contribution to pauses in our study remained limited . Complex spikes had a larger average AI than simple spikes ( complex: 0 . 36 ± 0 . 12 , simple: 5 . 9 × 10-4 ± 7 . 7 × 10-4 ) , but their standard deviations were comparable ( complex: 0 . 36 ± 0 . 05 , simple: 0 . 31 ± 0 . 04 ) . This implies that AI values were widely distributed in both simple and complex spikes . However , durations of complex-spike pauses did not increase as rapidly with AI as those of simple-spike pauses . In all cells , correlations between AI and log10 ( ISI ) were significantly smaller for complex spikes than simple spikes ( complex: 0 . 24 ± 0 . 15 , simple: 0 . 66 ± 0 . 03; p<2 . 84 × 10-12 , Wilcoxon rank-sum test ) , which made complex-spikes pauses poorly predicted by AI ( Figure 2—figure supplement 1A ) . Therefore , if we select ~10% of all ISIs as representative 'pauses' , based on ISI size and the AI value of a preceding spike ( see below ) , only 4 . 3 ± 2 . 5% of all pauses would be caused by complex spikes . For this reason , we did not include complex spikes and their associated pauses in our analysis beyond this point . Our first question was whether the relationship between spikes of individual PCs and the activity of the local network vary with specific spike categories , i . e . , regular-firing or pause-related . To examine this , we first selected three subsets of simple spikes from each cell , representing regular , pause-initiating , and pause-terminating spikes , based on a criterion combining AI and ISI duration ( see Materials and methods ) that selected ~10% of all spikes in each category ( Figure 2—figure supplement 1B ) . Then , we computed the spike-triggered average LFP ( STALFP ) ( Gray and Singer , 1989; Soteropoulos and Baker , 2006 ) for each . We found that the STALFP of spikes that initiated or terminated pauses was sharply different from those of regular spikes: the pause-related STALFP was characterized by strong and sharp peaks around the spike time , whereas the regular spike-related STALFP showed only a weak modulation ( Figure 3A left , B ) . For comparison , if computed with randomly selected , or all simple spikes , the STALFP showed a level of peak correlation similar to that of regular spikes ( Figure 3A right , B ) . This suggests that pause spikes are more strongly and more precisely coupled to network activity than regular spikes , but this fact goes unnoticed if the temporal structure of the spike train is ignored when computing the STALFP ( Figure 3A right ) , because pauses are relatively rare events . The larger amplitude of the STALFP for pause-related spikes was readily observed when we varied the criterion for pause spike selection ( Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 13810 . 007Figure 3 . Pause and regular spikes correlate differently with the LFP . ( A ) Left: STALFP of the pause-initiating ( red ) , -terminating ( cyan ) , and regular spikes ( black ) . The grey region is the 99% confidence interval . h is the peak-to-peak amplitude and shown for pause-terminating spikes . Right: STALFP of randomly selected spikes ( black ) and all spikes ( green ) . Data are mean ± SEM . ( B ) Relative STA amplitude , h , to that of all spikes ( hAll ) for each spike category . ( C ) Phase distribution of the β/γ LFP at the pause , regular , and randomly selected spikes ( thick: histogram , thin: kernel-estimated density ) . φ denotes location of the peak , shown for the pause-initiating spike . ( D ) Pair phase consistency ( PPC ) for each spike category . ( E ) Peak phases for the pause-initiating ( x-axis ) and pause-terminating spikes ( y-axis ) in all data . Grey and red lines represent φPauseT = φPauseI and φPauseT = φPauseI + 278 . 3º , respectively . In ( C ) and ( E ) , *p<10–5 ( Wilcoxon rank-sum test ) . Data in ( A ) , ( D ) , and ( E ) are the same as in Figure 1A , B . DOI: http://dx . doi . org/10 . 7554/eLife . 13810 . 00710 . 7554/eLife . 13810 . 008Figure 3—figure supplement 1 . STALFP with different selection criteria for pause spikes . We used a similar scheme to choose pause and regular spikes as in Figure 3 , but varied the AI and CV2 thresholds to change the number of spikes in each group from 7 . 5% to 37 . 5% of all spikes in increments of 7 . 5% . ( A ) STALFP of the pause-initiating , -terminating , and regular spikes . Note that the fast fluctuating part of the STALFP remains robust in the case of pause spikes . Data are the same as in Figure 3A . ( B ) Amplitude of the STALFP of the β/γ band LFP vs . the number of spikes in each group , for all cells . Amplitudes are all normalized by those of STALFP from all spikes ( dotted line ) . Data are mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 13810 . 00810 . 7554/eLife . 13810 . 009Figure 3—figure supplement 2 . STALFP in other data sets . In a minority of cases ( n = 7 ) , STALFP of the regular spikes ( AI ≈ 0 ) had a comparable ( >60% ) amplitude to that of pause spikes ( A ) . However , the amplitude significantly diminished when STALFP was recomputed with the LFP in the β/γ band ( B ) . The amplitude of the regular spike STALFP in the β/γ band ( hβ/γRegular ) was negatively correlated with the original amplitude ( hRegular ) ( C ) suggesting that , when regular spikes are significantly coupled to the LFP , they prefer time scales slower than the β frequency . In C , the red circle corresponds to data in A and B while the cyan circle is a single , exceptional case where hRegular remained relatively large ( D ) . For pause spikes , the original amplitude is mostly retained as hβ/γ/horig = 0 . 86 ± 0 . 13 ( mean ± SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13810 . 00910 . 7554/eLife . 13810 . 010Figure 3—figure supplement 3 . Spike-LFP phase locking in other frequency bands . PPC is computed with the band-passed LFP in the β/γ ( 15–42 Hz ) , low β ( 10–15 Hz ) , θ ( 4–10 Hz ) , and δ ( <4 ) band , and all normalized by values in the β/γ case . DOI: http://dx . doi . org/10 . 7554/eLife . 13810 . 01010 . 7554/eLife . 13810 . 011Figure 3—figure supplement 4 . STALFP depends on LFP spectral properties . ( A–C ) STALFP based on the LFP from the same electrode as the spike signal ( A ) and on the LFP from two other electrodes . ( B , C ) In B , STALFP for pause and regular spikes has distinct waveforms similar to A , but not C . Coherence between LFPs in ( A ) and ( B ) is significantly higher than between ( A ) and ( C ) . Note that we omitted the final stage of LFP filtering ( < 42Hz ) in ( B ) and ( C ) , nonetheless , the spike waveform is negligible . ( D ) Coherence between LFPs from different locations . The LFP shown in B is significantly more coherent with the LFP in ( A ) than the LFP in ( C ) in the higher frequency range ( >4 Hz ) , which underlies the difference in STALFP shown in ( A–C ) . Electrodes for ( B ) and ( C ) were both 1 mm distant from the spike electrode in A . The difference between ( B ) and ( C ) suggests that the relevant signal also depends highly on the vertical electrode position . Coherences and their confidence intervals ( 99% , light color ) were computed with the chronux toolbox ( http://chronux . org ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13810 . 011 In a few exceptional PCs , the STALFP of regular spikes had a modulation amplitude comparable to that of pause spikes ( >70% in amplitude , n = 5; Figure 3—figure supplement 2A ) , whereas these cells did not show difference in correlation with eye movements ( measured by max \|CCFSpike-EV\| ) from the rest ( p>0 . 16 , Wilcoxon rank-sum test ) . However , in most of these cases ( n = 4 ) , the STALFP of regular spikes developed on a much slower time scale , mostly below the high β frequency ( 15 Hz ) . Therefore , when the STALFP was recomputed with the β/γ band ( 15–42 Hz ) LFP , STALFP amplitudes were mostly retained for pause spikes , but significantly diminished for regular spikes , particularly if the original amplitude was large ( Figure 3—figure supplement 2B–D ) . This was observed even when the LFP power spectrum did not have a distinct peak in the β/γ band , implying that it is a property of pause spikes rather than a product of LFP dynamics that generates the β/γ band oscillation . The preferential coupling of pause spikes to the higher frequency LFP band , not the lower , led us to examine whether pause spikes are phase-locked to the β/γ LFP . This was true in the entire dataset , where pause spikes strongly preferred certain phases of the β/γ LFP while regular spikes showed only a weak dependence on the phase ( Figure 3C ) . To estimate how reliably spikes fired at certain LFP phases , we computed the pairwise phase consistency ( PPC ) , which is an average coincidence between any two spikes in the LFP phase space ( see Materials and methods for more details ) and provides an unbiased measure of phase locking ( Vinck et al . , 2010 ) . Again , we found a large difference between pause and regular spikes in their PPC ( Figure 3D ) , specifically for the β/γ LFP ( Figure 3—figure supplement 3 ) . Notably , pause-initiating spikes preferentially occurred near the climbing phase , i . e . , <φPauseI> = 48 . 7º ± 45 . 7º where φPauseI was the most preferred phase for each data set , and pause-terminating spikes were shifted by about 3/4 cycles on average ( <φPauseT> = <φPauseI> + 278 . 3º ± 35 . 3º ) ( Figure 3E ) . These results clearly show that pauses are not randomly occurring events . On the contrary , their initiating and terminating spikes are temporally locked to specific patterns of network activity , which generate significant and temporally precise fluctuations in the LFP . If pause spikes in single PCs are related to specific patterns of network activity , what information do they jointly encode about eye motion ? To answer this question , we first probed how the phase of the β/γ LFP evolves during each saccade in recordings for each cell . In many of them , the β/γ LFP was significantly coherent across all saccades , with a phase locking to saccade onsets ( Figure 4A , B ) . We quantified this by computing the average cross-saccade phase coherence during a time window around saccade onset ( from −100 ms to 150 ms ) , which we called LFP phase reliability . Significantly high LFP phase reliability was found in 80% of the data ( p<0 . 01 , n = 28 , one sided t-test to time-shifted control data; <RLFP> = 0 . 20 ± 0 . 02 , mean ± SEM ) . 10 . 7554/eLife . 13810 . 012Figure 4 . Pause spikes and β/γ LFP encode motion timing . ( A ) Eye speed ( Top ) and β/γ LFP ( Bottom ) aligned with saccade onset . Black lines represent an average over all saccades ( grey , n = 865 ) . ( B ) Phases of the β/γ LFP at saccade onset for the same data . ( C ) Top: Spike trains for pause-initiating , -terminating , and regular spikes aligned with onsets of randomly subselected saccades ( n = 289 ) . Light colored regions represent periods with significantly reliable firing ( jp ( t ) <0 . 05 , see Materials and methods ) . Bottom: Z-score for spike occurrence , smoothed by a gaussian kernel ( σ = 5 ms ) . ( D ) Spike reliability of pause and regular spikes in all data . p = 0 . 0229 , 0 . 0193 ( Wilcoxon rank-sum test ) . ( E ) LFP phase reliability versus spike reliability . The two were significantly correlated ( red and black line ) for pause-initiating and regular spikes ( p = 9 . 3074 × 10-6 , 0 . 0012; Fisher’s z-test ) , but not for pause-terminating spikes ( p = 0 . 0912 ) , due to some recordings with low spike reliability but high LFP reliability . Data in ( A–C ) are the same as in Figure 1A , B . DOI: http://dx . doi . org/10 . 7554/eLife . 13810 . 01210 . 7554/eLife . 13810 . 013Figure 4—figure supplement 1 . Peak firing of pause and regular spikes during saccades . ( A ) Distribution of the peak Z-score for the firing probability of the pause versus regular spikes . We used a time window from −100 ms to 150 ms around saccade onset and the baseline was estimated with 200 randomized controls . Only significant peaks ( colored dots ) were used to compute the distributions ( p<0 . 01 , t-test; n = 27 , 19 , and 19 for pause-initiating , -terminating , and regular spikes , respectively ) . Gray dots represent insignificant data sets . p = 0 . 0061 ( Wilcoxon rank-sum test ) ( B ) Distribution of Z-score peak timing . The mean timing ± SD for pause-initiating , -terminating , and regular spikes are 1 . 48 ± 24 . 5 ms , 15 . 6 ± 43 . 0 ms , and 13 . 9 ± 70 . 6 ms . Colored and gray dots represent significant and insignificant peaks , respectively . Only the distribution of pause-initiating spikes is significantly different from a uniform distribution ( p = 6 . 7353 × 10-4 , Kolmogorov-Smirnov test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13810 . 013 This predicted that pause spikes , which are phase-locked to the β/γ LFP , should reliably code eye movement timing . The most significant pattern that we observed was that pause-initiating spikes encode saccade onsets with a significant and sharp increase in average firing ( Figure 4C and Figure 4—figure supplement 1 ) . This could be caused either by a firing rate increase of pause spikes during some saccades , or by a sharp spike-time correlation across most saccade trials , i . e . reliable spiking . To ascertain which , we estimated how many spike coincidence events occurred beyond the prediction from the firing rate , from all possible pairings of spike trains and saccades , which we called the spike reliability: Briefly , for each pair of saccade trials , we counted spike coincidences ( \|Δt\|≤3 ms ) between two spike trains within a 50 ms-long moving time window , centered at t ( −100 ms ≤t≤150 ms , relative to saccade onset ) . Then we estimated the probability , jp ( t ) , that two random spike trains with the same firing rates could have the same or more spike coincidences , looking for significant ones ( jp ( t ) <0 . 05 ) ( Riehle et al . , 1997; Denker et al . , 2011; Ito et al . , 2011 ) . The fraction of those significant coincidences from all cells , summed over all trials and time , became the spike reliability . Significantly coincident spikes were indeed found during significant peaks of firing probability ( colored regions in Figure 4C ) ; therefore , spike reliability was significantly higher for pause spikes ( Figure 4D ) , further emphasizing their role in encoding the onset of eye motion . Finally , the LFP phase and spike reliability were much more steeply related to each other for pause-initiating spikes than for regular spikes ( Figure 4E ) . This demonstrates that the fidelity of temporal coding by pauses in individual PCs crucially reflects the reliability of activity and coding of the local network . So far we have focused on pause spikes , rare events in the spike train that provide timing information . On the other hand , eye velocity-spike correlation ( and similarly saccade-triggered average rate ) showed firing rate modulation by eye movement kinematics such as direction , duration , etc . ( Figure 1B–C and Figure 1—figure supplement 1A–F ) . This suggests that a firing rate code may also be present in PC spike trains . To answer this question , we constructed an inverse model that predicts firing rate from eye movements in each cell , based on the linear-nonlinear ( LN ) model framework ( Victor and Shapley , 1980 ) . LN models are widely used in sensory system studies , but have not been applied to PCs to date . First , the eye velocity profile is compared with a preferred pattern ( motion feature ) estimated from the linear cross-correlation ( CCFs in Figure 1B ) to generate a linear prediction m of the firing rate , which goes through an additional nonlinear transformation to fit the actually observed firing rate ( Figure 5A , B ) . 10 . 7554/eLife . 13810 . 014Figure 5 . PC firing rate linearly encodes motion kinematics . ( A ) Schematics of the eye motion-to-rate inverse model . ( B ) Example motion feature . ( C ) Predicted vs . measured firing rate for all ( green ) , regular ( black ) , and pause ( red ) spikes . The rates of pause and regular spikes are rescaled to match those of all spikes to compensate for subsampling . The dotted line represents equality . ( D ) Goodness of fit R2 for the linear prediction to actual rate . p<10-30 ( Wilcoxon rank-sum test ) . ( E ) R2 for all cells . Error bars are omitted and box plots are for means . p<10-6 , **p<10-11 ( Wilcoxon rank-sum test ) . Error bars represent SEM . Data in B–D are the same as in Figure 1A , B . DOI: http://dx . doi . org/10 . 7554/eLife . 13810 . 01410 . 7554/eLife . 13810 . 015Figure 5—figure supplement 1 . Spikes with high regularity have a spike-eye motion correlation that is very similar to that of the full spike train . Black traces are correlation coefficients of the CCFSpike-EV , computed in eight eye movement directions , between the full spike train and regular spikes selected based on the CV2 threshold ( x-axis ) . The blue line is the fraction of regular spikes out of all spikes . The inset contains the same correlation coefficients for pause spikes vs . the fraction of pause spikes out of all spikes . Varying numbers of pause spikes were selected by varying the AI threshold as in Figure 3—figure supplement 1 . Data are mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 13810 . 015 We found that the linear rate prediction m followed the actual rate very closely ( R2 = 0 . 92 ± 0 . 07 , mean ± SD ) ; therefore , the effect of the nonlinear transformation is small ( Figure 5C–E ) . This made the linear rate prediction m alone a good predictor of the time-course of average firing rate modulation during saccades and it was used exclusively in the rest of the analysis . Importantly , correlation with the firing rate remained high ( R2 = 0 . 81 ± 0 . 17 ) when we computed the linear prediction with only regular spikes , which comprised about 20% of all spikes . Conversely , the firing rate of pause spikes ( both pause-initiating and –terminating spikes ) did not modulate as linearly or steeply as for regular spikes ( R2 = 0 . 35 ± 0 . 25 ) . This suggests that information encoded by the whole firing rate is very different from eye movement features related to pause spikes such as onsets ( Figure 4C ) , but this information can be captured well by a subset of the full spike train , regular spikes . This led us to further inquire into the similarity between rate coding by regular spikes and the full spike train . For this , we imposed different thresholds on CV2 ( =2\|AI\| ) for selecting the regular spikes , computed the direction-dependent CCFSpike-EV of those selected spikes ( Figure 1B ) , and estimated its similarity to that of the full spike train by computing their correlation coefficient ρ . We found that ρ quickly grew as we increased the CV2 threshold , whereas the fraction of regular spikes increased more slowly ( Figure 5—figure supplement 1 ) . For example , when we select ~47% of all spikes as regular spikes by imposing CV2 < 0 . 4 , their CCFSpike-EV is very similar ( ρ ≈ 0 . 93 in average ) to that of the full spike train . Even at 20% , CCFSpike-EV of these regular spikes had ρ ≈ 0 . 85 on average , similar to the linear coding property ( Figure 5C ) . Note that ρ for pause spikes behaved very differently; it was not significant for a reasonable fractions of pause spikes ( Figure 5—figure supplement 1 inset ) . The combination of Figure 5 and Figure 5—figure supplement 1 demonstrates that regular spikes dominate rate coding . Regular spike and full spike trains respond to similar motion features and also share the linear coding property . Our results resemble those of a previous report in which 'patterns' of regularly firing spikes dominate the rate responses of PCs to sensory stimuli ( Shin et al . , 2007 ) . Nevertheless , pause spikes can make small contributions to rate coding . With all spikes , the LN model is empirically linear , but the rate coding of regular spikes predicted by the model is slightly less linear ( Figure 5D , E ) . Pause spikes , which by definition represent fast rate changes , must compensate for this . One regular spike is inexpensive from the standpoint of information content , because a small and very regular subset can already provide a good approximation of the full rate modulation ( Figure 5—figure supplement 1 ) . Furthermore , regular spikes are weakly coupled to network activity ( Figure 3 ) . This evidence strongly suggests linear rate coding as the primary role of regular spikes . Examples in sensory and motor systems ( Riehle et al . , 1997; Panzeri et al . , 2010; Gire et al . , 2013 ) and theoretical analysis ( Ratté et al . , 2013 ) have shown that neurons can use multiplexed coding strategies , where each spike in a single spike train can differentially couple to local network activity and encode different information about sensory stimuli or behavior . This study presents the first evidence for multiplexed coding in cerebellar Purkinje cells . Specifically , spikes that initiate pauses are strongly coupled to the β/γ band of the LFP and are therefore probably synchronized among nearby PCs . These spikes form a temporally reliable signal to initiate saccadic eye motion . Conversely , regular spikes in the same spike trains are desynchronized among nearby PCs and form a rate code that predicts direction selective eye kinematics . Use of a multiplexed code resolves the perceived contradiction between temporal and rate coding that has dominated recent discussions about PC spiking ( De Zeeuw et al . , 2011; Heck et al . , 2013 ) . Pauses in the spike train are observed in many tonically firing neurons in various contexts . In striatal cholinergic interneurons , synchronized pauses after bursts encode a salient stimulus ( Aosaki et al . , 1995 ) . Since pause-initiating and -terminating spikes can synchronize sharply ( Jaeger , 2003; Shin and De Schutter , 2006 ) , PCs can potentially operate by a similar coding mechanism . In fact , spike synchronization by PCs is a powerful mechanism to control their postsynaptic targets in the cerebellar nucleus ( CN ) . With exceptionally fast GABAergic synapses ( Person and Raman , 2012 ) , CN neurons can reliably generate time-locked rebound spikes in response to synchronized inputs followed by simultaneous disinhibition , even at moderate levels of synchrony ( ≤50% ) ( Person and Raman , 2012 ) and spike time jitter ( ≤20 ms ) ( Gauck and Jaeger , 2000; Sudhakar et al . , 2015 ) . Furthermore , recent optogenetic experiments have shown that synchronous pauses induced either by direct/indirect inhibition or at the offset of direct excitation , can reliably trigger firing in CN neurons , and importantly , at movement onset ( Heiney et al . , 2014; Lee et al . , 2015 ) . Crucially , excitation-induced pauses were short ( ~35 ms ) , but effective , and even shorter pauses from direct inhibition caused similar effects ( Lee et al . , 2015 ) . Synchronized pause spikes ( Shin and De Schutter , 2006 ) can not only represent timing information , but also can complement a rate code . Previous studies have analyzed PC firing collected over many trials and neurons , and found that a collective representation of time and motion emerges as a form of burst firing alone ( Thier et al . , 2000 ) or together with suppressed/paused firing of many PCs ( Catz et al . , 2008; Arnstein et al . , 2015; Herzfeld et al . , 2015 ) . Our results suggest that pause spikes can enhance temporal fidelity of such population-level representations since their temporal consistency across different PCs ( e . g . Figure 4—figure supplement 1B ) can offer a reliable representation , despite large heterogeneity in rate coding schemes of different PCs . For example , Herzfeld et al . ( 2015 ) found that population PC coding of saccade direction depends critically on the timing of pause onset , which varies only up to ~10 ms . Here we showed that pause spikes of individual PCs can indeed fire with a reliability of a few milliseconds . This can be particularly important at the single trial/saccade level , but averaging over multiple trials and cells may not be a suitable approach to probe it . Instead , we evaluated the correlation between simultaneously recorded PC spikes and LFP , and also trial-to-trial correlation ( reliability ) of those signals , for each cell . Our results show that synchronized pause spikes constitute the most significant population signal in PCs despite their sparse appearance in individual spike trains , suggesting that they can be a specific signaling mechanism in the PC-CN part of the motor pathway . The spike-LFP relationship indicates that pause spikes are generated by the local network . While firings by presynaptic afferents , granule cells , Golgi cells , as well as by the postsynaptic targets , are clearly related to the LFP ( Soteropoulos and Baker , 2006; Dugué et al . , 2009; Ros et al . , 2009 ) , PCs only occasionally or weakly modulate their simple spike firing with the LFP ( Courtemanche et al . , 2002; Ros et al . , 2009 ) , except for the very high frequency component ( ~200 Hz ) ( de Solages et al . , 2008 ) . Here we found that the β/γ LFP robustly and preferentially couples to pause spikes and that both can reliably encode time information in a correlated way . Considering that cerebellar LFP is coherent with neocortical LFP ( Courtemanche and Lamarre , 2005; Soteropoulos and Baker , 2006; Ros et al . , 2009 ) and also plays a critical role in maintaining LFP coherence between the sensory and motor cortex , particularly in the low γ band ( Popa et al . , 2013 ) , pause spikes may also have a special relationship with the LFP in the cerebral cortex . We did not attempt to resolve the origin of the time encoding LFP signal due to limitations of the experimental setup . However , there are multiple possible primary sources . One is the dense activation of mossy fibers and granule cells that accompanies the significant LFP in the granular layer ( Morissette and Bower , 1996; Roggeri et al . , 2008; Diwakar et al . , 2011 ) . Because our electrodes are probably too far from the granular layer to detect the signal directly , it is likely that that localized massive activity propagates via ascending and parallel fibers to activate many interneurons and ultimately PCs . In particular , molecular layer interneurons ( MLI ) could provide significant feedforward inhibition ( Mittmann et al . , 2005 ) , causing PCs to pause ( Mittmann and Häusser , 2007 ) . If simultaneous activation of local MLIs ( and/or their synaptic inputs to the local population of PCs ) contributes to the LFP , this would explain our observed correlation of the fast LFP signal and pauses in PCs . However , the time-encoding signal components we observed seem to be highly localized in the cerebellar cortex . In LFPs recorded ~1 mm horizontally from the spike electrode , STALFP amplitude , particularly of pause spikes was greatly diminished to absent ( Figure 3—figure supplement 4 ) . This suggests that the pause-related β/γ LFP signal originates from localized sources and decays quickly with distance , spreading at most a few hundred microns . This is consistent with the fact that PCs are rarely synchronized unless they are very close to each other ( <100 μm ) ( Ebner and Bloedel , 1981; Jaeger , 2003; Shin and De Schutter , 2006 ) . We also found that the effect of complex spikes was minimal since pauses triggered by complex spikes ( Latham and Paul , 1970 ) had a distinct distribution compared to simple-spike pauses ( Figure 2—figure supplement 1A ) . The overall impact of complex spikes was negligible , most probably because none of our tasks involved sensorimotor learning where complex spikes are crucial ( Catz et al . , 2005; Medina and Lisberger , 2008 ) , as they tend to occur after saccades , triggered by significant saccade errors ( Herzfeld et al . , 2015 ) . Many studies have reported significant correlated spiking in similar settings . In the motor and visual cortex , synchronized spikes have larger STALFP and better phase locking to the motion-related LFP β oscillation ( Denker et al . , 2011; Ito et al . , 2011 ) . In another part of the cerebellum , Medina and Lisberger found that firing rate variability and cross-correlation of PC firing both peaked at the onset of smooth-pursuit eye motion ( Medina and Lisberger , 2007 ) . Our findings are consistent with their observations since pause spikes , which fire reliably with respect to motion onset , have higher ISI variability , and significant coupling to the LFP implies correlated firing . Medina and Lisberger also suggested that such correlated spiking is due to common input to PCs ( Medina and Lisberger , 2007 ) , which can significantly contribute to the LFP signal ( Denker et al . , 2011 ) and trigger synchronized pauses ( Jaeger , 2003 ) . On the other hand , PCs also use linear rate coding of eye movements and here non-pause regular spikes are predominant . Linear coding of motion kinematics by the PC firing rate has been repeatedly observed ( Shidara et al . , 1993; Roitman et al . , 2005; Medina and Lisberger , 2007; Herzfeld et al . , 2015 ) . If correlations with other PCs are weak , as is the case for regular spikes , rate coding by individual PCs can precisely control continuous movements since CN neurons receive inputs from many PCs and can average away noise in individual inputs . Because different stages of a movement may demand that some aspects of the motion be controlled more precisely than others , it is useful for PCs to use different spike codes . Our study demonstrates that PCs can use both temporal- and rate-coding schemes to multiplex their population output with different types of information ( De Schutter and Steuber , 2009; Ratté et al . , 2013 ) , so that precision in motion timing and continuous control can be managed adaptably . We conclude that multiplexed coding is used for sensorimotor coordination in the cerebellar cortex . All animal experiments were approved by the local animal care committee ( Protocol number: Regierungpräsidium Tübingen N1/08 and N6/13 ) , conducted in accordance with German law and the National Institutes of Health's Guide for the Care and Use of Laboratory Animals and carefully monitored by the veterinary administration ( Regierungspräsidium and Landratsamt Tübingen ) . Three adult male rhesus ( Macaca mulatta ) monkeys ( E , H , and N; 10 , 11 , and 15 years old , respectively ) were subjects in this study . They were implanted with a magnetic scleral search coil to record the eye position ( Judge et al . , 1980 ) , a titanium head post to painlessly immobilize the head during experiments , and a circular titanium recording chamber located over the midline of the cerebellum to allow electrophysiological recordings ( Thier and Erickson , 1992 ) . Position and orientation of the implants were carefully planned using pre-surgical MRI and confirmed using postsurgical MRI that helped to direct electrodes to the oculomotor vermis . All surgical procedures were conducted using aseptic techniques under the full anesthesia consisting of isofluorane supplemented with remifentanil ( 1–2 μg/kg/min ) . All relevant physiological parameters such as body temperature , heart rate , blood pressure , pO2 , and pCO2 were monitored . Postoperatively , buprenorphine was given until no sign of pain was evident . Animals were allowed to fully recover before starting the experiments . Animals made saccades either without ( n = 16 ) or with visual cues ( n = 18 ) , but we analyzed all data combined . In the spontaneous saccade paradigm , animals made eye movements freely in the absence of any given visual stimulus . In the visually guided task , animals first focused on a fixation spot at the center of the monitor for periods of time that varied randomly between 1000 and 1500 ms . As soon as the fixation spot disappeared , a target for the primary saccade appeared randomly in one of eight possible locations in the periphery with the angle θ = 0º , 45º , . . . , 315º at a constant distance from the center , varying between 2 . 5º and 20º in separate blocks . Extracellular potentials were initially high- and low-pass filtered online and recorded separately . The band pass-filtered ( 300 Hz–3 kHz ) channel was used to identify single-unit activity . Single-PC units were distinguished by the presence of simple and complex spikes via online sorting , but all spikes were resorted offline and semi-automatically via a neural network trained on manually selected spike waveform prototypes as in ( Yartsev et al . , 2009 ) . The LFP from the low-pass filtered ( <150 Hz ) channel went through a series of additional filtering steps to remove influences from spike waveforms . First , an online notch filter at 50 Hz was applied to reduce line noise . An additional offline low-pass filter at 42 Hz was applied . Then , the LFP was resampled at 90 Hz and oversampled back to 1 kHz to minimize the effect of spike waveforms . Eye velocity was computed from recorded eye position using a Savitzky-Golay filter of the fifth order with a 25 ms time window . Saccade onsets were detected from eye speed using a custom adaptive detection algorithm described in Appendix 1 . We first selected data based on recording quality both in neural and eye motion recordings and then based on statistical significance of CCFEV-LFP and CCFEV-Spike ( see below ) . When the signal for the quantity A and B are x ( t ) and y ( t ) , respectively , the cross-correlation function CCFAB is computed byCCFAB ( t ) =1ZAB∑s=0Lx ( s+t ) y ( s ) −1Mshift∑m=1Mshift ( 1ZAB∑s=0Lx ( s+t ) ymT ( s ) ) , where L is the signal length . ymT ( t ) is y ( t ) shifted by m·T , ymT ( t ) = y ( t + m·T ) . The second term is a shift-correction that estimates the baseline and average of possible uncontrolled correlation . We used T = 1 s and Mshift ~ 200 depending on L and whether the variance of the shift corrections stabilized . Assuming this variance corresponds to the standard error of the CCF both for the original and shifted data , we computed the t-score for the one-sided test for whether the CCF becomes significant ( p<0 . 01 ) , particularly from t = −100 ms to t = 100 ms . For the normalization ZAB , we used two different schemes depending on whether A and B were both continuous ( eye velocity and LFP ) or one of them was a spike train . In the former , the 'CCF-like' convention , ZAB = ( Var[x]·Var[y] ) 1/2 ( L−\|t\| ) , was used . In the latter , we used the 'STA-like' normalization ZAB = Var[x]1/2·Nspike where Nspike was the number of spikes . Therefore , STALFP ( t ) = CCFLFP-Spike ( t ) where y ( s ) was a spike train with 1 ms wide time bins , up to a constant factor . We estimated dependence of the LFP and spikes on saccade angle θ by computing their CCF ( θ ) with the component of the eye velocity vector v , vθ = ( v·eθ ) + where eθ = [cos θ , sin θ] ( θ = 0º , 45º , … , 315° ) and ( ) + represents rectification . Then , we computed the matrix CCF = [CCF ( 0º ) ; CCF ( 45º ) ; …] and the noise-to-signal ratio , NSR = Tr Cov[CCF]/\|\|Mean[CCF]\|\|2 , which was used as an estimate of θ-dependence in CCF ( θ ) . About 200 control CCF ( θ ) were made from time-shifted data ( see above ) and their variance was used in the t-test for differences in NSR between the LFP and spikes . The ISI-asymmetry index ( AI ) was computed as in Figure 2A . For each data set , we generated 200 rate-matching artificial spike trains in a similar way to Shin et al . ( 2007 ) . We first computed ISIs , {ISIk} ( k = 1 , …Nspike-1 ) , from experimental data . At each k , we computed the local firing rate from the nearest five ISIs as rk = 5/ ( ISIk-2 + ISIk-1+ . . . ISIk+2 ) . The k-th artificial ISI , Tk , was drawn from the gamma distribution P ( Tk ) ~ rk2 Tk exp ( -rk Tk ) with a refractory period 4 ms imposed . From these , we computed the mean histogram and its variance . The tail shape of the ISI distribution was tested using the powerlaw Python package ( http://pypi . python . org/pypi/powerlaw ) , which provided a fitting algorithm to the power-law tail and statistical tests to compare the result with the alternative hypothesis of an exponential tail . For pause-initiating spikes , we first selected 15% of spikes having the largest AI values . Then , we removed 25% of spikes with the shortest pause ISI . Pause-terminating spikes were selected similarly , but with the smallest AI . Minimal pause duration varied with the average firing rate of the cell , but was in general ~20% larger than the ISI corresponding to the mean firing rate . Regular spikes were selected based only on their CV2 , and their number was matched to those in comparison groups . The β/γ LFP was obtained by band-pass filtering the LFP between 15–42 Hz . Phases were extracted by the Hilbert transformation ( MATLAB function hilbert ) . From the phase and spike times , we computed pairwise phase consistency ( PPC ) in the following way ( Vinck et al . , 2010 ) . When the phase of the β/γ LFP at spike times is {φn} ( n = 1 , … , Nspike ) , PPC is given byPPC=2Nspike ( Nspike−1 ) ∑m=1Nspike−1∑n=m+1Nspikeexp⁡ ( i ( ϕm−ϕn ) ) LFP phase reliability RLFP is measured by phase coherence of the β/γ LFP across saccades . When φi ( t ) is the LFP phase at t , time relative to saccade onset ( t = 0 at the onset ) , for the i-th of N saccades , RLFP is given byRLFP=1Te−Tb∫TbTeZ ( t ) dt , Z ( t ) =\|1N∑i=1Kexp⁡ ( iϕi ( t ) ) \| , where we used Tb = −100 ms and Te = 150 ms . We also computed RLFP of time-shifted data in the same way as CCF and their mean and variance were used for the one-sided t-test . Spike reliability Rspike was evaluated using the fraction of significant cross-saccade synchronization events , following Denker et al . ( 2011 ) . We first collected spike trains during the same time window as for RLFP . Then , within a 50 ms-wide moving window centered at t in each saccade period ( t = 0 at the onset ) , we counted the number of spike coincidences up to ± 3 ms time difference , nemp ( t ) . In the same moving window , rate prediction of the coincidence probability and expected number of coincidences nexp ( t ) were computed from firing rates based on summing all possible spike train/saccade pairings . With a null hypothesis of Poisson statistics , the probability of nemp ( t ) isjp ( t ) =P ( nemp ( t ) \|nexp ( t ) ) =∑r=0nemp ( t ) nemp ( t ) rr ! exp⁡ ( −nexp ( t ) ) . The criterion for significant synchronization was jp ( t ) <0 . 05 and if this was satisfied , all coincidence events in the window were regarded as spike synchronization from 'unit events' ( Riehle et al . , 1997; Denker et al . , 2011; Ito et al . , 2011 ) . Then , Rspike is a fraction of synchronization events versus all coincidences , Rspike=∑jp ( t ) <0 . 05nemp ( t ) ∑tnemp ( t ) . Our inverse model consists of two parts: a motion feature f acts as a receptive field that linearly transforms eye velocity history to the linear rate prediction , m . Then , a nonlinear function P ( spike\|m ) gives the spiking probability . f was estimated from the spike-triggered average of the rectified eye velocity in four directions ( θ = 0º , 90º , etc . ) from −300 ms to 300 ms around the spike time . We computed the rate prediction , m , by linear filtering the rectified eye velocity by f , and estimated P ( spike\|m ) by comparing m with actual firing rate . See Appendix 2 for more mathematical details . When z- and t-tests were used , we checked normality of sample distribution using D'Agostino's K2 test ( significance level = 0 . 05 ) . In some cases that failed the normality test , we estimated a p-value by computing an empirical upper bound of the type I error rate by using control data sets . We increased 400~2000 control data sets , depending on computing time , evaluated the quantity of interest , and made one-sided comparisons with an estimate from original data . The p-value was estimated to be an error rate of the comparisons . For each statistical test , we computed statistical power by estimating an upper bound of the type II error rate from the resampled data sets: we generated 400~2000 randomly resampled data sets with replacement , performed the same statistical test on them , and counted how many passed the test at a given significance level , which gave our estimate of the power . We regarded the original test result as significant only when the power is sufficiently high ( >0 . 8 ) , All analyses were done with MATLAB 2012a ( Mathworks , VA ) and custom scripts in Python 2 . 7 , which will be available on our homepage ( http://groups . oist . jp/cnu ) .	The cerebellum is a part of the brain that uses information from the senses to coordinate movement . Cells called Purkinje neurons in the cerebellum produce the final ‘output’ of its cortex . Therefore , Purkinje neurons have to communicate precise information about different aspects of the movement , such as its speed and timing . This information is likely to be represented by patterns of electrical activity within Purkinje neurons , but these patterns are still not fully understood . Hong et al . recorded and analyzed electrical ‘spikes’ , the output activity of Purkinje neurons , while monkeys made rapid eye movements . The recordings showed that occasional pauses in the otherwise regularly firing spikes of Purkinje neurons signaled the start of the eye movements . The pauses were accompanied by a sharp change in the local field potential , another electrical signal that comes from many neurons in the neighborhood . In the same cells , the rate of regularly firing spikes increased and decreased with the direction and speed of eye movements , following a simple relationship and independently of the local field potential . Purkinje neurons therefore appear to use both the timing and the rate of their spiking activity to represent movement . This resolves conflicting reports in the literature claiming that either rates of spiking or their timing code essential information about movements: both are important . This way of representing information by combining more than one source is known as multiplexed coding . Next , experiments recording electrical activity from many cells in the cerebellum at the same time are needed to find out how multiple Purkinje neurons can pause their spiking activity at the same time . Future experiments should also uncover how pauses in spiking and firing rates change with learning .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Multiplexed coding by cerebellar Purkinje neurons
The specification of cartilage requires Sox9 , a transcription factor with broad roles for organogenesis outside the skeletal system . How Sox9 and other factors gain access to cartilage-specific cis-regulatory regions during skeletal development was unknown . By analyzing chromatin accessibility during the differentiation of neural crest cells into chondrocytes of the zebrafish head , we find that cartilage-associated chromatin accessibility is dynamically established . Cartilage-associated regions that become accessible after neural crest migration are co-enriched for Sox9 and Fox transcription factor binding motifs . In zebrafish lacking Foxc1 paralogs , we find a global decrease in chromatin accessibility in chondrocytes , consistent with a later loss of dorsal facial cartilages . Zebrafish transgenesis assays confirm that many of these Foxc1-dependent elements function as enhancers with region- and stage-specific activity in facial cartilages . These results show that Foxc1 promotes chondrogenesis in the face by establishing chromatin accessibility at a number of cartilage-associated gene enhancers . Cartilage is the first skeletal type to be specified in the vertebrate body , providing important templates for later bone development and providing flexibility at joint surfaces and within the nose , ear , ribs , and larynx . The transcription factor Sox9 is essential for chondrogenic differentiation in all vertebrates examined ( Bi et al . , 1999; Lefebvre et al . , 1997; Mori-Akiyama et al . , 2003; Yan et al . , 2005 ) , yet it also has widespread roles outside the skeletal system , including the reproductive system , kidney , liver , and skin ( Jo et al . , 2014 ) . How Sox9 is directed to a chondrogenic program in trunk mesoderm and cranial neural crest-derived cells ( CNCCs ) has remained unclear . Sox9 is known to directly bind to a number of cis-regulatory elements adjacent to chondrogenic genes , including Col2a1 , Col10a1 , and Acan ( Askary et al . , 2015; Dy et al . , 2012; Lefebvre et al . , 1997; Ohba et al . , 2015 ) . It is , however , dispensable for chromatin accessibility at these same elements ( Liu et al . , 2018 ) , suggesting that other unknown factors may first open chromatin at chondrogenic enhancers for later activation by Sox9 . Forkhead-domain ( Fox ) family transcription factors are excellent candidates for establishing chromatin accessibility at chondrogenic enhancers . In the endoderm lineage , HNF3/FoxA binds closed chromatin at enhancers and makes these more accessible ( Cirillo et al . , 2002 ) . Foxd3 has been similarly proposed to establish chromatin accessibility in the early neural crest lineage ( Lukoseviciute et al . , 2018 ) . In mouse , loss of Foxc1 results in widespread cartilage and bone defects ( Kume et al . , 1998 ) , including impaired tracheal and rib cartilages ( Hong et al . , 1999 ) , loss of calvarial bone due to premature ossification ( Rice et al . , 2003; Sun et al . , 2013; Vivatbutsiri et al . , 2008 ) , syngnathia ( Inman et al . , 2013 ) , and disruption of endochondral bone maturation ( Yoshida et al . , 2015 ) . In zebrafish , loss of both Foxc1 paralogs ( foxc1a and foxc1b ) results in severe reductions of dorsal cartilages of the upper face , which are preceded by reduced expression of several Sox9 targets , including col2a1a , acana , matn1 , and matn4 ( Xu et al . , 2018 ) . Chromatin immunoprecipitation followed by deep sequencing ( ChIP-Seq ) using a Sox9 antibody in dissected mouse rib and nose cartilage revealed enrichment of Fox binding motifs within Sox9-bound cis-regulatory sequences near chondrogenic genes ( Ohba et al . , 2015 ) . Here , we use profiling of chromatin accessibility in wild-type and mutant zebrafish facial cartilages and find that Foxc1 paralogs are required for accessibility and activity of a number of cartilage enhancers . These findings support a model in which Foxc1 promotes chondrogenesis by establishing selective accessibility of cartilage enhancers in CNCC-derived mesenchyme . In order to identify potential cis-regulatory elements important for facial cartilage development , we performed a genome-wide analysis of chromatin accessibility in chondrocytes from 72 hr post-fertilization ( hpf ) zebrafish . We labeled chondrocytes by co-expression of sox10:Dsred and col2a1a:GFP transgenes ( Figure 1A ) and isolated double-positive and control double-negative cells by fluorescence-activated cell sorting ( FACS ) . We then subjected these cells to a modified ‘micro’ version of the assay for transposase-accessible chromatin followed by next-generation sequencing ( µATACseq ) . In order to focus on potential distal cis-regulatory elements , we excluded accessible regions within 1 kb upstream or 0 . 5 kb downstream of transcription start sites . This analysis yielded 33 , 679 distal accessible elements , with 5736 elements over-enriched in chondrocytes and 8955 elements under-enriched in chondrocytes ( Figure 1B , C ) . As confirmation of sequence quality , the cartilage-specific R2 enhancer of the Collagen Type II alpha 1 a gene ( col2a1a ) is enriched in chondrocyte-specific elements ( Figure 1—figure supplement 1A ) . De novo motif analysis of the top 2000 chondrocyte-enriched regions using HOMER recovered motifs for Sox , Fox , Nfat , Zfx , and Nkx transcription factor families ( Figure 1D; Supplementary file 1A ) . Sox9 ChIP-Seq of mouse chondrocytes had previously revealed Nfat and Fox motifs as the second and third most co-enriched with Sox motifs ( Ohba et al . , 2015 ) , and the Nkx motif might reflect the role of Nkx3 . 2 in cartilage differentiation ( Provot et al . , 2006 ) . Consensus sequences for Sox , Fox , and Nfat motifs were highly similar between zebrafish and mouse ( Figure 1—figure supplement 1B ) , despite our zebrafish analysis focusing on all accessible regions in facial chondrocytes and mouse analysis focusing on only Sox9-bound regions in rib chondrocytes . These striking motif similarities indicate strong conservation of the cartilage gene regulatory network between fish and mammals , and between the face and rib , and strongly suggest that many of the identified chondrocyte-specific accessible regions in zebrafish likely function as chondrocyte enhancers . In addition , Gene Ontology ( GO ) analysis of the nearest genes to the chondrocyte-enriched elements revealed cartilage development as one of the top six associated terms ( Figure 1E ) . We also recovered terms for neural crest cell migration and dorsal/ventral pattern formation , likely reflecting retention of enhancer accessibility linked to the neural crest origins and later dorsoventral arch patterning of the precursors of facial cartilage . We next investigated when cartilage-associated chromatin accessibility is established in relation to CNCC development . CNCCs are first specified at ~10 . 5 hpf at the border of the neural keel , finish their migration into the pharyngeal arches by ~20 hpf , and then show the first histological signs of cartilage development in the jaw at ~52 hpf ( Schilling and Kimmel , 1997 ) . CNCC-derived arch ectomesenchyme cells can be uniquely identified by co-expression of fli1a:GFP and sox10:Dsred transgenes at 36 and 48 hpf ( Askary et al . , 2017 ) , stages just prior to cartilage differentiation ( Figure 2A ) . We therefore performed µATACseq on FACS-purified fli1a:GFP+; sox10:Dsred+ cells at these stages . 4915 of 14 , 623 elements gaining accessibility from 36 to 48 hpf were linked to GO terms including skeletal system development and cartilage development , and were enriched for Sox and Fox transcription factor binding motifs . 4323 elements with decreasing accessibility were linked to several GO terms related to cellular migration and enriched for motifs of neural crest-associated transcription factors ( Nr2f , Lhx , Olig2 , Hox , Ets ) , consistent with decommissioning of enhancers involved in earlier neural crest migration ( Figure 2—figure supplement 1 ) . To more specifically understand the timing of cartilage-associated chromatin accessibility , we next analyzed chondrocyte-specific accessible elements from our 72 hpf dataset for their accessibilities at 36 and 48 hpf ( Figure 2B ) . Of the 5 , 736 elements enriched in chondrocytes at 72 hpf , only 6% ( 356 ) had peak accessibility at 36 hpf with no further increase in accessibility by 48 hpf ( ‘Group I’ ) . In contrast , 48% ( 2741 ) displayed increased accessibility between 36 and 48 hpf ( ‘Group II’ ) , and 46% ( 2639 ) between 48 and 72 hpf ( ‘Group III’ ) . For Group I , de novo motif enrichment revealed predicted binding sites for members of the Nfat , Fox , Lhx , Nr2f , Meis , and Pax families , and significant GO terms included neural crest migration , cell migration in general , and dorsal/ventral pattern formation ( Figure 2C , D ) . Combined with the known involvement of Foxd3 ( Montero-Balaguer et al . , 2006; Stewart et al . , 2006 ) , Lhx6/8 ( Denaxa et al . , 2009 ) , Nr2f1/2/5 ( Barske et al . , 2018 ) , Meis2 ( Machon et al . , 2015 ) , and Pax9 ( Nakatomi et al . , 2010 ) in CNCC specification , migration , and dorsal–ventral arch patterning , many Group I elements likely represent retention of cis-regulatory elements involved in the earlier specification , migration , and regional patterning of CNCCs . Group II and Group III elements share many common predicted transcription factor binding motifs , including Sox dimer , Fox , Nfat , and Ap1 motifs previously described for mouse cartilage ( Figure 2E , G; Supplementary file 1B; Ohba et al . , 2015 ) . An Nkx motif was recovered only for Group II ( p=10−58 , 36% of targets ) , an Egr motif was enriched for Group II ( p=10−65 , 55% of targets ) versus Group III ( p=10−20 , 16% of targets ) , and a Tead motif was enriched for Group III ( p=10−63 , 36% of targets ) versus Group II ( p=10−20 , 3% of targets ) . GO analysis for linked genes also revealed terms related to skeletal system development ( Group II ) and cartilage development ( Group III ) , as well as more general terms such as transcription and cell differentiation ( Figure 2F , H ) . We therefore conclude that the majority of chondrocyte-specific elements gain accessibility after pharyngeal arch formation and that transcription factor binding motifs change during cartilage differentiation . For example , enrichment of Nkx motifs in Group II elements might reflect the role of Nkx3 . 2 in limiting chondrocyte maturation ( Provot et al . , 2006 ) and promoting joint formation ( Miller et al . , 2003 ) , while the preferential enrichment of the Tead motif , which is linked to growth-associated Hippo signaling ( Ota and Sasaki , 2008 ) , in Group III elements might reflect the later proliferative expansion of chondrocytes . We had previously found that Foxc1 genes are essential for cartilage development in the upper face ( Xu et al . , 2018 ) , and both our µATACseq analysis of zebrafish chondrocytes and published Sox9 ChIP-seq analysis in mouse ( Ohba et al . , 2015 ) reveals co-enrichment of Sox and Fox motifs in accessible regions near known cartilage genes . In zebrafish Foxc1 ( foxc1a−/−; foxc1b−/− ) mutants , cartilages of the upper/dorsal face fail to develop ( Xu et al . , 2018 ) . In order to isolate the dorsal arch CNCC precursors affected in Foxc1 mutants , we used a pou3f3b:Gal4; UAS:nlsGFP ( pou3f3b>GFP ) dorsal CNCC transgenic line ( Barske et al . , 2020 ) along with the pan-CNCC sox10:Dsred transgenic line . In situ hybridization for foxc1a and foxc1b at 36 hpf showed partial overlap with the pou3f3b>GFP line in the dorsal-intermediate regions of the first and second arches ( Figure 3—figure supplement 1A ) . Consistently , we observed reductions of cartilage in the pou3f3b>GFP domain of mutants at 6 days post-fertilization ( dpf ) , except in the dorsal-most regions ( close to the ear ) that are pou3f3b>GFP-positive yet foxc1a/foxc1b-negative ( Figure 3C , D; Figure 3—figure supplement 1B ) . As pou3f3b>GFP+; sox10:Dsred+ CNCCs were present at 48 hpf ( Figure 3A , B ) , we performed µATACseq on these cells after FACS from Foxc1 mutants and controls at 36 and 48 hpf . A comparison of the 15 , 781 accessible regions in dorsal CNCCs ( pou3f3b>GFP+; sox10:Dsred+ ) and pan-CNCCs ( fli1a:GFP+; sox10:Dsred+ ) at 36 hpf revealed 79% with similar accessibility , 11% with greater accessibility in dorsal CNCCs , and 10% with greater accessibility in pan-CNCCs ( Figure 3—figure supplement 2A ) . Of 22 , 323 elements at 48 hpf , 72% were similarly accessible , 10% more accessible in dorsal CNCCs , and 18% more accessible in pan-CNCCs ( Figure 3—figure supplement 2B ) . A comparison of the 5 , 736 regions with specific accessibility in cartilage at 72 hpf revealed high correlation between dorsal- and pan-CNCCs at 36 hpf , with 96% displaying similar accessibility ( r = 0 . 92 , Figure 3—figure supplement 3A , C ) . By 48 hpf , however , we observed notable differences between accessibility of cartilage-specific elements , with 43% displaying greater accessibility in pan-CNCCs and only 0 . 2% displaying greater accessibility in dorsal CNCCs ( r = 0 . 71 , Figure 3—figure supplement 3B , D ) . The decreased accessibility of cartilage-specific elements in dorsal CNCCs at 48 hpf supports previous studies that dorsal chondrocytes develop later than other chondrocytes in the zebrafish face ( Barske et al . , 2016; Schilling and Kimmel , 1997 ) . Analysis of cartilage-associated elements in Foxc1 mutants revealed that 10% ( 120/1221 ) of elements that have established peak accessibility by 36 hpf ( i . e . Group I ) and 41% ( 636/1556 ) of elements that increase accessibility between 36 and 48 hpf ( i . e . Group II ) had reduced accessibility in Foxc1 mutants ( Figure 3E , Figure 3—figure supplement 4 ) . De novo motif analysis of Foxc1-dependent and Foxc1-independent elements in Group I showed enrichment of Sox ( p=10−12 , p=10−76 ) , Tead ( p=10−14 , p=10−20 ) , Mtf ( p=10−12 , p=10−22 ) , and Ets ( p=10−12 , p=10−61 ) motifs in both ( Figure 3F , Supplementary file 1C ) . Whereas several types of Fox motifs were uncovered in both , a closer analysis revealed enrichment of Foxa2 and Foxd3 motifs only in Foxc1-dependent elements ( p=10−45 , 38% of targets; p=10−16 , 18% of targets ) , and a Foxo1 motif only in Foxc1-independent elements ( p=10−42 , 49% of targets ) . Similarly in Group II elements , a Foxa1 motif was enriched only in Foxc1-dependent elements ( p=10−59 , 38% of targets ) , and Foxh1 and Foxp1 motifs only in Foxc1-independent elements ( p=10−22 , 22% of targets; p=10−19 , 7% of targets ) ( Figure 3G , Supplementary file 1C ) . Although Foxc1 motifs were not in the database used for motif predictions , the selective presence of Foxa1/2 motifs in Foxc1-dependent elements is consistent with previous reports that the consensus sequence for Foxc1-bound peaks is nearly identical to Foxa1/2 motifs ( Wang et al . , 2016 ) . Sox motifs were similarly enriched in both Group I and Group II Foxc1-dependent and -independent elements , and Nfat and Zfx motifs in Group II Foxc1-dependent and -independent elements . An Insm ( p=10−39 , 30% of targets ) motif was uncovered only in Foxc1-independent Group I elements and an Nkx motif ( p=10−35 , 52% of targets ) only in Foxc1-independent Group II elements . Whereas 33% ( 207/636 ) of Foxc1-dependent Group II elements had both Sox and Fox predicted motifs , only 16% ( 151/920 ) of Foxc1-independent Group II elements had both . These findings suggest that Foxc1-dependent and -independent cartilage elements may be commonly bound by Sox9 but likely differ in co-binding by Foxc1 and additional co-factors . For example , the presence of the Nkx motif only in Foxc1-independent elements suggests that it could be an alternative co-factor for Sox9 , in line with the known roles for Nkx3 . 2 in chondrocyte biology ( Provot et al . , 2006 ) . To verify whether Foxc1-dependent cartilage elements identified by µATACseq are chondrogenic enhancers , we tested the ability of individual elements in combination with an E1b minimal promoter to drive cartilage expression of green fluorescent protein ( GFP ) in zebrafish transgenic assays ( Figure 4—figure supplement 1A ) . We tested 22 Foxc1-dependent Group II elements near 15 different genes , which included elements linked to genes with known cartilage function ( ucmab , matn4 , matn1 , lect1 , epyc , col9a1a , col9a3 , sox10 , acana , foxa3 , mia ) and others with unknown cartilage function ( si:dkey33i1l . 4 , gas1b , lefty2 , slc35d1a ) ( Supplementary file 2 ) . We observed that 59% ( 13/22 ) of elements drove GFP expression in facial chondrocytes at 6 dpf . These included intronic elements within sox10 , lect1 , col9a3 , col9a1a , and slc35d1a; distal 5′ elements near ucmab , epyc , mia , acana , and matn4; a distal 3′ element near gas1b; and a promoter-associated element for lect1 . We confirmed cartilage expression in independent stable transgenic lines for all 13 positive elements ( Figure 4A; Figure 4—figure supplement 1; Supplementary file 2 ) . Whereas an element in the first intron of sox10 drove uniform cartilage-specific expression , most elements drove expression in specific sub-regions or particular differentiation stages of cartilage . A distal 5′ element of ucmab drove expression in chondrocytes of multiple joints in the zebrafish head , an intronic element of lect1 drove chondrocyte expression only in the jaw joint and hyomandibular-otic connection , a promoter-associated element of lect1 drove expression more strongly in the hyoid joint and hyomandibular-symplectic connection ( though also more broadly in chondrocytes ) , a distal 5′ element of acana drove restricted expression at the hyoid joint , and a distal 3′ element of gas1b drove restricted expression at the Meckel’s–Meckel’s joint and connection between the hyomandibular cartilage and opercular bone . Reciprocally , elements associated with col9a3 , epyc , mia , matn4 , and col9a1a were expressed in chondrocytes but generally excluded from joint regions ( particularly apparent at the hyoid joint and hyomandibular-symplectic connection ) . An intronic element for slc35d1a drove expression in both pharyngeal cartilage and muscle . Furthermore , we found that three enhancer transgenes with diverse expression patterns ( broad sox10 , joint-restricted ucmab , and joint-excluded epyc enhancers ) all displayed reduced activity specifically in the dorsal cartilage regions affected in Foxc1 mutants ( Figure 4B ) . We also confirmed by in situ hybridization that four genes linked to Foxc1-dependent enhancers ( sox10 , lect1 , col9a3 , epyc ) showed reduced expression in cartilage-forming regions of the dorsal arches of Foxc1 mutants ( Figure 4C ) , similar to our previous results for col2a1a , matn1 , matn4 , and acana ( Xu et al . , 2018 ) . Testing of four Foxc1-independent Group II elements revealed two that drove cartilage expression ( distal 5′ elements of gas1b and matn4 ) , one that drove cartilage , bone , and ligament expression ( distal 5′ element of sparc ) , and one with no activity ( promoter-associated element of sox10 ) ( Figure 4—figure supplement 1 ) . Thus , the majority of tested Foxc1-dependent and -independent Group II elements are equally capable of driving cartilage expression . We also tested five Foxc1-independent Group I elements and found that two showed arch CNCC expression at 36 hpf and minimal cartilage expression at 6 dpf ( prrx1a and emx3 elements ) , one no expression at 36 hpf and ligament expression at 6 dpf ( satb2 ) , and two no expression at either stage ( prrx1a , cd248a ) ( Figure 4—figure supplement 2 ) . In contrast , we did not observe 36 hpf arch CNCC expression in any of the 16 Group II elements that drove cartilage expression at 6 dpf . As most of the Group I elements tested had decreased accessibility from 36 to 48 hpf , it seems likely that these elements represent either neural crest elements in the process of decommissioning ( e . g . prrx1a and emx3 ) or elements prefiguring non-cartilage fates ( e . g . satb2 ) . Thus , most cartilage enhancers appear to gain accessibility after 36 hpf , with activity of late-opening cartilage enhancers in diverse locations , indicating that global cartilage expression patterns are achieved in part through the summation of enhancers with more restricted activity . Our findings indicate that Foxc1 function is required for the accessibility of close to half of chondrocyte enhancers in the zebrafish face . Given expression and function of Foxc1 in diverse cartilages ( e . g . limb , rib , tracheal ) in mouse , it seems likely that Foxc1 has a similar function in chondrocyte enhancer accessibility throughout the body . The co-enrichment of predicted Sox and Fox binding sites in 33% of chondrocyte enhancers suggests a model in which Foxc1 promotes Sox9 binding to these same enhancers by increasing chromatin accessibility , though this would require direct Sox9 binding studies to validate . However , not all Foxc1-dependent enhancers have predicted Fox binding motifs , which might suggest indirect regulation of chromatin accessibility in at least some cases . In addition , many enhancers do not appear to require Foxc1 activity . It is possible that other members of the Fox family compensate , such as Foxf1/2 in the facial midline ( Xu et al . , 2018 ) and Foxa2/3 during later hypertrophic maturation ( Ionescu et al . , 2012 ) . Alternatively , there may be other co-factors that mediate chondrocyte enhancer accessibility and activation , such as Nkx3 . 2 . We did not detect any obvious differences in the types of enhancer-proximal genes or the patterns and stages associated with Foxc1-dependent versus -independent enhancer activity . Further work will also be needed to understand the mechanism by which Foxc1 promotes chondrocyte enhancer accessibility , and the extent to which this reflects direct binding of Foxc1 to chondrogenic enhancers . Foxc1 lacks a chromatin modifying domain ( Yoshida et al . , 2015 ) and therefore would need to interact with a co-factor to directly open chromatin . Foxc1 could also act to maintain open chromatin , as shown for Foxa1 in the liver ( Reizel et al . , 2020 ) . Given that both Foxc1 and Sox9 have expression in many tissues outside the skeletal system , it will also be important to determine whether additional factors help to further restrict their activity to chondrocyte enhancers within skeletogenic mesenchyme . The Institutional Animal Care and Use Committee of the University of Southern California approved all experiments on zebrafish ( Danio rerio ) ( Protocol #10885 ) . Existing mutant and transgenic lines used in this study include foxc1ael542 and foxc1bel620 ( Xu et al . , 2018 ) ; Tg ( sox10:Dsred ) el110 and Tg ( fli1a:EGFP ) y1 ( Askary et al . , 2017 ) ; Tg ( col2a1aBAC:GFP ) el483 ( Paul et al . , 2016 ) ; and Tg ( UAS:nlsGFP;α-crystallin:Cerulean ) el609 and pou3f3bGal4ff-el79 ( Barske et al . , 2020 ) . For enhancer transgenic lines , we synthesized accessible elements with flanking attB4 and attB1 sequences using IDT gBlocks and cloned these into pDONR-P4-P1R using the Gateway Tol2kit ( Invitrogen ) to create p5E enhancer constructs ( Kwan et al . , 2007 ) . We then combined p5E constructs with pME-E1b-GFP , p3E-polyA , and pDestTol2AB2 using LR clonase . Final DNA constructs were microinjected with transposase RNA ( 30 ng/µl each ) into one cell stage zebrafish embryos . In most cases , multiple independent stable alleles per construct were analyzed in the F1 generation ( Supplementary file 2 ) . In situ hybridization and immunohistochemistry were performed as described ( Xu et al . , 2018 ) . In addition to foxc1a and foxc1b probes ( Xu et al . , 2018 ) , partial cDNAs were PCR-amplified with Phusion High-Fidelity DNA polymerase ( New England Biolabs , Ipswich , MA ) , cloned into pCR_Blunt_II_Topo ( ThermoFisher Scientific , Waltham , MA ) , linearized , and then synthesized with Sp6 or T7 RNA polymerase ( Roche Life Sciences , Indianapolis , IN ) as specified ( Supplementary file 3 ) . For combined in situ hybridization and immunohistochemistry using foxc1a or foxc1b probes , we used rabbit anti-GFP antibody ( TP401 , Torrey Pines Biolabs , Secaucus , NJ ) to detect pou3f3b>GFP followed by goat anti-rabbit secondary antibody ( A11008 , Invitrogen , Carlsbad , CA ) . In situ hybridization and transgenic embryos were imaged with a Zeiss LSM800 confocal microscope . Maximum-intensity projections are shown for stable transgenic lines , and single representative sections are shown for injected embryos . Image levels were modified consistently across samples in Adobe Photoshop CS6 . Wild-type embryos double positive for fli1a:EGFP and sox10:Dsred ( 36 and 48 hpf ) , or col2a1a:GFP and sox10:Dserd ( 72 hpf ) , were sorted under a fluorescent dissecting microscope ( Leica M165FC ) before dissociation . For Foxc1 mutant analysis , we performed incrosses of pou3f3b:Gal4+/−; UAS:nls-GFP+/−; sox10:Dsred+/−; foxc1a+/−; foxc1b+/− fish and then selected for GFP+/Dsred+ embryos on a fluorescent dissecting microscope . Genotyping was then performed on tail lysates collected from individual embryos at 27 hpf . We then pooled foxc1a−/−; foxc1b−/− double mutants and separate sibling controls ( foxc1a+/−; foxc1b+/+ , and foxc1a+/+; foxc1b+/+ embryos ) for FACS . To facilitate embryo collection at the 36 hpf time point , embryos were moved at 27 hpf to an incubator set at 22°C to delay their development such that they reached 36 hpf the following morning . Cell dissociation and FACS were performed as previously described ( Askary et al . , 2017 ) . Around 5 , 000 cells of each sample were centrifuged at 500 g for 20 m at 4°C , and the pellet was suspended with 20 μL of lysis buffer ( 10 mM Tris–HCl [pH 7 . 4] , 5 mM MgCl2 , 10% DMF , 0 . 2% N-P40 ) by pipetting 6–10 times to release the nuclei without purification . The cell lysate was then mixed with 30 μL reaction buffer ( 10 mM Tris–HCl [pH 7 . 4] , 5 mM MgCl2 , 10% DMF , and homemade Tn5 Transposase ) by vortexing for 5 s . The reaction was incubated at 37°C for 20 min , followed by DNA purification using a Qiagen MiniElute kit . Purified DNA fragments were used to construct μATACseq libraries as previously described ( Buenrostro et al . , 2013 ) and sequenced using the NextSeq 500 platform ( Illumina ) with a minimum of 50 million paired-reads/sample . Two biological replicates of μATACseq experiments were performed for each condition . The Encode analysis pipeline ( https://github . com/ENCODE-DCC/chip-seq-pipeline ) for ATACseq ( Davis et al . , 2018 ) was used with small modifications . The raw reads were trimmed to 37 bp and aligned to the zebrafish GRCz10 genome assembly by STAR aligner ( Dobin et al . , 2013 ) . PCR duplicates , and the reads that aligned to ‘blacklist regions’ ( Amemiya et al . , 2019 ) , were removed , and then peaks were called by model-based analysis of ChIP-Seq ( MACS2 ) ( Zhang et al . , 2008 ) , with p=10−7 cutoff and disabled dynamic lambda option ( --nolambda ) for individual replicates . Only peaks common for two biological replicates ( Irreproducibility Discovery Rate < 0 . 1 ) were kept for further analysis . For visualization , bigwig files were generated from duplicates-removed bam files with bedtools and bedgraphtobigwig , and normalization was based on total read numbers . Heatmaps were generated with deeptools2 ( Ramírez et al . , 2016 ) based on the normalized bigwig signal files . Individual genomic loci were examined by IGV ( Broad Institute ) . For quantitatively comparing the accessibility of distal regulatory elements between two conditions , peaks from two conditions were first merged , and only distal regulatory regions ( 1 kb upstream or 0 . 5 kb downstream from the transcription start sites ) were kept for comparison . Raw read counts from each replicate were computed using ‘bedtools multicov’ function ( bedtools multicov -bams input . bam -bed peak . bed > count . txt ) . Raw read count matrices from four replicates ( two replicates per condition ) were analyzed using DESeq2 package ( Love et al . , 2014 ) . FDR = 0 . 1 was used as the cut off to filter the peaks that were differentially accessible between two conditions . Volcano plots were generated using ggplot2 package in R . Fold change and adjusted p value were outputs from DESeq2 package in R , which is the same analysis behind the heatmaps in Figure 3 . We used a −log10 adjusted p value greater than one as the cutoff for determining significantly changed peaks . HOMER ( Heinz et al . , 2010 ) was used to identify de novo motifs and their associated p values . David 6 . 8 GO analysis was performed on the web interface ( https://david . ncifcrf . gov/ ) based on nearest neighbor genes for all differentially accessible elements . Pearson correlation was calculated using cor ( ) function in R . Requests for material should be directed to J . Gage Crump ( gcrump@usc . edu ) .	Animals with backbones ( or vertebrates ) have body shape determined , in part , by their skeletons . These emerge in the embryo in the form of cartilage structures that will then get replaced by bone during development . The neural crest is a group of embryonic cells that can become different tissues . In the head , it forms the cartilage scaffold for some of the facial bones and the base of the skull . During this process , a protein called Sox9 is required for neural crest cells to morph into cartilage . This transcription factor binds to regulatory sequences in the genome to turn cartilage genes on . But Sox9 is also required to form non-cartilage tissues in organs such as the liver , lung , and kidneys . How , then , does Sox9 only turn on the genes required for cartilage formation in the embryonic face ? This specificity can be controlled by which regulatory sequences Sox9 can physically access in a cell: controlling which regulatory sequences Sox9 can access determines which genes it can activate , and which type of tissue a cell will become . Xu , Yu et al . wanted to understand exactly how Sox9 switches on the genes that turn neural crest cells into facial cartilage . They studied the genomes of zebrafish embryos , which have a cartilaginous skeleton similar to other vertebrates , and found out which areas were accessible to transcription factors in the neural crest cells that became facial cartilage . Analyzing these regions suggested that sites where Sox9 could bind were often close to binding sites for another protein , called Foxc1 . When zebrafish embryos were genetically modified to inactivate Foxc1 proteins , many of the regulatory sequences in cartilage failed to become accessible , and the cartilaginous skeleton did not form properly . These results support a model in which Foxc1 opens up the genomic regions that Sox9 needs to bind for cartilage to form , as opposed to the regions that Sox9 would bind to make different organ cell types . The findings of Xu , Yu et al . uncover the stepwise process by which cartilage cells are made during development . Further research based on these results could allow scientists to develop new ways of replacing cartilage in degenerative conditions such as arthritis .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "developmental", "biology", "short", "report" ]	2021	Foxc1 establishes enhancer accessibility for craniofacial cartilage differentiation
Aging , for virtually all life , is inescapable . However , within populations , biological aging rates vary . Understanding sources of variation in this process is central to understanding the biodemography of natural populations . We constructed a DNA methylation-based age predictor for an intensively studied wild baboon population in Kenya . Consistent with findings in humans , the resulting ‘epigenetic clock’ closely tracks chronological age , but individuals are predicted to be somewhat older or younger than their known ages . Surprisingly , these deviations are not explained by the strongest predictors of lifespan in this population , early adversity and social integration . Instead , they are best predicted by male dominance rank: high-ranking males are predicted to be older than their true ages , and epigenetic age tracks changes in rank over time . Our results argue that achieving high rank for male baboons – the best predictor of reproductive success – imposes costs consistent with a ‘live fast , die young’ life-history strategy . Aging , the nearly ubiquitous functional decline experienced by organisms over time ( López-Otín et al . , 2013 ) , is a fundamental component of most animal life histories ( Jones et al . , 2014 ) . At a physiological level , age affects individual quality , which in turn affects the ability to compete for mates and other resources , invest in reproduction , and maintain somatic integrity . At a demographic level , age is often one of the strongest predictors of survival and mortality risk , which are major determinants of Darwinian fitness . In order for patterns of aging to evolve , individuals must vary in their rates of biological aging . Thus , characterizing variation in biological aging rates and its sources – beyond simply chronological age – is an important goal in evolutionary ecology , with the potential to offer key insight into the trade-offs that shape individual life-history strategies ( Monaghan et al . , 2008 ) . Recent work suggests that DNA methylation data can provide exceptionally accurate estimates of chronological age ( Horvath and Raj , 2018 ) . These approaches typically use supervised machine learning methods that draw on methylation data from several hundred CpG sites , identified from hundreds of thousands of possible sites , to produce a single chronological age prediction ( Hannum et al . , 2013; Horvath , 2013; Levine et al . , 2018 ) . Intriguingly , some versions of these clocks also predict disease risk and mortality , suggesting that they capture aspects of biological aging that are not captured by chronological age alone ( Declerck and Vanden Berghe , 2018 ) . For example , in humans , individuals predicted to be older than their true chronological age are at higher risk of Alzheimer’s disease ( Levine et al . , 2015 ) , cognitive decline ( Levine et al . , 2015; Marioni et al . , 2015 ) , and obesity ( Horvath et al . , 2014 ) . Accelerated epigenetic age is in turn predicted by environmental factors with known links to health and lifespan , including childhood social adversity ( Jovanovic et al . , 2017; Raffington et al . , 2020 ) and cumulative lifetime stress ( Zannas et al . , 2015 ) . These observations generalize to other animals . Dietary restriction , for instance , decelerates biological aging based on DNA methylation clocks developed for laboratory mice and captive rhesus macaques , and genetic knockout mice with extended lifespans also appear epigenetically young for age ( Maegawa et al . , 2017; Petkovich et al . , 2017; Stubbs et al . , 2017 ) . However , while DNA methylation data have been used to estimate the age structure of wild populations ( where birthdates are frequently unknown ) ( De Paoli‐Iseppi , 2018; Polanowski et al . , 2014; Thompson et al . , 2017; Wright et al . , 2018 ) , they have not been applied to investigating sources of variance in biological aging in the wild . To do so here , we coupled genome-wide data on DNA methylation levels in blood with detailed behavioral and life-history data available for one of the most intensively studied wild mammal populations in the world , the baboons of the Amboseli ecosystem of Kenya ( Alberts and Altmann , 2012 ) . First , we calibrated a DNA methylation-based ‘epigenetic clock’ and assessed the clock’s composition . Second , we compared the accuracy of this clock against other age-associated traits and between sexes . Third , we tested whether variance in biological aging was explained by socioenvironmental predictors known to impact fertility or survival in this population . Finally , we investigated an intriguing association between epigenetic age acceleration and male dominance rank . Our results show that predictors of lifespan can be decoupled from rates of epigenetic aging . However , other factors – particularly male dominance rank – are strong predictors of epigenetic clock-based age acceleration . These results are the first to establish a link between social factors and epigenetic aging in any natural animal population . Together , they highlight potential sex-specific trade-offs that may maximize fitness , but also compromise physiological condition and potentially shorten male lifespan . We used a combination of previously published ( Lea et al . , 2016 ) and newly generated reduced-representation bisulfite sequencing ( RRBS ) data from 245 wild baboons ( N = 277 blood samples ) living in the Amboseli ecosystem of Kenya ( Alberts and Altmann , 2012 ) to generate a DNA methylation-based age predictor ( an ‘epigenetic clock’; Hannum et al . , 2013; Horvath , 2013 ) . Starting with a data set of methylation levels for 458 , 504 CpG sites genome-wide ( Figure 1—figure supplement 1; Supplementary file 1 ) , we used elastic net regression to identify a set of 573 CpG sites that together accurately predict baboon age within a median absolute difference ( MAD ) of 1 . 1 years ± 1 . 9 s . d . ( Figure 1; Supplementary file 1; Pearson’s r = 0 . 762 , p=9 . 70×10−54; median adult life expectancy in this population is 10 . 3 years for females and 7 . 9 for males; Colchero et al . , 2016 ) . The choice of these sites reflects a balance between increasing predictive accuracy within the sample and minimizing generalization error to unobserved samples , using a similar approach as that used to develop epigenetic clocks in humans ( Hannum et al . , 2013; Horvath , 2013 ) ( see also Materials and methods and Figure 1—figure supplement 2 ) . Consistent with findings in humans ( Horvath , 2013 ) , clock sites are enriched in genes , CpG islands , promoter regions , and putative enhancers , compared to the background set of all sites we initially considered ( i . e . , the 458 , 504 CpG sites that were candidates for inclusion in the clock; in humans , this background set is the set of analyzable sites on the Illumina 27K methylation array [Horvath , 2013; Figure 1—figure supplement 3]; Fisher’s exact tests , all p<0 . 05 ) . Clock sites are also more common in age-associated differentially methylated regions in baboons ( Figure 1—figure supplement 3; sites that increase with age: log2 ( odds ratio [OR] ) =4 . 189 , p=3 . 64×10−9; sites that decrease with age: log2 ( OR ) =5 . 344 , p=1 . 54×10−8 ) ( Lea et al . , 2015a ) , such that many , but not all , of the clock sites also exhibit individual associations between DNA methylation levels and age ( Figure 1—figure supplement 4 and Figure 2—figure supplement 1; Supplementary file 3 ) . Additionally , clock sites were more likely to be found in regions that exhibit enhancer-like activity in a massively parallel reporter assay ( sites that increase with age: log2 ( OR ) =2 . 685 , p=1 . 22×10−2; sites that decrease with age: log2 ( OR ) =4 . 789 , p=1 . 78×10−5 ) ( Lea et al . , 2018a ) and in regions implicated in the gene expression response to bacteria in the Amboseli baboon population ( overlap of lipopolysaccharide [LPS] up-regulated genes and sites that increase with age: log2[OR]=0 . 907 , p=7 . 03×10−4; overlap of LPS down-regulated genes and sites that decrease with age: log2[OR]=1 . 715 , p=1 . 55×10−3 ) ( Lea et al . , 2018b ) . Our results thus suggest that the Amboseli baboon epigenetic clock not only tracks chronological age , but also captures age-related changes in blood DNA methylation levels that are functionally important for gene regulation , particularly in relation to the immune system . Overall , the clock performed favorably relative to other morphological or biomarker predictors of age in this population . The epigenetic clock generally explained more variance in true chronological age , resulted in lower median error , and exhibited less bias than predictions based on raw body mass index ( BMI ) or blood cell composition data from flow cytometry or blood smears ( traits that change with age in baboons; Altmann et al . , 2010; Jayashankar et al . , 2003 ) . Its performance was comparable to molar dentine exposure , a classical marker of age ( Galbany et al . , 2011; Figure 1—figure supplement 5 ) . For a subset of 30 individuals , we had two samples collected at different points in time . The predicted ages from these longitudinally collected samples were older for the later-collected samples , as expected ( Figure 1C , D; binomial test p=5 . 95×10−5 ) . Furthermore , the change in epigenetic clock predictions between successive longitudinal samples positively predicted the actual change in age between sample dates ( β = 0 . 312 , p=0 . 027 , controlling for sex; difference between actual change and predicted change: mean 3 . 11 years ± 3 . 25 s . d . ) . However , clock performance was not equivalent in males and females . Specifically , we observed that the clock was significantly more accurate in males ( Figure 1; males: N = 135; MAD = 0 . 85 years±1 . 0 s . d . ; Pearson’s r = 0 . 86 , p=5 . 49×10−41; females: N = 142; MAD = 1 . 6 years±2 . 4 s . d . ; r = 0 . 78 , p=6 . 78×10−30; two-sided Wilcoxon test for differences in absolute error by sex: p=4 . 37×10−9 ) . Sex differences were also apparent in the slope of the relationship between predicted age and chronological age . Males show a 2 . 2-fold higher rate of change in predicted age , as a function of chronological age , compared to females ( Figure 1A , B; chronological age by sex interaction in a linear model for predicted age: β = 0 . 448 , p=9 . 66×10−19 , N = 277 ) . Interestingly , sex differences are not apparent in animals <8 years , which roughly corresponds to the age at which the majority of males have achieved adult dominance rank and dispersed from their natal group ( Alberts and Altmann , 1995a; Alberts and Altmann , 1995b; Alberts et al . , 2003 ) ( N = 158 , chronological age by sex interaction β = −0 . 038 , p=0 . 808 ) . Rather , sex differences become apparent after baboons have reached full physiological and social adulthood ( N = 119 , chronological age by sex interaction β = 0 . 459 , p=9 . 74×10−7 in animals ≥ 8 years ) , when divergence between male and female life-history strategies is most marked ( Alberts and Altmann , 1995a; Alberts and Altmann , 1995b; Alberts et al . , 2003 ) and when aging rates between the sexes are predicted to diverge ( Clutton-Brock and Isvaran , 2007; Kirkwood and Rose , 1991; Williams , 1957 ) . Because of these differences , we separated males and females for all subsequent analyses . However , we note that the effects of age on DNA methylation levels at individual clock sites are highly correlated between the sexes ( Pearson’s r = 0 . 91 , p=3 . 35×10−204 ) , with effect sizes that are , on average , more precisely estimated in males ( paired t-test p=4 . 53×10−74 for standard errors of βage; Figure 1—figure supplement 4 ) . In other words , the sex differences in clock performance reflect changes in methylation that occur at the same CpG sites , but with higher variance in females . Lower accuracy in females compared to males therefore appears to result from the greater variability in DNA methylation change in older females ( Figure 1 ) . Although the baboon epigenetic clock is a good predictor of age overall , individuals were often predicted to be somewhat older or younger than their known chronological age . In humans and some model systems , the sign and magnitude of this deviation captures information about physiological decline and/or mortality risk beyond that contained in chronological age alone ( Maegawa et al . , 2017; Petkovich et al . , 2017; Stubbs et al . , 2017; Ryan et al . , 2020 ) . To test whether this observation extends to wild baboons , we focused on four factors of known importance to fitness in the Amboseli population . First , we considered cumulative early adversity , which is a strong predictor of shortened lifespan and offspring survival for female baboons ( Tung et al . , 2016; Zipple et al . , 2019 ) . We measured cumulative adversity as a count of major adverse experiences suffered in early life , including low maternal social status , early-life drought , a competing younger sibling , maternal loss , and high experienced population density ( i . e . , social group size ) . Second , we considered social bond strength in adulthood , which positively predicts longer adult lifespan in baboons , humans , and other wild social mammals ( Archie et al . , 2014a; Campos et al . , 2020; Holt-Lunstad et al . , 2010; Snyder-Mackler et al . , 2020 ) . Third , we considered dominance rank , which is a major determinant of access to mates , social partners , and other resources in the Amboseli baboons ( Archie et al . , 2014a; Alberts et al . , 2006; Gesquiere et al . , 2018; Lea et al . , 2015b ) . Finally , we considered BMI , a measure of body condition that , in the Amboseli baboons , primarily reflects lean muscle mass ( mean body fat percentages have been estimated at <2% in adult females and <9% in adult males ) ( Altmann et al . , 1993 ) . Because raw BMI ( i . e . , BMI not correcting for age ) also tracks growth and development ( increasing as baboons reach their prime and then declining thereafter [Altmann et al . , 2010; Figure 2—figure supplement 2]; Pearson’s r in males between rank and raw BMI = −0 . 56 , p=6 . 38×10−9 ) , we calculated BMI relative to the expected value for each animal’s age using piecewise regression , which also eliminates correlations between BMI and male rank ( Pearson’s r = −0 . 070 , p=0 . 504 ) . We refer to this adjusted measure of BMI as age-adjusted BMI . Because high cumulative early adversity and low social bond strength are associated with increased mortality risk in the Amboseli baboons , we predicted that they would also be linked to increased epigenetic age . For rank and age-adjusted BMI , our predictions were less clear: improved resource access could conceivably slow biological aging , but increased investment in growth and reproduction ( either through higher fertility in females or physical competition for rank in males ) could also be energetically costly . To investigate these possibilities , we modeled the deviation between predicted age and known chronological age ( Δage ) as a function of cumulative early adversity , ordinal dominance rank , age-adjusted BMI , and for females , social bond strength to other females . Social bond strength was not included in the model for males , as this measure was not available for a large proportion of males in this data set ( 53 . 8% ) . We also included chronological age as a predictor in the model , as epigenetic age tends to be systematically overpredicted for young individuals and underpredicted for old individuals ( Figure 1A , B; this bias has been observed in both foundational work on epigenetic clocks [Hannum et al . , 2013] and recent epigenetic clocks calibrated for rhesus macaques [Horvath , 2020] , as well as for elastic net regression analyses more generally [Engebretsen and Bohlin , 2019] ) . Including chronological age in the model , as previous studies have done ( Hannum et al . , 2013; Levine et al . , 2018 ) , controls for this compression effect . None of the predictor variables were strongly linearly correlated ( all Pearson’s r < 0 . 35; Supplementary file 4 ) . Surprisingly , despite being two of the strongest known predictors of lifespan in wild baboons , neither cumulative early-life adversity nor social bond strength explain variation in Δage ( Table 1 ) . In contrast , high male dominance rank is strongly and significantly associated with larger values of Δage ( β = −0 . 078 , p=7 . 39×10−4; Figure 2; Table 1; Figure 2—figure supplement 3 ) . Alpha males are predicted to be an average of 10 . 95 months older than their true chronological age – a difference that translates to 11 . 5% of a male baboon’s expected adult lifespan in Amboseli ( Colchero et al . , 2016 ) . In contrast , dominance rank did not predict Δage in females ( p=0 . 228; Table 1 ) . Finally , age-adjusted BMI also predicted Δage in males ( p=6 . 33×10−3 ) , but not in females ( p=0 . 682; Table 1 ) . These results are robust to inclusion of read depth and bisulfite conversion rate as covariates in the model ( Supplementary file 5; in males , read depth is correlated with chronological age [R2 = −0 . 409 , p=0 . 038] , but is not correlated with Δage [R2 = 0 . 003 , p=0 . 561] ) . Despite the tendency for high-ranking males to have higher raw BMI due to increased muscle mass , the effects of rank and age-adjusted BMI in males are distinct . Specifically , modeling dominance rank after adjusting for raw BMI also produces a significant effect of rank on Δage in the same direction ( p=9 . 93×10−4; Supplementary file 5 ) , as does substituting the age-adjusted BMI measure for either raw BMI or the residuals of raw BMI after adjusting for dominance rank ( rank p=1 . 52×10−2 and p=1 . 88×10−4 , respectively; Supplementary file 5 ) . In contrast , BMI is only a significant predictor of male Δage when corrected for age ( i . e . , age-adjusted ) and when rank is included in the same model ( Table 1; Supplementary file 5 ) . Furthermore , we obtain the same qualitative results if all low BMI males are removed from the sample ( BMI < 41; this cut-off was used because it drops all young males who have clearly not reached full adult size; p=7 . 14×10−3; Supplementary file 5 ) . Dropping these males also eliminates the age-raw BMI correlation ( Pearson’s r = −0 . 16 , p=0 . 21 ) . In baboon males , achieving high rank depends on physical condition and fighting ability ( Alberts et al . , 2003 ) . Consequently , rank in males is dynamic across the life course: males tend to attain their highest rank between 7 and 12 years of age and fall in rank thereafter ( Figure 2—figure supplement 4 ) . Thus , nearly all males in the top four rank positions in our data set were between 7 and 12 years of age at the time they were sampled ( however , because not all 7–12 year olds are high ranking , low-rank positions include males across the age range; Supplementary file 1 , Figure 2—figure supplement 4 ) . We therefore asked whether the association between high rank in males and accelerated epigenetic aging is a function of absolute rank values , regardless of age , or deviations from the expected mean rank given a male’s age ( i . e . , ‘rank-for-age’; Figure 2—figure supplement 4 ) . We found that including rank-for-age as an additional covariate in the Δage model recapitulates the significant effect of ordinal male rank ( p=0 . 045 ) , but finds no effect of rank-for-age ( p=0 . 819; Supplementary file 5 ) . Our results therefore imply that males incur the costs of high rank primarily in early- to mid-adulthood , and only if they succeed in attaining high rank . If attainment of high rank is linked to changes in epigenetic age within individuals , this pattern should be reflected in longitudinal samples . Specifically , males who improved in rank between samples should look older for age in their second sample relative to their first and vice versa . To assess this possibility , we calculated ‘relative epigenetic age’ ( the residuals of the best-fit line relating chronological age and predicted age ) for 14 males for whom we had repeated samples over time , 13 of whom changed ranks across sample dates ( N = 28 samples , two per male ) . Samples collected when males were higher status predicted higher values of relative epigenetic age compared to samples collected when they were lower status ( Figure 3; paired t-test , t = −2 . 99 , p=0 . 011 ) . For example , in the case of a male whom we first sampled at low status ( ordinal rank = 18 ) and then after he had attained the alpha position ( ordinal rank 1 ) , the actual time that elapsed between samples was 0 . 79 years , but he exhibited an increase in predicted age of 2 . 6 years . Moreover , the two males that showed a decrease in predicted age , despite increasing in chronological age ( Figure 1D ) , were among those that experienced the greatest drop in social status between samples . Thus , change in rank between samples for the same male predicts change in Δage , controlling for chronological age ( R2 = 0 . 37 , p=0 . 021 ) . Consistent with our cross-sectional results , we found a suggestive relationship between change in Δage and BMI ( R2 = 0 . 31 , p=0 . 08 ) . Here , too , the effect of dominance rank does not seem to be driven by BMI: while the association between change in Δage and change in rank is no longer significant when modeling rank after adjusting for raw BMI , the correlation remains consistent ( R2 = 0 . 20 , p=0 . 167 ) . In contrast , raw BMI adjusted for rank explains almost none of the variance in change in Δage ( R2 = 0 . 01 , p=0 . 779 ) . Together , our findings indicate that major environmental predictors of lifespan and mortality risk – particularly social bond strength and early-life adversity in this population – do not necessarily predict epigenetic measures of biological age . Although this assumption is widespread in the literature , including for epigenetic clock analyses ( Liu et al . , 2019; Shalev and Belsky , 2016 ) , our results are broadly consistent with empirical results in humans . Specifically , while studies of early-life adversity , which also predicts lifespan in human populations , find relatively consistent support for a relationship between early adversity and accelerated epigenetic aging in children and adolescents ( Jovanovic et al . , 2017; Raffington et al . , 2020; Brody et al . , 2016a; Brody et al . , 2016b; Davis et al . , 2017; Marini , 2018; Sumner et al . , 2019 ) , there is little evidence for the long-term effects of early adversity on epigenetic age in adulthood ( Zannas et al . , 2015; Austin et al . , 2018; Boks et al . , 2015; Lawn et al . , 2018; Simons et al . , 2016; Wolf et al . , 2018 ) . Thus , while DNA methylation may make an important contribution to the biological embedding of early adversity into adulthood ( Aristizabal et al . , 2020; Hertzman , 2012 ) , it does not seem to do so through affecting the epigenetic clock itself . Social and environmental effects on the clock instead seem to be most influenced by concurrent conditions , lending support to ‘recency’ models for environmental effects on aging that posit that health is more affected by the current environment than past experience ( Ben-Shlomo and Kuh , 2002; Shanahan et al . , 2011; Shanahan and Hofer , 2011 ) . Additional longitudinal sampling will be necessary to evaluate whether current conditions alone can explain accelerated epigenetic aging or whether it also requires integrating the effects of exposures across the life course ( the ‘accumulation’ model; Ben-Shlomo and Kuh , 2002; Shanahan and Hofer , 2011 ) . Alternatively , the effects of early-life adversity and social bond strength may act through entirely distinct pathways than those captured by our epigenetic clock ( including targeting tissues or cell types that we were unable to assess here ) . Indeed , the proliferation of alternative epigenetic clocks in humans has revealed that the clocks that best predict chronological age are not necessarily the clocks that most closely track environmental exposures , and the same is likely to be true in other species ( Levine et al . , 2018; Belsky et al . , 2020 ) . Notably , the functional significance of the clock – that is , whether it reflects the mechanisms that causally drive aging , or instead serves as a passive biomarker – also remains unclear . We found that the most robust socioenvironmental predictor of epigenetic age in the Amboseli baboons is male dominance rank , with a secondary effect of age-adjusted BMI observable when rank is included in the same model . Although high BMI also predicts accelerated epigenetic age in some human populations ( Ryan et al . , 2020 ) , high BMI in these human populations is related to being overweight or obese . In contrast , because wild-feeding baboons in Amboseli are extremely lean ( Altmann et al . , 1993 ) , the range of BMI in most human populations is distinct from the range exhibited by our study subjects ( importantly , BMI in humans is calculated differently than BMI in baboons [see Materials and methods] , and therefore the BMI scales are species specific ) . Instead , the higher BMI values in our dataset represent baboons in better body condition ( more muscle mass ) . Given that rank in male baboons is determined by physical fighting ability ( Alberts et al . , 2003 ) , these results suggest that investment in body condition incurs physiological costs that accelerate biological age . If so , the rank effect we observe may be better interpreted as a marker of competitiveness , not as a consequence of being in a ‘high rank’ environment . In support of this idea , work on dominance rank and gene expression levels in the Amboseli baboons suggests that gene expression differences associated with male dominance rank tend to precede attainment of high rank , rather than being a consequence of behaviors exhibited after high rank is achieved ( Lea et al . , 2018b ) . Consistent with potential costs of attaining or maintaining high status , alpha males in Amboseli also exhibit elevated glucocorticoid levels ( Gesquiere et al . , 2011 ) , increased expression of genes involved in innate immunity and inflammation ( Lea et al . , 2018b ) , and a trend toward elevated mortality risk ( Campos et al . , 2020 ) . Males who can tolerate these costs and maintain high rank are nevertheless likely to enjoy higher lifetime reproductive success , since high rank is the single best predictor of mating and paternity success in baboon males ( Alberts et al . , 2003 ) . This interpretation may also explain major sex differences in the effects of rank on epigenetic age , where dominance rank shows no detectable effect in females . Dominance rank in female baboons is determined by nepotism , not physical competition: females typically insert into rank hierarchies directly below their mothers , and hierarchies therefore tend to remain stable over time ( and even intergenerationally ) ( Hausfater et al . , 1982 ) . Our results contribute to an emerging picture in which dominance rank effects on both physiological and demographic outcomes are asymmetrical across sexes , and larger in males . Specifically , in addition to Δage , male rank is a better predictor of immune cell gene expression and glucocorticoid levels than female rank ( Lea et al . , 2018b; Gesquiere et al . , 2011; Levy et al . , 2020 ) . Recent findings suggest that high rank may also predict increased mortality risk in male Amboseli baboons , whereas neither high rank nor low rank predicts increased mortality risk in females ( Campos et al . , 2020 ) . Together , these results argue that social status/dominance rank effects should not be interpreted as a universal phenomenon . Instead , the manner through which social status is achieved and maintained is likely to be key to understanding its consequences for physiology , health , and fitness ( Simons and Tung , 2019 ) . Specifically , we predict that high status will be most likely to accelerate the aging process , including epigenetic age , in species-sex combinations where high status increases reproductive success or fecundity , and achieving status is energetically costly ( e . g . , male red deer , mandrills , and geladas; female meerkats Clutton-Brock et al . , 2006; Clutton-Brock and Huchard , 2013; Emery Thompson and Georgiev , 2014 ) . Expanding studies of biological aging to a broader set of natural populations , especially those for which behavioral and demographic data are also available , will be key to testing these predictions . This study focused on a longitudinally monitored population of wild baboons ( Papio cynocephalus , the yellow baboon , with some admixture from the closely related anubis baboon P . anubis Alberts and Altmann , 2001; Tung et al . , 2008 ) in the Amboseli ecosystem of Kenya . This population has been continuously monitored by the Amboseli Baboon Research Project ( ABRP ) since 1971 ( Alberts and Altmann , 2012 ) . For the majority of study subjects ( N = 242 of 245 individuals ) , birth dates were therefore known to within a few days’ error; for the remaining three individuals , birth dates were known within 3 months’ error ( Supplementary file 1 ) . All DNA methylation data were generated from blood-derived DNA obtained during periodic darting efforts , as detailed in Lea et al . , 2018b; Altmann et al . , 1996; Tung et al . , 2015 . Samples were obtained under approval from the Institutional Animal Care and Use Committee ( IACUC ) of Duke University ( currently #A044-21-02 ) and adhered to all the laws and regulations of Kenya . In brief , individually recognized study subjects were temporarily anesthetized using a Telazol-loaded dart delivered through a blow gun . Baboons were then safely moved to a new location where blood samples and morphometric data , including body mass and crown-rump length , were collected . Baboons were then allowed to recover from anesthesia in a covered holding cage and released to their group within 2–4 hr . Blood samples were stored at −20° C in Kenya until export to the United States . DNA methylation data were generated from blood-extracted DNA collected from known individuals in the Amboseli study population ( N = 277 samples from 245 animals; 13 females and 15 males were each sampled twice , and 1 female and 1 male were each sampled three times ) . Here , we analyzed a combined data set that included previously published RRBS ( Meissner et al . , 2005 ) data from the same population ( N = 36 samples ) ( Lea et al . , 2016 ) and new RRBS data from 241 additional samples . RRBS libraries were constructed following Boyle et al . , 2012 , using ~200 ng baboon DNA plus 0 . 2 ng unmethylated lambda phage DNA per sample as input . Samples were sequenced to a mean depth of 17 . 8 ( ±10 . 5 s . d . ) million reads on either the Illumina HiSeq 2000 or HiSeq 4000 platform ( Supplementary file 1 ) , with an estimated mean bisulfite conversion efficiency ( based on the conversion rate of lambda phage DNA ) of 99 . 8% ( minimum = 98 . 1% ) . Sequence reads were trimmed with Trim Galore ! ( Krueger , 2012 ) to remove adapters and low quality sequence ( Phred score < 20 ) . Trimmed reads were mapped with BSMAP ( Xi and Li , 2009 ) to the baboon genome ( Panu2 . 0 ) , allowing a 10% mismatch rate to account for the degenerate composition of bisulfite-converted DNA . We used autosomally mapped reads to count the number of methylated and total reads per CpG site , per sample ( Xi and Li , 2009 ) . To control for possible local genetic variation , we used BSMAP’s rescaled ‘effective total counts’ measures , which adjusts for the presence of possible CpG site disrupting genetic variants . Importantly , although our population consists of hybrids , previous work on DNA methylation variation across baboon species shows that species differences have a negligible effect on quantifying DNA methylation ( i . e . , the rate of incorrect calls differs by <0 . 4% between anubis and yellow baboons , the two species that contribute to ancestry in Amboseli; Vilgalys et al . , 2019 ) . Following Lea et al . , 2016; Lea et al . , 2015a , CpG sites were filtered to retain sites with a mean methylation level between 0 . 1 and 0 . 9 ( i . e . , to exclude constitutively hyper- or hypo-methylated sites ) and mean coverage of ≥5× . We also excluded any CpG sites with missing data for ≥5% of individuals in the sample . After filtering , we retained N = 458 , 504 CpG sites for downstream analysis . For the remaining missing data ( mean number of missing sites per sample = 1 . 4 ± 3 . 5% s . d . , equivalent to 6409 ± 16 , 024 s . d . sites ) , we imputed methylation levels using a k-nearest neighbors approach in the R package impute , using default parameters ( Hastie et al . , 2001 ) . We used the R package glmnet ( Friedman et al . , 2009 ) version 2 . 0 . 10 to build a DNA methylation clock for baboons . Specifically , we fit a linear model in which the predictor variables were normalized levels of DNA methylation at 458 , 504 candidate clock CpG sites across the genome and the response variable was chronological age . To account for the excess of CpG sites relative to samples , glmnet uses an elastic net penalty to shrink predictor coefficients toward 0 ( Friedman et al . , 2010 ) . Optimal alpha parameters were identified by grid searching across a range of alphas from 0 ( equivalent to ridge regression ) to 1 ( equivalent to Lasso ) by increments of 0 . 1 , which impacts the number of clock CpG sites by varying the degree of shrinkage of the predictor coefficients toward 0 ( Figure 1—figure supplement 2 ) . We defined the optimal alpha as the value that maximized R2 between predicted and true chronological age across all samples . We set the regularization parameter lambda to the value that minimized mean-squared error during n-fold internal cross-validation . To generate predicted age estimates for a given sample , we used a leave-one-out cross-validation approach in which all samples but the ‘test’ sample were included for model training , and the resulting model was used to predict age for the left-out test sample . To avoid leaking information from the training set into the test set , and to maximize the generalizability of the clock , we did not remove batch effects from the quantile normalized methylation ratio estimates . However , we confirmed that our results in the main model , for both males and for females , were robust if we added batch effect ( previously generated samples [n = 36] versus newly generated samples [n = 241] ) as a covariate . Training samples were scaled independently of the test sample in each leave-one-out model to avoid bleed-through of information from the test data into the training data . To do so , we first quantile normalized methylation ratios ( the proportion of methylated counts to total counts for each CpG site ) within each sample to a standard normal distribution . Training samples were then separated from the test sample and the methylation levels for each CpG site in the training set were quantile normalized across samples to a standard normal distribution . To predict age in the test sample , we compared the methylation value for each site in the test sample to the empirical cumulative distribution function for the training samples ( at the same site ) to estimate the quantile in which the training sample methylation ratio fell . The training sample was then assigned the same quantile value from the standard normal distribution using the function qnorm in R . To evaluate whether CpG sites included in the epigenetic clock , relative to the 458 , 504 CpG background sites , were enriched in functionally important regions of the baboon genome ( Lea et al . , 2015a; Vilgalys et al . , 2019 ) , we used two-sided Fisher’s exact tests to investigate enrichment/depletion of the 573 epigenetic clock sites in ( 1 ) gene bodies and exons , based on the Ensembl annotation Panu2 . 0 . 90; ( 2 ) CpG islands annotated in the UCSC Genome Browser; ( 3 ) CpG shores , defined as the 2000 basepairs flanking CpG islands ( following Lea et al . , 2015a; Vilgalys et al . , 2019; Irizarry et al . , 2009 ) ; and ( 4 ) promoter regions , defined as the 2000 basepairs upstream of the 5′-most annotated transcription start site for each gene ( following Lea et al . , 2015a; Vilgalys et al . , 2019 ) . We also considered ( 5 ) putative enhancer regions , which have not been annotated for the Panu2 . 0 assembly . We therefore used ENCODE H3K4me1 ChIP-seq data from human peripheral blood mononuclear cells ( PBMCs ) ( ENCODE Project Consortium , 2012 ) and the liftOver tool to define likely enhancer coordinates in Panu2 . 0 . We also tested for enrichment of clock sites in regions of the genome that have been identified by previous empirical studies to be of special interest . First , we considered regions that likely have regulatory activity in blood cells , defined as all 200 base-pair windows that showed evidence of enhancer activity in a recently performed massively parallel reporter assay ( Lea et al . , 2018a ) . We used liftOver to identify the inferred homologous Panu2 . 0 coordinates for these windows , which were originally defined in the human genome . Second , we defined age-related differentially methylated regions in the Amboseli baboons based on genomic intervals found , in previous analyses , to contain at least three closely spaced age-associated CpG sites ( inter-CpG distance ≤1 kb ) , as described in Lea et al . , 2015a . Third , because inflammatory processes involved in innate immunity are strongly implicated in the aging process , we defined LPS up-regulated and LPS down-regulated genes as those genes that were significantly differentially expressed ( 1% false discovery rate ) between unstimulated Amboseli baboon white blood cells and LPS-stimulated cells from the same individual , following 10 hr of culture in parallel ( Lea et al . , 2018b ) . To assess the utility of the DNA methylation clock relative to other data types , we compared its predictive accuracy to clocks based on three other age-related phenotypes: tooth wear ( percent molar dentine exposure; Galbany et al . , 2011 ) , body condition ( BMI; Altmann et al . , 2010 ) , and blood cell type composition ( blood smear counts and lymphocyte/monocyte proportions from flow cytometry performed on peripheral blood mononuclear cells , as in Lea et al . , 2018b; Snyder-Mackler et al . , 2016 ) . Leave-one-out model training and prediction were performed for each data type using linear modeling . To compare the relative predictive accuracy of each data type , we calculated the R2 between predicted and chronological age , the MAD between predicted and chronological age , and the bias in age predictions ( the absolute value of 1 − slope of the best-fit line between predicted and chronological age ) ( Figure 1—figure supplement 5 ) . We asked whether factors known to be associated with inter-individual variation in fertility or survival also predict inter-individual variation in Δage ( predicted age from the epigenetic clock minus known chronological age ) . To do so , we fit linear models separately for males and females , with Δage as the dependent variable and dominance rank at the time of sampling , cumulative early adversity , age-adjusted BMI , and chronological age as predictor variables ( Tung et al . , 2016 ) . For females , we also included a measure of social bond strength to other females as a predictor variable , based on findings that show that socially isolated females experience higher mortality rates in adulthood ( Archie et al . , 2014a; Silk et al . , 2010 ) . Samples with missing values for any of the predictor variables were excluded in the model , resulting in a final analysis set of 66 female samples ( from 59 females ) and 93 male samples ( from 84 males ) . The chronological ages of samples with complete data relative to samples with missing data were equivalent for females ( t-test , t = 1 . 95 , p=0 . 053 ) but were slightly lower for males ( t-test , t = −3 . 04 , p=0 . 003; mean chronological ages are 7 . 98 and 9 . 65 years for complete and missing samples , respectively ) . Predictor variables were measured as follows . To test whether changes in rank predict changes in relative epigenetic age within individuals , we used data from 11 males from the original data set and generated additional RRBS data for nine samples , resulting in a final set of 14 males who each were sampled at least twice in the data set , 13 of whom occupied different ordinal ranks at different sampling dates ( mean years elapsed between samples = 3 . 7 ± 1 . 65 s . d . ; mean absolute difference in dominance ranks = 1 . 29 ± 8 . 34 s . d . ) . This effort increased our total sample size to N = 286 samples from 248 unique individuals . To incorporate the new samples into our analysis , we reperformed leave-one-out age prediction with N-fold internal cross-validation at the optimal alpha selected for the original N = 277 samples ( alpha = 0 . 1 ) . For the 277 samples carried over from the original analysis , we verified that age predictions were nearly identical between the previous analysis and the expanded data set ( R2 = 0 . 98 , p=2 . 21×10−239; Supplementary file 1 ) . Based on the new age predictions for males in the data set ( N = 140 ) , we again calculated relative epigenetic age as the residual of the best-fit line relating predicted age to chronological age . We then used the 14 males with repeated DNA methylation profiles and rank measures in this data set to test whether , within individuals , changes in dominance rank or rank-for-age explained changes in relative epigenetic age between samples . In total , five males were sampled three times . For four of these five , we only included the two samples that were sampled the farthest apart in time ( i . e . , excluded the temporal middle sample ) to maximize the age change between sample dates . For the fifth male , BMI information was missing for the third sample , so we included the first two samples collected in time . All R code used to analyze data in this study is available at https://github . com/janderson94/BaboonEpigeneticAging; Anderson , 2021; with a copy archived at swh:1:rev:58ca836d3416c2a447cbd055aee66c11140aec86 .	For most animals , age is one of the strongest predictors of health and survival , but not all individuals age at the same rate . In fact , animals of the same species can have different 'biological ages' even when they have lived the same number of years . In humans and other mammals this variation in aging shows up in chemical modifications known as DNA methylation marks . Some researchers call these marks 'epigenetic' , which literally means 'upon the genes' . And some DNA methylation marks change with age , so their combined pattern of change is often called the ‘epigenetic clock’ . Environmental stressors , such as smoking or lack of physical activity , can make the epigenetic clock ‘tick’ faster , making the DNA of some individuals appear older than expected based on their actual age in years . These ‘biologically older’ individuals may also experience a higher risk of age-related disease . Studies in humans have revealed some of the reasons behind this fast biological aging , but it is unclear whether these results apply in the wild . It is possible that early life events trigger changes in the epigenetic clock , affecting health in adulthood . In primates , for example , adversity in early life has known effects on fertility and survival . Low social status also has a negative effect on health . To find out whether early experiences and the social environment affect the epigenetic clock , Anderson , Johnston et al . tracked DNA methylation marks in baboons . This revealed that epigenetic clocks are strong predictors of age in wild primates , but neither early adversity nor the strength of social bonds affected the rate at which the clocks ticked . In fact , it was competition for social status that had the most dramatic effect on the clock’s speed . Samples of males taken at different times during their lives showed that their epigenetic clocks sped up or slowed down as they moved up or down the social ladder , reflecting recent social experiences , rather than events early in their lives . On average , epigenetic clock measurements overestimated the age in years of alpha males by almost a year , showing that fighting to be on top comes at a cost . This study highlights one way in which the social environment can influence aging . The next step is to understand how health is affected by the ways that animals attain social status . This could help researchers who study evolution understand how social interactions and environmental conditions affect survival and reproduction . It could also provide insight into the effects of social status on human health and aging .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "genetics", "and", "genomics" ]	2021	High social status males experience accelerated epigenetic aging in wild baboons
A quantitative understanding of how sensory signals are transformed into motor outputs places useful constraints on brain function and helps to reveal the brain’s underlying computations . We investigate how the nematode Caenorhabditis elegans responds to time-varying mechanosensory signals using a high-throughput optogenetic assay and automated behavior quantification . We find that the behavioral response is tuned to temporal properties of mechanosensory signals , such as their integral and derivative , that extend over many seconds . Mechanosensory signals , even in the same neurons , can be tailored to elicit different behavioral responses . Moreover , we find that the animal’s response also depends on its behavioral context . Most dramatically , the animal ignores all tested mechanosensory stimuli during turns . Finally , we present a linear-nonlinear model that predicts the animal’s behavioral response to stimulus . An animal’s nervous system interprets sensory signals to guide behavior , including behaviors that are involved in evading predation . Investigating how the nervous system processes these signals is a critical step towards understanding neural function . Mechanosensation in the nematode Caenorhabditis elegans is an attractive platform for investigating sensorimotor processing . Six soft-touch mechanosensory neurons arranged throughout the body detect mechanical stimuli including those delivered either by a small probe in what is called a touch or by striking the petri dish containing the animal in what is called a tap ( Chalfie and Sulston , 1981 ) . Despite decades of investigation , however , the behavioral response to dynamic time-varying mechanosensory signals has not been fully explored . Here we provide new details about the mechanosensory response system by quantitatively exploring the animal’s detailed behavioral response to rich , dynamically varying signals . We find that the animal responds to the temporal features of signals in its mechanosensory neurons , such as its time-derivative ( i . e . rate of change ) , that extend over many seconds . Moreover , we find evidence that the animal’s sensorimotor response depends on the animal’s current behavior state . That we find evidence of temporal processing and context dependency , even in the nematode’s relatively simple touch circuit , raises the possibility that these features could be ubiquitous across sensory systems . Finally , we present a simple quantitative model that predicts the animal’s response to novel mechanosensory signals . Mechanosensation is important for C . elegans survival . Caenorhabditis elegans are preyed upon by nematophagous fungi , and touch-defective animals fail to detect and escape from the fungus ( Maguire et al . , 2011 ) . Much is already known about this critical circuit . The six soft-touch mechanosensory neurons detect both spatially localized and non-localized stimuli . Anterior touches are detected by anterior neurons ALML , ALMR and AVM and evoke reversal behaviors whereas posterior touches are detected by posterior neurons PLML and PLMR and evoke forward sprints ( Chalfie and Sulston , 1981; Chalfie et al . , 1985; McClanahan et al . , 2017; Mazzochette et al . , 2018 ) . Non-spatially localized plate taps are detected by both anterior and posterior soft-touch neurons and evoke reversals in young adult animals ( Chalfie and Sulston , 1981; Rankin et al . , 1990 ) ; on rare occasions , they also evoke forward acceleration ( Wicks and Rankin , 1995; Chiba and Rankin , 1990 ) . Owing in part to its ease of delivery and its inherent compatibility with high-throughput methods ( Swierczek et al . , 2011 ) , plate tap emerged early on as an assay for studying sensitization and habituation ( Rankin et al . , 1990 ) . Plate tap has been used in concert with the touch assay to study the development , circuitry ( Chalfie and Sulston , 1981 ) , genes , molecules and receptors ( Sanyal et al . , 2004; Kindt et al . , 2007 ) of the mechanosensory system . When the animal interacts with its environment or brushes up against a nematophagous fungi’s constricting ring , it necessarily receives time-varying stimuli . The response of an individual touch receptor neuron to force ( O'Hagan et al . , 2005 ) , including to time-varying stimuli , is well characterized ( Eastwood et al . , 2015 ) . The onset and offset of an applied force evokes strong excitatory currents that adapt with a timescale of a few tens of milliseconds ( O'Hagan et al . , 2005 ) and have a frequency response thought to peak in the 100 to 500 Hz range ( Eastwood et al . , 2015 ) . Intracellular calcium activity in individual soft touch neurons has also been well characterized in response to touch and this activity exhibits slower transients that occur with a timescale of seconds ( Suzuki et al . , 2003; Cho et al . , 2018 ) . In contrast to this detailed understanding at the single neuron level , the animal’s downstream response to rich temporally varying mechanosensory signals has been less well characterized . The animal’s behavior response to mechanosensory stimuli has primarily been studied using impulse stimuli . Specifically , the stimuli were either a brief application of touch , tap or optogenetic stimulation , and the most salient feature of these stimuli was their amplitude , not their temporal profile ( Petzold et al . , 2013; Stirman et al . , 2011; McClanahan et al . , 2017; Mazzochette et al . , 2018 ) . In the classical touch assay , for example , a saturating force that lasts just a few tenths of a second is applied ( Nekimken et al . , 2017 ) . Tap stimuli are even shorter in duration . To our knowledge , the only temporally varying stimuli used to investigate behavioral responses to mechanosensation are: trains of taps or touches ( Chiba and Rankin , 1990; Kitamura et al . , 2001 ) , trains of optogenetic pulses ( Porto et al . , 2017; Leifer et al . , 2011 ) , trains of ultrasound pulses ( Kubanek et al . , 2018 ) , the delivery of 100 Hz or 1 kHz acoustic vibration ( Nagy et al . , 2014a , 2014b; Sugi et al . , 2016 ) , and the delivery of sustained acoustic vibrations of different frequencies lasting many minutes to hours ( Chen and Chalfie , 2014 ) . The following behaviors have been extensively studied in response to mechanosensory stimulation . Early work scored the animal’s reversals ( Chiba and Rankin , 1990 ) and more recent work includes reversal distance ( Kitamura et al . , 2001 ) , rate of reversals ( Swierczek et al . , 2011 ) or pauses , reversal duration and reversal latency ( Ardiel et al . , 2017 ) . The effect of mechanosensory stimulation on accelerations has also been studied ( Wicks and Rankin , 1995 ) . Recent work , however , shows that the animal’s repertoire of behavior is larger ( Stephens et al . , 2008; Brown et al . , 2013 ) . Over short timescales , reversals or accelerations depend on the set of neurons stimulated and the stimulus strength . The location of an applied force determines which touch receptor neurons are activated and thus whether the animal accelerates or reverses , while the amplitude of the applied stimulus determines the probability that the animal responds at all ( Driscoll , 1997; Stirman et al . , 2011; Petzold et al . , 2013; McClanahan et al . , 2017; Mazzochette et al . , 2018 ) . Over longer timescales of minutes to hours , however , the picture has been shown to be more complicated . Habituation ( Rankin et al . , 1990 ) , quiescence ( Raizen et al . , 2008; Cho et al . , 2018 ) , and exposure to prolonged vibrations , salt or hypoxia , all modulate the animal’s sensitivity to mechanical stimuli ( Chen and Chalfie , 2014 , 2015 ) . More recently , evidence has also emerged that short timescale properties of the stimulus may also play a role in modulating the animal’s behavioral response . Porto et al . ( 2017 ) reported the use of reverse correlation and a binary optogenetic stimulus to present evidence that temporal processing is important for the animal’s behavioral response over a timescale of seconds . In our work here , we show that the nervous system does indeed process signals from the mechanosensory neurons over timeseries of many seconds . We find that the animal’s behavior response depends on higher-order temporal features such as the derivative of those mechanosensory signals , in addition to the stimulus amplitude and the animal’s own behavioral context . Here , we revisit the animal’s behavioral response to mechanosensory stimulation armed with high-throughput optogenetic methods for delivering time-varying stimuli and improved techniques for measuring animal posture ( Stephens et al . , 2008 ) and behavior ( Berman et al . , 2014 ) . Using reverse correlation ( Ringach and Shapley , 2004; Schwartz et al . , 2006; Gepner et al . , 2015 ) , we analyze over 8000 animal-hours of recordings and find new insights into the interplay between sensory processing and behavior . We first investigated the animal's response to plate tap , a spatially non-localized mechanosensory stimulus generated by tapping the dish containing the animals . Plate taps had previously been reported to evoke reverse locomotion ( Rankin et al . , 1990 ) and rarely forward accelerations ( Wicks and Rankin , 1995 ) in the young adult animals used here . A solenoid repeatedly delivered a tap stimulus every 60 s for 30 min to a plate of many young adult wild-type ( N2 ) worms , repeated across 22 plates , resulting in 40 , 409 total animal-tap presentations . The inter-stimulus interval was chosen to minimize the effects of habituation ( Rankin and Wicks , 2000 ) . The animal’s behavior was continuously measured and classified using a behavior-mapping technique similar to that described in Berman et al . ( 2014 ) . Briefly , statistical inference was performed on all of the animal’s posture dynamics to generate a single behavior map . Stereotyped posture dynamics that emerged from this map were defined as behaviors . Each individual animal’s posture dynamics were projected into this map at each point in time and automatically classified into one of nine behavior states , which were assigned labels such as 'Turn . ' See Figure 1 , Figure 1—figure supplements 1 and 2 and methods for a complete description of the behavior mapping . Also see example videos of behavioral mapping in Figure 1—video 2 and Figure 1—video 3 . Consistent with previous reports , we observed that taps most dramatically evoked the animal to transition to the 'Fast Reverse' state . Tap stimulus induced a 14-fold increase in the fraction of animals exhibiting 'Fast Reverse' immediately post stimuli , see Figure 2a and Figure 2—figure supplement 2 . In addition , animals that continued in forward locomotion exhibited an overall slowing down , transitioning from fast locomotion states to slower locomotion states , which to our knowledge had not been reported previously . We also observed a 4 . 5-fold increase in the fraction of animals exhibiting 'Turn' behavior approximately 5 s post stimulus . The fraction of animals exhibiting 'Slow Reverse' also increased slightly upon stimulation . These measurements suggest that plate tap evokes a wide-range of behavioral responses in the animal . We sought to activate the mechanosensory circuit optogenetically because optogenetic stimulation is more amenable to modulation and control . Optogenetic stimulation of the six mechanosensory neurons had previously been shown to evoke reversals and accelerations , similar to the response to tap ( Nagel et al . , 2005 ) . We wondered whether the details of the behavior response to tap that we observed are also present in response to optogenetic activation . Animals expressing the light-gated ion channel Chrimson in their soft touch mechanosensory neurons ( strain AML67 [Pmec-4::Chrimson::SL2::mCherry::unc-54] ) were illuminated with red light for 1 s with a 60 s inter-stimulus interval ( 2 , 444 stimulus-animal presentations , 20 μW mm–2 , selected to be in a region of high behavior sensitivity , see Figure 2c ) . Consistent with previous reports , light stimulation evoked a behavior response that was quantitatively similar to that of the plate tap ( see Figure 2b ) and required the cofactor all-trans retinal ( ATR ) , see Figure 2—figure supplement 2 . For both light and tap , the most salient response was a dramatic increase in animals in the 'Fast Reverse' state . Both light and tap also evoked an increase in 'Forward 3' behaviors and both evoked similar decreases in 'Forward 4' , '5' and '6' behaviors . Both light and tap also evoked an increase in 'Turn' behavior that peaked 5 s post-stimulus . Hence , optogenetic stimulation of mechanosensory neurons evoke detailed behavior responses similar to those resulting from a mechanical stimulus . This suggests that our optogenetic stimulation generates physiologically reasonable signals in the mechanosensory neurons and we therefore proceeded to explore the animal’s response to optogenetic stimulation . When the animal explores its natural environment , crawls through crevices , and interacts with other organisms , it probably experiences time-varying mechanical stimuli . Therefore , we sought to investigate the animal’s response to random temporally varying optogenetic stimulation . We find that the animal’s specific behavioral response correlates with higher-order temporal features of the stimulus , not merely the amplitude . To deliver rich temporally varying stimuli , we continuously presented a plate of transgenic animals with light modulated by broad frequency noise ( 7 Hz nyquist limit , 0 . 5 s correlation time , 25 μW mm–2 average intensity , min 0 , max 50 μW mm–2 , see power spectra in Figure 3—figure supplement 4 , and video in Figure 1—video 1 ) . Modulating light intensity has been shown to elicit graded potentials during optogenetic activation of other C . elegans neurons ( Liu et al . , 2009; Narayan et al . , 2011 ) , therefore we expect the time-varying light stimuli to result in a membrane potential that varies smoothly over time . Noise stimulation evoked a wide range of behavioral responses ( see Figure 3—figure supplement 1 ) . We used reverse correlation to identify the salient features of the stimulus that correlates with transitions into each behavior . Reverse correlation yields kernels that describe how a behavior is tuned to a stimulus . Kernels are particularly powerful in the context of the linear non-linear ( LN ) model , a simple and ubiquitous model in neuroscience that can be used to predict a neuron’s or animal’s response to stimulus ( Ringach and Shapley , 2004; Schwartz et al . , 2006; Coen et al . , 2014; Gepner et al . , 2015; Hernandez-Nunez et al . , 2015; Calhoun and Murthy , 2017; Clemens and Murthy , 2017 ) . See in particular ( Gepner et al . , 2015 ) . Briefly , the LN model treats the response to a stimulus as a stochastic process involving two steps: first the stimulus timeseries s ( t ) is convolved with a kernel A ( linear operation ) , and then it is transformed into a response probability P via a non-linear look-up function f ( non-linear operation ) , such that , ( 1 ) P[behavior] ( t ) =f[ ( A∗s ) ( t ) ]; ( A∗s ) =∫0∞ A ( τ ) s ( t−τ ) dτ . The shapes of the kernel and the non-linearity describe how a behavior response is tuned to the stimulus . Kernels can be estimated by finding the behavior-triggered average . Briefly , the stimulus in a time window centered on a behavior transition is averaged across all such behavior transitions . The mean subtracted and time-reversed behavior-triggered average is an estimate of the kernel , and so henceforth , we use the terms behavior-triggered average and kernel interchangeably . Once the kernels A are calculated , it is straightforward to estimate the non-linearities f from the observed behavior responses ( see 'Materials and methods' ) . Kernels and associated non-linearities were computed for transitions into each of the nine behavior states from over 50 , 000 behavior transition events per behavior ( see Figure 3 and Figure 3—figure supplement 2 ) . Kernels for six of the nine behaviors were found to be significant compared to a shuffled stimuli ( see 'Materials and methods' ) . By contrast , kernels computed from control animals grown without the necessary cofactor ATR all failed to pass our significance threshold ( see Figure 3—figure supplement 3 ) . Non-linearities calculated for the six behaviors were found to be mostly linear , suggesting that in our case the kernels themselves capture most of the information about how the nervous system responds to our stimulus . Our prior understanding of the mechanosensory circuit makes strong predictions about the shape of the kernels that we should expect . If the behavior depends only on which neurons are activated , then all kernels should have the same shape , scaled linearly , because we are always activating the same set of neurons . ( This assumes that all six neurons are activated in a linear regime , which seems reasonable given the approximately linear response observed in Figure 2c ) . Moreover , if the probability of response depends only on instantaneous stimulus amplitude , then we further expect all kernels to be narrow Gaussians . In contrast to these predictions , we see a wide diversity of kernels . Forward-locomotion kernels have biphasic waveforms , not at all like Gaussians . 'Forward 6' , for example , has the shape of a differentiator , suggesting that the transitions into 'Forward 6' correlate with decreasing stimuli on a 7 s timescale . Kernels for 'Slow Reverse' and 'Fast Reverse' , on the other hand , do look like Gaussians , consistent with prior reports that reversals do depend on the stimulus amplitude . Interestingly , the Gaussians are wide , which suggests that the animal may integrate the sensory signal over approximately 3 s in determining to reverse . Taken together , we conclude that the animal’s behavior response is not merely correlated with which neurons are stimulated and the stimulus amplitude . Instead different behaviors correlate with different temporal features of signals in the mechanosensory neurons , even though the same six neurons were always activated . The behavioral response correlates with properties of the stimulus such as the derivative or the integral , not just the amplitude . We wondered about the organization of the behavioral responses with respect to the stimuli to which they are tuned . One might expect animals to have evolved their behavioral response so that similar behaviors are tuned to similar stimuli . Indeed , we find that similar behaviors have quantitatively similar kernels . Hierarchical clustering was performed on the euclidian distance of the scaled kernels ( see Figure 3 ) . The two reverse locomotion states have similar kernels and were clustered together . Forward velocity states fell into two clusters that were based on speed: 'Forward 3' and 'Forward 4' are slower and clustered together , whereas 'Forward 5' and 'Forward 6' are faster and clustered together . That similarities in the kernels reflect the similarities in their associated behaviors , provides additional confidence in our reverse correlation analysis . To test causally whether specific signals in the mechanosensory neurons can bias the animal towards specific behaviors as predicted , we generated stimuli that were tailored to elicit specific behavioral responses . The kernels found in Figure 3 purport to describe how each behavioral response is tuned to the stimuli . Therefore , stimuli shaped like one of the kernels should drive an increase in transitions into its respective behavior . If , however , the behavioral response is tuned differently , then the kernel-shaped stimulus may evoke decreases in transitions to that behavior . ( We already know that the animal can respond to some stimuli by decreasing transitions to certain behaviors because we saw this with tap and 'Forward 6' , for example [see Figure 2—figure supplement 2] ) . We tested whether stimuli that are shaped like the kernels in Figure 3 increased transitions into their associated behaviors . Kernel waveforms were presented to a plate of animals in a randomized order ( six kernels , >13 , 500 animal-stimulus presentations per kernel; 40 s inter-stimulus interval ) . Five of six kernels elicited increased transitions to their respective behaviors as predicted , three of the six significantly so ( see Figure 4a ) . None significantly decreased transitions to their respective behaviors . We therefore conclude that the kernels correctly depict tuning of the behavioral responses . Consequently , we conclude that mechanosensory signals ( even in the same neurons ) can be tailored to evoke specific behaviors just by altering the stimulus waveform . The LN model provides an analytical framework to predict how an animal responds to a stimulus . The LN model correctly predicted that kernel-shaped waveforms should increase transitions into each kernel’s associated behavior state ( see Figure 4a ) . The kernel-shaped waveforms also evoked other behavioral responses . For example , stimuli that were shaped like the 'Forward 4' kernel increased transitions to both 'Forward 4' and 'Forward 3'; but decreased transitions to 'Forward 5' and 'Forward 6' ( see Figure 4b ) . How well , we wondered , does the LN model predict those responses ? We compared the observed behavioral responses ( colored lines ) to detailed time-dependent predictions made by the LN model ( black lines ) . To the resolution at which we could observe , we were reassured to find that the LN model correctly predicted the sign and temporal profile of changes in transition rates for all nine behavior states in response to each of the six kernel stimuli ( see Figure 4b and Figure 4—figure supplement 1 ) , suggesting that the LN model captures myriad details of the animal’s behavioral response . We further challenged our understanding of the animal’s behavioral response to stimulus by presenting an entirely novel stimulus , a triangle-wave ( 340 , 757 stimulus-animal presentations ) ( see Figure 5 and Figure 5—figure supplement 1 ) . How well does the LN model predict the animal’s behavior response to this novel stimulus ? The LN model captured the sign and general trend ( though not all features ) of the time-dependent change in the transition rate to all nine behaviors in response to the triangle wave . Moreover , the LN model provides a framework for understanding the animal’s response by inspecting features of the kernel waveform . For example , the 'Fast Reverse' kernel is symmetric in time and its mean-subtracted integral is positive . Therefore the shape of the 'Fast Reverse' kernel suggests that 'Fast Reverse' should be tuned to the overall stimulus intensity but not its derivative . Indeed we observe a very slight increase in the rate of transitions to 'Fast Reverse' during peak stimulus intensity . Conversely , the 'Forward 6' kernel is asymmetric in time and its biphasic waveform resembles that of the negative derivative of a Gaussian . Therefore , 'Forward 6' should be tuned to decreases in stimulus intensity , as we observe . Taken together , our experiments show that the animal can be driven to transition into different specific behavior states by modulating the temporal profile of signals in the same mechanosensory neurons , and that the LN model predicts the animal’s response . Caenorhabditis elegans are known to respond differently to the same stimuli when they are in different long-lived behavior states such as hunger ( Ghosh et al . , 2016 ) , quiescence ( Raizen et al . , 2008; Schwarz et al . , 2011; Nagy et al . , 2014b; Cho et al . , 2018 ) or arousal ( Cho and Sternberg , 2014 ) , or while undergoing Dauer formation ( Chen and Chalfie , 2014 , 2015 ) . We wondered whether mechanosensory processing might also be influenced by short-lived behavior states , like the 'Turn' , 'Reverse' or 'Forward' locomotory states measured here . To investigate tuning of the animal’s behavioral response conditional on its current behavior state , we calculated context-dependent kernels , one for each pairwise transition ( see Figure 6—figure supplement 1 ) . Of 72 possible pairwise transitions , 27 had kernels that passed our shuffled significance threshold ( compared to only four for our off-retinal control ( see Figure 6—figure supplement 2 ) . Transitions to some behavior states , such as 'Forward 4' , had kernels that changed dramatically depending on which behavior the animal originated from ( see columns in Figure 6—figure supplement 1 ) . The pairwise-specific kernels provided evidence of two types of context-dependent sensory processing in C . elegans that occur within short-time scales . In both cases , the animal appears to respond to the same stimuli differently depending on its current behavior . In the first , the animal responds to certain mechanosensory signals by speeding up or slowing down . In the second type , the animal suppresses its response to mechanosensory stimuli during turning behavior . These two types of context-dependency are described below . Behavior transitions that involve slowing down have similar tuning . For example , the 'Forward 5'→'Forward 4' kernel has a similar shape to the 'Forward 4'→'Forward 3' kernel ( see Figure 6 , left column ) . Likewise , transitions involving speeding up also have similar kernels . For example , 'Forward 3'→'Forward 4' and 'Forward 4'→'Forward 5' have similar kernels ( see Figure 6 , right column ) . Moreover , the two classes of kernels appear to be reflections of one another about the line of mean stimulus intensity . The stereotypy of the speed up and slow down kernels suggests that these nematodes have evolved to respond to certain stimuli by slowing down or speeding up in a relative way instead of transitioning to a stimulus-defined velocity . This is of interest because it implies a form of context dependency: it suggests that the same stimulus will drive the animal into forward locomotory states of different speeds depending on the animal’s current state . To determine whether the stereotyped speed-up or slow-down stimulus does indeed cause the animal to speed up or slow down , we again inspected the animal’s response to the kernel-shaped stimuli or the triangle-wave stimulus . Indeed , we found that the same stimulus drives the animal into a different forward locomotory state depending on the animal’s current state ( see Figure 7 ) . For example , animals in the slower 'Forward 4' state responded to a 'Forward 4' kernel-shaped stimulus by decreasing their transitions to 'Forward 5' . By contrast , animals in the faster 'Forward 6' state responded to the same stimulus by increasing their transitions into 'Forward 5' . This was one of multiple instances in which we observed the animal responding to the same stimuli with opposite responses depending on its current behavior . During triangle wave stimulation , for example , an increasing ramp causes slowing down , whereas a decreasing ramp causes speeding up ( see Figure 7 ) . We therefore conclude that stereotyped mechanosensory signals drive the animal to speed up or slow down . When the animal turns , it ignores all tested mechanosensory signals . This surprising observation is predicted by reverse-correlation analysis and confirmed by optogenetic and tap stimulation . Transitions out of 'Turn' are uncorrelated with stimulus , and kernels for those transitions all fail to pass our shuffled significance threshold ( see bottom row in Figure 6—figure supplement 1 ) . Consequently , the kernels predict that the animal should ignore mechanosensory stimuli during turns . By contrast , for every other behavior state , there is always at least one ( and often many ) transitions exiting out of the state whose kernels pass our significance threshold ( all rows other than 'Turn' have at least one significant kernel ) . To further test whether the animal does indeed ignore stimuli during turns , we investigated the animal’s context-dependent response to light pulses or tap . When the animal was in the 'Turn' state , neither a light pulse nor a tap evoked a significant change in the rate of transitions into any other behavior ( see bottom row Figure 8—figure supplements 1 and 2 ) ( multiple-hypothesis corrected E-test , see 'Materials and methods' ) . By contrast , when the animal was in other states , such as 'Forward 5' , both a tap and light pulses evoked significant changes in the transition rate into other behaviors . In fact , every other behavior state except for 'Forward 2' had at least one behavior transition exiting the state whose transition rate was significantly affected by either light or a tap . The 'Turn' behavior state was unique in that none of the kernels for transitions originating in 'Turn' passed the shuffled significance threshold , and no transition rates changed significantly in response to either light or a tap , ( see Figure 8—figure supplements 1 and 2 ) . We therefore conclude that in 'Turn , but not other states , the animal ignores mechanosensory stimuli . Transitions into 'Fast Reverse' provide an illustrative example ( see Figure 8 ) . When the animal is in the 'Turn' state , there is no significant difference in the rate of 'Turn'→'Fast Reverse' transition between shuffled and stimuli . But when the animal is in 'Forward 5' , light and taps caused a significant increase in 'Forward 5'→'Fast Reverse' . Taken together , we conclude that the animal attends to mechanosensory signals during most behavior states , such as 'Forward 5' , but ignores them during turns . This work provides new insights into C . elegans sensory processing . First , we show that the animal’s behavioral response is tuned to the temporal properties of mechanosensory signals , such as the derivative , that extend over many seconds in time . Moreover , by adjusting the waveform of a stimulus , mechanosensory signals in the same neurons can be tailored to elicit different behavioral responses . Second , mechanosensory signals influence a broad set of behaviors . Mechanosensation not only drives reversals and accelerations but can also evoke the animal to slow down . Third , even short timescale behavior states can influence the animal’s sensory processing . Earlier work has reported context-dependent sensory processing for behaviors with timescales of minutes to hours , such as hunger-satiety ( Ghosh et al . , 2016 ) , quiescence ( Raizen et al . , 2008; Schwarz et al . , 2011; Nagy et al . , 2014b; Cho et al . , 2018 ) , arousal ( Cho and Sternberg , 2014 ) or Dauer formation ( Chen and Chalfie , 2014 , 2015 ) . Here , we show that seconds-long timescale behaviors can also profoundly alter how the animal responds to a stimulus . Most dramatically , when the animal turns , it appears to ignore mechanosensory signals completely . A high throughput approach was crucial in revealing these new findings . Previously , we had probed the behavior response to mechanosensation using a targeted illumination system that allowed us to probe individual mechanosensory neuron pairs ( Leifer et al . , 2011 ) . That approach , however , is impractical for collecting the thousands of animal-hours of recordings needed here . In this work , we instead activate all mechanosensory neurons simultaneously , which allows us to study many animals in parallel . The set of neurons that is activated is determined by the opsin’s expression pattern . If opsins were expressed only in a single neuron , the current approach would also achieve single-neuron resolution . Although single-cell promoters are not known for any of the soft-touch mechanosensory neurons , intersectional approaches may allow the targeting of subsets ( Wei et al . , 2012; Schmitt et al . , 2012 ) . Future work is needed to explore the role of individual mechanosensory neurons in temporal processing . Automated behavior mapping was also critical for interpretation of the thousands of hours of animal behavior . We chose to classify behavior into discrete states , which are a natural description of discrete behaviors such as turns or reversal events . Alternatively , one could have chosen to use continuous description of behaviors , such as velocity , angular velocity or acceleration , which might be a more natural description of forward locomotion and speeding up or slowing down . The linear-nonlinear ( LN ) model was used to map out the relationship between sensory signals and behavior , and it predicts the animal’s response to stimuli satisfactorily . The LN model was chosen largely because of its ubiquity in neuroscience and simplicity of interpretation . We suspect that other models would yield similar findings . The LN model assumes a particular structure of linear and non-linear processing that is not inherently motivated by the biology , and it fails to take into account longer-timescale effects such as habituation . By contrast , the Gated Recurrent Unit ( GRU ) neural network model is one example of a model that is entirely non-linear and known to handle multi-timescale dependencies ( Cho et al . , 2014 ) . GRUs are just one of many alternative models with varying degrees of complexity and interpretability that could be used to probe temporal processing ( Glaser et al . , 2017 ) . In more complex sensory systems such as the retina , we have come to expect that the nervous system is carefully tuned to the temporal properties of sensory signals ( Meister and Berry , 1999 ) . Recently , it was shown that in drosophila , temporal processing is important for behavioral responses to odor , light and sound ( Behnia et al . , 2014; Coen et al . , 2014; Gepner et al . , 2015; Hernandez-Nunez et al . , 2015 ) . And in the much simpler C . elegans , temporal processing within timescales in the order of seconds has been observed in thermosensation ( Clark et al . , 2006 , 2007 ) , as well as in chemosensation ( Kato et al . , 2014 ) where it is known to be crucial for guiding thermotaxis or chemotaxis . In the C . elegans mechanosensory circuit , it had been shown previously that temporal processing occurs at the receptor level in order to convert applied forces into evoked currents , with a timescale of tens of milliseconds ( Eastwood et al . , 2015 ) , but it had remained unclear whether the nervous system used temporal information downstream to detemine the animal’s behavioral response . In this work , we now see evidence of temporal processing on seconds-long behavior-relevant timescales that guides the animal’s behavioral response . This temporal processing may arise from recurrent activity in the neural network downstream of the touch receptor neurons . The observation of such behavior-relevant temporal processing even in the simple mechanosensory circuit raises the possibility that temporal processing may be ubiquitous across sensory systems for driving behavior . Why might it be beneficial for the C . elegans nervous system to have evolved to tune its behavioral response to the temporal properties of mechanosensory signals , such as the derivative , over seconds ? The natural ecology of C . elegans is not well understood ( Félix and Braendle , 2010 ) and the statistics of the forces that it encounters in its natural environment are not known . We speculate that it could be useful for the worm to react differently if mechanosensory signals are increasing or decreasing , instead of making decisions solely on the overall stimulus strength . Note that we have characterized temporal processing to optogenetic signals , thus bypassing the animal’s mechanoelectro transduction machinery . Further work is needed to characterize the temporal processing of applied forces directly . It is is striking and surprising that the animal ignores mechanosensory inputs during turning . Why might the animal have evolved to ignore such signals during turns ? The turn is part of the C . elegans escape response , an avoidance behavior that shares some similarities with escape responses in other organisms , such as crayfish , mollusks or goldfish ( Pirri and Alkema , 2012 ) . Caenorhabditis elegans escape consists of reverse locomotion , followed by a turn and then forward locomotion in a new direction . The turn allows the animal to reorient and navigate away from a predator , and defects in this circuit have been shown to decrease survivability ( Maguire et al . , 2011 ) . Failing to complete the turn could inadvertently cause the animal to retrace its steps and return to danger . Ultimately , we see evidence of two kinds of internal processes that govern how the animal interprets sensory signals . First , the animal integrates mechanosensory information over a timescale of seconds . Second , the animal interprets these signals differently depending on the animal’s behavior state . An exciting future direction will be to identify the neural circuit mechanisms that allow the worm’s nervous system to integrate mechanosensory signals over time; and to alter its response rapidly depending on behavior state . This could shed insight into how internal brain states rapidly modulate sensory processing in a simple model system . The two strains used in this study were wild-type N2 Bristol animals ( RRID:WB-STRAIN:N2_ ( ancestral ) ) and AML67 ( RRID:WB-STRAIN:AML67 ) ( wtfIs46[pmec-4::Chrimson::SL2::mCherry::unc-54] ) , a transgenic strain that expresses the light-gated ion channel Chrimson and a fluorescent protein mCherry in mechanosensory neurons . To generate AML67 , 40 ng of plasmid ( pAL::pmec-4::Chrimson::SL2::mCherry::unc-54 ) were injected into N2 animals and integrated via UV irradiation ( Evans , 2006 ) . These animals were outcrossed with N2 six times . AML67 has been deposited in the public Caenorhabditis Gentics Center repository at the University of Minnesota . Plasmid pAL::pmec-4::Chrimson::SL2::mCherry::unc-54 ( https://www . addgene . org/107745/ ) was engineered using a HiFi Cloning Kit ( NEB ) . Chrimson was a kind gift from Ed Boyden of MIT . mCherry and backbone was amplified from pJIM20 , a gift from John Murray of the University of Pennsylvania . The promoter sequence ( mec-4 ) , splicing sequence ( SL2 ) and 3′-utr sequence ( unc-54 ) were amplified using primers as listed in Table 1 . The construct was sequenced confirmed before injection . Transgenic animals exhibited reduced sensitivity to a tap or touch compared to wild-type animals , presumably because Chrimson competes with endogenous MEC-4 protein for transcription ( see Figure 2—figure supplement 4 ) . From the alleles we had generated , we selected AML67 for use in this study because it was the most sensitive to tap and touch , despite being reduced compared to wild-type . Strains were maintained on 9 cm NGM agar plates seeded with OP50 Escherichia coli food at 20° C . Worms were bleached 3 days prior to the experiment to provide 1-day-old adults . For optogenetic experiments , bleached worms were placed on plates seeded with 1 ml of 0 . 5 mM all-trans-retinal ( ATR ) mixed with OP50 . Control plate lacked ATR . To avoid inadvertent optogenetic activation , plates were wrapped in aluminum foil , handled in the dark , and viewed under dissection microscopes using dim blue light . To harvest worms for high-throughput experiments , roughly 100 to 200 worms were cut from agar , washed and then spun-down in a 1 . 5 ml micro centrifuge tube . Worms at the bottom of the tube were placed on an unseeded 9 cm NGM agar plate via a micropipette . Excess liquid on the plate was carefully wicked away using tissue paper . Worms were allowed to adapt to their new environment for 25 min before recordings or stimulation were carried out . Experiments were conducted in a custom-built high-throughput imaging rig ( Figure 2—figure supplement 1 ) . Plates of animals were recorded while undergoing 30 min of optogenetic or tap stimulation . Imaging was performed as follows: the agar plate was illuminated by a ring of 850 nm infrared LEDs ( irrf850-5050-60-reel , environmentallights . com ) . A 2592 × 1944 pixel CMOS camera ( ACA2500-14um , Basler ) recorded worm movements at 14 frames per second and a magnification of 20 μm per pixel , so as to provide sufficient spatiotemporal resolution to capture posture dynamics . The field of view of the camera was centered on the plate and included approximately 50% of the plate surface . Custom LabVIEW software acquired images from the camera and controlled stimulus delivery as described below . Taps were delivered to the side of 9 cm plates containing the animals by means of a solenoid , following a method similar to that described by Swierczek et al . ( 2011 ) . An electric solenoid tapper ( Small Push-Pull Solenoid , Adafruit ) was driven with a 70 ms , 24 V , DC pulse under Labview control via a LabJack DAQ and a solid-state relay . During tap experiments , taps were delivered to the plate once per minute for 30 min ( see Table 2 ) . The 1 min inter-stimulus interval was chosen to minimize habituation ( Timbers et al . , 2013 ) . Experiments involving optogenetic stimulation are summarized in Table 2 . Optogenetic stimulation was delivered by three 625 nm LEDs ( M625L3 , Thorlabs ) positioned such that their light approximately tiles the agar plate visible in the camera’s field of view . LED’s were driven by a diode driver ( L2C210C , Thorlabs ) under the control of LabVIEW via an analog signal from a LabJack DAQ ( Model U3-HV with LJTick-DAC ) . The range of the light intensity for optogenetic stimulation averaged at the plate spanned from 0 to 80 μW mm–2 . Small spatial inhomogeneities in light intensity were characterized and accounted for in software so as to calculate the precise light intensity delivered to each animal . An infared long pass filter ( FEL0800 , Thorlabs ) in front of the camera blocked light from the stimulus LEDs and only permitted light from the infrared behavior LEDs . The unsupervised behavior mapping approach used in this work is adapted from work in drosophila ( Berman et al . , 2014 ) and is similar in spirit to work in rodents ( Wiltschko et al . , 2015 ) . It also builds upon decades of methodological advances quantifying C . elegans behavior ( Croll , 1975; Stephens et al . , 2008; Ramot et al . , 2008; Brown et al . , 2013; Yemini et al . , 2013; Gyenes and Brown , 2016; Gomez-Marin et al . , 2016 ) . Animal behavior was measured and classified using an analysis pipeline , summarized in Figure 1—figure supplement 1 . First , the worms were located and tracked , then their posture was extracted , and finally their posture dynamics were clustered and classified . Details of the pipeline are described below . The pipeline was written in MATLAB and run on the Princeton University’s high-performance parallel computing cluster . Source code is available at ( https://github . com/leiferlab/liu-temporal-processing ) ( Liu and Leifer , 2018; copy archived at https://github . com/elifesciences-publications/liu-temporal-processing ) . Reverse correlation was used to find a linear kernel and non-linearity that describe the relationship between the animal’s behavior transitions and an applied stimulus . When presented as a timeseries of rates , as in Figure 2—figure supplement 2 , transition rates were calculated according to the following: behavior timeseries from all recordings were cropped in a time window around each stimulus , commingled , and then time aligned to the stimulus . The fraction of all animals undergoing a transition was calculated at each time step . The fractions of animal were directly converted into a rate of transitions per animal per minute , yielding the timeseries of rates . Behavioral analysis and stimulation data for all tracked animals in all experiments in Table 2 are available at https://doi . org/10 . 6084/m9 . figshare . 5956348 . See dataset README for details . All recorded data , including raw images ( 2 TB ) , will be available at http://dx . doi . org/10 . 21227/H27944 .	A worm called Caenorhabditis elegans has a nervous system made up of only 302 neurons , far fewer than the billions of cells that comprise our own brains . And yet these few hundred neurons are enough for these worms to detect and respond to their surroundings . C . elegans is thus a popular choice for studying how nervous systems process sensory information and use it to control behavior . Yet , most experiments to date have used only simple stimuli , such as taps or pokes , and studied a handful of behaviors , such as whether or not a worm stops moving or backs up . This limits the conclusions it has been possible to draw . Liu et al . therefore set out to determine how the worm’s nervous system responds to more complex stimuli . These included physical stimuli , such as taps on the side of the dish containing the worms , as well as simulated stimuli . To generate the latter , Liu et al . used a technique called optogenetics to directly activate the neurons in the worm’s body that would normally detect information from the senses , by simply shining a light on the worms . Doing so gives the worm the sensation of a physical stimulus , even though none was present . Liu et al . then used mathematics to examine the relationships between the stimuli and the worms’ responses . The results confirmed that worms usually respond to simple stimuli , such as taps on the side of their dish , by backing up . But they also revealed more advanced forms of stimulus processing . The worms responded differently to stimuli that increased over time versus decreased , for example . A worm's response to a stimulus also varied depending on what the worm was doing at the time . Worms that were in the middle of turns , for instance , ignored stimuli to which they would normally respond . This suggests that an animal’s current behavior influences how its nervous system interprets sensory information . The discovery of relatively sophisticated responses to sensory stimuli in C . elegans indicates that even simple nervous systems are capable of flexible sensory processing . This lays a foundation for understanding how neural circuits interpret sensory signals . Building on this work will ultimately help us understand how more complicated nervous systems interpret and respond to the world .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "physics", "of", "living", "systems", "neuroscience" ]	2018	Temporal processing and context dependency in Caenorhabditis elegans response to mechanosensation
Septin proteins evolved from ancestral GTPases and co-assemble into hetero-oligomers and cytoskeletal filaments . In Saccharomyces cerevisiae , five septins comprise two species of hetero-octamers , Cdc11/Shs1–Cdc12–Cdc3–Cdc10–Cdc10–Cdc3–Cdc12–Cdc11/Shs1 . Slow GTPase activity by Cdc12 directs the choice of incorporation of Cdc11 vs Shs1 , but many septins , including Cdc3 , lack GTPase activity . We serendipitously discovered that guanidine hydrochloride rescues septin function in cdc10 mutants by promoting assembly of non-native Cdc11/Shs1–Cdc12–Cdc3–Cdc3–Cdc12–Cdc11/Shs1 hexamers . We provide evidence that in S . cerevisiae Cdc3 guanidinium occupies the site of a ‘missing’ Arg side chain found in other fungal species where ( i ) the Cdc3 subunit is an active GTPase and ( ii ) Cdc10-less hexamers natively co-exist with octamers . We propose that guanidinium reactivates a latent septin assembly pathway that was suppressed during fungal evolution in order to restrict assembly to octamers . Since homodimerization by a GTPase-active human septin also creates hexamers that exclude Cdc10-like central subunits , our new mechanistic insights likely apply throughout phylogeny . Septin proteins are found in nearly every eukaryotic lineage , with the exception of land plants ( Auxier et al . , 2019; Nishihama et al . , 2011; Onishi and Pringle , 2016; Pan et al . , 2007 ) . In most extant organisms studied to date , multiple septin proteins co-assemble into linear , rod-shaped hetero-oligomers with at least some potential to polymerize into filaments ( Fung et al . , 2014; Mostowy and Cossart , 2012; Oh and Bi , 2011 ) . Septin hetero-oligomers are now known to contribute functionally to a wide variety of cellular processes ( Dolat et al . , 2014; Saarikangas and Barral , 2011 ) . A number of human diseases and disorders , ranging from cancer to male infertility , have been linked to septin dysregulation or mutation ( Dolat et al . , 2014; Saarikangas and Barral , 2011 ) , but within our understanding of septin roles in cellular function one common theme emerges: the ability to assemble into quaternary complexes is key . All distinct septins share clear homology , but phylogenetic analysis reveals septin groups and subgroups ( Pan et al . , 2007 ) that can accurately predict which septin occupies which subunit position within hetero-oligomers ( Nakahira et al . , 2010 ) . However , the subtleties of the distinctions between septins that drive assembly of hetero-oligomers with precise subunit organization remain incompletely understood . Moreover , the oligomerization interfaces have not diverged drastically from the ancestral form , as nearly all septins retain the ability to homo-oligomerize in vitro ( Farkasovsky et al . , 2005; Versele et al . , 2004 ) whether or not they do so in vivo . How cells assemble specific septin hetero-oligomers to meet functional demands represents a major focus of septin research . More generally , septins provide a powerful model to study the mechanisms by which multisubunit complexes are assembled with high efficiency and fidelity in living cells . Septins clearly evolved from an ancestral GTPase ( Leipe et al . , 2002 ) but multiple septins subsequently lost the ability to hydrolyze GTP . Unlike the building blocks of polymers of the other cytoskeletal NTPases actin and tubulin ( and their prokaryotic counterparts ) , in native septin hetero-oligomers most , if not all , of the septin nucleotide-binding pockets are buried within an oligomerization interface , called the G interface ( Sirajuddin et al . , 2007 ) . Particularly for septin complexes in budding yeast , where hetero-octamers polymerize into filaments via the other , so-called ‘NC’ septin-septin interface ( Bertin et al . , 2008; Sirajuddin et al . , 2007 ) , burial of GTP in G interfaces within highly stable complexes severely limits the potential for cycles of GTP binding , hydrolysis , and exchange to modulate the assembly and disassembly of septin filaments . Nonetheless , most septins retain sequences clearly corresponding to regions/motifs in other small GTPases that contact bound nucleotide and/or change conformation upon GTP hydrolysis . These include the G1/Walker/P-loop , residues of which contact each of the three phosphates; the G2/Switch I and G3/Switch II loops , which are in proximity to the γ phosphate when it is present but are in distinct locations when it is absent; and the G4 motif , which contacts the base and dictates specificity for guanosine nucleotides . The question of what roles GTP binding and hydrolysis play in septin biogenesis and function has been a major focus in septin research for over two decades ( Mitchison and Field , 2002 ) . The baker’s yeast , Saccharomyces cerevisiae , has been at the forefront of septin research , for a number of reasons . The repertoire of septin proteins expressed in yeast ( seven ) is much simpler than that of humans ( 13 ) . Two yeast septins are expressed and function specifically during the process of sporulation ( De Virgilio et al . , 1996; Fares et al . , 1996; Garcia et al . , 2016; Heasley and McMurray , 2016; Pablo-Hernando et al . , 2008 ) , a version of gametogenesis , leaving five septins expressed in nearly stoichiometric amounts ( Kulak et al . , 2014 ) in mitotically dividing cells: Cdc3 , Cdc10 , Cdc11 , Cdc12 , and Shs1 . Yeast septin localization during the mitotic cell division cycle is primarily restricted to the cortex , where septins take the form of filamentous rings at the mother-bud neck that undergo a series of discrete structural transitions as the bud emerges and grows and cytokinesis and cell separation take place ( Oh and Bi , 2011 ) . Septin filament assembly is essential for proliferation ( McMurray et al . , 2011 ) , hence defects in septin folding and oligomerization translate into clear defects in yeast colony/culture growth . A combination of molecular genetic ( Garcia et al . , 2011; Iwase et al . , 2007; McMurray et al . , 2011; Nagaraj et al . , 2008; Versele et al . , 2004 ) , biochemical ( Bertin et al . , 2008; Farkasovsky et al . , 2005; Garcia et al . , 2011; Versele et al . , 2004 ) , and structural ( Bertin et al . , 2008; Farkasovsky et al . , 2005; Garcia et al . , 2011; Sirajuddin et al . , 2007 ) approaches have demonstrated that two kinds of hetero-octamers co-exist , with a linear hexameric core of the order Cdc12–Cdc3–Cdc10–Cdc10–Cdc3–Cdc12 ‘capped’ at each end with either Cdc11 or Shs1 . While the functional basis for yeast octamer diversity remains largely unknown , at the level of quaternary structure septins are best understood in S . cerevisiae . We are most interested in the mechanisms of assembly of septin hetero-oligomers , and previously used mutants isolated in unbiased genetic screens to elucidate roles in this process for nucleotide binding ( Weems et al . , 2014 ) and interactions with cytosolic chaperones ( Johnson et al . , 2015 ) , including the fungal disaggregase Hsp104 , a hexameric AAA+ ATPase ( Sweeny and Shorter , 2016 ) . Here we report that the function of specific septin mutants is affected in unexpected ways by guanidine hydrochloride ( GdnHCl ) , a small molecule used for decades to inhibit Hsp104 ATPase activity in vivo ( Derkatch et al . , 1997; Jung and Masison , 2001 ) . Our findings reveal important roles for specific residues and structural motifs in septin hetero-oligomer assembly , suggest a series of key events during fungal septin evolution that enforced incorporation of Cdc10 subunits in hetero-oligomers , and provide what is , to our knowledge , the first evidence that guanidinium ( Gdm ) has the potential to functionally replace Arg residues in vivo . We previously reported interactions between an ATPase-dead mutant of Hsp104 and a temperature-sensitive ( TS ) mutant of Cdc10 , in which overexpression of the mutant Hsp104 inhibited the proliferation of cdc10 ( D182N ) cells at an otherwise permissive low temperature ( 27°C ) ( Johnson et al . , 2015 ) . We expected that the addition of 3 mM GdnHCl to the culture medium would mimic these inhibitory effects in cells with wild-type ( WT ) Hsp104 . Instead , we found complete rescue by GdnHCl of the cdc10 ( D182N ) TS proliferation defect at all temperatures tested ( up to 37°C ) ( Figure 1A ) . GdnHCl did not have the same effect on cells carrying TS alleles of other septins , although in its presence a cdc3 ( G365R ) strain grew noticeably worse at 27°C , 30°C and 34°C , and a cdc12 ( G247E ) strain grew slightly worse at 30°C and slightly better at 37°C ( Figure 1A ) . To our surprise , GdnHCl rescue of cdc10 ( D182N ) did not require Hsp104 ( Figure 1B ) . Others have suggested that a mitochondrial Hsp104 homolog , Hsp78 , might share functions with Hsp104 ( Erives and Fassler , 2015 ) , and Hsp78 and Hsp104 share the same residues within the ATP binding pocket that contact Gdm in the bacterial Hsp104 homolog ClpB ( Figure 1C ) , but Hsp78 was also dispensable for GdnHCl rescue of cdc10 ( D182N ) high-temperature growth ( Figure 1B ) . Inhibition of Hsp104 ( i . e . , curing of prions ) or Hsp78 ( i . e . , induction of cytoplasmic petites ) by GdnHCl in vivo requires concentrations in the medium of ≥1 mM ( Tuite et al . , 1981 ) , but GdnHCl was sufficient to provide partial rescue of 37°C growth by cdc10 ( D182N ) cells at concentrations < 0 . 1 mM ( Figure 1D , E ) . We conclude that GdnHCl suppresses septin defects in cdc10 ( D182N ) mutants via a mechanism that does not involve Hsp104 or its homolog Hsp78 . We previously noted that the cdc12 ( G247E ) mutation appears to decrease the levels of WT Cdc10 ( Johnson et al . , 2015 ) . Cells tolerate loss of Cdc10 by assembling filaments via hexameric building blocks in which Cdc3 forms a non-native G homodimer , but cdc10∆ cells are TS , presumably because at high temperature Cdc3 fails to adopt the homodimerization-competent conformation and/or the non-native Cdc3 homodimer is unstable ( McMurray et al . , 2011 ) . Rescue of the TS defects at 37°C of the cdc10 ( D182N ) and cdc12 ( G247E ) mutants by GdnHCl could reflect improved assembly of Cdc10-less hexamers , rather than any specific effect on the cdc10 ( D182N ) allele . Consistent with this idea , GdnHCl also fully suppressed the TS phenotype resulting from a G100E substitution in Cdc10 ( Figure 1—figure supplement 1A ) , as well as that of cdc10∆ ( Figure 2A ) . Clear but partial GdnHCl rescue was also observed for cdc10 ( G44D ) cells ( Figure 1—figure supplement 1A ) . We confirmed via immunofluorescence with anti-Cdc11 antibodies ( Figure 1—figure supplement 1B ) and electron microscopy ( Figure 1—figure supplement 1C ) that at 37°C GdnHCl-treated cdc10 cells assembled filamentous rings composed of other septins . Thus GdnHCl does not simply bypass the need for septin filaments . Given these results , we wondered if in cells carrying point-mutant cdc10 alleles GdnHCl bypasses altogether the incorporation of the mutant Cdc10 subunits during septin hetero-oligomer assembly . Indeed , even at temperatures permissive in the absence of GdnHCl for septin ring incorporation of Cdc10 ( D182N ) -GFP , in the presence of 3 mM GdnHCl the fluorescence of Cdc10 ( D182N ) -GFP was restricted to the cytoplasm ( Figure 2B ) . WT Cdc10-GFP continued to localize to the bud neck despite the presence of GdnHCl ( Figure 2B ) . Similarly , in diploid cells co-expressing Cdc10-mCherry and Cdc10 ( D182N ) -GFP , the addition of GdnHCl specifically eliminated Cdc10 ( D182N ) incorporation in septin rings , with no obvious effect on WT Cdc10 ( Figure 2C ) . Our results suggest that when mutations within the Cdc10 G interface perturb the ability of Cdc10 to acquire a conformation that binds tightly to Cdc3 , GdnHCl allows other Cdc3 molecules to outcompete the mutant Cdc10 proteins for occupancy of the Cdc3 G interface , ultimately co-assembling with Cdc11 , Cdc12 , and Shs1 into Cdc10-less hexamers capable of robust septin functions . As predicted from this model , GdnHCl was unable to rescue cdc10∆ growth when Cdc3 carried a mutation ( W364A ) previously shown ( McMurray et al . , 2011 ) to block Cdc3 homodimerization ( Figure 2D ) . Yeast septin function is regulated ( in poorly defined ways ) by a large number of pathways ( Longtine and Bi , 2003 ) , which might be altered by GdnHCl to indirectly promote Cdc3 homodimerization . GdnHCl might also alter other properties of Cdc10-less yeast cells to promote septin function despite inefficient Cdc3 homodimerization . For example , deletion of DPL1 , which encodes a dihydrosphingosine phosphate lyase , improves the proliferation of cdc10∆ cells at the moderate temperature of 30°C ( Michel et al . , 2017 ) . The authors of that study speculated that the absence of Dpl1 alters the properties of the plasma membrane in ways that mimic the effects of low temperature and allow Cdc10-less septin complexes to better function in cytokinesis ( Michel et al . , 2017 ) . Accordingly , Cdc10-less septin function at high-temperature in the presence of GdnHCl might reflect GdnHCl inhibition of Dpl1 . However , cdc10∆ dpl1∆ cells were only able to proliferate at 37°C when GdnHCl was present in the growth medium ( Figure 1—figure supplement 1D ) . While these data do not exclude the possibility that GdnHCl alters membrane properties in ways independent of Dpl1 , they demonstrate that the GdnHCl effect on septin function in cdc10∆ cells is Dpl1-independent . To ask if GdnHCl is able to promote Cdc10-less septin filament assembly outside of yeast cells and away from membranes , we co-expressed Cdc3 , Cdc11 , and hexahistidine-tagged Cdc12 ( 6xHis-Cdc12 ) in E . coli , which lacks native septins , and purified septin complexes as described previously ( McMurray et al . , 2011 ) . We then used negative staining and electron microscopy ( EM ) to ask if these purified complexes appeared as hexamers in high-salt solution and polymerized into filaments upon salt dilution . We previously used the same approach to demonstrate that a mutation in the Cdc3 G interface , G261V , promotes septin hexamer and filament assembly in the absence of Cdc10 ( McMurray et al . , 2011 ) . To test effects of GdnHCl , we either grew the bacteria in the presence of 50 mM GdnHCl and kept GdnHCl in all buffers thereafter , or we only introduced the GdnHCl during purification . In the absence of GdnHCl , Cdc10-less complexes were detected during purification via high-salt size-exclusion chromatography ( SEC ) as a single peak ( in addition to a peak of aggregates eluting in the void volume ) which , when analyzed by EM , was composed of particles containing two or three septins ( Figure 3A–C ) . These complexes were unable to polymerize into filaments in low salt ( Figure 3D ) , consistent with our earlier work ( McMurray et al . , 2011 ) . Two-septin particles likely represent Cdc3–Cdc12 hetero-dimers , since in these conditions Cdc11 occasionally dissociates from the ends of hetero-octamers containing Cdc10 ( Bertin et al . , 2008 ) . By contrast , in SEC purification of complexes synthesized in the presence of GdnHCl a new , higher-molecular weight peak appeared ( Figure 3A ) , which EM analysis revealed to contain trimers , tetramers , pentamers and hexamers ( Figure 3B , C ) . The larger complexes were frequently ‘kinked’ at a location three subunits from one end ( Figure 3B ) , consistent with weak septin-septin contacts at this point , which we also previously observed with the Cdc3 ( G261V ) mutation ( McMurray et al . , 2011 ) . Upon salt dilution , the Cdc10-less complexes synthesized in the presence of GdnHCl robustly polymerized into bundles of laterally-associated filaments ( Figure 3D ) , very similar to the behavior of WT hetero-octamers that contain Cdc10 ( Bertin et al . , 2008 ) . Fourier transform analysis of the striations in these filament bundles revealed a repeating unit of ~27 nm , the length of a septin hexamer ( Figure 3D ) . Adding GdnHCl to complexes synthesized in its absence did not allow filament formation ( Figure 3D ) . We conclude from these data that GdnHCl is able to promote septin filament assembly by recombinant Cdc10-less septin complexes produced in a heterologous system that lacks septin-regulatory pathways . Thus GdnHCl likely acts directly on septin proteins , and only when it is present during septin folding/assembly . We identified the G261V mutation in Cdc3 using an unbiased genetic selection for suppressors of the TS phenotype of cdc10∆ cells , and concluded that the introduction of Val in this position stabilizes a conformation of the G interface that self-associates better than the major conformation populated by WT Cdc3 at high temperatures ( McMurray et al . , 2011 ) . We reasoned that Gdm might bind in the Cdc3 G interface and similarly promote the homodimerization-competent conformation . For clues as to where Gdm might bind , we first used the structure of human SEPT2 bound to the non-hydrolyzable nucleotide analog 5'-guanylyl imidodiphosphate ( GppNHp ) ( Sirajuddin et al . , 2009 ) to generate a homology model of the globular domain of monomeric Cdc3 , and then performed unbiased in silico docking simulations to ask for the lowest-energy predicted sites of Gdm binding . Four sites had binding free energies of ≤−3 kcal/mol . One site , between the side chains of Thr302 and His262 , lies in the G interface , and had the second-lowest predicted free energy of binding ( Figure 4A ) . As a parallel approach , we considered that ( i ) Gdm mimics the distal end of an Arg side chain , and ( ii ) a number of previous in vitro studies of diverse non-septin proteins – including T4 lysozyme ( Baldwin et al . , 1998 ) , β-galactosidase ( Dugdale et al . , 2010 ) , and carboxypeptidase A ( Phillips et al . , 1992 ) – demonstrated that Gdm is able to functionally occupy molecular vacancies that are created by the replacement of an Arg residue with a residues having a smaller side chain ( e . g . Ala ) . We thus hypothesized that one or more Arg residues might contribute to G interface contacts for other septins that are , unlike S . cerevisiae Cdc3 , capable of robust G homodimerization . To identify candidate Arg residues , we performed a phylogenetic analysis of Cdc3 homologs from different species , which we separated into two categories based on the severity of phenotypes accompanying deletion of the CDC10 homolog . If non-Saccharomyces species possess an Arg that favors G homodimerization by Cdc3 , then we predicted this group of species should easily form Cdc10-less septin hexamers and be better able to tolerate deletion of the Cdc10 homolog . Such functional information was available for non-Saccharomyces species from eleven distinct fungal genera . In Aspergillus ( fumigatis ( Vargas-Muñiz et al . , 2015 ) or nidulans Hernández-Rodríguez et al . , 2012 ) , Candida albicans ( Warenda and Konopka , 2002 ) , Cryptococcus neoformans ( Kozubowski and Heitman , 2010 ) , Fusarium graminearum ( Chen et al . , 2016 ) , Neurospora crassa ( Berepiki and Read , 2013 ) , Schizosaccharomyces pombe ( An et al . , 2004 ) , and Ustilago maydis ( Alvarez-Tabarés and Pérez-Martín , 2010 ) , deletion of the Cdc10 homolog has distinctly milder phenotypic consequences compared to deletion of the Cdc3 homolog . In Magnaporthe oryzae , septin ring assembly is perturbed by deletion of any septin , but higher-order , filamentous structures persist in cells lacking the Cdc10 homolog Sep4 , whereas in the other mutants septin localization is almost exclusively diffuse ( Dagdas et al . , 2012 ) . Finally , in Coprinopsis cinerea , a UV-induced mutant defective in fruiting body development was rescued by a gene encoding CcCdc3 , but the expression of CcCdc10 remained about 100-fold decreased in the mutant following rescue , suggesting that the mutant was defective in both CcCdc3 and CcCdc10 expression , and that rescue of CcCdc3 expression was sufficient to restore function ( Shioya et al . , 2013 ) . We interpret these results as evidence that in C . cinerea , as well , loss of Cdc10 is better tolerated than loss of Cdc3 . Non-Saccharomyces yeasts in the family Saccharomycetaceae behave differently ( see Figure 4—figure supplement 1A ) . Only the Shs1 homolog is non-essential for proliferation in Kluyveromyces lactis ( Rippert and Heinisch , 2016 ) . In Ashbya gossypii , deletion of any mitotic septin prevents septin ring assembly; loss of the Cdc10 homolog is as severe as loss of the Cdc3 homolog ( Helfer and Gladfelter , 2006 ) . According to our hypothesis , if there is a key Arg residue that promotes Cdc3 homodimerization , and S . cerevisiae lacks it , then K . lactis and A . gossypii should also lack it . Figure 4—figure supplement 1B shows an alignment of protein sequences for Cdc3 homologs from the fungal species listed above . As ScCdc3 is predicted to be unable to hydrolyze GTP ( Sirajuddin et al . , 2009 ) , we also included for comparison human SEPT2 and SEPT9 , because for these septins crystal structures are available of non-native G homodimers in the ‘GTP state’ ( bound to the non-hydrolyzable GTP analog GppNHp or GTPγS , respectively ) . Only 16 residues were distinctly different between a group including S . cerevisiae and the two species that cannot tolerate Cdc10 loss , and the group of species that tolerate CDC10 deletion . For six of these residues , the changes reflect substitutions with strong predicted effects on amino acid properties ( polarity , charge , size , etc . ) . One of these non-conservative variants lies in the C-terminal extension , which is disordered in septin crystal structures . Based on the crystal structures of SEPT2•GppNHp and SEPT9•GTPγS , all of the other five non-conservative variants cluster near the G interface ( Figure 4B ) , and one of these residues – corresponding to Thr302 in ScCdc3 – is an Arg in eight of the nine species that tolerate Cdc10 loss ( Figure 4—figure supplement 1B ) . Moreover , in the SEPT2•GppNHp structure the Gdm group of the corresponding Arg in SEPT2 is located in the same place where our in silico docking results predicted that Gdm binds to Cdc3 , between Thr302 and His262 ( Figure 4 ) . Together , these findings strongly support the idea that Gdm promotes Cdc3 homodimerization by occupying the same site in the Cdc3 G interface that in other septins is occupied by the Gdm moiety of an Arg side chain . This Arg residue was likely substituted to Thr during the evolution of the fungal lineage that includes S . cerevisiae , A . gossypii , and K . lactis , concomitant with loss of Cdc10-less hexamers in these yeasts . Our phylogenetic comparisons predicted that Gdm should act similarly on the A . gossypii Cdc3 homolog to promote AgCdc3 homodimerization in the absence of Cdc10 . To test this prediction , we exposed A . gossypii cells lacking CDC10 to GdnHCl and assessed higher-order septin assembly by monitoring the localization of Shs1-GFP expressed in the same cells . As reported previously ( Helfer and Gladfelter , 2006 ) , in cdc10∆ cells in the absence of GdnHCl Shs1-GFP fluorescence was cytoplasmic and diffuse ( Figure 5 ) . Addition of GdnHCl restored the localization of Shs1-GFP to branch points and hyphal tips ( Figure 5 ) , the same pattern observed in WT cells ( Helfer and Gladfelter , 2006 ) . Thus the ability of GdnHCl to promote septin function in the absence of Cdc10 , and the loss of Arg at the position corresponding to ScCdc3 Thr302 , are both conserved over >100 m years of evolution . To better understand the molecular details of Gdm ‘rescue’ of Arg-substituted mutants , earlier in vitro Gdm studies also tested other small molecules with similar chemical properties , including urea and the GdnHCl derivatives aminoguanidine ( Pimagedine , here ‘aGdnHCl’ ) and N-ethylguanidine ( ‘eGdnHCl’ ) ( Rynkiewicz and Seaton , 1996 ) . We undertook an analogous approach to study GdnHCl rescue of the cdc10 ( D182N ) TS phenotype . First , we tested urea , which differs from Gdm by only a few atoms ( Figure 6A ) . Both GdnHCl and urea denature proteins at high concentrations ( >5 M ) and , at subdenaturing concentrations ( 0 . 05–4 M ) , both can non-specifically stabilize proteins in vitro , likely by lowering conformational entropy ( Bhuyan , 2002 ) . If GdnHCl rescues cdc10 mutants via non-specific conformational stabilization , urea should do the same . We grew WT or cdc10 ( D182N ) cells on solid rich medium containing 0 . 64 M urea , a concentration we found to slow , but not prevent , growth of WT cells , and saw no cdc10 ( D182N ) rescue ( Figure 6A–B ) . To test a range of concentrations , we also placed filter disks soaked with GdnHCl or urea on lawns of WT or cdc10 ( D182N ) cells . We saw a clear zone of rescue around the GdnHCl disk , but not with urea ( Figure 6C ) . These results point to a mechanism of cdc10 rescue by GdnHCl that is mechanistically distinct from non-specific protein stabilization at subdenaturing conditions . We next exposed WT or cdc10 ( D182N ) cells to 0 . 375 mM GdnHCl , aGdnHCl , eGdnHCl , or combinations of GdnHCl plus eGdnHCl or GdnHCl plus aGdnHCl ( 0 . 375 mM each ) . At 0 . 375 mM GdnHCl provided only a partial rescue ( Figure 6D ) , allowing us to detect subtle effects of the GdnHCl derivatives when combined with GdnHCl . At 22°C , 0 . 375 mM of any single drug had no noticeable effect on growth; slight growth inhibition of both WT and mutant cells was observed when the total concentration of GdnHCl plus derivative was 0 . 75 mM ( Figure 6D ) . At 37°C , in addition to the expected partial cdc10 ( D182N ) rescue by GdnHCl , there was a very slight rescue by eGdnHCl , and an even less pronounced rescue by aGdnHCl ( Figure 6D ) . By contrast , mixing GdnHCl with aGdnHCl or eGdnHCl to a total concentration of 0 . 75 mM resulted in full or nearly full rescue of cdc10 ( D182N ) 37°C growth , respectively ( Figure 6D ) . We interpret these findings to mean that aGdm and eGdm ( the guanidinium ion derivatives ) occupy the position between Cdc3 Thr302 and His262 differently than does Gdm . They may not provide the appropriate molecular contacts across the G dimer interface , or they may bind less well , or even too well ( if Gdm acts only transiently during Cdc3 folding ) . Indeed , aGdm and eGdm were predicted by in silico modeling to bind in the same location as Gdm but with lower free energies , with the additional moieties projecting in various directions ( Figure 6E ) in ways that might further alter Cdc3 conformation . It is also possible that yeast cells are less permeable to these GdnHCl derivatives . Urea did not rescue at all , possibly because it cannot form a cation . Finally , as would be expected if only small molecules like Gdm can fit into the pocket provided by Thr302 and surrounding residues , 5 mM arginine hydrochloride in the medium provided no rescue to cdc10 ( D182N ) or cdc10 ( G100E ) mutants ( Figure 6F ) . To test if the absence of Arg at ScCdc3 position 302 is sufficient to explain the inability of S . cerevisiae cdc10∆ cells to proliferate at 37°C , we replaced Thr302 with Arg . The double-mutant cdc10∆ cdc3 ( T302R ) cells were , like cdc10∆ CDC3+ , TS , but unlike cdc10∆ CDC3+ the addition of GdnHCl failed to restore proliferation at 37°C ( Figure 7 ) . By contrast , when we replaced Thr302 with Val – an amino acid with properties similar to Thr , including a short side chain – we saw rescue by Gdm ( Figure 7 ) . We interpret these data as evidence that Gdm binds to ScCdc3 near Thr302 ( or Val302 ) in order to promote homodimerization , but does so in a way that is not recapitulated by an Arg side chain . Instead , Arg302 blocks functional Gdm binding , providing further support for the idea that Gdm acts ‘locally’ , and not via a ‘global’ structural stabilization mechanism . We conclude that the T302R substitution alone cannot ‘reverse evolution’ , yet GdnHCl does . How ? ScCdc3 Thr302 is predicted to lie within the septin α4 helix , which is located near both the nucleotide binding pocket and the G interface ( Figure 4 ) . To begin to understand the molecular mechanism by which Gdm at this position promotes Cdc3 homodimerization , we first examined 13 available septin crystal structures and asked which other residues are nearby the residue in the position equivalent to Thr302 . The only non-α4-helix residue within 5 Å in every structure was a highly conserved His ( Table 1 ) residing in a loop at the end of the β4 strand previously dubbed the ‘trans loop 1’ ( Weirich et al . , 2008 ) , the same His predicted by our in silico modeling to contact bound Gdm in Cdc3 ( Figure 4A ) . In SEPT2•GppNHp , His158 contacts a highly conserved Asp within the Switch II loop of the other protomer ( Figure 4B ) ( Sirajuddin et al . , 2009 ) . By contrast , in the sole Chlamydomonas reinhardtii septin , CrSEPT , ( Pinto et al . , 2017 ) or in human SEPT9•GTPγS , the trans loop 1 His instead contacts the nucleotide bound by the other protomer ( Figure 4B ) . Thus , depending on which septins are involved , this key His residue makes either of two distinct contacts across the septin G dimer interface . If during hetero-oligomerization two alternative septins ( e . g . Cdc3 or a mutant Cdc10 ) present to Cdc3 ‘competing’ G interfaces in distinct conformations , then the position of the trans loop 1 His may dictate which is ultimately chosen for incorporation , and Gdm could influence this choice . In SEPT2•GppNHp , His158 is positioned in cis by contacts ( ~3 Å ) with a Glu residue ( Glu202 ) ( Sirajuddin et al . , 2009 ) , which lies one turn away from Arg198 in the α4 helix ( Figure 4B ) . The Gdm moeity of SEPT2 Arg198 ( residue corresponding to Thr302 in Cdc3 ) is also within ~ 3 Å of the backbone amide carbonyl of His158 ( Figure 4B , Table 1 ) , and thus is equally well located to position His158 . In CrSEPT and SEPT9•GTPγS , the Arg198 equivalent instead contacts the Glu202 equivalent ( Figure 4B ) . Thus in some but not all septins the Gdm moiety of the α4 Arg is ideally positioned to contact the trans loop 1 His and thereby potentially bias partner choice during septin G dimerization . By contacting His262 , Gdm bound near Thr302 in Cdc3 could bias Cdc3 towards homodimerization and away from heterodimerization with Cdc10 . To look for additional evidence in support of this model , we next asked which other residues are located in the vicinity ( ≤5 Å ) of the trans loop 1 His in 13 available septin crystal structures . The α4 helix was the only region to meet this criterion in every structure , and , as expected , when Arg was present at the position corresponding to ScCdc3 Thr302 , it was always within 5 Å ( Table 1 ) . Apart from the α4 helix , the Switch II loop , and adjacent trans loop 1 residues , residues within 5 Å of the His fell within only two other regions: the P-loop and the Switch I loop ( Table 1 ) . These are precisely the regions wherein we found non-conservative G interface substitutions between fungal species that tolerate CDC10 deletion and those that do not ( Figure 4B ) . These observations provide strong support for the idea that Gdm binding between the ScCdc3 α4 helix and His262 mimics an ancestral evolutionary state and favors interaction with other similarly disposed Cdc3 molecules . Our data support a model in which Gdm influences whether or not septin complexes incorporate a central homodimer of mutant Cdc10 molecules by influencing whether the Cdc3 trans loop 1 contacts the Switch II loop of another Cdc3 molecule , or points toward the site where nucleotide is normally bound by Cdc10 . This model is reminiscent of the conclusions of a previous study , in which we proposed that the conformation of the Switch II loop in Cdc12 biases choice of its G-dimer partner ( Cdc11 or Shs1 ) ( Weems and McMurray , 2017 ) . In that study , we showed that a Switch II mutation in Cdc12 is sufficient to bias partner choice . In an independent study ( Weems et al . , 2014 ) , we found that a spontaneous mutation ( D210G ) in the equivalent Switch II residue in Cdc3 restores the ability of Cdc3 to interact with a nucleotide-free mutant Cdc10 at high temperatures . Like Gdm , the cdc3 ( D210G ) mutation rescues the TS phenotype of cdc10 ( D182N ) cells , but unlike Gdm it does so by restoring , rather than bypassing , incorporation of the mutant Cdc10 subunits; in other words , the Switch II mutation D210G biases Cdc3 partner choice towards nucleotide-free Cdc10 , whereas Gdm biases partner choice away from it . Our model predicts that in cdc3 ( D210G ) cells Gdm should be less able to exclude the mutant Cdc10 than in cells expressing WT Cdc3 , because the mutant Cdc3 Switch II will be unable to accomplish the molecular contact ( s ) with the Cdc3 trans loop 1 that Gdm promotes . Indeed , Gdm slightly reduced , but did not eliminate , Cdc10 ( D182N ) -GFP localization to septin rings in cdc3 ( D210G ) cells ( Figure 8A ) . These data are consistent with an important role for contacts between the Switch II loop and the trans loop 1 during Cdc3 homodimerization promoted by Gdm . Since Cdc12 is an active GTPase , and the Switch II changes conformation upon GTP hydrolysis , we interpreted our previous results with Switch II-mutant Cdc12 as evidence that the Switch II conformation normally ‘communicates’ across the G interface the phosphorylation state of Cdc12’s bound nucleotide ( Weems and McMurray , 2017 ) . We further bolstered this argument by mutating in Cdc12 a Thr residue in the Switch I loop that is required for septin GTPase activity ( Sirajuddin et al . , 2009; Weems and McMurray , 2017 ) . Notably , Cdc3 lacks Thr in this position . How could Gdm operate via Switch II conformation on Cdc3 , a ‘dead’ GTPase ? We noticed in our phylogenetic analysis of Cdc3 homologs in other fungal species that many possess the ‘catalytic Thr’ ( Figure 4—figure supplement 1B ) . In fact , there was a perfect correlation between the presence of the catalytic Thr ( or Ser ) and the α4 Arg whose Gdm group is presumably mimicked by Gdm in Cdc3 ( Figure 4—figure supplement 1B ) . Analogous to our model of Cdc12 G-partner choice via slow GTP hydrolysis ( Weems and McMurray , 2017 ) , we reasoned that in a species with an active ‘Cdc3’ GTPase , a transient ‘Cdc3’•GTP molecule might prefer to dimerize with ‘Cdc10’ in that species , and ‘Cdc3’•GDP might prefer to form a homodimer , bypassing ‘Cdc10’ incorporation into septin complexes . According to this model , because Saccharomyces , Ashbya , and Kluyveromyces species express catalytically dead Cdc3 molecules , only Cdc3•GTP exists in these cells , and only Cdc10-containing hetero-octamers are made . In Aspergillus nidulans , on the other hand , the Cdc3 homolog , AspB , possesses the catalytic Thr ( Figure 4—figure supplement 1B , Figure 8—figure supplement 1 ) , and this species is known to produce a mix of octamers containing AspD ( the Cdc10 homolog ) and AspD-less hexamers ( Hernández-Rodríguez et al . , 2014 ) . We replaced the GTPase domain of ScCdc3 ( residues 95–341 ) with that of AspB ( residues 6–234; Figure 8—figure supplement 1 ) and assessed Cdc10 incorporation into complexes by measuring the ratio of Cdc10-mCherry to Shs1-GFP in septin rings of living cells . Consistent with our model , Cdc10-mCherry incorporation was markedly reduced compared to cells with WT Cdc3 ( Figure 8B ) , indicating that the GTPase properties of AspB were sufficient to confer Aspergillus-like pathways of septin hetero-oligomer assembly , that is , frequent bypass of Cdc10 incorporation . S . cerevisiae cells expressing the Cdc3-AspB chimera as the only source of Cdc3 were unable to proliferate at 37°C ( Figure 8C ) . We hypothesized that this defect reflects misfolding at high temperature of the Cdc3-AspB chimera to a conformation incapable of interacting with itself or with Cdc10 , and searched for spontaneous suppressors of the TS phenotype . We obtained a suppressor in which proliferation at 37°C was restored ( Figure 8C ) . Sequencing the coding region of Cdc10 revealed a single nucleotide change causing the amino acid substitution Q265H ( Figure 8D ) ; the Cdc3-AspB chimera was unchanged . In most septin structures , His in this position makes a critical contact across the G interface with a highly conserved Trp residue in the ‘Sep4’ motif ( Pan et al . , 2007 ) ( Figure 8E ) , the same Trp we mutated in the cdc3 ( W364A ) mutant ( see Figure 2D ) . Since Gln replaces His here in ScCdc10 ( and in the other fungal species with putatively GTPase-dead Cdc3 homologs; Table 2 ) , the Cdc3•GTP–Cdc10•GDP interface must involve a different kind of interaction . We interpret these findings as evidence that stable association of Cdc10•GDP with GDP-bound Cdc3-AspB , rather than the Cdc3•GTP with which Cdc10 co-evolved , requires His–Trp contacts between Sep4 motifs of the sort found in other dimers between two GDP-bound septins . Indeed , incorporation of Cdc10-mCherry into septin rings was partially restored in cdc3-aspB cdc10 ( Q265H ) cells ( Figure 8B ) . Finally , Gdm was unable to fully rescue the TS phenotype of the cdc3-aspB CDC10+ strain ( Figure 8C ) , as expected if the site of Gdm action is occluded by the α4 Arg present in the Cdc3-AspB chimera . These data provide further support for the idea that during evolution ScCdc3 lost the ability to hydrolyze GTP to GDP and , consequently , the option of assembling septin complexes without a central Cdc10 homodimer . The GTPase domains of Cdc3 and AspB are < 50% identical , and differ at many more positions than the five identified by our phylogenetic analysis as co-varying with the α4 Arg and the catalytic Thr ( Figure 8—figure supplement 1 ) . To ask if those five differences are sufficient to direct Cdc10 bypass during septin assembly , we used CRISPR-Cas9 to cut the endogenous CDC3 coding sequence and , via homologous recombination , replace most of it with a ‘recoded’ gene encoding the same polypeptide sequence but using numerous synonymous codons , or with a similarly ‘recoded’ gene additionally encoding the five substitutions ( P127E D128S K181T T302R Q306D ) . Recoding allowed us to obtain transformants in which recombination occurred at the ends of our donor templates , rather than in sequences immediately flanking the cut site , thus promoting incorporation of all five substitutions . We also obtained a transformant in which only two substitutions were incorporated , T302R Q306D , and we included this strain in our subsequent analysis . The recoded control strain and the P127E D128S K181T T302R Q306D and T302R Q306D mutant strains were indistinguishable from the parental CDC3+ strain in terms of growth at all tested temperatures , and also with regard to the relative amounts of Cdc10-mCherry incorporated into septin rings ( Figure 8—figure supplement 2 ) . Thus , although we cannot rule out the possibility that recoding with synonymous codons alters Cdc3 co-translational folding in some way that masks effects of the amino acid changes we introduced , it appears that additional sequence changes are needed to ‘reverse evolution’ and restore the ability of Cdc3 to homodimerize robustly in vivo . We propose that Gdm binding adjacent to ScCdc3 Thr302 provides molecular contacts that position His262 to interact in trans with an Asp residue from a septin Switch II loop in a ‘GTP state’ , such as that provided by another Cdc3•GTP . The spontaneous mutation we previously found to promote Cdc3 homodimerization ( McMurray et al . , 2011 ) alters a trans loop 1 residue immediately adjacent to Cdc3 His262 , Gly261 . Val in this position probably also ‘tunes’ the configuration of His262 to favor Cdc3 homodimerization via interaction with Cdc3 Asp201 . Specific configurations of the Switch II loop clearly dictate G dimer partner choice , as mutating the residue corresponding to Cdc3 Asp210 promotes both stable Cdc3 heterodimerization with nucleotide-free Cdc10 ( Weems et al . , 2014 ) and biased selection of Shs1 by Cdc12 ( Weems and McMurray , 2017 ) . Mutating a nearby Asp in the Switch II loop of human SEPT7 also allows homodimerization when only a non-hydrolyzable GTP analog is available ( Zent and Wittinghofer , 2014 ) . ‘Tuning’ by Gdm via cis interactions presumably occurs prior to G dimerization and is likely a transient event: once the trans loop 1 His has engaged either the Switch II Asp or nucleotide ( or nearby where nucleotide normally binds , in the case of nucleotide-free mutant Cdc10 ) , the Arg–His interaction probably need not persist post-dimerization . It follows that Gdm may not remain bound to Cdc3 following G dimerization , and thus falls under a broad definition of a ‘pharmacological’ or ‘chemical’ chaperone . Since in otherwise WT cells Gdm promotes bypass of mutant Cdc10 molecules with substitutions in various locations in the G interface and nucleotide-binding pocket , but not WT Cdc10 , additional contacts between WT Cdc10•GDP and Cdc3•GTP likely favor Cdc3–Cdc10 heterodimerization over Cdc3 homodimerization despite the presence of Gdm . Poorer GdnHCl rescue of cdc10 ( G44D ) ( P-loop mutant ) relative to cdc10 ( D182N ) ( G4 mutant ) and cdc10 ( G100E ) ( Switch II mutant ) may reflect a comparatively better ability of Cdc10 ( G44D ) molecules to , despite temperature-induced misfolding , occupy the Cdc3 G interface and resist exclusion , thereby interfering with hexamer assembly . Bound nucleotide itself may provide the key contacts that distinguish WT Cdc10 from the mutants , with Cdc10 ( G44D ) subunits being more successful than the other mutants at binding nucleotide . At moderate temperatures ( 22–27°C ) Cdc3 molecules carrying the G365R substitution are slow to acquire the conformation competent for interaction with Cdc10 ( Schaefer et al . , 2016 ) . Our observation that GdnHCl exacerbated the TS phenotypes of cdc3 ( G365R ) cells ( Figure 1A ) thus provides further evidence of a bias imposed by Gdm on Cdc3 toward conformations that are suboptimal for interaction with Cdc10 , the effects of which are masked in WT cells but become obvious when combined with additional folding biases like the G365R substitution . Whereas Gdm was unable to drive Cdc10 bypass in WT cells , swapping the GTPase domain of Cdc3 for that of AspB was sufficient to promote bypass of WT Cdc10 ( Figure 8B ) , consistent with the idea that Cdc3•Gdm only partially mimics the ancestral homodimer configuration . Introducing into Cdc3 the five residues ( including the catalytic Thr and α4 Arg ) that are found in Cdc3 homologs in species that tolerate loss of the Cdc10 homolog was unable to drive exclusion of Cdc10 from septin complexes ( Figure 8—figure supplement 2 ) . Cdc12 natively possesses four of these five residues ( Figure 8—figure supplement 1 ) , yet forcing septin hexamer assembly via non-native Cdc12 homodimers ( by deleting SHS1 and CDC11 ) results in S . cerevisiae cells that , like cdc10∆ , cannot proliferate at 37°C ( McMurray et al . , 2011 ) . Clearly GTPase activity and the presence of these residues is insufficient for robust septin homodimerization , and additional changes are likely key . Other factors promote Cdc3 homodimerization in similar ways as GdnHCl , as illustrated by the collective ability of the G261V mutation , excess GTP and a synthetic lipid monolayer to drive filament formation by Cdc10-less complexes in vitro ( Bertin et al . , 2010 ) . This effect requires high protein concentration ( 0 . 15 mg/mL ) ( Bertin et al . , 2010 ) . Even higher protein concentration ( 1 mg/mL ) bypasses the need for Cdc3 mutations and some extrinsic cofactors , though excess GTP is still critical ( Farkasovsky et al . , 2005 ) . These observations support a ‘conformational ensemble’ model of Cdc3 folding , in which at any given time a small proportion of Cdc3 molecules adopt the homodimerization-competent conformation . While this proportion can be increased in various ways , another way to populate even a very rare conformation is to simply increase the total number of Cdc3 molecules . Although septins are absent from land plants , protein transport across the outer envelope membrane of chloroplasts is mediated by TOC complexes containing a small GTPase subunit , Toc34 , that shares high structural homology to septins . Others have noted ( Weirich et al . , 2008 ) that the septin trans loop 1 is in a similar position to a Toc34 interface loop that makes key contacts with nucleotide bound by the partner protein across a homodimer interface ( Figure 9A ) . In both pea ( Pisum sativum ) and Arabidopsis thaliana an Arg residue in the Toc α4 helix projects towards the homodimer interface , to within 5 Å of the trans loop 1 equivalent ( Figure 9A ) . Arg in this position is conserved among all Toc34 homologs ( Figure 9B ) . We conclude that Arg-mediated α4 helix tuning of the trans loop 1 for specific dimerization events across the nucleotide-binding-pocket-containing interface is an evolutionarily ancient mechanism that predated the appearance of septins per se . Following loss of the α4 Arg in the fungal lineage giving rise to S . cerevisiae , A . gossypii , and K . lactis , the amino acid at the position equivalent to ScCdc3 Thr302 diverged rapidly . Ser or Ala appear in its place in other species within the genus Saccharomyces , which share >86% sequence identity with ScCdc3 ( Figure 9C ) . By comparison , within the genus Candida Arg is invariant among six species sharing as little as 35% identity with C . albicans Cdc3 ( Figure 9C ) . These observations are consistent with the notion that Thr in this position does not make specific structural contributions to septin assembly . One clear prediction from this model is that in the two fungal lineages distinguished by the presence or absence of the Cdc3 ‘catalytic’ Thr , the Cdc10 subunit should have evolved differently , to recognize in one case only Cdc3•GTP and , in the other , either Cdc3•GTP and Cdc3•GDP . Our unbiased identification of the cdc10 ( Q265H ) mutation in cells expressing the Cdc3-AspB chimera provided one example of how Cdc10 adapted to interacting with a Cdc3 subunit ‘fixed’ in the GTP-bound state . Others previously noted ( Sirajuddin et al . , 2009 ) that ScCdc10 is unusual among septins in that it has Lys rather than His in the trans loop 1 . If the residue in this position acts to promote interaction with a G dimer partner in a specific nucleotide state , this may explain why Lys in this position is shared by AgCdc10 and KlCdc10 , but not any of the Cdc10 homologs in the fungal species in which the Cdc3 homolog retains the ‘catalytic Thr’ ( Table 2 ) . Co-variation of these residues likely reflects co-evolution . Figure 9—figure supplement 1A illustrates a possible pathway of such changes during septin evolution . Were Cdc10-less hexamers specifically selected against during evolution of a fungal lineage ? Compared to other closely-related fungi , Ashbya , Kluyveromyces , and Saccharomyces share little in common with regard to the cellular functions in which septins are known to operate , such as polarity , morphogenesis , cytokinesis , or sporulation . Hence it is hard to imagine how losing septin hexamers could provide a specific selective advantage to a common ancestor , though this may reflect our incomplete understanding of what septin hexamers do differently than septin octamers . On the other hand , the assembly of hexamers liberates the Cdc10 subunits to act elsewhere , and specific functions of ‘free’ Cdc10 homologs could represent an evolutionary pressure point to maintain GTPase activity in Cdc3 homologs . Indeed , in U . maydis ( Alvarez-Tabarés and Pérez-Martín , 2010 ) and A . nidulans ( Hernández-Rodríguez et al . , 2014 ) the Cdc10 homologs are found in distinct cellular structures independently of the other septins , where they may function independently , as well . Many human cell types also natively produce both septin hexamers and octamers ( Abbey et al . , 2016; Kim et al . , 2011; Sellin et al . , 2014; Sellin et al . , 2011a; Sellin et al . , 2011b ) . Incorporation of a subunit from the SEPT3 group ( SEPT3 , SEPT9 , or SEPT12 ) appears to confer functions to human octamers that hexamers lack , such as microtubule bundling ( Bai et al . , 2013 ) , vesicle trafficking in neurons ( Karasmanis et al . , 2018 ) , directing proper spermatogenesis ( Kuo et al . , 2015 ) , and membrane abscission at the end of cytokinesis ( Estey et al . , 2010 ) . Human and yeast septin octamers were thought to differ in where within the complex the ‘extra’ subunits incorporate , with SEPT3-group septins like SEPT9 occupying the ‘terminal’ positions at both octamer ends ( Kim et al . , 2011 ) . However , two recent reports ( Mendonça et al . , 2019; Soroor et al . , 2019 ) suggest that in fact SEPT3-group septins form a central homodimer flanked by SEPT7 , with SEPT6 and SEPT2 occupying the penultimate and terminal positions , respectively ( McMurray and Thorner , 2019 ) . In this ‘revised’ organization , the GTPase-dead subunit position ( the SEPT6 group ) corresponds to that of Cdc12 , and the Cdc3 position is occupied by SEPT7 , an active GTPase ( see Figure 9—figure supplement 1B ) . Notably , a mutation ( S63N ) that doubles SEPT7 GTPase activity in vitro shifts assembly in vivo from primarily hetero-octamers to smaller complexes ( Abbey et al . , 2016 ) , consistent with our model that septins in the GDP-bound state are prone to homodimerization across the G interface and exclusion of intervening subunits . SEPT9 and SEPT3 are active GTPases and , like Cdc10 ( Versele and Thorner , 2004 ) , as monomers they hydrolyze GTP faster than other septins ( Macedo et al . , 2013; Zent and Wittinghofer , 2014 ) . We previously proposed that rapid GTP hydrolysis by monomeric Cdc10 allows it to homodimerize prior to interacting with Cdc3 , such that Cdc10•GDP encounters Cdc3•GTP during hetero-oligomerization ( Weems and McMurray , 2017 ) . If human septins assemble similarly , then hexamers lacking SEPT3-family subunits are not made by eviction of central homodimers from pre-existing hetero-octamers . Instead , the outcome of hexamer or octamer is decided upon de novo assembly and depends upon the phosphorylation state of the nucleotide bound by SEPT7 . Indeed , induced overexpression of tagged SEPT3-family septins in human K562 cells demonstrated that preexisting hexamers gradually disappeared by dilution during cell division and were replaced by newly-made octamers ( Sellin et al . , 2014 ) . In that study , the ‘excess’ SEPT9 persisted for days ( Sellin et al . , 2014 ) . Given subsequent results pointing to a role for ‘free’ SEPT9 in vesicle trafficking in mouse hippocampal neurons ( Karasmanis et al . , 2018 ) , in specific cellular circumstances slow SEPT7 GTP hydrolysis may be crucial for generating ‘free’ SEPT3-family septins that perform independent functions . Cdc10 and SEPT9 fall into one ancient evolutionary clade , whereas Cdc3 , Cdc12 , SEPT6 and SEPT7 all appear to have distinct origins ( Auxier et al . , 2019 ) . Thus the loss of septin GTPase activity occurred multiple times independently . By demonstrating that GTPase activity facilitates promiscuity during dimerization , and thereby expands the repertoire of septin hetero-oligomers , our findings provide the first molecular rationale for the evolution of GTPase-dead septins: to narrow the variety of building blocks available for higher-order assemblies . A similar logic may explain the most famous example of loss of GTPase activity of a cytoskeletal protein , that of the α subunit of the tubulin heterodimer . The ability of Gdm to occupy vacancies created by the mutation of Arg residues is well documented in vitro for a wide variety of enzymes ( Baldwin et al . , 1998; Barnett et al . , 2010; Boehlein et al . , 1997; Dugdale et al . , 2010; Goedl and Nidetzky , 2008; Guillén Schlippe and Hedstrom , 2005; Hung et al . , 2014; Lehoux and Mitra , 2000; Phillips et al . , 1992; Rynkiewicz and Seaton , 1996 ) , but effects of GdnHCl on living cells were limited to Gdm binding at sites in WT proteins or nucleic acids: in the ATP-binding pocket of Hsp104 and its relatives ( Zeymer et al . , 2013 ) , in the intracellular pore of voltage-gated potassium channels ( Kalia and Swartz , 2011 ) , in the enteroviral 2C protein ( Sadeghipour et al . , 2012 ) , and , most recently , in bacterial riboswitches that sense intracellular Gdm and induce a cellular response to detoxify it ( Nelson et al . , 2017 ) . The function of a rationally-designed Arg-mutant kinase can be rescued in vitro by GdnHCl ( Williams et al . , 2000 ) , and the same mutant is rescued by imidazole in vivo ( Qiao et al . , 2006 ) , but our findings that GdnHCl can functionally replace an Arg in Cdc3 represent the first documented case of ‘chemical rescue’ by GdnHCl in vivo . The apparent requirement for GdnHCl to be present during Cdc3 folding illustrates how our case of GdnHCl mimicry of a ‘missing’ Arg is fundamentally distinct from the other examples of ‘chemical rescue’ by GdnHCl . In the other cases , a protein that had evolved to adopt a specific conformation was mutated at a single Arg in order to perturb this conformation ( or , in many cases , interaction with a substrate molecule ) , and Gdm was able to restore the native fold/interaction . In those prior studies , the mutated protein was already near the native conformation and only a small effect of GdnHCl was required to achieve the native conformation . In our case , a WT protein that had evolved to avoid a particular conformation was induced to reach that non-native conformation only when GdnHCl was present during its folding . During Cdc3 folding in the absence of Gdm , other intramolecular interactions drive acquisition of the low-energy native conformation that disfavors homodimerization . Gdm would have to act early , in a manner equivalent to how canonical chaperones direct the folding pathways of their clients . The ability of the T302R mutation to block , but not to mimic , the effect of GdnHCl could be explained by an ability of Gdm to make intramolecular contacts within Cdc3 that , because it is tethered to the polypeptide backbone , the Gdm moiety of an Arg side chain cannot . Our findings do not exclude the possibility that additional Gdm molecules bind elsewhere in Cdc3 and contribute to Cdc3 folding , with occupancy of the site near Thr302 being necessary but not sufficient . In yeast cells cultured in 1 mM GdnHCl the intracellular concentration is ~ 20 mM ( Jones et al . , 2003 ) . Thus in the conditions in which we observed cdc10 rescue Gdm likely binds to hundreds or thousands of proteins , with few adverse effects . Indeed , GdnHCl is approved by the Food and Drug Administration of the USA for use in humans , and at low doses has few serious side effects ( Oh et al . , 1997 ) . Exploiting the ability of Gdm to replace Arg in living cells represents a compelling area for future research , particularly as a possible therapeutic ‘pharmacoperone’ ( Conn et al . , 2014 ) . Finally , given the evidence that bacteria evolved ways to manage intracellular Gdm ( including actively exporting it [Kermani et al . , 2018] ) , Gdm may , like the naturally occurring osmolyte trimethylamine N-oxide ( Bandyopadhyay et al . , 2012 ) and the chaperone Hsp90 ( Lindquist , 2009 ) , influence evolution by buffering against the phenotypic consequences of mutations . Yeast were transformed using the Frozen-EZ Yeast Transformation II Kit ( Zymo Research ) . Genetic manipulations were otherwise performed according to standard methods ( Amberg et al . , 2005 ) , except for the creation of the cdc3-aspB chimera and the CDC3 ( P127E D128S K181T T302R Q306D ) and CDC3 ( T302R Q306D ) strains and associated ‘recoded’ control strain , which were made in yeast strain JTY5397 using CRISPR-Cas9 cleavage of the CDC3 locus and repair with PCR products as donor templates , following an established protocol ( Akhmetov et al . , 2018 ) and using plasmid pEM-CDC3-CRISPR1 , a gift of Ed Marcotte , which encodes Cas9 and a guide RNA targeting nucleotides 1039–1069 of the CDC3 ORF ( 5’ GATATTGTAGAGAACTACAG 3’ ) . To create the donor template for the cdc3-aspB chimera , a portion of the aspB coding sequence from the aspB plasmid pRL10 , which was a gift of Michelle Momany and is based on the yeast two-hybrid vector pGBKT7 ( Takara Bio/Clontech ) , was used as template for a PCR reaction with Q5 polymerase ( New England Biolabs ) and primers Cdc3AspBfw and Cdc3AspBextend1 according to the polymerase manufacturer’s instructions . The resulting product , which included the sequence encoding part of AspB flanked by CDC3 sequences , was used as template for a second Q5 reaction with primers Cdc3AspBfw and Cdc3GTPase_extend2 , in order to extend the CDC3 homology to span the site of the Cas9 cut . To create the donor template for the CDC3 ( P127E D128S K181T T302R Q306D ) strain and its ‘recoded’ control , an 870-nucleotide segment of the Cdc3 ORF corresponding to the GTPase domain was synthesized ( Integrated DNA Technologies ) with multiple synonymous codons replacing the native sequence , representing 67 nucleotide changes but no amino acid substitutions . An otherwise identical sequence also including the P127E D128S K181T T302R Q306D mutations was also synthesized . Donor template PCRs were done with Q5 and primers G1KTTRQDfw and Cdc3recodere . Transformants were screened by PCR of the CDC3 locus and subsequent sequencing . S . cerevisiae media: Rich growth medium was YPD ( 1% yeast extract ( #Y20020 , Research Products International Corp . , Mount Prospect , IL ) , 2% peptone ( #P20241 , RPI Corp . ) , 2% dextrose ( #G32045 , RPI Corp . ) ) . Synthetic growth medium was based on YC ( 0 . 1 g/L Arg , Leu , Lys , Thr , Trp and uracil; 0 . 05 g/L Asp , His , Ile , Met , Phe , Pro , Ser , Tyr and Val; 0 . 01 g/L adenine; 1 . 7 g/L Yeast Nitrogen Base ( YNB ) without amino acids or ammonium sulfate; 5 g/L ammonium sulfate; 2% dextrose ) with individual components ( from Sigma Aldrich , St . Louis , MO , or RPI Corp . ) eliminated as appropriate for plasmid selection . For solid media , agar ( #A20030 , RPI Corp . ) was added to 2% . For counterselection against URA3 , 5-fluoro-orotic acid monohydrate ( #F5050 , United States Biological , Salem , MA ) was added to modified YC medium ( 1 g/L Pro in place of ammonium sulfate , 0 . 02 g/L uracil ) to final 0 . 6 g/L . For counterselection against LYS2 , α-aminoadipate ( #A1374-09 , US Biological ) at 2 g/L replaced ammonium sulfate in YC medium containing only YNB , uracil , His , Leu , Lys , Met , Trp , and dextrose . G-418 sulfate ( Geneticin , #G1000 , US Biological ) was added to YPD at 200 μg/mL for selection of kanMX . A . gossypii media: Ashbya strains were grown in 10 mL Ashbya Full Medium ( AFM , 1% casein peptone , 1% yeast casein extract , 2% dextrose and 0 . 1% myo-inositol ) containing ampicillin ( 100 μg/mL ) , CloNAT ( 50 μg/mL ) and G-418 ( 200 μg/mL ) . For imaging , cells were washed into 2X low-fluorescence minimal medium ( per liter: 3 . 4 g YNB without amino acids or ammonium sulfate , 2 g Asp , 2 g myo-inositol , 40 g dextrose , 20 mg adenine , 21 g 3- ( N-morpholino ) propanesulfonic acid ( MOPS ) , pH adjusted to 7 . 0 with sodium hydroxide ) . GdnHCl ( #G4505 , Sigma Aldrich , St . Louis , MO ) , ArgHCl ( #A5131 , Sigma Aldrich ) , urea ( Bio-Rad # 1610730 ) , aminoguanidine hydrochloride ( #sc-202931 , Santa Cruz Biotechnology , Santa Cruz , CA ) and N-ethylguanidine hydrochloride ( #sc-269833 , Santa Cruz Biotechnology ) were dissolved in water . Growth curves were generated using a Cytation 3 plate reader ( BioTek , Winooski , Vermont ) as described previously ( Schaefer et al . , 2016 ) using a starting overnight culture of the cdc10 ( D182N ) strain CBY06417 grown at 22°C prior to dilution into 200 µL YPD medium in each well . Immunofluoresence was performed as described previously ( Pringle et al . , 1989 ) using an anti-Cdc11 primary antibody ( #sc7170 , Santa Cruz Biotechnology , Santa Cruz , CA ) and Alexa-Fluor 488-labeled anti-rabbit secondary antibody ( #20012 , Biotium , Inc , Fremont , CA ) at a 1:5000 and 1:1000 dilution , respectively . BL21 Star ( DE3 ) E . coli cells ( ThermoFisher Scientific # C601003 ) were transformed with plasmids pMVB121 ( encoding 6xHis-Cdc12 ) and pMVB133 ( encoding untagged Cdc3 and Cdc11 ) and grown overnight in 3 mL LB medium ( per L: 10 g tryptone , 5 g yeast extract , 10 g NaCl ) with ampicillin ( 50 µg/mL ) , chloramphenicol ( 34 µg/mL ) , and varying amounts of GdnHCl ( from 0 mM to 96 mM ) at 37°C in glass culture tubes with agitation . Growth was assessed qualitatively and was equivalent for all cultures , so 50 mM was chosen as the concentration for subsequent experiments . 1 L 2XTY ( per L: 16 g Tryptone , 10 g yeast extract , 5 g NaCl ) cultures with ampicillin , chloramphenicol and with or without 50 mM GdnHCl were grown at 37°C with shaking to OD600 between 0 . 6 and 1 . 0 , at which point IPTG was added to 0 . 1 mM and the culture was incubated overnight . Cells were collected by centrifugation , resuspended in a minimal volume of lysis buffer ( 300 mM NaCl , 2 mM MgCl2 , 40 µM GTP , 1 mM EDTA , 5 mM β-mercaptoethanol , 0 . 5% Tween-20 , 12% glycerol , 50 mM Tris-HCl , pH 8 . 0 , with or without 50 mM GdnHCl ) and frozen by dripping into liquid nitrogen . Cells were thawed in 30 mL lysis buffer containing protease inhibitors ( Complete EDTA-free , Roche #11836170001 ) , lysozyme and benzonase ( Sigma-Aldrich #E8263 ) . Cells were sonicated for 5 min total ( 30 s on , 30 s off ) with a tip sonicator . Lysates were clarified by centrifugation at 12 , 000xg at 4°C and mixed with 200 µL of a slurry of Ni2+-NTA agarose ( Qiagen #30210 ) that had been equilibrated in lysis buffer . After rotating for 1 hr at 4°C , beads were collected by centrifugation , washed two times with 5 mL of 300 mM NaCl , 20 mM imidazole , 0 . 1% Tween-20 , and 50 mM Tris-HCl , pH 8 . 0 , with or without 50 mM GdnHCl . The beads were poured into a column and the septin complex was eluted in fractions from the beads with 500 mM NaCl , 40 µM GTP , 500 mM imidazole , 50 mM Tris-HCl , pH 8 . 0 , with or without 50 mM GdnHCl . Aliquots of fractions were analyzed by SDS-PAGE and Coomassie staining , and imidazole was removed from the fraction containing the peak of septin complex via buffer exchange into 500 mM NaCl , 40 µM GTP , 50 mM Tris-HCl , pH 8 . 0 using a PD-10 column ( GE Healthcare LifeSciences #17085101 ) . The sample was further purified by size exclusion chromatography using a Superdex 200 column and 50 mM Tris-HCl , 300 mM NaCl pH 8 , supplemented with or without 50 mM GdnHCl ) . The fractions were analyzed by SDS-PAGE and Coomassie staining to identify samples for electron microscopy . For EM of yeast cells , cells were cultured at 37°C in YPD medium with 3 mM GdnHCl and collected by vacuum filtration , then vitrified by high pressure freezing using a HPM100 ( Leica Microsystems , Wetzlar , Germany ) apparatus . The frozen cells were cryo-substituted using a Leica ASF2 device in a medium containing osmium tetroxide ( 1% ) , water ( 5% ) and uranyl acetate ( 0 . 1% ) in acetone using a protocol described elsewhere ( Bertin and Nogales , 2016 ) . The cells were embedded into Epon before being sectioned into 50 nm sections using an ultramicrotome UC6 ( Leica ) equipped with a 4 . 5 mm diamond knife ( Diatome , Hatfield , PA ) . The resulting sections were deposited onto copper electron microscopy grids ( mesh size of 100 ) . The images were collected with a 120 kV Lab6 microscope ( Technai Spirit , FEI , Eindhoven , Netherlands ) equipped with a CCD Quemesa camera ( Olympus , Tokyo , Japan ) . For EM of septin complexes purified from E . coli , 4 µL of sample at a final concentration of 0 . 01 mg/mL for high salt conditions or 0 . 1 mg/mL for low salt conditions were absorbed for 30 s on a glow-discharged carbon coated grid ( Electron Microscopy Sciences , ref CF300-CU ) . The grids were then negatively stained for 1 min using 2% uranyl formate . Data was collected using either a Tecnai Spirit microscope ( Thermofischer , FEI , Eindhoven , The Netherlands ) operated at an acceleration voltage of 80Kv and equipped with a Quemesa ( Olympus ) camera or with a Lab6 G2 Tecnai ( ThermoFisher , FEI , Eindhoven , the Netherlands ) operated at an acceleration voltage of 200 kV . The data was acquired using a 4k × 4k F416 CMOS camera ( TVIPS ) in an automated manner using the EMTools software suite ( TVIPS ) . Square boxes were hand-picked from the images respectively using the boxer tool from the EMAN software suite ( Ludtke et al . , 1999 ) . 4233 boxes of 135 pixels were picked from the control sample while 3247 boxes of 203 pixels were picked from images collected of the sample grown in the presence of GdnHCl . Subsequent processing was carried out using SPIDER ( Frank et al . , 1996 ) . After normalization of the particles , a non-biased reference-free algorithm was used to generate 20 classes . Multi-reference alignment followed by classification was carried out to generate about 50 classes . Fast Fourier transform analysis and power spectrum generation were performed in ImageJ ( Schneider et al . , 2012 ) . All images of S . cerevisiae were captured using an EVOSfl all-in-one microscope ( Advanced Microscopy Group , Mill Creek , Washington ) using a 60X objective and Texas Red , RFP , or GFP LED/filter cubes , as described previously ( Johnson et al . , 2015 ) . Bud neck fluorescence was quantified using line scans as described previously ( Johnson et al . , 2015; Weems and McMurray , 2017 ) . We developed a new macro for ImageJ/FIJI , called ‘Get plot profile Min-Max’ , to facilitate such quantification . Intensity values or ratios thereof were plotted using GraphPad Prism 8 . 0 , using the medium smoothing ( kernel density ) setting for violin plots . When necessary for presentation , images were inverted and brightness- and contrast-adjusted using Adobe Photoshop ( Adobe Systems Incorporated , San Jose , California ) or ImageJ ( Schneider et al . , 2012 ) , always the same way for every image of the same type . Ashbya cells in minimal low-fluorescence medium were mounted onto 2% agarose gel pads and the edges were sealed with Valap ( a 1:1:1 mixture of vaseline , lanolin , and paraffin ) . Images were acquired using a Zeiss Axioimager-M1 upright light microscope ( Carl Zeiss , Jenna , Germany ) equipped with a Plan-Apochromat 63X/1 . 4 numerical aperture oil objective and an Exfo X-Cite 120 lamp . Fluorescence imaging was performed using Zeiss 38HE filter cubes ( GFP ) . Images were acquired using an Orca-AG charge-coupled device ( CCD ) camera driven by μManager . Visualization and rendering of structures in PDB format was performed using MacPyMol ( Schrödinger , New York , NY ) . Residues within 5 Å of other residues were identified using the ‘around’ command . Structure prediction for ScCdc3 was performed using the I-TASSER server ( http://zhanglab . ccmb . med . umich . edu/I-TASSER/ ) ( Zhang , 2008 ) and ScCdc3 sequence truncated to amino acids 116–411 to match the template model , SEPT2•GppNHp ( PDB 3FTQ ) . Molecular docking was performed using a local installation of AutoDock Vina ( Trott and Olson , 2010 ) via Chimera 1 . 11 ( http://www . rbvi . ucsf . edu/chimera ) ( Pettersen et al . , 2004 ) , using a search volume containing the entire ‘receptor’ model . The fungal protein sequences listed in the legend to Figure 4—figure supplement 1 were analyzed using the Phylogeny . fr platform online ( http://www . phylogeny . fr/ ) ( Dereeper et al . , 2008 ) to generate a phylogenetic tree using the ‘One Click’ method with default settings . Multiple alignments were performed using the COBALT tool at the NCBI server ( https://www . ncbi . nlm . nih . gov/tools/cobalt/ ) or , for sequences of yeast species not available via NCBI , using the ‘Fungal Alignment’ function at the Yeast Genome Database ( Saccharomyces Genome Database , https://www . yeastgenome . org/ ) . In some cases , sequences obtained via the Yeast Genome Database were aligned with other sequences via COBALT . Genomic DNA from the cdc3-AspB suppressor strain was isolated as described previously and used as template in Q5 PCR reactions with primers 5’cdc3fw and 3’cdc3re , or 5’cdc10fw and 3’cdc10re performed according to the polymerase manufacturer’s instructions . Following treatment with alkaline phosphatase and Exonuclease I ( Thermo Scientific Fermentas ) , the purified PCR product was directly sequenced at the sequencing facility of the Barbara Davis Center for Childhood Obesity , with the same primers used for amplification and , for CDC10 , with primer cdc10midfw . To confirm correct strain construction , sequencing of other loci was performed in an equivalent manner with appropriate primers ( S2 Table ) . Sample sizes ( n , number of cells ) for fluorescence microscopy were chosen based on our ability to detect decreases in bud neck localization in our previous studies ( Johnson et al . , 2015; Weems and McMurray , 2017 ) . Similarly , sample sizes ( n , number of particles ) for single-particle EM analysis were chosen based on our ability to detect changes in septin complex subunit composition in our previous study of the Cdc3 ( G261V ) mutation ( McMurray et al . , 2011 ) , and sample sizes ( n , number of replicate wells ) for growth curve analysis were chosen based on our ability to detect changes in the growth rates of septin-mutant strains in our previous study ( Schaefer et al . , 2016 ) . Each individual experiment was performed at least once , with all experimental samples and relevant controls prepared and analyzed in parallel . Analysis of effect sizes and generation of effect size plots were performed using DABEST ( data analysis with bootstrap-coupled estimation ) ( Ho et al . , 2019 ) via the online server at http://www . estimationstats . com . Other violin plots were generated with Prism 8 . 0 ( Graphpad ) using the ‘medium’ smoothing/kernel density setting .	For a cell to work and perform its role , it relies on molecules called proteins that are made up of chains of amino acids . Individual proteins can join together like pieces in a puzzle to form larger , more complex structures . How the protein subunits fit together depends on their individual shapes and sizes . Many cells contain proteins called septins , which can assemble into larger protein complexes that are involved in range of cellular processes . The number of subunits within these complexes differs between organisms and sometimes even between cell types in the same organism . For example , yeast typically have eight subunits within a septin protein complex and struggle to survive when the number of septin subunits is reduced to six . Whereas other organisms , including humans , can make septin protein complexes containing six or eight subunits . However , it is poorly understood how septin proteins are able to organize themselves into these different sized complexes . Now , Johnson et al . show that a chemical called guanidinium helps yeast make complexes containing six septin subunits . Guanidinium has many similarities to the amino acid arginine . Comparing septins from different species revealed that one of the septin proteins in yeast lacks a key arginine component . This led Johnson et al . to propose that when guanidinium binds to septin at the site where arginine should be , this steers the septin protein towards the shape required to make a six-subunit complex . These findings reveal a new detail of how some species evolved complexes consisting of different numbers of subunits . This work demonstrates a key difference between complexes made up of six septin proteins and complexes which are made up of eight , which may be relevant in how different human cells adapt their septin complexes for different purposes . It may also become possible to use guanidinium to treat genetic diseases that result from the loss of arginine in certain proteins .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2020	Guanidine hydrochloride reactivates an ancient septin hetero-oligomer assembly pathway in budding yeast
Individuals with neurofibromatosis type 1 ( NF1 ) frequently exhibit cognitive and motor impairments and characteristics of autism . The cerebellum plays a critical role in motor control , cognition , and social interaction , suggesting that cerebellar defects likely contribute to NF1-associated neurodevelopmental disorders . Here we show that Nf1 inactivation during early , but not late stages of cerebellar development , disrupts neuronal lamination , which is partially caused by overproduction of glia and subsequent disruption of the Bergmann glia ( BG ) scaffold . Specific Nf1 inactivation in glutamatergic neuronal precursors causes premature differentiation of granule cell ( GC ) precursors and ectopic production of unipolar brush cells ( UBCs ) , indirectly disrupting neuronal migration . Transient MEK inhibition during a neonatal window prevents cerebellar developmental defects and improves long-term motor performance of Nf1-deficient mice . This study reveals essential roles of Nf1 in GC/UBC migration by generating correct numbers of glia and controlling GC/UBC fate-specification/differentiation , identifying a therapeutic prevention strategy for multiple NF1-associcated developmental abnormalities . Neurofibromatosis type 1 ( NF1 ) is a genetically inherited disorder that afflicts 1 in 2700 newborns ( Evans et al . , 2010 ) . NF1 is caused by loss-of-function mutations in the NF1 tumor suppressor gene , which encodes neurofibromin , a negative regulator of proto-oncogene RAS ( Cichowski and Jacks , 2001; Upadhyaya and Cooper , 2012 ) . RAS mediates multiple signaling pathways including extracellular signal-regulated kinase ( ERK ) subfamily of mitogen-activated protein kinases ( MAPK ) , phosphatidylinositol 3-kinase ( PI3K ) and mammalian target of rapamycin complex 1 ( mTORC1 ) ( Schubbert et al . , 2007; Mendoza et al . , 2011 ) . In addition to the development of tumors in the peripheral and central nervous system ( CNS ) , neurodevelopmental deficits are highly prevalent among children with NF1 , negatively impacting cognition , motor function , and social interaction ( Hyman et al . , 2005 , 2006; Johnson et al . , 2010; Krab et al . , 2011; Lorenzo et al . , 2011; Lehtonen et al . , 2013; Walsh et al . , 2013; Garg et al . , 2013a , 2013b; Adviento et al . , 2014; Champion et al . , 2014; Plasschaert et al . , 2014 ) . While cognitive impairments associated with NF1 have been well documented , motor dysfunction , social and behavioral deficits including autism spectrum disorders ( ASD ) have only recently been established as common features of NF1 in childhood ( Johnson et al . , 2010; Krab et al . , 2011; Lorenzo et al . , 2011; Walsh et al . , 2013; Garg et al . , 2013a , 2013b; Champion et al . , 2014 ) . Approximately 50–80% of children with NF1 have impairments in fine and gross motor function , which can be identified as early as at the toddler stage ( Johnson et al . , 2010; Krab et al . , 2011; Lorenzo et al . , 2011 ) . One recent study has shown that impairments of gross motor skills and cognitive functioning in NF1 are often co-morbidities , suggesting the potential existence of a common pathological mechanism underlying both motor and cognitive impairments ( Champion et al . , 2014 ) . The cerebellum is traditionally known as a motor organ , which controls both motor coordination and motor learning ( Sillitoe and Joyner , 2007 ) . However , it has been increasingly recognized that the cerebellum also plays a critical role in higher-order brain functions such as cognition , learning , affect and behavior ( Schmahmann and Caplan , 2006; Strick et al . , 2009; Basson and Wingate , 2013 ) . About 80% of individuals with ASD exhibit anatomical abnormalities in the cerebellum , among which gliosis and Purkinje cell loss are most commonly identified ( Bailey et al . , 1998; Palmen et al . , 2004; Vargas et al . , 2005; Fatemi et al . , 2012 ) . Magnetic resonance imaging ( MRI ) studies on children with ASD discovered increased white matter and reduced gray matter volumes in the cerebellum ( Courchesne et al . , 2001; Bloss and Courchesne , 2007 ) . In addition , T2-weighted hyperintensities in the cerebellum are one of the most consistent brain abnormalities observed in individuals with NF1 ( Feldmann et al . , 2010; Payne et al . , 2014 ) . These findings suggest that individuals with NF1 , particularly those with co-morbidity of cognitive , motor and social deficits , might have developmental abnormalities of the cerebellum . During development , major cerebellar cell populations are derived from two germinal zones . Radial glial ( RG ) stem cells in the ventricular zone of the fourth ventricle ( IV–VZ ) give rise to all the GABAergic neuronal lineages—Purkinje cells and GABAergic interneurons including small deep cerebellar nuclei ( DCN ) , Golgi , basket , and stellate cells ( Sillitoe and Joyner , 2007; Buffo and Rossi , 2013 ) . Cerebellar astrocytes including Bergmann glia ( BG ) are also derived from RG cells in the IV-VZ . A secondary germinal zone in the anterior portion of the rhombic lip ( RL ) generates glutamatergic neuronal lineages , including large DCN , unipolar brush cells ( UBCs ) , and granule cells ( GCs ) ( Englund et al . , 2006; Sillitoe and Joyner , 2007; Mugnaini et al . , 2011; Buffo and Rossi , 2013 ) . At the onset of mouse cerebellar neurogenesis , projection neurons ( Purkinje cells and DCN ) are first generated between embryonic day 10 . 5–11 . 5 ( E10 . 5–11 . 5 ) ( Hatten , 1999 ) . At E12 . 5 , Math1 ( also known as Atoh1 ) -expressing proliferating precursors migrate out of the RL and spread across the surface of the cerebellum to form the external granule layer ( EGL ) ( Hatten , 1999 ) . Around postnatal day 4–5 ( P4–5 ) , GC precursors ( GCPs ) in the EGL start to exit cell cycle and form a pre-migratory zone along the edge of the EGL . Post-mitotic cells then migrate along the BG scaffold , across the molecular layer ( ML ) , and form the internal granule layer ( IGL ) , where they become mature GCs and form synapses with Purkinje cells through mossy fibers ( Hatten , 1999; Sillitoe and Joyner , 2007 ) . Besides GCs , RL-derived Math1+ precursors also generate the second type of glutamatergic excitatory interneurons , UBCs . UBCs exit the RL at perinatal stages and migrate through the white matter ( WM ) into the IGL of the cerebellum ( Englund et al . , 2005; Mugnaini et al . , 2011 ) . The molecular and cellular mechanisms regulating migration and lineage specification of GC/UBC precursors remain largely unknown . It has been shown that the components of RAS/ERK and PI3K signaling pathways play diverse and critical roles in different cell types and at distinct stages during cerebellar development ( Yue et al . , 2005; Fogarty et al . , 2007; Lin et al . , 2009; Yu et al . , 2011; Müller Smith et al . , 2012; Li et al . , 2014; Meier et al . , 2014 ) . However , recent studies have shown that Nf1 heterozygous mice exhibit no defect in cerebellum-specific motor tests and that children with NF1 do not appear to have a general dysfunction of the cerebellum ( Krab et al . , 2011; van der Vaart et al . , 2011 ) . These observations suggest that mono-allelic NF1/Nf1 inactivation is not sufficient to cause developmental and functional deficits in the cerebellum . Thus , we hypothesize that bi-allelic Nf1 inactivation in neural stem and progenitor cells during development is required for producing cellular pathologies in the cerebellum . In this study , we utilize four genetically engineered mouse ( GEM ) models that target bi-allelic Nf1 inactivation into different cell lineages or in the same lineage ( s ) at different stages during cerebellar development . We identify a specific time window of susceptibility of glial and neuronal precursors to Nf1 mutation during early , but not late stages of cerebellar development . Accordingly , we develop a transient therapeutic strategy during a neonatal window , which prevents cerebellar developmental defects and provides long-term benefits on motor functions . We first utilized a Cre transgenic strain controlled by human glial fibrillary acidic protein promoter ( hGFAP-cre ) to inactivate Nf1 in RG stem cells in the IV-VZ around E12 . 5 ( Figure 1—figure supplement 1A , A′ ) . RG cells targeted by hGFAP-cre in the E12 . 5 IV-VZ give rise to glutamatergic neuronal precursors populating the RL and astrocyte precursors producing astrocytes in cerebellar parenchyma ( Zhu et al . , 2005; Sillitoe and Joyner , 2007; Yang et al . , 2008 ) . Of note , the majority of oligodendrocytes in the cerebellum do not arise from RG cells in the IV-VZ , but from sources in the ventral-lateral mid/hindbrain ( that was also targeted by hGFAP-cre ) ( Yang et al . , 2008; Grimaldi et al . , 2009; Mecklenburg et al . , 2011 ) . Glutamatergic neurons ( including GCs ) and glia , but not GABAergic Purkinje cells , were targeted by hGFAP-cre-mediated recombination , revealed by β-galactosidase ( β-gal ) expression of the Rosa26-LacZ reporter and PCR analysis ( Figure 1—figure supplement 1A , B , B′ ) ( Soriano , 1999 ) . The hGFAP-cre driven Nf1 conditional knockout mice ( hereafter referred to as Nf1hGFAPCKO ) displayed severe defects in motor behaviors including abnormal hind limb crossing and vertical projection of tails . Most of the Nf1hGFAPCKO mice had ataxic hind limbs and some exhibited a unique ‘handstand’ behavior during postnatal stages ( Figure 1—figure supplement 1C; Video 1 ) . The motor dysfunction observed in neural-specific Nf1hGFAPCKO mice suggests possible defects in the cerebellum . As revealed by Hematoxylin & Eosin ( H&E ) staining , neuronal lamination in all folia of the Nf1hGFAPCKO cerebellum was disrupted with a varying degree of severity ( Figure 1A ) . In the severely affected areas ( e . g . , folia V/VI ) , the patterning of Calbindin+ Purkinje cells was disrupted , and failed to form the single Purkinje cell layer ( PCL ) observed in controls ( Figure 1B ) . Since hGFAP-cre was not targeted to Purkinje cells , the disorganized Purkinje cells still expressed neurofibromin ( Figure 1C; Figure 1—figure supplement 1B ) ( Zhu et al . , 2005; Yang et al . , 2008 ) . Thus , the disrupted patterning of Purkinje cells must be non-cell-autonomous , and secondary to Nf1 inactivation in other cerebellar neurons and/or glia . To further confirm the notion , we utilized L7-cre to inactivate Nf1 specifically in Purkinje cells ( Figure 1—figure supplement 1D ) ( Tsai et al . , 2012 ) . No overt defect was identified in the Nf1L7CKO cerebellum , confirming that Nf1 is relatively dispensable in the development of Purkinje cells ( Figure 1—figure supplement 1E , F ) . 10 . 7554/eLife . 05151 . 003Video 1 . Behavioral defects of Nf1hGFAPCKO mice before and after MEKi treatment . This video illustrates the typical phenotypes seen in vehicle-treated and untreated Nf1hGFAPCKO mice compared to controls including scruffy fur , unbalanced gait , vertical projection of the tail , hyperactivity , hind-limb crossing , and hand-standing . Following MEKi treatment , the G-responders appear similar to the controls except the vertical projection of the tail , while the P-responders are more similar to the vehicle-treated mutants . All the mice in this video are P21 except the mouse with the hand-standing phenotype , which was videotaped at 4 months of age . DOI: http://dx . doi . org/10 . 7554/eLife . 05151 . 00310 . 7554/eLife . 05151 . 004Figure 1 . Nf1 inactivation during early cerebellar development disrupts neuronal lamination and causes BG abnormalities . ( A ) Sagittal sections from control and Nf1hGFAPCKO cerebella at 2 months of age were stained with H&E and imaged at three increasing magnifications . Each cerebellar folium is numbered by Roman numerals . Arrows in ( A , B ) highlight folia V/VI where mutants exhibited extra foliation and arrows in ( B′′ ) point to the cells clustered in the ML and on the pial surface of the Nf1hGFAPCKO cerebellum . The total number of cells in the ML and IGL were quantified in ( A′ ) and ( A′′ ) . ( B ) Cerebellar sections were stained for Calbindin ( CalB ) /NeuN . Arrows point to mispositioned CalB+ Purkinje cells . ( C ) Sections from adult Nf1hGFAPCKO cerebellum were stained with β-gal/CalB and Nf1 . Purkinje cells ( arrows ) were β-gal− and expressed Nf1 . ( D ) Cerebellar sections were stained for NeuN and GABARα6 . The insets are the high magnification images of the boxed areas highlighting the co-localizing cells . The number of NeuN+ cells in the ML and IGL were quantified in ( D′ ) and the percentage of NeuN+ cells in the IGL among total NeuN+ cells ( ML + IGL ) is shown . The percentage of GABARα6+NeuN+/NeuN+ cells in the ML was quantified in ( D′′ ) . ( E ) The cell bodies ( arrows ) and processes of BG in control and mutant cerebella were labeled by BLBP staining . The total number of BLBP+ BG cells was quantified in ( E′ ) . ( F ) Sections were stained for CalB/BLBP . Boxed areas compare the Purkinje cell alignment in the less affected areas ( lower box ) and severely affected areas ( upper box ) in the Nf1hGFAPCKO cerebellum . Note the correlation between increased number/misalignment of BG and the severity of the disruption of Purkinje cell layer . ( G ) Sections were stained for CalB/GFAP . Boxed areas highlight the severely disrupted BG alignment and Purkinje cell patterning . All the quantification data are presented as mean ± SEM . DAPI labels the nuclei . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05151 . 00410 . 7554/eLife . 05151 . 005Figure 1—figure supplement 1 . Purkinje cell defects in the Nf1hGFAPCKO cerebellum are non-cell-autonomous . ( A ) The developmental lineages of the major cerebellar cell types are illustrated . Cell types expressing distinct Cre recombinases ( hGFAP , L7 , Math1 ) are highlighted by red dashed boxes . Of note , these Cre drivers except for L7-cre also target cells in the other regions of the brain . ( A′ ) The laminar structure , the alignment and marker expression of Bergmann glia , Purkinje cells and granule cells in adult cerebellum are illustrated . ( Ba-d ) X-gal staining was performed on cerebellar sections from P0 . 5 ( A , B ) and P60 ( C , D ) double-transgenic hGFAP-cre+; R26LacZR mice to determine the pattern of Cre-mediated recombination . Arrows point to Purkinje cells , which were not stained . Inset in ( D ) shows a positive control for the X-gal staining in the Purkinje cells that were targeted by Synapsin I-cre ( Zhu et al . , 2001 ) . PCZ , Purkinje cell zone; PCL , Purkinje cell layer . EGL , external granule cell layer ( B′ ) DNA extracts of the whole cerebellum and tail from hGFAP-cre+;Nf1flox/+ mice were analyzed by PCR analysis , which distinguishes Nf1 wildtype ( + ) , flox ( x1 ) and deleted ( Δ ) alleles . Note that in the cerebellum the flox allele was completely recombined and became deleted allele . ( C ) Adult Nf1hGFAPCKO mice exhibited a unique ‘handstand’ behavior . ( D–F ) Cerebellar sections from adult control and Purkinje-cell-specific Nf1L7CKO mice were analyzed . ( D ) The specificity of L7-cre-mediated recombination in Purkinje cells was confirmed by CalB and β-gal staining . ( E , F ) CalB+ Purkinje cells , BLBP+ BG cells and NeuN+ granule cells were compared between control and Nf1L7CKO cerebella and no difference was identified . DAPI labels the nuclei . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05151 . 005 In contrast to the normal cerebellum with three distinct laminar layers—ML , PCL and IGL , the boundary between these three layers in the Nf1hGFAPCKO cerebellum was less clearly defined , and the cell number in the ML was dramatically increased with a concomitant reduction of cells in the IGL ( Figure 1A–A′′; Figure 1—figure supplement 1A′ ) . Some of the abnormally accumulated cells in the Nf1hGFAPCKO ML were found on the pial surface of the cerebellum , corresponding to the EGL—a structure that transiently exists during development , but disappears after the completion of GC migration ( arrows , Figure 1Ab′′ ) . No significant difference was found in the total number of NeuN+ neurons in the ML and IGL ( per surface area ) between the control and Nf1hGFAPCKO cerebellum . Instead , the distribution of NeuN+ neurons was greatly altered between the control and Nf1hGFAPCKO cerebellar cortex—the number of NeuN+ neurons was increased over 20-fold in the mutant ML ( Figure 1D , D′ ) . While no cells in the control ML expressed GABA receptor alpha 6 subunit ( GABARα6 ) , a marker for mature GCs , 60% of abnormally accumulated neurons in the Nf1hGFAPCKO ML were positive for GABARα6 ( Figure 1D , D′′ ) . These observations are most consistent with a model wherein a subset of Nf1-deficient GCs fail to migrate inward from the EGL , across ML and PCL , to the IGL . The arrested migration of GCs leads to abnormal neuronal patterning and lamination of the Nf1hGFAPCKO cerebellum . We next investigated the integrity of the BG scaffold—an essential glial structure for GC migration during development ( Sillitoe and Joyner , 2007 ) . In the control cerebellum , BG cells were located either immediately below or interspersed between Purkinje cell bodies , forming a Purkinje cell and BG ( PC/BG ) monolayer ( arrows , Figure 1E , F ) ( Buffo and Rossi , 2013 ) . In contrast , Nf1-deficient BG cells failed to form a PC/BG monolayer , which appeared to be caused by excess number and mispositioning of BG cells ( arrows , Figure 1E , E′ , F ) . A strong correlation was observed between the severity of neuronal patterning defects and the abnormalities in the number and position of BG cells and the integrity of the PC/BG monolayer ( Figure 1F ) . Importantly , the capacity of Nf1-deficient BG cells to extend their processes to the pial surface was relatively uncompromised—this was even true for some of the BG cells that were located in the areas outside the PC/BG layer in the Nf1hGFAPCKO cerebellum ( Figure 1G ) . Based on these results , we propose a model wherein inactivation of Nf1 during early embryonic stages overproduces BG cells , thus impairing the formation of the PC/BG monolayer , and consequently , the alignment of the BG scaffold . The defective BG scaffold non-cell-autonomously arrests GC migration and subsequently disrupts Purkinje cell patterning during cerebellar development . We investigated the mechanism by which Nf1 inactivation during early cerebellar development overproduced BG cells in the Nf1hGFAPCKO cerebellum . No defect was observed in the overall structure and the distribution of RL-derived Math1+ neuronal precursors in the EGL , Purkinje cells or BLBP+ cells in the E17 . 5 Nf1hGFAPCKO cerebellum ( Figure 2—figure supplement 1A ) . Furthermore , no difference was identified in the number of BLBP+ RG cells ( including proliferating BLBP+Ki67+ population ) in the IV-VZ ( Figure 2—figure supplement 1B ) . However , an increased number of both total BLBP+ and proliferating BLBP+Ki67+ cells was observed in the prospective white matter ( PWM ) of the E17 . 5 Nf1hGFAPCKO cerebellum compared to controls ( Figure 2A , A′ ) . More importantly , mitotic index , determined by the ratio of proliferating BLBP+Ki67+ cells to the total BLBP+ cells , was significantly increased in the E17 . 5 Nf1hGFAPCKO PWM ( Figure 2A , A′ ) . These observations suggest that Nf1 specifically constrains the proliferation of astrocyte precursors in the PWM , but not multipotent RG cells of the IV-VZ in the E17 . 5 cerebellum . At P0 . 5 , both Purkinje cells and BLBP+ BG cells ( except for those populating future folia IX and X ) were situated into a more defined Purkinje cell plate ( or rudimentary PC/BG layer ) ; BG cells , particularly those in the anterior cerebellum , underwent differentiation extending their GFAP+BLBP+ fibers to the pial surface ( Figure 2—figure supplement 1C ) . Unexpectedly , the increased mitotic index was no longer detected in the more differentiated BLBP+ BG cells in the anterior folia , but only in the posterior folia ( IX and X ) of the Nf1hGFAPCKO cerebellum where BG cells had not formed a rudimentary PC/BG layer at P0 . 5 ( Figure 2B , B′ , 2C , C′; Figure 2—figure supplement 1D ) . The differentiation of BG cells was accompanied by the transition of ( 1 ) the anatomical location from the PWM to rudimentary PC/BG layer and ( 2 ) the subcellular expression of BLBP from nuclei to cytoplasm . Specifically , BLBP staining in the anterior folia of the P0 . 5 control cerebellum became largely cytoplasmic , while a robust nuclear BLBP staining was typically found in less differentiated BLBP+ cells in both the E17 . 5 control and Nf1hGFAPCKO cerebellum ( arrows , Figure 2A ) . Consistently , an increase of cells with strong nuclear BLBP staining in the P0 . 5 Nf1hGFAPCKO cerebellum was identified , suggesting a persistent increase of proliferating BG cells into postnatal stages ( Figure 2D ) . Similarly , BLBP+ cells in the PWM , the precursors of WM astrocytes , still exhibited an increased mitotic index in the P0 . 5 Nf1hGFAPCKO cerebellum ( Figure 2—figure supplement 1D–D′′ ) . At P8 when BG cells had formed a rudimentary PC/BG layer throughout the cerebellum , the mitotic index of BLBP+ BG cells was no longer increased in the Nf1hGFAPCKO cerebellum , despite a significant increase of both total BLBP+ and proliferating BLBP+Ki67+ BG cells ( Figure 2—figure supplement 1E , E′ ) . Together , these results suggest that Nf1 constrains the proliferation of the BG/astrocyte lineage cells specifically at an intermediate precursor stage located in the PWM , but not at an earlier RG stem cell stage or a more differentiated BG stage , thereby identifying a window of susceptibility of astrocyte precursors to Nf1 mutation . 10 . 7554/eLife . 05151 . 006Figure 2 . Nf1 inactivation leads to increased proliferation of intermediate glial precursors and subsequent glial/neuronal defects during perinatal development . ( A ) Sections of control and Nf1hGFAPCKO cerebella were stained for BLBP and Ki67 . Boxed areas in ( A ) are shown in high magnification on the right . Dashed lines delineate the boundary between the prospective white matter ( PWM ) and cerebellar cortex . IV , fourth ventricle . The number and mitotic index of BLBP+ cells were quantified in ( A′ ) . ( B , C ) Cerebellar sections were stained for BLBP/Ki67 and imaged at anterior fissure ( B ) and posterior fissure ( C ) . Dashed lines delineate the boundary of the PWM . The mitotic index of BLBP+ cells in respective fissures was quantified in ( B′ ) and ( C′ ) . ( D ) Sections were stained for GFAP and BLBP . BLBP in the mutant cerebellum exhibit strong nuclear staining ( arrows ) . ( E ) Sections were stained for Cre/GFAP ( A–B′ ) and APC/BLBP ( C , D ) . White bars mark the rudimentary PC/BGL in the control and the Nf1hGFAPCKO cerebellum . The arrow points to an abnormal BG cell located close to the EGL in the mutant cerebellum . ( C , D ) Arrowheads point to BG cell bodies . ( F ) Sections were stained for CalB/GFAP . Arrowheads point to BG cell bodies . ( G ) Sections were stained by H&E . Lower panels are the high magnification view of folia V/VI . Arrows point to folia V/VI ( upper panels ) where the mutants exhibit the extra foliation . In lower panels , arrows point to the cell clusters in the mutant ML . The total cell number in the ML was quantified in ( G′ ) . ( H ) Sections were stained for NeuN . The total number of NeuN+ cells in the ML was quantified in ( H′ ) . All the quantification data are presented as mean ± SEM . DAPI labels the nuclei . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05151 . 00610 . 7554/eLife . 05151 . 007Figure 2—figure supplement 1 . Nf1 regulates cerebellar astrocyte proliferation during an immediate precursor stage . ( A ) Cerebellar sections from E17 . 5 control and Nf1hGFAPCKO cerebella were stained for H&E , Math1 ( in situ hybridization ) , CalB and BLBP ( immunohistochemistry ) . ( B ) The total number and mitotic index of BLBP+ cells in the IV-VZ ( shown in Figure 2A ) were quantified . ( C–D′′ ) Cerebellar sections from P0 . 5 control and Nf1hGFAPCKO cerebella were stained for H&E , Calbindin , GFAP , BLBP and BLBP/Ki67 and imaged at low magnification . Dashed boxes in ( D ) represent the anterior fissure , PWM and posterior fissure , which were compared at high magnification in ( D′ ) and Figure 2B , C . The total number and mitotic index of BLBP+ cells was quantified in ( D′′ ) . ( E–H′′ ) Control and Nf1hGFAPCKO cerebella were analyzed at P8 . ( E ) Sections were stained for BLBP and Ki67 . Arrows point to representative co-localizing cells . The total number and mitotic index of BLBP+ cells were quantified in ( E′ ) . ( F ) Sections were stained for GFAP and CalB . Arrows point to Bergmann glial nuclei . ( G ) Sections were stained for Math1 and Zic1 by in situ hybridization . Zic1 is a marker for the granule neuron lineage in the postnatal cerebellum . ( H ) Control and Nf1hGFAPCKO mice were analyzed 2h after BrdU pulse at P8 . Sections were stained for BrdU . Lower panels are the high magnification view of folia V and VI . Red arrows ( upper panels ) point to folia V/VI where the mutants exhibit the extra foliation . In lower panels , arrows point to the proliferating cells in the BG layer . ( H′ ) High magnification view of BrdU staining in the EGL shows the focal reduction of BrdU+ cells in the mutants . The total number and percentage of BrdU+ cell number in the EGL were quantified in ( H′′ ) . All the quantification data are presented as mean ± SEM . DAPI labels the nuclei . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05151 . 007 At P8 , control BG cells labeled by nuclear Cre staining formed a rudimentary 2 to 3 cell-layer lining within the APC+ Purkinje cell layer ( Figure 2E ) . In contrast , excess BG cells failed to form a similar PC/BG layer , but were aberrantly distributed throughout the Nf1hGFAPCKO cerebellar cortex ( Figure 2E , F ) . Consequently , some Nf1-deficient BG cells extended their processes from ectopic locations of the cerebellum including the WM , ML and IGL , accompanied by misalignment of the BG scaffold ( Figure 2E , F; Figure 2—figure supplement 1F ) . Similar to the Nf1hGFAPCKO cerebellum in adulthood , disrupted patterning of GCs and Purkinje cells often co-existed with the misalignment of BG cells at P8 ( Figure 2F; Figure 2—figure supplement 1F ) . Indeed , all major defects observed in BG cells , GCs and Purkinje cells of the adult Nf1hGFAPCKO cerebellum were identified at P8 , suggesting that these defects manifest during neonatal stages ( Figure 2G–H′; Figure 2—figure supplement 1F–H′′ ) . These results supports a model wherein overproduction of astrocyte precursors during embryonic stages is the primary defect causing both glial and neuronal abnormalities in the Nf1hGFAPCKO cerebellum . To rule out a role of Nf1 in late stages of glial development , we sought to more specifically inactivate Nf1 in glial cell lineages during late embryonic stages by employing an inducible system . This inducible system utilized a Nestin-creER ( NCreER ) transgene , which can be activated in Nestin-expressing neural stem and progenitor cells by tamoxifen ( TM ) treatment ( Figure 3—figure supplement 1A ) ( Burns et al . , 2007; Wang et al . , 2012 ) . Upon a single TM pulse at E17 . 5 , NCreER-targeted cells analyzed at P21 were restricted to glial lineage cells in both the control and Nf1NCreERCKO cerebellum with only one exception in folium X , where a significant number of both glia and GCs in the IGL were targeted ( Figure 3A , B; Figure 3—figure supplement 1B , C ) . Given the sequential differentiation from the anterior to posterior cerebellum , this targeting strategy would inactivate Nf1 in more differentiated glial cells in the anterior folia compared to those in the posterior folia . No alteration in the number of BLBP+ cells or the BG scaffold was observed in any of the anterior folia ( I to V ) between the control and Nf1NCreERCKO cerebellum ( Figure 3A ) . Furthermore , neurons did not abnormally accumulate in the ML of the anterior Nf1NCreERCKO folia ( Figure 3A , A′; Figure 3—figure supplement 1D , D′ ) . In folium X , the number of Nf1-deficient BLBP+ cells ( BLBP+β-gal+ ) was significantly increased , though this was subtle compared to the Nf1hGFAPCKO cerebellum ( Figure 3B , C ) . Indeed , the other posterior folia ( e . g . , VI–IX ) exhibited a slightly , but significantly increased number of Nf1-deficient BLBP+ cells ( Figure 3—figure supplement 1C , C′ ) . The overproduction of BLBP+β-gal+ BG cells in the posterior folia of the Nf1NCreERCKO cerebellum appeared to be caused by cell-autonomous Nf1 inactivation , as the number of non-NCreER-targeted BLBP+β-gal- cells in the same areas was not altered ( Figure 3B , C ) . Importantly , the overproduction of BLBP+ BG cells was accompanied by focal misalignment of the BG scaffold and abnormal accumulation of both Nf1-deficient ( NeuN+β-gal+ ) and Nf1-intact ( NeuN+β-gal− ) neurons in the ML of the Nf1NCreERCKO cerebellum ( Figure 3B , D; Figure 3—figure supplement 1E ) . These results suggest that overproduction of this particular astrocytic lineage causes impaired formation of the PC/BG monolayer and BG scaffold , non-cell-autonomously arresting GC migration . 10 . 7554/eLife . 05151 . 008Figure 3 . Nf1 is not required for BG function in late developmental stages . Control and Nf1NcreERCKO mice were tamoxifen ( TM ) -induced at E17 . 5 and analyzed at P21 . Adjacent sections were triple-stained for BLBP/β-gal/NeuN or GFAP/β-gal/NeuN staining , and were imaged at folium V ( A ) and folium X ( B ) . NeuN staining ( * , Alexa 647 ) was artificially converted to green . The insets in ( B ) highlight BLBP+β-gal+ cells and NeuN+β-gal+ cells in the boxed areas . The number of BLBP+ cells ( A′ , C ) and NeuN+ cells ( D ) was quantified and their co-localization with β-gal was compared . Note that in folia V , none of the NeuN+ cells co-localized with β-gal . Arrows point to representative co-localizing cells , and arrowheads label non-co-localizing cells . All the quantification data are presented as mean ± SEM . DAPI labels the nuclei . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05151 . 00810 . 7554/eLife . 05151 . 009Figure 3—figure supplement 1 . Inactivation of Nf1 in Nestin-expressing progenitor cells at E17 . 5 induces overproduction of glia cells in the posterior cerebellar folia . ( A ) Control and Nf1NcreERCKO mice were tamoxifen ( TM ) -induced at E17 . 5 and analyzed at P21 ( see Materials and Methods ) . ( B ) X-gal staining shows the distribution of recombined cells in the TM-induced control and mutant cerebella . ( C ) The E17 . 5 TM-induced control and Nf1NcreERCKO cerebella were triple-labeled by BLBP , β-gal and NeuN . In Nf1NcreERCKO cerebellum , white colored Roman numerals ( I–V ) represent folia that exhibited minimal phenotype , yellow ones exhibited mild phenotype ( VI–IX ) and the red one ( X ) had the most severe defects . ( C′ ) High magnification images at folia VIII/IX are shown on the right . ( D , E ) Low magnification images of GFAP/β-gal/NeuN staining are shown to compare the overall glial/neuronal phenotypes in folium V ( D ) and folium X ( E ) , respectively . Red arrows point to the white matter of folium X . White arrows point to representative co-localizing cells , and arrowheads label non-co-localizing cells . ( D′ ) The number of NeuN+ cells in the ML of folium V was quantified . DAPI labels the nuclei . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05151 . 009 To determine whether Nf1 also plays a cell-autonomous role in the glutamatergic neuronal lineages , we utilized Math1-cre , which specifically targets Math1+ neuronal precursors in the developing RL that give rise to both GCs and UBCs ( Englund et al . , 2006; Schuller et al . , 2008 ) . Math1-cre-mediated recombination was almost exclusively restricted to the cells in the IGL of the adult cerebellum ( Figure 4A ) . Double staining of β-gal and other lineage markers showed that β-gal+ cells were restricted to GCs in the IGL , but not Purkinje cells or glial cells , including BG cells ( Figure 4B , B′ ) . No defect in Purkinje cells or glial cells , particularly the BG scaffold , was observed in the adult Nf1Math1CKO cerebellum ( Figure 4B , B′ ) . Consequently , unlike the severely disrupted Nf1hGFAPCKO cerebellum , the Nf1Math1CKO cerebellum was relatively unimpaired ( Figure 4—figure supplement 1A ) . However , the Nf1Math1CKO cerebellum did exhibit one major defect observed in the Nf1hGFAPCKO cerebellum—abnormal accumulation of cells in the ML including some on the pial surface ( arrows , Figure 4A , B–B′′; Figure 4—figure supplement 1A , A′ ) . The number of NeuN+β-gal+ , but not NeuN+β-gal− neurons , was significantly increased , suggesting that the increased neuronal density in the ML of the Nf1Math1CKO cerebellum was caused by abnormal accumulation of Nf1-deficient NeuN+ neurons in a cell-autonomous manner ( Figure 4A , B , B′′ ) . Despite that no defect in the BG scaffold was observed , abnormal accumulation of neurons was evident in the ML of the P8 Nf1Math1CKO cerebellum , suggesting a developmental defect during neonatal stages ( Figure 4—figure supplement 1B ) . To investigate the mechanism underlying the arrested migration of Nf1-deficient GCs , we performed BrdU pulse-chase experiments to determine the time course of the differentiation and migration of newly generated GCs during neonatal stages . We analyzed P8 control and Nf1Math1CKO cerebella 24 , 32 and 48 hr after BrdU pulse by determining the location and differentiation status of BrdU-labeled cells . 24 hr after pulse , BrdU+ cells in both the control and Nf1Math1CKO cerebellum remained exclusively in the EGL and exhibited similar cell cycle profiles , suggesting that Nf1 is not required for proliferation and cell-cycle exit of GC precursors ( GCPs ) ( Figure 4—figure supplement 1C–C′′ ) . However , 32 hr after BrdU pulse , a subset of BrdU+ cells in the control cerebellum started departing from the EGL and entered the ML , not yet reaching the IGL ( Figure 4C ) . While migrating GCs in the control ML did not express MAP2 , a marker for mature GCs in the IGL , and had nuclei with a spindle-shaped migrating morphology , many BrdU+ cells in the Nf1Math1CKO ML already acquired MAP2 expression with rounded nuclei—typical neuronal differentiation features for mature GCs in the IGL ( arrows , Figure 4C , C′ ) . After 48-hr BrdU pulse , BrdU+MAP2+ cells in the Nf1Math1CKO cerebellum progressively accumulated in the ML and EGL whereas control BrdU+ cells did not express MAP2 until they reached the IGL ( Figure 4D; Figure 4—figure supplement 1D , D′ ) . Newly generated neurons were abnormally accumulated in the ML/EGL of the Nf1Math1CKO cerebellum , leading to an increased number of differentiated BrdU+Ki67− and p27+ cells in the ML/EGL with a concomitant reduction in the IGL ( Figure 4—figure supplement 1E , E′ , F , F′ ) . These observations suggest that a subset of Nf1-deficient GCPs undergo accelerated and premature neuronal differentiation , thereby arresting their migration in the EGL and ML . This conclusion was further supported by in vitro studies . 2 days after being plated in culture , significantly more Nf1-deficient GCPs expressed MAP2 than control cells , confirming an accelerated neuronal differentiation rate in a cell-autonomous manner ( Figure 4—figure supplement 1G , G′ ) . Importantly , Nf1-deficient GCs could efficiently interact with wild-type glia in a co-culture system and migrated normally in a trans-well migration assay ( Figure 4—figure supplement 1H , H′ ) . Together , both in vivo and in vitro assays argue that instead of causing defects directly on migration , Nf1 inactivation promotes premature neuronal differentiation of a subset of GCPs , leading to abnormal accumulation of differentiated neurons in the ML and EGL . 10 . 7554/eLife . 05151 . 010Figure 4 . Neuron-specific Nf1 inactivation leads to cell-autonomous defects in the glutamatergic interneuron lineages . ( A ) X-gal staining on P21 control and Nf1Math1CKO cerebella labeled cells that underwent Math1-cre-mediated recombination . Arrows point to the abnormally accumulated cells in the ML of the mutant cerebellum . ( B , B′ ) Cerebellar sections were stained for NeuN/β-gal and Cal/β-gal , and GFAP/β-gal and BLBP/β-gal . The total number of NeuN+/β-gal+ and NeuN+/β-gal− cell per high magnification field was quantified in ( B′′ ) . ( C–D ) Control and Nf1Math1CKO mice were analyzed 32 hr and 48 hr after BrdU-pulse at P8 . Cerebellar sections were stained for BrdU/MAP2 and imaged at folium V . High magnification images comparing the nuclear morphology of BrdU+ cells are shown in ( C′ ) and ( D′ ) . ( E , E′ ) Cerebella sections from P8 control and Nf1Math1CKO mice were stained for Tbr2/MAP2 ( E ) and Tbr2/NeuN ( E′ ) and imaged at folium V . The inset in ( E ) shows an example of Tbr2+MAP2+ cells in the mutant ML . ( F , G ) 24 hr and 48 hr after BrdU-pulse , cerebellar sections from P8 control and Nf1Math1CKO mice were stained for Tbr2/BrdU . The distribution of Tbr2+BrdU+ cells ( arrows ) in the EGL-ML-IGL was quantified 24 hr and 48 hr after BrdU-pulse in ( F′ ) and ( G′ ) , respectively . ( H ) Golgi staining was performed on P8 cerebellar sections . Cells with UBC morphology are highlighted by red arrows and shown in high magnification ( lower panels ) . Dashed red lines delineate the ML . PC , Purkinje cell; GC , granule cell; bt , the brush tip of UBCs . ( I ) Cerebellar sections from adult control and Nf1Math1CKO cerebella were stained for Tbr2/NeuN and imaged at folium V . ( I′ ) The number of Tbr2+NeuN+ ( arrows ) and Tbr2+NeuN− cells ( arrowheads ) in the ML of folium V per high magnification image was quantified and compared with Nf1hGFAPCKO mice . The percentage of Tbr2+NeuN+ cells among total NeuN+ cells was also quantified . ( J ) A proposed model summarizes the role of Nf1 in preventing the ectopic differentiation of GCPs into UBCs . As a comparison , the timeline of normal UBC genesis is shown in ( J′ ) . All the quantification data are presented as mean ± SEM . DAPI labels the nuclei . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05151 . 01010 . 7554/eLife . 05151 . 011Figure 4—figure supplement 1 . Neuron-specific Nf1 inactivation does not impact on the proliferation or migration of GCPs . ( A ) Cerebellar sections from adult control and Nf1Math1CKO cerebella were stained by H&E . Lower panels are the high magnification view of folia V/VI in upper panels . ( A′ ) The total number of cells in the ML of folia V/VI was quantified and compared with Nf1hGFAPCKO mice . ( B ) P8 cerebellar sections were stained for BLBP/NeuN and imaged at folium V . Dashed lines delineate the ML . ( C ) Sections of P8 control and Nf1Math1CKO cerebella were analyzed and stained for BrdU/Ki67 24 hr after BrdU pulse . ( C′ ) The mitotic index of EGL precursors was quantified as the percentage of Ki67+ cells of total DAPI+ cells in the EGL . ( C′′ ) The cell cycle exit index ( fraction of cells exiting cell cycle ) was quantified as the ratio of BrdU+/Ki67− cells ( cells exiting cell cycle ) to BrdU+/Ki67+ cells ( cells remaining or reentering cell cycle ) in the EGL . ( D–E ) P8 control and Nf1Math1CKO mice were analyzed 48 hr after BrdU pulse . ( D ) Cerebellar sections were stained for BrdU/MAP2 . The number of MAP2+BrdU+ cell in the ML was quantified in ( D′ ) . ( E ) Sections were stained for BrdU/Ki67 . The number of BrdU+/Ki67− cells ( newly differentiated cells ) in the inner EGL/ML and IGL was quantified in ( E′ ) . oEGL , outer EGL; iEGL , inner EGL . ( F ) Cerebellar sections from P8 control and Nf1Math1CKO mice were stained for p27 and the number of p27+ cells in the inner EGL was quantified in ( F′ ) . ( G ) GCPs were isolated from control and Nf1Math1CKO cerebella at P8 and cultured in differentiation media for 2 days . Cells were then stained for MAP2 and quantified for the percentage of cells expressing MAP2 among total DAPI+ cells ( G′ ) . ( H ) Purified P6 granule cells from control and Nf1Math1CKO cerebella were cultured with wildtype cerebellar glia at a ratio of 1:2 . Cells were then stained for Tuj1 and GFAP or subjected to Transwell migration assay . The number of granule cells that migrated through the membrane was quantified , normalized by the membrane area ( per mm2 ) and compared ( H′ ) . All the quantification data are presented as mean ± SEM . DAPI labels the nuclei . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05151 . 011 Strikingly , many of the prematurely differentiated neurons in the ML of the P8 Nf1Math1CKO cerebellum expressed Tbr2 , a marker specific for UBCs in the cerebellum ( Figure 4E , E′ ) . UBCs are the other known RL-derived glutamatergic interneurons that are enriched in folia responsible for vestibular motor function ( folia IX and X ) ( Mugnaini et al . , 2011 ) . While GCs arise from GCPs in the EGL and migrate across ML to IGL between P4–P21 , Tbr2+ UBCs exit the RL during late embryonic stages , and then migrate through the PWM into the IGL until P10 ( Englund et al . , 2006; Sillitoe and Joyner , 2007; Mugnaini et al . , 2011 ) . While Tbr2+ UBCs were rarely identified in the control ML ( Englund et al . , 2006 ) , Tbr2+ cells including the newly generated Tbr2+BrdU+ cells were readily identified in the P8 Nf1Math1CKO EGL and ML ( Figure 4E , E′ , F , F′ ) . Tbr2 expression was ectopically acquired in a subset of newly generated cells even before migrating out of the EGL , and over time these Tbr2+BrdU+ double-positive cells progressively accumulated from the EGL , ML to IGL ( Figure 4F , F′ , G , G′ ) . These observations demonstrate that Nf1 inactivation ectopically generates Tbr2+ UBCs from GCPs in the EGL , which appear to migrate along the same pathway as GCs . More importantly , cells with the classic morphology of UBCs were identified in the ML of the P8 Nf1Math1CKO cerebellum , but not in the control cerebellum , indicating that these Tbr2+ cells not only acquire UBC-specific Tbr2 expression , but also undergo morphological differentiation of UBCs ( Figure 4H ) ( Mugnaini and Floris , 1994 ) . Consequently , approximately 30–40% of abnormally accumulated NeuN+ neurons in the ML of the adult Nf1Math1CKO cerebellum expressed Tbr2 ( Figure 4I , I′ ) . Together , these observations identify cell-autonomous Nf1 functions in suppressing GCPs to ectopically differentiate into UBCs ( Figure 4J , J′ ) . Ectopic generation of Tbr2+ UBCs from Nf1-deficient GCPs prompted us to investigate whether GABARα6-negative neurons abnormally accumulated in the ML of the Nf1hGFAPCKO cerebellum were UBCs ( Figure 1D , D′′ ) . Indeed , a dramatic increase of Tbr2+ cells was also observed in the ML throughout the adult Nf1hGFAPCKO cerebellum , which was most evident in folia V/VI ( Figure 5A–A′′ ) . The increase of Tbr2+ cells in the ML ( 98% ) and IGL ( 94% ) of the Nf1hGFAPCKO cerebellum was almost entirely contributed by UBCs that expressed a moderate level of Tbr2 as well as NeuN ( Tbr2+NeuN+ ) ( arrows , Figure 5B , B′ ) . Almost all UBCs in the control folium V expressed high levels of Tbr2 , but no NeuN ( Tbr2highNeuN− ) and were mainly distributed in the IGL ( arrowheads , Figure 5B , B′ ) . It should be noted that the number of Tbr2highNeuN− UBCs was not significantly different between the control and the Nf1hGFAPCKO cerebellum . The distinction between Tbr2+NeuN+ and Tbr2highNeuN− UBC populations was further supported by the different percentage of these two types of UBCs expressing a high level of Calretinin ( CRhigh ) . CRhigh is a marker for a subset of UBCs due to its low expression in granule cells ( Mugnaini et al . , 2011 ) . Tbr2+NeuN+ UBCs in the ML and IGL of the Nf1hGFAPCKO cerebellum expressed CRhigh at a significantly lower percentage ( 36% and 38% ) than Tbr2highNeuN− UBCs ( 67% ) typically observed in the IGL ( Figure 5—figure supplement 1A , A′ ) . Importantly , the absolute number of Tbr2+NeuN+ UBCs was similar in the ML of folium V between the Nf1Math1CKO and Nf1hGFAPCKO cerebellum , further supporting the notion that ectopic generation of these UBCs results from loss of cell-autonomous Nf1 function in the glutamatergic neuronal lineages ( Figure 4I′ ) . Similar to the Nf1Math1CKO cerebellum , ectopic Tbr2+NeuN+ UBCs were continuously generated in the ML and IGL of the Nf1hGFAPCKO cerebellum during postnatal stages ( arrows , Figure 5—figure supplement 1B , B′ , C , C′ ) , well beyond the developmental window of the generation of control Tbr2highNeuN− UBCs ( Figure 4J , J′ ) . Together , these observations demonstrate that the Tbr2+ UBCs ectopically generated in both the Nf1Math1CKO and Nf1hGFAPCKO cerebellum are different from those previously described in the normal cerebellum in 6 critical aspects: ( 1 ) the marker expression , Tbr2+NeuN+ vs Tbr2highNeuN− , ( 2 ) the low vs high percentage expressing CRhigh , ( 3 ) the location , ML and IGL vs IGL , ( 4 ) the developmental origin , precursors in the EGL vs RL , ( 5 ) the timing of birth , neonatal vs embryonic stages , and ( 6 ) the migratory pathway , EGL-ML-IGL vs RL-PWM-IGL . Thus , we suggest the existence of two UBC sublineages . Importantly , the majority of GABARα6-negative neurons were Tbr2+NeuN+ cells , while no significant difference was observed in GABAergic neurons between the control and Nf1hGFAPCKO cerebellum ( Figure 5—figure supplement 1D , E , E′ ) . These observations suggest that Tbr2+NeuN+ UBCs are the major contributors to the non-GCs abnormally accumulated in the ML of the Nf1hGFAPCKO cerebellum ( Figure 1D–D′′ ) . 10 . 7554/eLife . 05151 . 012Figure 5 . Glia-independent and -dependent UBC abnormalities in the Nf1hGFAPCKO cerebellum . ( A ) The overall expression pattern of Tbr2 in the adult control and Nf1hGFAPCKO cerebella was compared . High magnification images were merged to provide a high-resolution view . ( A′ ) Tbr2 staining in folium V/VI was compared . Arrows point to Tbr2+ cells in the ML . The total number of Tbr2+ cell in the ML of each folium was quantified in ( A′′ ) . ( B ) Sections were stained for Tbr2/NeuN and imaged at folium V . The number of Tbr2+NeuN+ cells and Tbr2+NeuN− cells per high magnification field was quantified in ( B′ ) . The percentage of Tbr2+NeuN+ cells among Tbr2+ cells is illustrated . ( C ) Tbr2 staining in folium X was compared . The total number of Tbr2+ cells in the ML , IGL and WM of folia X was quantified in ( C′ ) . ( D ) Sections were stained for Tbr2/NeuN and imaged at folium X . The number of Tbr2+NeuN+ cells and Tbr2+NeuN− cells per high magnification field was quantified in ( D′ ) and ( D′′ ) . The percentage of Tbr2+NeuN+ cells is illustrated . ( E ) Sections were stained for GFAP and BLBP and imaged at folium X . The total number of GFAP+ cells in the WM was quantified in ( E′ ) . ( F ) Cerebellar sections from control and Nf1Math1CKO were stained for Tbr2/NeuN and compared . ( G–G′′ ) Control and Nf1NcreERCKO mice were TM-induced at E17 . 5 and analyzed at P21 . Tbr2/β-gal staining was imaged at folium X ( G ) and high magnification views of the white matter ( G′ ) and ventral folium X ( G′′ ) were compared . Dashed lines mark the border between the WM and IGL in folium X . Of note , ectopic Tbr2+ cells in the WM of Nf1NcreERCKO were exclusively β-gal-negative ( Nf1-wildtype ) ( G′ ) . In contrast , the increase of Tbr2+ cells in the mutant ML was mostly contributed by Tbr2+β-gal+ cells ( Nf1-deficient ) ( arrows , G′′ ) . Arrows label co-localizing cells and arrowheads label non-co-localizing cells . All the quantification data are presented as mean ± SEM . DAPI labels the nuclei . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05151 . 01210 . 7554/eLife . 05151 . 013Figure 5—figure supplement 1 . The ectopic accumulation of Tbr2+ UBC in the ML of Nf1hGFAPCKO cerebella occurs during postnatal development . ( A ) Cerebellar sections were stained for Tbr2/CR and imaged at folium V . The number of Tbr2+CR+ cells and Tbr2+CR− cells per high magnification field was quantified in ( A′ ) . The percentage of Tbr2+CR+ cells is illustrated . ( B ) Sections of P8 control and Nf1hGFAPCKO cerebella were stained for Tbr2/NeuN and imaged at folium V . The number of Tbr2+ cells in the ML of folium V was quantified in ( B′ ) . ( C , C′ ) Control and Nf1hGFAPCKO mice were pulsed with BrdU at P8 and sacrificed at P18 . Cerebellar sections were stained for Tbr2/BrdU ( C ) and NeuN/BrdU ( C′ ) to label newly generated Tbr2+BrdU+ UBCs and NeuN+BrdU+ granular cells ( arrows ) ( D ) Adjacent sections from the same Nf1hGFAPCKO adult cerebellum were stained for GABAα6/NeuN and Tbr2/NeuN . * labels a blood vessel structure found in both sections , which was used as a reference to match these two sections . Boxed areas highlight a group of cells that express a significant level of GABAα6 but are mostly Tbr2− . ( E ) Sections of adult control and Nf1hGFAPCKO cerebella were stained for NeuN/GABA and the number of GABA+ GABAergic neurons in the ML were quantified in ( E′ ) . DAPI labels the nuclei . All the quantification data are presented as mean ± SEM . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05151 . 01310 . 7554/eLife . 05151 . 014Figure 5—figure supplement 2 . A subpopulation of UBCs in the Nf1hGFAPCKO cerebella fail to migrate into the IGL and are ectopically present in the WM during postnatal development . ( A ) The low magnification view of Figure 5D is shown to compare the overall distribution of Tbr2+ cells in folia IX and X . ( B ) Sections of adult control and Nf1hGFAPCKO cerebella were stained for Tbr2/CR and imaged at folium X . Dashed lines mark the border of white matter in folium X . The number of Tbr2+CR+ cells ( arrows ) and Tbr2+CR− cells ( arrowheads ) per high magnification field were quantified in ( B′ , B′′ ) . The percentage of double-labeled Tbr2+CRhigh cells among total Tbr2+ cells was illustrated . ( C , C′ ) The distribution of Tbr2+ cells in E17 . 5 ( C ) and P0 . 5 ( C′ ) control and Nf1hGFAPCKO cerebella was compared . Boxed areas in ( C′ , upper panels ) are shown at higher magnification in lower panels to indicate the absence of Tbr2+ cells in the EGL of both control and mutant cerebella . ( D ) Sections of P8 control and Nf1hGFAPCKO cerebella were stained for Tbr2/NeuN and imaged at folia IX/X . The number of Tbr2+ cells in the WM of folium X was quantified in ( D′ ) . DAPI labels the nuclei . All the quantification data are presented as mean ± SEM . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05151 . 014 As observed in the rest of mutant folia , the number of Tbr2+ cells was also significantly increased in the ML of the folium X of the Nf1hGFAPCKO cerebellum ( Figure 5C , C′; Figure 5—figure supplement 1E ) . However , the number of Tbr2+ cells was unexpectedly reduced in the IGL of folium X of the Nf1hGFAPCKO cerebellum concomitant with ectopic presence of Tbr2+ cells in the surrounding WM ( Figure 5C , C′; Figure 5—figure supplement 2A ) . As in Folium V , the Tbr2+NeuN+ UBC sublineage contributed to almost the entire UBC population in the ML of both the control ( 97% ) and Nf1hGFAPCKO ( 99% ) folium X ( Figure 5D , D′ ) . Similarly , Tbr2+NeuN+ UBCs were indeed increased in the IGL of Nf1hGFAPCKO folium X , and thus the reduction of Tbr2+ UBCs was exclusively attributed to a decreased number of Tbr2highNeuN− UBCs in the IGL ( Figure 5D , D′ ) . Consistently , UBCs in both the Nf1hGFAPCKO WM and IGL were the Tbr2highNeuN− sublineage , which was normally located in the IGL of the control cerebellum with a higher percentage expressing CRhigh ( Figure 5—figure supplement 2B–B′′ ) . More importantly , the total number of Tbr2highNeuN− UBCs was almost identical , when the number of the Tbr2highNeuN− UBCs in the WM and IGL of the Nf1hGFAPCKO cerebellum was combined to compare with the number of Tbr2highNeuN− UBCs in the IGL of the control cerebellum , suggesting arrested migration in the WM ( Figure 5D–D′′ ) . Consistently , no alteration in the number and distribution of Tbr2+ UBCs was observed at early stages of UBC development before they migrated out of the RL or entered the PWM during perinatal stages ( Figure 5—figure supplement 2C , C′ ) . Instead , abnormal accumulation of Tbr2highNeuN− UBCs in the Nf1hGFAPCKO WM was evident at P8 , a time point when normal UBCs had already migrated through the PWM and entered the IGL ( Figure 5—figure supplement 2D , D′ ) . Together , these results suggest that the abnormal accumulation of Tbr2highNeuN− UBCs in the WM of the Nf1hGFAPCKO cerebellum was caused by arrested migration of UBCs in the PWM on the way from the RL to the IGL during neonatal development ( Englund et al . , 2006 ) . Importantly , no ectopic Tbr2+ UBCs were observed in the WM of the Nf1Math1CKO cerebellum , suggesting a non-cell-autonomous mechanism ( Figure 5F ) . Consistently , abnormally increased numbers of GFAP+BLBP− astrocytes were observed in the enlarged WM of the adult Nf1hGFAPCKO cerebellum , raising the possibility that excess astrocytes in the WM could impede the migration of UBCs from the WM to the IGL ( Figure 5E , E′ ) . In the TM-induced Nf1NCreERCKO cerebellum where the number of glial cells was increased specifically in the posterior folia , a subpopulation of Tbr2high UBCs , all of which were wild-type for Nf1 ( not labeled by β-gal expression ) , ectopically accumulated in the WM of folium X ( arrowheads , Figure 5G , G′ ) . Of note , excess Tbr2+ UBCs in the ML of the TM-induced Nf1NCreERCKO cerebellum appears to result from overproduction of Nf1-deficient Tbr2+β-gal+ UBCs—a cell-autonomous mechanism as described in the Nf1Math1CKO cerebellum ( Figure 5G , G′′ ) . These results suggest that overproduction of astrocytes in the WM of the Nf1hGFAPCKO and Nf1NCreERCKO cerebella produces a physical barrier that exerts non-cell-autonomous effect , arresting migration of Tbr2highNeuN− UBCs in the WM . To develop a therapeutic strategy to rescue cerebellar defects caused by Nf1 inactivation , we analyzed three major downstream effectors of Ras-mediated signaling pathways: Erk/MAPK , PI3K/Akt and mTORC1 . Although Western blot analysis showed a significant reduction of neurofibromin expression , Erk/MAPK was the only downstream signaling pathway that was consistently hyperactivated in Nf1Math1CKO cerebella ( Figure 6A ) . Consistent with the specificity of Math1-cre transgene , cell cultures with enriched P6 GCPs showed almost a complete loss of neurofibromin expression with a robust increase of Erk activation ( Figure 6A′ ) . A high level of phosphorylated Erk ( p-Erk ) expression was observed in a subset of IGL cells , while a low level in the ML and a complete absence of p-Erk in the EGL was found in the control cerebellum at P8 ( Figure 6B ) . The pattern of p-Erk expression is consistent with a possible role of a high level of Erk signaling in neuronal differentiation . A robust increase of p-Erk expression was observed in the P8 Nf1Math1CKO cerebellum including the inner edge of the EGL ( pre-migratory zone ) , the ML , and IGL ( Figure 6B ) . Importantly , the NeuN+ neurons abnormally accumulated in the ML of the P8 Nf1Math1CKO cerebellum expressed a high level of p-Erk ( arrows , Figure 6B ) . These results demonstrate that Nf1 is a major regulator that suppresses Erk signaling in migrating GCs in the pre-migratory EGL and migratory zone ( ML ) of the developing cerebellum . To test whether hyperactive Erk signaling is responsible for Nf1-deficient GCPs to undergo premature differentiation and ectopically adopt a UBC fate , we treated newborn control and Nf1Math1CKO pups with a MEK inhibitor ( PD0325901 ) or vehicle from P0 . 5 to P21 for 3 weeks ( Sebolt-Leopold and Herrera , 2004; Wang et al . , 2012 ) . This 3-week treatment protocol exhibited no adverse effect on normal cerebellar development , but reduced p-Erk expression in both MEKi-treated control and Nf1Math1CKO cerebella ( data not shown ) . Strikingly , MEK inhibition completely eliminated ectopic Tbr2 expression in the ML of MEKi-treated Nf1Math1CKO cerebella ( Figure 6C ) . MEK inhibition consistently reduced over 60% of abnormally accumulated neurons in the ML of the Nf1Math1CKO cerebellum ( Figure 6C , C′ ) . Since an increase of p-S6 staining was also identified in purified GCPs from the P6 Nf1Math1CKO cerebellum compared to controls ( Figure 6A′ ) , we treated control and Nf1Math1CKO mice with rapamycin , an mTORC1 inhibitor ( Guertin and Sabatini , 2009 ) . Rapamycin treatment had no benefit on neuronal defects in the Nf1Math1CKO cerebellum , but caused adverse effects on neuronal migration in both the control and mutant cerebellum ( Figure 6—figure supplement 1 ) . Together , these results demonstrate that Nf1-mediated MEK/Erk inhibition is the major mechanism underlying the suppression of a subset of GCPs that otherwise would undergo premature differentiation and ectopically adopt a UBC fate in the developing cerebellum . 10 . 7554/eLife . 05151 . 015Figure 6 . Glia-independent neuronal defects in the Nf1Math1CKO cerebellum are rescued by P0 . 5–P21 MEKi-treatment . Western blot analysis was performed on P8 whole cerebellar lysates ( A ) or P6 cultured GCP lysates ( A′ ) from control and Nf1Math1CKO mice . ( B ) Cerebellar sections from P8 control and Nf1Math1CKO mice were stained for NeuN and p-Erk . Arrows highlight the NeuN+p-Erk+ cells in the mutant EGL and ML . ( C ) Control and Nf1Math1CKO mice were treated with vehicle or MEKi ( 5 mg/kg ) from P0 . 5–P21 and analyzed at P21 . Cerebellar sections were stained for Tbr2/NeuN , and imaged at low magnification ( upper panels ) and folium V ( lower panels ) . ( C′ ) The number of NeuN+ cells in the ML was quantified and compared . Of note , the yellow triangles represent 3 MEKi-treated mutant cerebella that still displayed cell clusters near the pial surface . Individual data points are presented , as well as mean ± SEM for each group . DAPI labels the nuclei . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05151 . 01510 . 7554/eLife . 05151 . 016Figure 6—figure supplement 1 . Rapamycin treatment during neonatal stages does not rescue the neuronal defects in the Nf1Math1CKO cerebellum , but causes adverse effects . ( A ) Sections of P8 control and Nf1Math1CKO cerebella were stained for p-S6 and NeuN . ( B ) Control and Nf1Math1CKO mice were treated with vehicle or Rapamycin from P0 . 5–P21 and analyzed at P21 . Arrows point to p-S6+NeuN+ cells , which are also shown in the inset . ( C ) Rapamycin-treated control and Nf1Math1CKO mice exhibited significant cerebellar defects . Sections from vehicle- or Rapamycin-treated controls were stained for NeuN and p-S6 . Arrows point to folia where there is abnormal neuronal clustering in the Rapamycin-treated cerebellum , but not in the vehicle-treated control . DAPI labels the nuclei . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05151 . 016 Based on the success of MEK inhibition on the developing Nf1Math1CKO cerebellum , we sought to investigate whether MEK inhibition could also rescue the more severely affected Nf1hGFAPCKO cerebellum . The MEKi-treated Nf1hGFAPCKO mice displayed a varying degree of therapeutic responses in glial and neuronal defects in the developing cerebellum as well as their general behaviors ( Video 1 ) . Of 14 MEKi-treated Nf1hGFAPCKO cerebella analyzed , 9 mice ( good-responders or G-responders ) exhibited a significant rescue in the number of neurons abnormally accumulated in the ML of the developing cerebellum at a level similar to what was achieved in the MEKi-treated Nf1Math1CKO mice ( green color , Figure 7A–A′′ ) . However , the remaining 5 of the MEKi-treated Nf1hGFAPCKO mice ( poor-responders or P-responders ) only exhibited marginal improvement in the neuronal defect ( yellow color , Figure 7A–A′′ ) . Importantly , G- and P-responders could be readily distinguished by the presence or absence of rescues ( to the control levels ) in the number of BG cells , alignment of the BG scaffold , the number of mispositioned Purkinje cells and arrested UBCs in the WM ( Figure 7A , B , B′ , C , C′ ) . It should be noted that both G-responders and P-responders exhibited a complete rescue of ectopic Tbr2 expression in the ML of the MEKi-treated Nf1hGFAPCKO mice ( Figure 7D , D′ ) . Therefore , the fundamental difference in the cerebella of G- and P-responders lies in the presence or absence of rescue in glia-dependent cerebellar defects , which could result from inconsistent inhibition of much higher levels of hyperactive Erk signaling in Nf1-deficient glial lineage cells than that in Nf1-deficient immature neurons in the ML ( Figure 7E; Figure 7—figure supplement 1A , B ) . Consistently , when we increased the dose of MEKi from 5 mg/kg to 20 mg/kg to lactating females , more consistent rescues were observed in the cerebella of MEKi-treated Nf1hGFAPCKO pups ( Figure 7F; Figure 7—figure supplement 2 ) . 10 . 7554/eLife . 05151 . 017Figure 7 . MEKi treatment rescues glia-dependent cerebellar defects in the Nf1hGFAPCKO cerebellum . ( A ) Cerebellar sections of vehicle- and MEKi-treated ( 5 mg/kg ) control and Nf1hGFAPCKO mice were stained for GFAP/NeuN and CalB/BLBP and imaged at folium V . 9 out 14 MEKi-treated mutant mice ( G-responder ) exhibited a significant improvement , although the remaining 5 ( P-responder ) still had large number of cells in the EGL/ML and displayed severe laminar disruption . The number of NeuN+ cells in the ML was quantified and compared in ( A′ ) . For MEKi-treated mutants , green and yellow color represents G-responders and P-responders , respectively . G-responders and P-responders were separated into two groups and compared in ( A′′ ) . ( C , D ) Cerebellar sections were stained for Tbr2/NeuN and imaged at folia V and X . ( B , B′ , C′ , D′ ) The correlation between the number of NeuN+ cells in the ML , and the number of BLBP+ BG cells ( B ) , mispositioned Purkinje cells ( B′ ) , ectopic Tbr2+ cells in the WM ( C′ ) and ML ( D′ ) was plotted and compared . ( E ) Sections of P8 control and Nf1hGFAPCKO cerebella were stained for GFAP and p-Erk . Arrows highlight the GFAP+p-Erk+ cells in the ML , IGL and WM of the Nf1hGFAPCKO cerebellum . ( F ) Three Nf1hGFAPCKO mice treated with MEK inhibitor at 20 mg/kg from P0 . 5–P21 were analyzed at P60 . GFAP/NeuN and CalB/BLBP staining show consistent rescue of both neuronal and glial defects . Individual data points are presented , as well as mean ± SEM for each group . DAPI labels the nuclei . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05151 . 01710 . 7554/eLife . 05151 . 018Figure 7—figure supplement 1 . Ras/Erk signaling is differentially activated in neuronal and glial precursors of the developing Nf1hGFAPCKO cerebellum . ( A , B ) Sections of P8 control and Nf1hGFAPCKO cerebella were stained for GFAP and p-Erk . Arrows highlight the GFAP+p-Erk+ cells in the ML , IGL and WM of the Nf1hGFAPCKO cerebellum . ( B ) is the high magnification view of ( A ) . The different p-Erk level in the neuronal vs glial precursors Nf1hGFAPCKO cerebella is more evident in the images with lower exposure in ( B ) . Boxed images are also shown in Figure 7E . DAPI labels the nuclei . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05151 . 01810 . 7554/eLife . 05151 . 019Figure 7—figure supplement 2 . High-dose MEKi-treatment produces more consistent phenotypic rescue . ( A ) Control and Nf1hGFAPCKO mice were treated with vehicle or MEKi ( 20 mg/kg ) from P0 . 5–P21 and analyzed at P60 . H&E staining of the overall cerebellar structure and folium V are presented . Low magnification view of NeuN staining , as well as high magnification merged images of NeuN/GFAP , CalB/BLBP/DAPI are compared . Arrows point to misaligned Purkinje cells in the ML or IGL . Boxed images are also shown in Figure 7F . The number of NeuN+ cells in the ML , the total number of BLBP+ cells and the number of mispositioned Purkinje cells are quantified in ( B ) , ( C ) and ( D ) , respectively . DAPI labels the nuclei . All the quantification data are presented as mean ± SEM . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05151 . 019 We next sought to determine whether the transient P0 . 5–P21 treatment protocol could exert a long-term and sustained therapeutic benefit . When analyzed at P45–P60—1 month after MEKi treatment was terminated , MEKi-treated Nf1hGFAPCKO G-responders exhibited the rescue effects on both glial and neuronal defects , while P-responders continued to have defects similar to those observed after acute treatment at P21 ( Figure 8A , A′ ) . To investigate whether the rescue of the cerebellar cellular and structural defects could improve motor functions , we performed a Rotarod test on control and Nf1hGFAPCKO mice for 4 consecutive days immediately after 3-week MEKi treatment . The MEKi-treated Nf1hGFAPCKO mice stayed significantly longer on the Rotarod compared to vehicle-treated mutant mice ( Figure 8B ) . Moreover , while vehicle-treated Nf1hGFAPCKO mice did not improve motor performance over the 4-day training course , the MEKi-treated mutant mice exhibited a significant improvement similar to controls , suggesting MEKi treatment also enhanced motor learning ( Figure 8B ) . Importantly , the improved motor performance was maintained at least 1 month after MEKi treatment was terminated , though the rescue in motor learning was lost ( Figure 8B′; Video 2 ) . G-responders stayed significantly longer on Rotarod than P-responders , providing a nice correlation between recues in cerebellar developmental defects and motor functions of MEKi-treated Nf1hGFAPCKO mice ( Figure 8C ) . Together , these results demonstrate that transient MEKi treatment during neonatal stages prevents cerebellar developmental defects and provides long-term benefits on motor functions of Nf1hGFAPCKO mice . 10 . 7554/eLife . 05151 . 020Figure 8 . Transient MEKi treatment provides long-term rescue of the cerebellar defects and improves the motor performance of Nf1hGFAPCKO mice . ( A ) Control and Nf1hGFAPCKO mice were treated with vehicle or MEK inhibitor ( MEKi ) from P0 . 5–P21 and analyzed for BLBP/NeuN staining between P50 and P60 . The number of NeuN+ cells in the ML and BLBP+ BG cells were quantified , respectively ( A′ ) . ( B , B′ ) Rotarod tests were performed on transiently vehicle- and MEKi-treated control and Nf1hGFAPCKO mice for consecutive days from P21–P24 ( B ) or P45–P48 ( B′ ) and the duration time for the mice to stay on the Rotarod was documented and compared . The correlation between the number of NeuN+ cells in the ML and duration time on the Rotarod at P48 was plotted in ( C ) . ( D ) A model is proposed to summarize cell-autonomous and non-cell-autonomous defects in Nf1hGFAPCKO cerebellum and the underlying mechanisms . See main text . ( E ) A schematic diagram illustrates the differential Erk activation in the P8 control and Nf1hGFAPCKO cerebella . In controls , p-Erk expression is mostly restricted in the IGL neurons and a subset of WM glia , while in mutants , high level of p-ERK expression is identified in migrating neurons in the ML and the majority of WM glia . Note that in Nf1hGFAPCKO cerebella , p-ERK level is much higher in glia compared to neurons . Individual data points are presented , as well as mean ± SEM for each group . DAPI labels the nuclei . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05151 . 02010 . 7554/eLife . 05151 . 021Video 2 . Rotarod testing . Representative P48 control and Nf1hGFAPCKO mice were videotaped on the final ( fourth ) day of rotarod performance testing . During the test , the vehicle-treated mutants had difficulty walking in a constant forward direction , rotated with the rotarod , and had frequent slippage with difficulty maintaining a grip on the rotarod ultimately resulting in a fall before 5 min . The MEKi-treated mice all had improved performance on the rotarod test , staying on the rotarod for the entire duration , but had a variable response to the treatment , as some of the MEKi-treated mice continued to show evidence of motor dysfunction ( e . g . , more likely to cling onto the rotarod instead of walking ) . The video includes clips from different time points and speeds during the rotarod test as a representation of the mouse performance during the task . DOI: http://dx . doi . org/10 . 7554/eLife . 05151 . 021 Astrocytes , the most abundant cells in the brain , play diverse roles in both the developing and mature brain ( Barres , 2008; Molofsky et al . , 2012; Freeman and Rowitch , 2013 ) . Compared to the other two major CNS cell types , neurons and oligodendrocytes , the identity of astrocyte precursors remains poorly understood . This is largely due to a lack of known markers that can reliably define astrocytes at distinct developmental stages ( Molofsky et al . , 2012; Freeman and Rowitch , 2013 ) . BG cells have a unique and well-defined migratory pathway coupled with differentiation that allows us to define the role of Nf1 in this particular astrocyte lineage at different developmental stages ( Buffo and Rossi , 2013 ) . Before becoming terminally differentiated in the PC/BG monolayer , BG precursors proliferate at three distinct stages of differentiation that can be distinguished by different anatomical locations during development ( Figure 8D ) . These three stages include: Stage I , RG/early-precursor stage in the IV-VZ; Stage II , BG intermediate-precursor stage in the PWM; Stage III , BG late-precursor stage in the rudimentary PC/BG layer . In the Nf1hGFAPCKO cerebellum , increased proliferation was not observed in E17 . 5 IV-VZ RG cells ( Stage I ) , or in BG late-precursors following migration into the rudimentary PC/BG layer in P0 . 5 anterior folia , or throughout P8 folia ( Stage III ) . In contrast , a significant increase of proliferation index was specifically identified in Nf1-deficient BG intermediate-precursors in the E17 . 5 PWM and P0 . 5 posterior folia ( Stage II ) , leading to an increased number of both total BLBP+ and proliferating BLBP+Ki67+ cells . These observations suggest that Nf1 plays a robust role in constraining BG proliferation specifically at an intermediate precursor stage ( Stage II ) , but not during earlier or later stages ( Stage I or Stage III ) . This conclusion is further confirmed by acute inactivation of Nf1 in the E17 . 5 developing cerebellum using an inducible Nf1NCreERCKO model . Thus , the temporal and regional phenotypic differences upon Nf1 loss observed in the BG lineage can be strictly attributed to differences in the differentiation status of this particular astrocyte lineage . Moreover , Nf1-deficient astrocytes in the developing WM also exhibited increased proliferation . Since gliogenesis occurs at different time points in different regions of the developing brain , this ‘differentiation-based’ model might provide an explanation for conflicting results on the role of Nf1 in neural stem and progenitor cells among previously published studies ( Dasgupta and Gutmann , 2005; Zhu et al . , 2005; Hegedus et al . , 2007; Becher and Holland , 2010; Lee da et al . , 2010; Lee da et al . , 2012; Wang et al . , 2012 ) . It should be noted that region-specific mechanisms underlying glial pathologies induced by Nf1 loss do exist within the dorsal brain . While increased proliferation in Nf1-deficient astrocyte precursors appears to be responsible for the increase of glial cells in the dorsal hindbrain ( e . g . , cerebellum ) , altered fate-specification of Nf1-deficient neural stem and progenitor cells in the subventricular zone ( SVZ ) of the lateral ventricle causes the increase of glial cells in the corpus callosum by promoting gliogenesis at the expense of neurogenesis in the olfactory bulb ( Wang et al . , 2012 ) . It is not entirely clear why Nf1 plays a critical role in suppressing astrocyte proliferation at a specific precursor stage . Loss of function studies in mice have shown that receptor tyrosine kinase ( RTK ) signaling , particularly fibroblast growth factor receptor ( FGFR ) signaling , is essential for the transition from RG to BG cells ( Lin et al . , 2009; Muller Smith et al . , 2012; Meier et al . , 2014 ) . Most strikingly , loss of Shp2 , also known as Ptpn11 ( Protein tyrosine phosphatase non-receptor type 11 ) , completely blocks the transition from RG to BG cells , which can be rescued by constitutively active MEK ( Li et al . , 2014 ) . These studies suggest that activation of Shp2-dependent FGF/MEK/Erk signaling is essential for the transition from RG to BG cells during cerebellar development ( from Stage I to II ) . Accordingly , our study suggests that Erk signaling must be constrained by pathway inhibitors such as Nf1 to ensure the generation of the correct number of BG cells . Consistently , the most robust Erk/MAPK activation upon Nf1 loss was observed in astrocyte precursors at Stage II and early Stage III ( Figure 8E ) . Our study provides several lines of evidence supporting the notion that the correct number of astrocytes is essential for the formation of the PC/BG monolayer and the BG scaffold , and subsequently glutamatergic interneuronal migration . First , in the Nf1hGFAPCKO cerebellum , the severity of glial and neuronal defects is not uniform , but highly correlated with the excess number of BG cells in the affected areas . Indeed , the capacity of Nf1-deficient BG cells to extend their radial fibers appeared relatively uncompromised even for those projected from the areas outside of the PC/BG layer . Second , considering the time required for TM induced , Cre-mediated Nf1 deletion at E17 . 5 and the half-life of neurofibromin protein , we expect that a complete loss of Nf1 would be fully achieved in targeted cells around P0 . 5 when BG cells in the anterior folia ( I–V ) are at Stage III , while BG cells in the posterior folia ( IX and X ) are still proliferating in the PWM at Stage II . Consistently , a significant increase of Nf1-deficient BG cells was observed in the posterior folia ( IX , X ) , but not in the anterior folia of the same Nf1NCreERCKO cerebellum . Moreover , failed formation of the PC/BG monolayer , misalignment of the BG scaffold , and abnormal accumulation of both Nf1-intact and Nf1-deficient neurons in the ML were only observed in the posterior folia of the E17 . 5 TM-induced Nf1NCreERCKO cerebellum . It should be noted that glial and neuronal defects in the posterior folia of the E17 . 5 TM-induced Nf1NCreERCKO cerebellum are significantly less severe than those observed in the Nf1hGFAPCKO cerebellum . This is likely due to a relatively greater window for the expansion of Nf1-deficient glial precursors in the Nf1hGFAPCKO cerebellum ( from E12 . 5 on ) than E17 . 5 TM-induced Nf1NCreERCKO cerebellum . Third , transient treatment of MEKi exhibited a robust long-term rescue of glial and neuronal defects in the cerebella of G-responders in which neurons and glia were still deficient for Nf1 after the treatment . Taken together , these results suggest that as long as the excess number of glia is therapeutically fixed during a specific window of cerebellar development , a long-term rescue could be achieved , as Nf1 is relatively dispensable for glial cell proliferation during late stages of development and adulthood . Our study also shows that the correct number of astrocytes in the developing WM of the cerebellum is essential for the migration of UBCs from the PWM to the IGL . This neuronal migratory defect appears to result from loss of a non-cell-autonomous Nf1 function in glial cells . This conclusion is based on the observations that ( 1 ) neuron-specific Nf1Math1CKO cerebellum exhibited no such defect in the WM , and ( 2 ) Nf1-intact UBCs were arrested in the WM of the E17 . 5 TM-induced Nf1NCreERCKO cerebellum where excess astrocytes were present . Furthermore , the migratory defect of UBCs was completely rescued in the transiently MEKi-treated cerebella of G-responders , which contained a relatively normal number of glial cells . Together , our study demonstrates that Nf1 not only plays a critical role in suppressing neoplastic transformation of glial cells , but also is essential for controlling the generation of the correct number of normal astrocytes—which is required for the migration of both types of glutamatergic interneurons ( GCs and UBCs ) during cerebellar development . Analysis of the neuron-specific Nf1Math1CKO cerebellum supports the existence of two UBC sublineages . One UBC sublineage characterized by Tbr2highNeuN− is relatively well studied , which exit cell cycles during perinatal stages and migrate from the RL via the PWM to the IGL ( Englund et al . , 2006; Mugnaini et al . , 2011 ) . Although Nf1-regulated glial number is essential for the migration of this UBC sublineage through the PWM , Nf1 is not required for the generation and migration of the Tbr2highNeuN− UBC sublineage . In contrast , the other rare Tbr2+NeuN+ UBC sublineage appears to be suppressed by Nf1-regulated Erk signaling in GCPs in a cell-autonomous mechanism . Thus , our study demonstrates that Nf1 is indirectly required for GC migration by suppressing GCPs to undergo premature terminal differentiation and ectopic differentiation to Tbr2+NeuN+ UBCs in the EGL and ML . It is still possible that Nf1 loss causes abnormal migration of the Tbr2highNeuN− sublineage from the RL to the EGL , and then switches them to the Tbr2+NeuN+ UBC sublineage in the EGL . However , no Tbr2+ cells were found in the EGL of the mutant cerebellum at E17 . 5 or P0 . 5 , when Tbr2highNeuN− UBCs migrate out of the RL into the PWM ( Figure 5—figure supplement 2C , C′ ) . This observation argues against the possibility of abnormal migration of Tbr2highNeuN− UBCs into the EGL . Indeed , Nf1 loss has little or no impact on the generation , migration and total number of Tbr2highNeuN− UBCs in the mutant cerebellum . Furthermore , in contrast to the precursors of Tbr2highNeuN− UBCs that exit cell cycles during perinatal stages , Nf1-deficient precursors of Tbr2+NeuN+ UBCs continue proliferating in the EGL at least until P8 at the peak of GCP proliferation . Together , these observations suggest that Nf1 plays a specific role in the Tbr2+NeuN+ UBC sublineage by controlling the fate-specification of GCPs in differentiating into GCs vs Tbr2+NeuN+ UBCs . The involvement of UBCs in cerebellar vestibular functions raises the possibility that the UBC defects in these Nf1 GEM models could contribute to the impairments of motor and balance observed in individuals with NF1 ( Mugnaini et al . , 2011; Champion et al . , 2014 ) . Our previous study showed that MEKi could be delivered to neonatal pups through milk from MEKi-treated lactating females ( Wang et al . , 2012 ) . This ‘MEKi in milk’ treatment protocol allows a low dose of MEKi to get into pups without introducing any physical distress . When lactating females received MEKi at 5 mg/kg ( body weight ) , a mild 30–40% of Erk inhibition was observed in the brains of Nf1hGFAPCKO pups ( MB , EJ and YZ , unpublished observations ) . Strikingly , this mild Erk inhibition robustly and consistently rescued developmental defects in the cerebella of 100% of MEKi-treated Nf1Math1CKO mice , as well as over 60% of the MEKi-treated Nf1hGFAPCKO mice ( G-responders ) . It should be noted that the remaining P-responders are not ‘non-responders’ , as these MEKi-treated Nf1hGFAPCKO mice still exhibited a complete rescue of ectopic Tbr2 expression in their cerebella , and consistent rescues in the defects in the corpus callosum and SVZ , as described previously ( Wang et al . , 2012 ) . The inconsistent rescue in the glia-dependent defects observed in the cerebella of MEKi-treated Nf1hGFAPCKO mice is likely due to the extremely high levels of Erk activation in Nf1-deficient glial cells in the developing cerebellum compared to other cells . Indeed , when the dose of MEKi was increased to 20 mg/kg , a more consistent rescue in cerebellar defects was observed in the cerebella of the MEKi-treated Nf1hGFAPCKO mice ( Figure 7F ) . More importantly , this ‘MEKi in milk’ treatment protocol provides robust long-term rescues on cerebellar developmental defects and motor performance . It should be emphasized that this transient MEKi treatment protocol also rescued structural defects in the other regions of the Nf1hGFAPCKO brain including the enlarged corpus callosum , a brain structural defect associated with severe cognitive impairments in a subset of NF1 patients ( Wang et al . , 2012 ) . Thus , motor performance improvement by this protocol is a collective reflection of rescues of multiple NF1-associated brain structural defects including those in the cerebellum and corpus callosum . Given that cognitive and motor impairments are often co-morbidities in children with NF1 ( Champion et al . , 2014 ) , our findings suggest that this transient ‘MEKi in milk’ protocol can serve as a common prevention strategy for multiple NF1-associated neurodevelopmental disorders . Three independent Cre transgenic strains under the control of hGFAP , Math1 and L7 promoters were utilized to target an Nf1 mutation into different cell lineages during cerebellar development . The control mice for Nf1hGFAPCKO mice are a pool of phenotypically indistinguishable mice: hGFAP-cre-;Nf1flox/flox , hGFAP-cre−;Nf1flox/+ and hGFAP-cre+;Nf1flox/+ . The control mice with similar genetic configurations were used for Nf1Math1CKO and Nf1L7CKO mice . All the CKO mice have genotypes of Cre+;Nf1flox/flox . Nf1hGFAPCKO , Nf1Math1CKO , Nf1L7CKO and control mice were maintained in the mixed backgrounds of C57Bl6 , 129Svj and FVB , which improved overall health and lifespan of Nf1hGFAPCKO mice . Age- and littermate-matched control and mutant mice were used for analysis . Tamoxifen inducible strain Nf1NcreERCKO is described in later sections along with the tamoxifen administration protocol . All mice in this study were cared for according to the guidelines that were approved by the Animal Care and Use Committees of the University of Michigan at Ann Arbor , MI and the Children's National Medical Center in Washington , DC . MEK inhibitor ( MEKi , PD0325901 , Sigma-Aldrich , St . Louis , MO ) was dissolved in DMSO at a concentration of 25 mg/ml and resuspended in vehicle ( 0 . 5% hydroxypropyl methyl-cellulose with 0 . 2% Tween80 , Sigma-Aldrich ) at a concentration of 1 mg/ml . The solution was administered by oral gavage at the dosage of 5 mg/kg or 20 mg/kg ( body weight ) daily to lactating females for the treatment of P0 . 5–P21 mice . MEKi-treated mice were collected and compared to littermate control mice and vehicle-treated Nf1hGFAPCKO and Nf1Math1CKO mice . Both paraffin and frozen sections were utilized for histological analysis . Control and mutant littermates at various time points were perfused with Phosphate buffered saline ( PBS ) followed by 4% paraformaldehyde ( PFA ) ( Sigma-Aldrich ) . Brains were divided into two hemispheres along the midline and each hemisphere was processed for either frozen blocks ( for in situ hybridization and X-gal staining , see below ) or paraffin-embedded blocks . For paraffin sections , brains were dissected and post-fixed in 4% PFA overnight at 4°C . Fixed brains were sagittally sectioned at 5 μm thickness . Slides from histologically comparable positions were stained by hematoxylin and eosin ( H&E ) and imaged with a light microscope ( Olympus BX51 , Olympus America Inc . , Center Valley , PA ) . Immunohistochemistry was performed on paraffin sections as previously described ( Wang et al . , 2009 ) . The visualization of primary antibodies was performed with the avidin-biotin horseradish peroxidase system ( Vectastain ABC kit , Vector Labs , Burlingame , CA ) . The dilutions of primary antibodies used on paraffin sections in this study were: BLBP ( 1:2000 , rabbit , a kind gift from Dr . Todd Anthony ) , Calbindin ( 1:1000 , mouse , Sigma-Aldrich ) , GFAP ( 1:2000 , rabbit , DAKO , Carpinteria , CA ) and BrdU ( 1:1000 , rat , Abcam , Cambridge , MA ) . Immunofluorescence was performed on both paraffin and frozen sections . Primary antibodies were visualized by Cy2 ( or Alexa 488 ) , Cy3 ( or Alexa 555 ) and Cy5 ( or Alexa 647 ) -conjugated secondary antibodies ( 1:200 , Cy2/Cy3/Cy5 , Jackson Immunoresearch , West Grove , PA; 1:400 , Alexa 488/555/647 , Invitrogen , Grand Island , NY ) . The dilutions of primary antibodies used in this study were as follows: GFAP ( 1:2000 , mouse , BD Pharmingen , San Jose , CA ) , GFAP ( 1:2000 , rabbit , DAKO ) , NeuN ( 1:400 , mouse , Millipore , Billerica , MA ) , BLBP ( 1:500 , rabbit , a gift from Nathaniel Heintz ) , Calbindin ( 1:1000 , mouse , Sigma-Aldrich ) , GABARα6 ( 1:500 , rabbit , Millipore ) , BrdU ( 1:1000 , rat , Abcam ) , Ki-67 ( 1:500 , mouse , BD Pharmingen ) , Tbr2 ( 1:1000 , rabbit , Abcam ) , Calretinin ( 1:1000 , mouse , Millipore ) , MAP2 ( 1:1000 , mouse , Sigma-Aldrich ) , P27 ( 1:250 , rabbit , Santa Cruz Biotechnology , Dallas , TX ) , β-gal ( 1:1000 , rabbit , 5 prime-3 prime , Gaithersburg , MD ) , β-gal ( 1:3000 , chicken , Abcam ) , p-Erk ( 1:200 , rabbit , Cell Signaling , Beverly , MA ) , p-S6 ( 1:1000 , rabbit , Cell Signaling ) . Sections were examined under a light/fluorescence microscope ( Olympus BX51 ) . Brains were removed immediately after decapitation , frozen in a bed of crushed dry ice , sectioned in a sagittal plane at 14 µm with a cryostat , mounted on RNase-free slides , and treated with 4% paraformaldehyde , and incubated in 0 . 1 M triethanolamine with acetic anhydrate . Radioactive riboprobes were synthesized from template DNA and labeled by S35-UTP . For hybridization , an 800 bp Math1 antisense probe was made from a plasmid pLA1-HindIII containing 600 bp of the Math1-coding region and approximately 200 bp of the 5′-flanking sequence ( Helms and Johnson , 1998 ) . The Zic1 antisense probe was a 690-bp XhoI–SalI segment from the mouse Zic1 cDNA ( Aruga et al . , 1994 ) . After hybridization , slides were dipped in photographic emulsion , counterstained , dehydrated and cover-slipped . Tissue were visualized and photographed with darkfield using an Olympus BX51 microscope . Detailed protocol was described previously ( Martin-Zanca et al . , 1990 ) . Tamoxifen ( Sigma-Aldrich ) was dissolved in corn oil ( Sigma-Aldrich ) at a concentration of 20 mg/ml and stored at −20°C . Pregnant female mice were intra-peritoneal ( IP ) injected once with tamoxifen at a dosage of 100 μg/g body weight at E17 . 5 . Their offspring carrying Nestin-creER transgene were collected at P21 for analysis . Nestin-creER+; Nf1flox/+; R26LacZR mice were used as controls and Nestin-creER+; Nf1flox/flox; R26LacZR mice were used as mutants ( Nf1NcreERCKO ) . The R26LacZR allele was introduced to the control , Nf1hGFAPCKO , Nf1Math1CKO , Nf1L7CKO and Nf1NcreERCKO mice as a reporter to monitor Cre-mediated recombination . Dissected brains were prepared for frozen sections and sliced at 12 μm and subjected to X-gal staining for 1 hr to overnight , according to the signal intensity . X-gal staining was performed as described previously ( Zhu et al . , 1998 , 2001 ) . Co-localization of β-gal expressing cells with other lineage markers was obtained by double immunofluorescence labeling with anti-β-gal antibody . Brains from P8 mice were dissected and subjected to Golgi staining for 10 days using a previously published protocol ( Glaser and Van der Loos , 1981 ) . Brains were then sagittally sectioned at 50 μm on a vibrotome and mounted on slides ( Leica , Germany ) . Sections were then developed in the darkroom ( 20 min 20% Ammonium hydroxide , Sigma-Aldrich + 10 min 20% Kodak fixer , Kodak , Rochester , NY ) and dehydrated in ethanol and xylene . Neonatal mice received a single injection of 50 μg/g ( gram , body weight ) BrdU ( Sigma-Aldrich ) . For proliferation assay , mice were perfused with 4% PFA 2 hr after the last pulse . For cell cycle and differentiation assay , neonatal mice were injected with BrdU 24 , 32 and 48 hr prior to analysis at P8 . For P8–P18 BrdU differentiation assay , mice were sacrificed 10 days after the initial pulse . In both assays , brains were dissected and processed for paraffin-embedded sections . BrdU immunofluorescence was performed as described previously ( Wang et al . , 2009 ) . Cerebella from P6 control and Nf1Math1CKO were dissected and dissociated into single-cell suspension using a previously published protocol ( Lee et al . , 2009 ) . Granule cells and glia were separated and cultured differently . For neuron/glia co-cultures , cultured glial cells were released by trypsination . Following trypsin inhibition , glial cells were counted , spun down , re-suspended in culture medium , and added to 1-day-old cultures of purified cerebellar neurons at a ratio of 2:1 . Trans-well migration assay was performed as described previously ( Watkins and Sontheimer , 2011 ) with the following modifications . Millicell cell culture inserts ( Millipore ) were placed into 24-well culture plate ( Corning Inc . , Corning , NY ) and granule cells ( 5 . 0 × 103 ) were added in each upper inserts . The bottom wells contained 10 μg/ml fibronectin as chemo-attractant . They were then incubated for 4 hr in a 37°C humidified CO2 incubator . At the end of the experiment , remaining cells in the well ( top ) were stripped with Q-tips . Cells on the opposite side ( bottom ) of the insert membrane were fixed with 4% PFA for 30 min and stained with DAPI . The number of cells migrated through the membrane was counted based on high-power images of membranes and the average cell number of 5 areas ( up , down left , right , center ) and was subjected to statistical analysis . For in vitro differentiation assay , dissociated cerebellar granule cell progenitors were plated on Poly-D-Lysine-coated cell culture plates in growth media without EGF and FGF . Cells were fixed after 72 hr and stained for MAP2 and DAPI . Both trans-well migration assay and in vitro differentiation assay were repeated with cells obtained from at least three mice . Snap-frozen tissues were homogenized in lysis buffer for 20 min on ice and subjected to centrifugation at 14 , 000 rpm for 10 min at 4°C . Equal amounts of protein samples were mixed with 1X SDS loading buffer [50 mM Tris-HCL ( pH6 . 8 ) , 2% SDS , 0 . 05% bromophenol blue , 10% glycerol , 100 mM β-mercaptoethanol] ( Biorad , Hercules , CA ) . Samples were separated on SDS-PAGE gel and transferred onto NC membranes ( Millipore ) . The blots were then blocked in 5% non-fat milk in TBST , followed by incubation of primary antibodies at 4°C overnight . After washing , the blots were incubated in horseradish peroxidase ( HRP ) -conjugated secondary antibodies at room temperature for 1 hr . Signals were detected using ECL or ECL plus ( GE healthcare , UK ) followed by film development . The primary antibodies used were as follows: p-Erk , Erk , p-S6 , S6 , p-Akt , Akt ( 1:1000 , rabbit , Cell Signaling ) and Nf1 ( 1:1000 , rabbit , Upstate , Lake Placid , NY ) . Rapamycin ( Millipore ) was dissolved in 100% sterile-filtered ethanol at a concentration of 10 mg/ml and resuspended in vehicle ( 5 . 2% PEG400 with 5 . 2% Tween80 , Sigma-Aldrich ) at a final concentration of 0 . 4 mg/ml . The solution was administered by intraperitoneal injection at the dosage of 4 mg/kg ( body weight ) every other day from P0 . 5–P21 . Rapamycin-treated mice were collected and compared to littermate control and Nf1hGFAPCKO mice treated with vehicle . Motor function and motor learning were evaluated by rotarod apparatus . 1 hr before the rotarod test , mice were transferred from the holding room to the experimental room . Mice were tested for 4 consecutive days at P21–P24 and P45–P49 . The speed of the rotarod was constantly accelerated from 4 to 34 rpm over a 5-min period . The latency to fall from the rod was recorded . The test was stopped at 12 min . Anatomically comparable sections from control and mutant brains were visualized under an Olympus BX51 microscope . Images were captured and subjected to analysis . Lengths , areas and the number of cells were quantified using the ImageJ software package . Statistical analysis was carried out using unpaired two-tailed Student's t-test . Comparisons across more than two groups were based on Anova test . At least three animals from each group were used for quantification . Data were presented as mean ± Standard Error Mean ( SEM ) . p < 0 . 05 was considered to be statistically significant . When quantifying cells in specific folia , we used the following criteria to define the areas to quantify . ‘Folia V/IV’ refers to the junction between folia V and IV at 20× magnification ( i . e . , Figure 1A middle panels ) and the quantification covered this entire area . Similarly , ‘Folia IX/X’ refers to the junction between folia IX and X at 20× magnification ( i . e . , Figure 5—figure supplement 2A ) . ‘Folium V’ or ‘Folium X’ refers to 40× images in these folia close to the center of the cerebellum that were imaged at comparable positions in controls and mutants .	Neurofibromatosis type 1 is a condition characterized by the growth of tumors along the nerves of the body . It is caused by mutations in a gene called NF1 , which codes for a protein that normally works to inhibit the activity of another protein called Ras . In healthy cells , Ras is needed to stimulate the cells to grow and divide . However , if the Ras protein is not turned off at the right time or if it is activated at the wrong time , it can force cells to keep growing and dividing; this leads to the growth of tumors . Along with being prone to developing cancer , individuals with neurofibromatosis type 1 also develop a range of neurodevelopmental disorders that alter their learning , motor skills and social interactions . Some also exhibit behaviors that are associated with autism . This led Kim , Wang et al . to investigate whether a region of the brain—called the cerebellum—that has recently been associated with autism is also affected in a mouse model of neurofibromatosis type 1 . The cerebellum is best known for its role in coordinating movement , although it also has functions in cognition , behavior and other processes . Ras is involved in the development of the cerebellum; and so Kim , Wang et al . asked whether the loss of the Nf1 gene from cells in the mouse cerebellum might cause the neurodevelopmental defects associated with neurofibromatosis type 1 . Loss of Nf1 during early ( but not in late ) development of the cerebellum disrupted the normal organization of the nerve cells ( or neurons ) into specific cell layers . These defects were caused , in part , by the over-growth of a type of supporting cell—called glia cells—at a specific developmental stage—that would normally form a scaffold to help neurons migrate to their correct position . Nf1 also controls the generation of the correct types of neurons in the right time and at right location during the early development of the cerebellum . Next , Kim , Wang et al . treated newborn mice with a compound that inhibits Ras signaling via their mother's milk for 3 weeks . In mice with an inactive Nf1 gene , the treatment helped to prevent some defects in the cerebellum and the mice had improved motor coordination several months later . Whether this could form the basis of a preventative treatment for neurodevelopmental disorders associated with neurofibromatosis type 1 in humans remains a question for future work .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2014	Transient inhibition of the ERK pathway prevents cerebellar developmental defects and improves long-term motor functions in murine models of neurofibromatosis type 1
Despite extensive research on the role of the rodent medial and lateral entorhinal cortex ( MEC/LEC ) in spatial navigation , memory and related disease , their human homologues remain elusive . Here , we combine high-field functional magnetic resonance imaging at 7 T with novel data-driven and model-based analyses to identify corresponding subregions in humans based on the well-known global connectivity fingerprints in rodents and sensitivity to spatial and non-spatial information . We provide evidence for a functional division primarily along the anteroposterior axis . Localising the human homologue of the rodent MEC and LEC has important implications for translating studies on the hippocampo-entorhinal memory system from rodents to humans . The entorhinal cortex ( EC ) —defining the interface between the hippocampus and the neocortex ( Munoz and Insausti , 2005 ) —plays a pivotal role in the integration of different sensory inputs into higher order mnemonic representations ( Eichenbaum et al . , 2007; Moser and Moser , 2013 ) . In rodents—and on the basis of cytoarchitectonics—the EC is typically ( Kerr et al . , 2007; Canto et al . , 2008; Van Strien et al . , 2009 ) subdivided into two major subregions , the medial- and the lateral entorhinal cortex ( MEC and LEC , respectively ) . The MEC receives inputs about spatial information from parahippocampal cortex ( PHC ) and the LEC receives item-related information from perirhinal cortex ( PRC ) ( Van Strien et al . , 2009; Deshmukh and Knierim , 2011; Ranganath and Ritchey , 2012; Knierim et al . , 2013 ) . Similar functional roles of the PHC and PRC have been described in humans ( Epstein and Kanwisher , 1998; Davachi et al . , 2003; Eichenbaum et al . , 2007; Ekstrom and Bookheimer , 2007; Litman et al . , 2009; Duarte et al . , 2011; Staresina et al . , 2011; Martin et al . , 2013; Vilberg and Davachi , 2013 ) and relate to distinct visual processing streams ( Kravitz et al . , 2011 ) . The differential input pattern into the rodent LEC and MEC also dovetails with a cell-type specific functional specialisation ( Eichenbaum and Lipton , 2008 ) . The MEC contains a high proportion of head-direction and grid cells , whose activity is modulated by running direction and spatial location , respectively ( Hafting et al . , 2005; Sargolini et al . , 2006 ) . In contrast , cells in the LEC respond to individual objects in the environment rather than to specific locations ( Deshmukh and Knierim , 2011; Tsao et al . , 2013; Knierim et al . , 2013 ) . Despite a wealth of data and marked differences in structure and function of the rodent MEC and LEC evidence for their human homologue remains elusive . This hampers translational studies , which is particularly relevant in the case of Alzheimer's disease ( AD ) with AD pathology starting in the EC ( Braak and Braak , 1992 ) . Within the EC , the vulnerability to AD-related pathology is not homogeneously distributed and differs between medial and lateral strips in humans , which has been related to similar findings in the rodent MEC and LEC , respectively ( Khan et al . , 2014 ) . However , the localization of the human homologue of the rodent MEC and LEC remains unclear . A source of considerable confusion is the fact that ‘MEC’ and ‘LEC’ are referring to cytoarchitectonically defined areas and not to anatomical locations . Hence , they do not circumscribe strips of medial and lateral EC . Rather , the MEC is located medially in the septal ( posterior ) part of the EC and the LEC is located laterally in the temporal ( anterior ) part of the EC in rodents ( Van Strien et al . , 2009 ) . Furthermore , tracing studies on PHC and PRC pathways in non-human primates suggest a dominant anterior-posterior division ( Suzuki and Amaral , 1994; Insausti and Amaral , 2008 ) , as do single-unit recordings that show activity consistent with the rodent LEC in the anterior EC in primates ( Killian et al . , 2012 ) . In contrast , neuroimaging studies on memory in healthy participants ( Schultz et al . , 2012; Reagh and Yassa , 2014 ) and participants with preclinical AD ( Khan et al . , 2014 ) suggest that the rodent MEC and LEC map on medial and lateral strips of EC in humans . To resolve this discrepancy in the literature , one needs to investigate the relatively small EC ( 25–30 mm2 in humans ) ( Krimer et al . , 1997 ) with high anatomical precision . An earlier study investigated entorhinal connectivity with high-resolution functional magnetic resonance imaging ( fMRI ) , but averaged signal changes over the entire region ( Lacy and Stark , 2012 ) . To achieve higher resolution imaging , here we leveraged high-field , sub-millimetre fMRI at 7 T and sought to identify the human homologue of the rodent MEC and LEC by informing our analysis by well-known functional and structural properties of the EC . Specifically , it has been shown that MEC and LEC exhibit differential connectivity with cortical regions ( Witter and Groenewegen , 1989; Kerr et al . , 2007; Van Strien et al . , 2009 ) . The differential fingerprints of anatomical connectivity should lead to differences in functional connectivity identifiable with fMRI ( Johansen-Berg et al . , 2004; Buckner et al . , 2013; Wang et al . , 2014 ) . To test patterns of functional connectivity , we measured whole-brain activity while participants performed a virtual reality task with spatial and non-spatial components and validated the results in publicly available resting-state data from the WU-Minn Human Connectome Project ( Van Essen et al . , 2013; Smith et al . , 2013 ) ( HCP—www . humanconnectome . org ) . In addition , differential sensitivity to spatial and non-spatial stimuli could provide converging evidence to identify the human homologue of the rodent MEC and LEC , which we tested in a third , independent dataset . A complementary approach to the global network perspective presented here is given by Maass et al . ( Maass et al . , 2015 ) who scrutinized the fine-grained connectivity pattern of medial temporal lobe regions with the EC . A recent model on cortical memory networks ( Ranganath and Ritchey , 2012 ) posits that an anterior-temporal ( AT ) system converges on the PRC and a posterior-medial ( PM ) system on the PHC . Based on studies in rodents , the two networks are hypothesised to connect to either the LEC or the MEC , respectively ( Witter and Groenewegen , 1989; Kerr et al . , 2007; Van Strien et al . , 2009 ) . Studies in non-human primates predict that the entorhinal projections of the two systems show a strong anteroposterior division ( Suzuki and Amaral , 1994 ) . In order to test this prediction and to elucidate the role of the EC , we first applied a model-based approach on fMRI data acquired while participants were performing a virtual-reality navigation task to directly mimic studies in rodents ( see ‘Materials and methods’ for details ) . This task targeted all entorhinal systems , because it involved both navigation-related spatial components and processing of non-spatial stimuli . We created spherical regions-of-interest ( ROIs ) with 4 mm radius around coordinates pertaining to either of the networks ( Libby et al . , 2012; Ranganath and Ritchey , 2012 ) ( see Table 1 ) , as well as ROIs for both the medial and lateral half , and anterior and posterior half of the EC to ensure comparable number of voxels per parcel and therefore comparable signal-to-noise ratio ( SNR ) properties . Then we computed seed-based connectivity from the two neocortical networks to either sets of EC ROIs , see Figure 1 . We found a main effect of network on entorhinal connectivity ( repeated-measures ANOVA: F ( 1 , 21 ) = 10 . 0 , p = 0 . 005 ) . Post-hoc t-tests revealed that the lateral parts of EC connected stronger to the AT compared to the PM network ( T ( 21 ) = 2 . 6 , p = 0 . 015; medial parts of EC: T ( 21 ) = −0 . 2 p = 0 . 83 ) . However , in contrast to previous suggestions , we additionally observed a connectivity difference along the anteroposterior axis ( repeated-measures ANOVA: main effect of network , F ( 1 , 21 ) = 13 . 2 , p = 0 . 001 ) and post-hoc t-tests showed that the anterior parts of EC connected more with the AT compared to the PM network ( T ( 21 ) = 2 . 7: p = 0 . 01; posterior EC: T ( 21 ) = −0 . 49 , p = 0 . 63 ) . 10 . 7554/eLife . 06738 . 003Table 1 . Selection of regions associated with the posterior-medial ( PM ) and the anterior-temporal ( AT ) system ( Libby et al . , 2012 ) . The coordinates of the PM system reflect peak voxel coordinates of a seed-based connectivity contrast of right parahippocampal cortex > right perirhinal cortex connectivity reported by Libby et al . ( Libby et al . , 2012 ) . The coordinates of the AT system reflect peak voxel coordinates of a seed-based connectivity contrast of right perirhinal cortex > right parahippocampal cortex connectivity . Coordinates are in MNI space . DOI: http://dx . doi . org/10 . 7554/eLife . 06738 . 003Left hemisphereLeft hemispherexyzxyzPM System Medial posterior occipital cortex ( BA 18 ) –––14−728 Occipital pole ( BA 17 ) −16−9622––– Parahippocampal cortex−12−42−822−32−8 Posterior cingulate cortex ( BA 29 ) −4−46410−4410 Posterior hippocampus−20−30−218−360 Posterior thalamus−20−34022−306 Retrosplenial cortex ( BA 30 ) −16−52−422−460AT System Dorsolateral prefrontal cortex ( BA 9 ) −246024185824 Dorsomedial prefrontal cortex ( BA 8 ) −2−6034––– Frontal polar cortex ( BA 10 ) –––4060−2 Lateral precentral gyrus ( BA 6 ) –––54410 Medial prefrontal cortex ( BA 8 ) −2−6034––– Orbitofrontal cortex ( BA 11/47 ) −616−22822−20 Postcentral gyrus ( BA 4 ) –––62−1016 Posterior superior temporal gyrus ( BA 22 ) −62−3414––– Rostrolateral prefrontal cortex ( BA 10 ) –––3860−12 Temporal polar cortex ( BA 38 ) –––3422−36 Ventrolateral prefrontal cortex ( BA 44/45 ) −56618–––10 . 7554/eLife . 06738 . 004Figure 1 . Subdivisions of entorhinal cortex ( EC ) and connectivity to anterior-temporal ( AT ) and posterior-medial ( PM ) cortical networks . ( A ) Schematic of the AT and PM system . Spherical regions-of-interest ( ROIs ) were centred on MNI coordinates associated with either of the two systems ( Libby et al . , 2012 ) , normalised to the group-specific template of the navigation study and then masked to include only gray matter voxels . The AT system included medial-prefrontal and orbitofrontal regions , whereas the PM system included occipital and posterior-parietal regions , see Table 1 for all selected regions . ( B ) Right parasagittal slice showing voxel-wise seed-based connectivity of the PM system restricted to the EC . Note the PM peak . ( C ) Right parasagittal slice showing voxel-wise seed-based connectivity of the AT system restricted to the EC . Note a peak in the anterior-lateral EC . ( D ) ROI-based connectivity estimates . Left panel: Connectivity strength ( partial correlation coefficient ) of anterior ( left ) and posterior EC ( right ) is plotted separately for the AT system ( red ) and the PM system ( blue ) . The systems differ in their entorhinal connectivity: the anterior EC connects stronger to the AT compared to the PM network . Right panel: Connectivity strength with lateral ( left ) and posterior EC ( right ) is plotted separately for the AT system ( red ) and the PM system ( blue ) . Lateral EC connected stronger to the AT compared to the PM network . Error bars show S . E . M . over subjects . See Figure 1—figure supplement 1 for additional slices . DOI: http://dx . doi . org/10 . 7554/eLife . 06738 . 00410 . 7554/eLife . 06738 . 005Figure 1—figure supplement 1 . Results of the model-based connectivity analyses ( additional slices ) . ( A ) Schematic of the AT and PM system . ( B ) Top row: left parasagittal slice showing connectivity with the PM network . Bottom rows: coronal slices showing connectivity with the PM network . ( C ) Top row: left parasagittal slice showing connectivity with the PM network . Bottom rows: coronal slices showing connectivity with the PM network . DOI: http://dx . doi . org/10 . 7554/eLife . 06738 . 005 In a second step , we wanted to overcome potential limitations of the seed-based analysis . For example , the selected volume and location of neocortical seed regions could introduce biases ( e . g . , spatial proximity of the seeds ) and imperfect normalisation procedures could affect the results particularly in the frontal lobes where projections from both the rodent LEC and MEC are neighbouring ( Kerr et al . , 2007 ) . In addition , manual subdivision of the EC along cardinal axes likely misrepresents cytoarchitectonic boundaries . Therefore , we adopted a complementary approach to trace the dominant modes of functional connectivity change within the EC in a fully data-driven manner ( Haak et al . , 2014 ) ( see ‘Materials and methods’ ) . In brief , for every voxel in the EC , we determined its functional connectivity fingerprint with respect to the rest of cortex and used these fingerprints to compute the pair-wise similarities among all voxels within the ROI . The ensuing ( voxels-by-voxels ) similarity matrix was then fed to the Laplacian Eigenmaps ( LE ) algorithm ( Belkin and Niyogi , 2003 ) , which has previously also been successfully applied to trace changes in white-matter tractography ( Johansen-Berg et al . , 2004; Cerliani et al . , 2012 ) and resting-state fMRI connectivity ( Haak et al . , 2014 ) . The LE algorithm projects the high-dimensional , voxel-wise connectivity data onto a series of one-dimensional vectors , with the requirement that the similarities among the connectivity fingerprints are maximally preserved ( in the vein of e . g . , multidimensional scaling ) . These vectors represent multiple , spatially overlapping maps ( as revealed by colour-coding the EC voxels according to the vectors' values ) and are sorted according to how well they preserve the similarities among the original , high-dimensional connectivity fingerprints . Thus , the first vector represents the dominant mode of connectivity change in the EC , the second represents the second-dominant mode , and so on . Applied to the fMRI data acquired while subjects performed the virtual-reality task , we observed that the dominant mode of functional connectivity change extended along the long-axis of the EC , approximately from the posterior to the anterior end ( Figure 2 and Figure 2—figure supplement 1 ) , while the orientation of the second was largely perpendicular ( Figure 3 and Figure 3—figure supplement 1 ) . Both modes of connectivity change could also be reliably detected using an independent resting-state fMRI dataset ( 60 subjects of the WU-Minn Human Connectome Project; see ‘Materials and methods’ ) , suggesting that the organization of EC functional connectivity is largely task-independent ( Figure 4 ) . Both the first and second-dominant modes of functional connectivity change were highly reproducible across resting-state sessions ( Pearson's R = 0 . 99 , p < 0 . 001 and Pearson's R = 0 . 98 , p < 0 . 001 , for the dominant and second-dominant modes , respectively ) . 10 . 7554/eLife . 06738 . 006Figure 2 . Dominant mode of functional connectivity change within EC and sensitivity to spatial and non-spatial information . ( A ) Dominant mode of functional connectivity change at the group-level ( Spearman's R = 0 . 53 ) . Similar colours indicate similar connectivity with the rest of the brain . ( B ) 3D rendering of the two clusters derived from the dominant mode of functional connectivity change ( displayed in red and blue ) and the outlines of the group-specific template . Upper panel: right side view . Lower panel: top view ( see Figure 2—figure supplement 1 for coronal views of the two clusters ) . ( C ) Upper panel: Map shows results of a non-parametric randomisation test of the spatial and non-spatial stimulation experiment restricted to the EC for display purposes ( see Figure 2—figure supplement 2 for whole-brain maps ) . The ‘scenes > objects’ contrast is displayed in blue to light-blue , the ‘scenes < objects’ contrast in red to yellow . Note that voxels in pmEC are sensitive to scenes , whereas voxels in alEC are sensitive to objects . Lower panel: The clusters from panel B exhibit antagonistic responses to spatial and non-spatial stimuli . Beta estimates for the contrast ‘scenes > objects’ ( averaged across participants ) are shown for clusters A and B . T ( 20 ) = 4 . 9 , p = 0 . 0001 . Error bars show S . E . M . over participants . ( D ) Whole-volume functional connectivity with clusters A and B . Regions connecting more with cluster A ( p < 0 . 05 , FWE corrected ) , such as occipital and posterior-parietal cortex that form part of the PM system are shown in blue . Regions connecting more with cluster B ( p < 0 . 05 , FWE corrected ) , such as medial-prefrontal and orbitofrontal cortex which form part of the AT system are displayed in red . DOI: http://dx . doi . org/10 . 7554/eLife . 06738 . 00610 . 7554/eLife . 06738 . 007Figure 2—figure supplement 1 . Coronal views of the two clusters . Top: sagittal slice . Dashed lines indicate position of the coronal slices shown below . Bottom: The posterior slice ( left ) exclusively contains cluster A . In the middle slice cluster A is located dorsomedially , proximal to the hippocampus and cluster B is located ventrolaterally , distal to the hippocampus . The anterior slice ( right ) contains mostly cluster B . A = anterior; P = posterior . DOI: http://dx . doi . org/10 . 7554/eLife . 06738 . 00710 . 7554/eLife . 06738 . 008Figure 2—figure supplement 2 . Whole-brain modulation by spatial and non-spatial stimuli . ( A ) Left parasagittal slice showing peak effects of two contrasts on the data from the spatial and non-spatial stimulation experiment . The ‘scenes > objects’ contrast is displayed in blue-lightblue , the ‘scenes < objects’ contrast in red-yellow . Note that voxels in the posterior EC are sensitive to scenes , whereas voxels in the anterior EC are sensitive to objects . ( B ) Coronal slice anterior-posterior location indicated by dashed line in ( A ) . Note the medial peak in the medial temporal lobe for the ‘scenes > objects’ contrast and the lateral peak for the ‘scenes < objects’ contrast . Images are thresholded at T > 2 for display purposes . DOI: http://dx . doi . org/10 . 7554/eLife . 06738 . 00810 . 7554/eLife . 06738 . 009Figure 2—figure supplement 3 . Signal-to-noise ratios ( SNRs ) in the alEC and pmEC . ( A ) Bar plots of the ratio between the mean signal intensity and signal standard deviation across voxels . The SNR across voxels was higher in the pmEC than the alEC ( T ( 21 ) = 15 . 2721 , p < 0 . 001 ) . This was associated with a larger signal in the pmEC ( mean signal alEC = 4 . 97; mean signal pmEC = 7 . 76; T ( 21 ) = 38 , p < 0 . 001 ) in the absence of differences in spatial standard deviation ( T ( 21 ) = 1 . 5 , p = 0 . 144 ) . Note , that mean signal was subtracted from time-series prior to all connectivity analyses ( see ‘Materials and methods’ ) , which makes it unlikely that signal intensity differences affected the connectivity results . ( B ) Bar plots of the ratio between the mean signal intensity and the signal standard deviation across time . Temporal SNR ( tSNR ) did not differ between alEC and pmEC ( T ( 21 ) = 0 . 2 , p = 0 . 83 ) . Error bars show S . E . M . over participants . DOI: http://dx . doi . org/10 . 7554/eLife . 06738 . 00910 . 7554/eLife . 06738 . 010Figure 2—figure supplement 4 . Homologous and non-homologous connectivity of the alEC and the pmEC . Connectivity between subregions identified using the data-driven connectivity analysis . Connectivity reflects partial Pearson correlation coefficients after Fisher Z transformation . Laterality is indicated by the last letter: left = ‘l’ , right = ‘r’ . ( B ) Connectivity for three conditions . ‘non-hom’ refers to non-homologous connectivity , ‘contra’ refers to contralateral connectivity and ‘hom’ refers to homologous connectivity . Note that homologous connectivity ( across hemispheres ) exceeded connectivity with the neighbouring non-homologous region in the same hemisphere ( T ( 21 ) = 4 . 05 , p = 0 . 0006 ) . Connectivity between non-homologous regions was strongest within hemispheres ( T ( 21 ) = 3 . 66 , p = 0 . 0015 ) . Error bars show S . E . M . over participants . *p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 06738 . 01010 . 7554/eLife . 06738 . 011Figure 3 . Second-dominant mode of functional connectivity change within EC and sensitivity to spatial and non-spatial information . ( A ) Second-dominant mode of functional connectivity change at the group-level ( Spearman's R = 0 . 28 ) . Similar colours indicate similar connectivity with the rest of the brain . ( B ) 3D rendering of the two clusters derived from the second-dominant mode of functional connectivity change ( displayed in red and blue ) and the outlines of the group-specific template . Upper panel: right side view . Lower panel: top view ( see Figure 3—figure supplement 1 for coronal views of the two clusters ) . ( C ) The clusters shown in panel B exhibit no antagonistic responses to spatial and non-spatial stimuli . Beta estimates for the contrast ‘scenes > objects’ ( averaged across participants ) are shown for cluster A and B ( T ( 20 ) = −0 . 26 , p = 0 . 8 ) . Error bars show S . E . M . over participants . ( D ) Regions connecting more with cluster A ( p < 0 . 05 , FWE corrected ) are shown in blue . Regions connecting more with cluster B ( p < 0 . 05 , FWE corrected ) are shown in red . Cluster A connected more with most of the neocortex . DOI: http://dx . doi . org/10 . 7554/eLife . 06738 . 01110 . 7554/eLife . 06738 . 012Figure 3—figure supplement 1 . Coronal views of the two clusters . Top: sagittal slice . Dashed lines indicate position of the coronal slices shown below . Bottom: All three coronal slices contain both clusters . Cluster A is located ventrally , distal to the hippocampus and cluster B dorsally , proximal to the hippocampus . A = anterior; P = posterior . DOI: http://dx . doi . org/10 . 7554/eLife . 06738 . 01210 . 7554/eLife . 06738 . 013Figure 4 . Dominant and second-dominant modes of functional connectivity change on the basis of resting-state functional magnetic resonance imaging . Results of analysis of the first 60 participants of the WU-Minn Human Connectome Project ( HCP ) , acquired on two different days ( Smith et al . , 2013 ) . Top row: day one . Bottom row: day two . ( A , C ) The dominant mode of functional connectivity change follows an anteroposterior trajectory . ( B , D ) . The second-dominant mode of functional connectivity change follows a mediolateral trajectory . Both modes were highly reproducible across different scanning days ( dominant mode: Pearson's R = 0 . 99 p < 0 . 001 , second-dominant mode: R = 0 . 98; p < 0 . 001 ) . Topology preservation—dominant mode , day one: Spearman's R = 0 . 62; day two: R = 0 . 61; second-dominant mode , day one: R = 0 . 44; day two: R = 0 . 46 . DOI: http://dx . doi . org/10 . 7554/eLife . 06738 . 013 Furthermore , in order to identify the potential human homologues of the rodent LEC and MEC , we clustered the vectors representing the dominant and second-dominant modes of functional connectivity change ( separately ) through a median-split approach ( see ‘Materials and methods’ ) . Hence , each cluster comprises 50% of voxels in the EC . 3D-rendering of the two clusters derived from the dominant mode of connectivity change revealed a consistent topology across hemispheres ( Figure 2B , Video 1 ) . One division contained the posterior EC ( Figure 2—figure supplement 1—displayed in blue ) . The other division included most of the anterior EC ( Figure 2—figure supplement 1—displayed in red ) . In addition to the dominant anterior-posterior distinction , the posterior cluster was located more medially ( and to some extent more dorsally ) and the anterior cluster was located more laterally ( and to some extent more ventrally ) . Hereafter , we refer to the clusters as posterior-medial EC ( ‘pmEC’ ) and anterior-lateral EC ( ‘alEC’ ) , respectively , consistent with Maass et al . ( Maass et al . , 2015 ) . Clusters derived from the second-dominant mode were less consistent across hemispheres , but showed an approximately orthogonal orientation relative to the first ( Figure 3 and Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 06738 . 014Video 1 . 3D rendering of the two clusters derived using the dominant mode of functional connectivity change . Cluster A is shown in red and cluster B in blue . DOI: http://dx . doi . org/10 . 7554/eLife . 06738 . 014 If the two clusters ( i . e . , EC halves ) derived from the dominant mode of functional connectivity change correspond to the homologues of the rodent LEC and MEC , their whole-brain connectivity profiles should correspond to the known connectivity profiles in rodents and resemble the AT and PM system proposed by Ranganath and Ritchey ( Ranganath and Ritchey , 2012 ) . To test this hypothesis , we computed whole-volume connectivity maps of the two clusters . Group-level contrasts showed peaks in the medial-prefrontal and orbitofrontal cortex for the alEC , regions associated with the AT system . In contrast , occipital and posterior-parietal cortex was dominated by connectivity with the pmEC , areas associated with the PM system ( Ranganath and Ritchey , 2012 ) . In addition , the pmEC showed increased connectivity with frontal regions ( see Figure 2D ) . These findings are in line with the patterns of reciprocal connections of the rodent LEC and MEC , respectively ( Kerr et al . , 2007 ) . Notably , this was not the case for the connectivity maps of the two clusters derived from the second-dominant mode of connectivity change ( Figure 3D ) that were widely dominated by only one of the clusters . Furthermore , we examined both spatial and temporal SNR ( tSNR ) of the alEC and the pmEC ( Figure 2—figure supplement 3 ) . tSNR did not differ between alEC and pmEC ( T ( 21 ) = 0 . 2 , p = 0 . 83 ) but spatial SNR did ( T ( 21 ) = 9 . 7 , p < 0 . 001 ) . This was associated with higher signal in the pmEC compared to the alEC ( mean signal: alEC = 4 . 97; mean signal pmEC = 7 . 76; T ( 21 ) = 38 , p < 0 . 001 ) in the absence of differences in spatial standard deviation ( T ( 21 ) = 1 . 5 , p = 0 . 144 ) . Note , that mean signal was subtracted from time-series prior to all connectivity analyses ( see ‘Materials and methods’ ) , which makes it unlikely that signal intensity differences affected the connectivity results . We used an independent component analysis ( ICA ) -based method for data cleaning ( see ‘Materials and methods’ ) that has been shown to efficiently remove residual effects of head motion . However , we additionally repeated the data-driven connectivity analysis after excluding time periods with large head movements ( motion scrubbing [Power et al . , 2012 , 2015] ) . Motion scrubbing had only minimal effects on the results . Pearson correlation coefficients of the pre- and post scrubbing results were close to 1 and highly significant ( gradient one: left R = 0 . 9964 , right R = 0 . 9958; gradient two: left = 0 . 9348 , right = 0 . 8816; all p values <0 . 001 ) . We also tested if the alEC and pmEC exhibited stronger connectivity with their potential homologue region in the contralateral hemisphere compared to each other . Here we observed that homologous connectivity indeed exceeded non-homologous connectivity ( Figure 2—figure supplement 4 ) . Importantly , this was also the case if non-homologous connectivity was assessed within the same hemisphere , between adjacent parts of the EC ( T ( 21 ) = 4 . 05 , p = 0 . 0006 ) . Finally , studies on rodent electrophysiology ( Deshmukh and Knierim , 2011; Knierim et al . , 2013 ) predict that the LEC and its human homologue should respond preferentially to non-spatial stimuli , whereas the MEC and its human homologue should be involved in processing spatial information . We tested this prediction by conducting a second fMRI study at 7 T in which an independent group of participants was presented with spatial ( pictures of scenes ) and non-spatial stimuli ( pictures of objects ) , see ‘Materials and methods’ for details . We contrasted fMRI responses to spatial and non-spatial stimuli . Here , we observed higher responses to spatial than non-spatial stimuli in the posterior EC , while the inverse contrast ( objects vs scenes ) showed higher responses in the anterior EC ( Figure 2C ) . ROI analyses using the clusters derived from the dominant mode of connectivity change revealed that the anterior-lateral cluster showed higher sensitivity to non-spatial stimuli compared to the PM cluster ( T ( 20 ) = 4 . 9 , p = 0 . 0001 , Figure 2C ) . This dissociation was not present for the clusters that were derived from the second-dominant mode of connectivity change within the EC ( T ( 20 ) = −0 . 26 , p = 0 . 8; Figure 3C ) . In sum , our results suggest that the human homologue of the rodent MEC maps predominantly on the human posterior parts of the EC , while the homologue of the rodent LEC maps predominantly on the anterior parts of the EC . The EC , in concert with the hippocampus , plays a crucial role in memory and learning ( Eichenbaum et al . , 2007 ) and is the core of the brain's navigational system ( Moser and Moser , 2013 ) . While the shape and location of the EC differs between rodents and primates ( Witter and Groenewegen , 1989 ) , the anatomical organisation and connectivity patterns are largely conserved across species ( Canto et al . , 2008 ) . However , translational studies on EC subregions faced the problem of identifying homologous regions across species . For example , recent neuroimaging studies on mnemonic processing ( Schultz et al . , 2012; Reagh and Yassa , 2014 ) and Alzheimer's pathology ( Khan et al . , 2014 ) directly related medial and lateral strips of EC in humans to the rodent MEC and LEC . However , the anatomical locations of these cytoarchitectonically defined regions in rodents differ along more than only the mediolateral axis . More specifically , the LEC is situated more anterior-ventrally , whereas the MEC is situated more posterior-dorsally in rodents ( Van Strien et al . , 2009 ) . Therefore , it is unlikely that medial and lateral strips of EC in humans correspond to the rodent MEC and LEC , respectively . Furthermore , in primates the characteristic projections from the PRC and PHC strongly map onto the anteroposterior axis ( Suzuki and Amaral , 1994 ) . Here , we leveraged the distinct connectivity fingerprints and functional roles ( such as complementary computation of scene and object information ) of the rodent LEC and MEC to find their human homologues with fMRI with three complementary methods and three independent datasets . Both model-based and data-driven connectivity analyses , as well as sensitivity to non-spatial vs spatial stimuli provide evidence for an anterior-lateral and a PM localisation of the homologues of the rodent LEC and MEC , respectively . Maass et al confirmed these findings in a study with two high-resolution , high-field fMRI datasets by focusing on local connectivity between regions of the medial temporal lobes . They found preferential connectivity of PRC and proximal subiculum to anterior-lateral parts of the EC , whereas posterior-medial parts of the EC were more connected to PHC and distal subiculum . This corresponds well with our findings ( Figure 2—figure supplement 1 ) . In line with the present study , Maass et al ( Maass et al . , 2015 ) report local connectivity fingerprints of the human anterior-lateral and posterior-medial EC that mimicked those of the rodent LEC and MEC , respectively . In addition to the change in functional connectivity from PM to anterior-lateral , our data-driven connectivity analysis also revealed a second organisation structure approximately perpendicular to the first ( Figure 3—figure supplement 1 ) , which might reflect bands of intra-entorhinal projections that are known to cross the LEC/MEC boundary in a roughly orthogonal orientation in rodents ( Canto et al . , 2008 ) and in primates ( Chrobak and Amaral , 2007 ) . The selective sensitivity to spatial and non-spatial information , or ‘context vs content’ more broadly ( Knierim et al . , 2013 ) , points towards fundamental difference in computations of the LEC and MEC . How to characterise those differences most accurately remains an open question ( Knierim et al . , 2013 ) , but our results can help to inform future studies on the role of the human alEC and pmEC in higher-level cognition . Notably , the present findings confirm three out of four complementary criteria for the definition of cortical areas that have traditionally been advocated ( Van Essen , 1985 ) , namely topographic organization , connectivity and functional properties ( the fourth one being cyto- and myeloarchitectonic organization ) . Previous neuroimaging studies in humans reported differences between medial and lateral aspects of EC that mimicked differences between the rodent MEC and LEC and assumed that both subregions are present on coronal slices of the EC ( Schultz et al . , 2012; Khan et al . , 2014; Reagh and Yassa , 2014 ) , that is , that the MEC and LEC correspond to medial and lateral strips of the EC . In light of our findings , these reports could be explained by a partial overlap of the medial and lateral divisions with the pmEC and the alEC , respectively . For example , we noticed a mediolateral difference of responses to spatial and non-spatial stimuli on some coronal slices ( Figure 2—figure supplement 2B ) . However , our results suggest that coronal slices through the most posterior EC exclusively harbour the human homologue of the rodent MEC . Similarly , anterior slices appear to contain mostly the homologue of the rodent LEC . Hence , improved mapping of homologous regions between rodents and humans should lead to increased effect sizes and more accurate interpretations . In summary , the present findings can help to inform future translational research on the role of entorhinal subregions in fields ranging from clinical neuroscience , such as on the early progression of Alzheimer's disease , to cognitive neuroscience , for example , nature and mechanisms of different forms of memory and their integration into higher order representations ( Eichenbaum and Lipton , 2008 ) . For the 7 Tesla experiments ( navigation experiment and spatial and non-spatial stimulation experiment ) , an EC ROI was manually defined with ITK Snap 3 . 2 ( Yushkevich et al . , 2006 ) ( http://www . itksnap . org ) on the group-specific high-resolution mean EPI template from the navigation experiment . Manual segmentation on the group-specific EPI template circumvented registration problems between structural images and partly distorted functional images . Based on anatomical landmarks as described by Insausti et al . , ( 1998 ) and Frankó et al . , ( 2014 ) , the posterior border of the EC was set to ∼1 . 5 mm posterior to the gyrus intralimbicus , the anterior border to ∼1 . 5 mm posterior to the limen insulae , the lateral border to the midpoint of the medial bank of the collateral sulcus and the medial border to the hippocampal fissure at the level of the uncus . A medial EC and a lateral EC ROI of roughly equal volume were created by dividing the main ROI mediolaterally ( which roughly corresponded to a proximo-distal division in relation to the hippocampus ) on consecutive coronal slices along the entire anteroposterior axis . An anterior EC and a posterior EC ROI of roughly equal volume were created by dividing the main ROI mid-way between the most anterior and the most posterior coronal slice of the EC . For the ‘model-based’ analysis , one ROI mask was created for the PM system and another one for the AT system . For this purpose we placed spheres of 4 mm radius on MNI coordinates associated with each system by Libby et al . , ( Libby et al . , 2012 ) ( see Table 1 for the selected coordinates ) . For the control analysis using the resting-state fMRI data from the Human Connectome Project , a probabilistic entorhinal ROI was generated using the Freesurfer toolbox ( Augustinack et al . , 2013 ) . Freesurfer's cortical reconstruction algorithm ( ‘recon-all’ ) was used on a T1 weighted MNI template image to generate an ROI mask of the EC with a probability threshold of 90% that was binarized and converted to NiFTI file format . For the quantification of the model-based analyses of connectivity between the AT and PM networks and the manually segmented medial and lateral or anterior and posterior EC , singular-value decomposition ( SVD ) and subtraction of the mean was performed on the voxel-wise time-series within each ROI . Next , functional connectivity was estimated by means of partial correlation analysis between the time-series of groups of ROIs . Finally , Pearson's partial correlation coefficients were Fisher-Z transformed and used for across-subject comparisons ( Figure 1D ) . For the estimation of voxel-wise connectivity maps we used the ‘dual_regression’ function implemented in FSL 5 . 0 . 4 ( http://fsl . fmrib . ox . ac . uk/fsl/ ) . This involved using the time-series of pairs of seed regions as regressors in a GLM to estimate functional connectivity to voxels in a target region ( Figure 1B , Figure 1C and Figure 1—figure supplement 1 ) or to the rest of the brain ( Figure 2D and Figure 3D ) . Finally , group-level statistics were computed with non-parametric randomisation tests ( FSL 5 . 0 . 4 , http://fsl . fmrib . ox . ac . uk/fsl/ ) and threshold-free cluster enhancement for null-hypothesis testing and the computation of p-value maps ( Figure 2D and Figure 3D ) . All connectivity analyses were restricted to gray matter voxels . To overcome limitations of the seed-based connectivity analysis , we employed ConGrads ( Haak et al . , 2014 ) , which allows for tracing the dominant modes of functional connectivity change within a pre-specified region of the brain in a fully data-driven manner . First , the fMRI time series from the EC were rearranged into a time-by-voxels matrix , which was also done for the fMRI time series of all gray-matter voxels outside the EC . For reasons of stability and computational tractability , we losslessly reduced the dimensionality of the data outside the EC using SVD . We determined the connectivity fingerprints of each voxel inside the EC by computing the correlation between the voxel-wise time series and the SVD-transformed data , and then used the η2 coefficient to quantify the similarities among the voxel-wise fingerprints ( Cohen et al . , 2008 ) . Next , we fed the ensuing similarity-matrix to the LE algorithm ( Belkin and Niyogi , 2003 ) , resulting in a series of vectors that represent the dominant modes of functional connectivity change . The LE algorithm and variants thereof have previously been successfully applied to trace changes in probabilistic tractography connectivity ( Johansen-Berg et al . , 2004; Cerliani et al . , 2012 ) , while in the context of resting-state fMRI , ConGrads has been shown to generate highly reproducible results in regions such as the human motor strip , both across sessions and participants ( Haak et al . , 2014 ) . Note that group-level results were obtained by running the LE algorithm on the average of the individual similarity matrices and that the analysis was performed separately for left and right hemispheric EC ROIs . To quantify how well the ensuing modes of connectivity change preserved the order of the similarities among the original , high-dimensional connectivity fingerprints ( topology preservation ) we used Spearman's rank correlation coefficient . Because the LE algorithm maximizes topology preservation , we report the correlation coefficients without p-values . Clustering was performed on the high-resolution data from the navigation experiment by grouping voxels above and below the median value of each mode of functional connectivity change , which resulted in two equally sized clusters per mode of connectivity change . For the ROI analyses on the data of the spatial and non-spatial stimulation experiment , the clusters were warped into MNI space . tSNR was determined by dividing the mean signal within a region by the standard deviation of that signal over time . Conversely , spatial SNR was determined by dividing the mean signal within a region by the standard deviation of the signal across voxels . This was done for each time point and spatial SNR was then averaged over time . First , we determined time points where instantaneous movement ( Power et al . , 2012 , 2015 ) exceeded a threshold of 0 . 5 mm . Then we excluded these time points ( volumes ) including the one preceding and the two thereafter from subsequent analyses . Hence , per instantaneous movement above threshold , four volumes were removed . On average this resulted in 121 volumes being removed from participant's time-series ( range: 4–300 ) . The data from the spatial and non-spatial stimulation experiment were analysed in native ( subject-specific ) space with general linear models that included regressors of interest for object and scene trials . Six movement regressors were included to account for movement-related noise . The regressors were convolved with the canonical hemodynamic response function and fitted to the time-series at each voxel in a whole-brain analysis . Single-subject contrast images were first normalised to a group-specific template and then to the MNI space . Group-level statistics were computed with non-parametric randomisation tests ( FSL 5 . 0 . 4 , http://fsl . fmrib . ox . ac . uk/fsl/ ) on the contrast images ( objects vs scenes and scenes vs objects ) using variance smoothing with a 5 mm3 kernel—to improve the estimation of the variance that feeds into the final t-statistic .	In the early 1950s , an American named Henry Molaison underwent an experimental type of brain surgery to treat his severe epilepsy . The surgeon removed a region of the brain known as the temporal lobe from both sides of his brain . After the surgery , Molaison's epilepsy was greatly improved , but he was also left with a profound amnesia , unable to form new memories of recent events . Subsequent experiments , including many with Molaison himself as a subject , have attempted to identify the roles of the various structures within the temporal lobes . The hippocampus—which is involved in memory and spatial navigation—has received the most attention , but in recent years a region called the entorhinal cortex has also come to the fore . Known as the gateway to the hippocampus , the entorhinal cortex relays sensory information from the outer cortex of the brain to the hippocampus . In rats and mice the entorhinal cortex can be divided into two subregions that have distinct connections to other parts of the temporal lobe and to the rest of the brain . These are the medial entorhinal cortex , which is the subregion nearest the centre of the brain , and the lateral entorhinal cortex , which is to the left or right of the centre . For many years researchers had assumed that human entorhinal subregions were located simply to the center or to the sides of the brain . However , it was difficult to check this as the entorhinal cortex measures less than 1 cm across , which placed it beyond the reach of most brain-imaging techniques . Now , two independent groups of researchers have used a technique called functional magnetic resonance imaging to show a different picture . The fMRI data—which were collected in a magnetic field of 7 Tesla , rather than the 1 . 5 Tesla used in previous experiments—reveal that the entorhinal cortex is predominantly divided from front-to-back in humans . One of the groups—Navarro Schröder , Haak et al . —used three different sets of functional MRI data to show that the human entorhinal cortex has anterior-lateral and posterior-medial subregions . In one of these experiments , functional MRI was used to measure activity across the whole brain as subjects performed a virtual reality task: this task included some components that involved spatial navigation and other components that did not . The other group—Maass , Berron et al . —used the imaging data to show that the pattern of connections between the anterior-lateral subregion and the hippocampus was different to that between the posterior-medial subregion and the hippocampus . The discovery of these networks in the temporal lobe in humans will help to bridge the gap between studies of memory in rodents and in humans . Given that the lateral entorhinal cortex is one of the first regions to be affected in Alzheimer's disease , identifying the specific properties and roles of these networks could also provide insights into disease mechanisms .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	Functional topography of the human entorhinal cortex
While the mammalian macrophage phenotypes have been intensively studied in vitro , the dynamic of their phenotypic polarization has never been investigated in live vertebrates . We used the zebrafish as a live model to identify and trail macrophage subtypes . We generated a transgenic line whose macrophages expressing tumour necrosis factor alpha ( tnfa ) , a key feature of classically activated ( M1 ) macrophages , express fluorescent proteins Tg ( mpeg1:mCherryF/tnfa:eGFP-F ) . Using 4D-confocal microscopy , we showed that both aseptic wounding and Escherichia coli inoculation triggered macrophage recruitment , some of which started to express tnfa . RT-qPCR on Fluorescence Activated Cell Sorting ( FACS ) -sorted tnfa+ and tnfa− macrophages showed that they , respectively , expressed M1 and alternatively activated ( M2 ) mammalian markers . Fate tracing of tnfa+ macrophages during the time-course of inflammation demonstrated that pro-inflammatory macrophages converted into M2-like phenotype during the resolution step . Our results reveal the diversity and plasticity of zebrafish macrophage subsets and underline the similarities with mammalian macrophages proposing a new system to study macrophage functional dynamic . Behind the generic name ‘macrophage’ hides various cell types with distinct phenotypes and functions . Currently , it is well established that macrophages are not just important immune effector cells but also cells with critical homeostatic roles , exerting a myriad of functions in development , homeostasis , and tissue repair and playing a pivotal role in disease progression ( Wynn et al . , 2013 ) . Therefore , there is a high interest in a better characterization of these cells to establish an early and accurate diagnosis . The wide variety of macrophage functions might be explained by the outstanding plasticity and versatility of macrophages that efficiently respond to environmental challenges and changes in tissue physiology by modifying their phenotype ( Mosser and Edwards , 2008 ) . Although there is a consensus that macrophages are a diversified set of cells , macrophage subtypes are still poorly characterized . Indeed , although these cell populations have been extensively investigated in mouse and human , these studies were mostly performed in vitro using monocyte-derived macrophages induced under specific stimuli . A comprehensive characterization of macrophage subsets that takes into account their specific behaviour , phenotypic diversity , functions , and modulation shall rely on a real-time tracking in the whole organism in response to environmental challenges . Mouse and human macrophages have been classified according to their polarization state . In this classification , M1 macrophages , also referred as classically activated macrophages , are pro-inflammatory cells associated with the first phases of inflammation , while M2 macrophages , also known as alternatively activated macrophages , are involved in the resolution of inflammation and tissue remodelling ( Gordon , 2003; Biswas and Mantovani , 2010; Sica and Mantovani , 2012 ) . Differential cytokine and chemokine production and receptor expression define the polarization state of macrophages . However , it is worthwhile to note that such binary naming does not fully reflect the diversity of macrophage phenotypes in complex in vivo environments in which several cytokines and growth factors are released and adjust the final differentiated state ( Chazaud , 2013; Thomas and Mattila , 2014 ) . Macrophages might adopt intermediate activation phenotypes classified by the relative levels of macrophage subset-specific markers . Therefore , macrophage plasticity results in a full spectrum of macrophage subsets with a myriad of functions ( Mosser and Edwards , 2008; Xue et al . , 2014 ) . Although the possible phenotype conversion of macrophages from M1 to M2 has been suggested in in vitro studies , a recent study argues for the sequential homing of M1 and M2 macrophages to the site of injury ( Stout et al . , 2005; Sica and Mantovani , 2012; Shechter et al . , 2013 ) . Such controversies highlight the lack of accurate real-time tracing of macrophage subtypes in vivo in the entire animal . Inflammation is a model of choice to study the wide range of macrophage subsets involved from its initiation to its resolution . Therefore , in the present study , we propose to decipher in vivo in real time the kinetic of macrophage subset recruitment , their behaviour and their phenotypic plasticity at the molecular level during a multiple-step inflammatory process . We used the zebrafish larvae model for its easy genetic manipulation , transparency , and availability of fluorescent reporter lines to track macrophages ( Ellett et al . , 2011 ) . While the existence of macrophage subtypes in zebrafish embryos has been suggested , they have not been fully characterized ( Herbomel et al . , 1999; Ellett et al . , 2011; Cambier et al . , 2013; Petrie et al . , 2014 ) . Here , we report a new reporter transgene for TNFa , a central inflammatory cytokine and well-established marker of M1 macrophages , instrumental to discriminate macrophage subsets during intravital imaging . Fin wounding-induced inflammation and Escherichia coli inoculation in zebrafish larvae of 3 dpf are two well-established models triggering macrophage recruitment . Using in situ hybridization , we observed that the expression of the tumour necrosis factor alpha ( tnfa ) , a consensus marker of M1 macrophages , was induced in cells accumulated in the caudal fin and the muscle following amputation ( nlarvae = 29/33 ) and E . coli inoculation ( nlarvae = 12/12 ) , respectively ( Figure 1A–E ) . To study the cells that express the tnfa transcripts , we established the Tg ( tnfa:eGFP-F ) transgenic zebrafish line expressing a farnesylated ( membrane-bound ) eGFP ( eGFP-F ) under the control of the tnfa promoter . While eGFP-F was undetectable in intact fins of Tg ( tnfa:eGFP-F ) larvae ( nlarvae = 10/10 ) , it was expressed in cells recruited to the wound at 6 hr post-amputation ( hpA ) ( nlarvae = 16/16 , Figure 1F , G ) . Similarly , eGFP-F expression was upregulated in cells accumulated in the muscle of Tg ( tnfa:eGFP-F ) larvae at 16 hr post-inoculation ( hpi ) of DsRed-expressing E . coli ( nlarvae = 8/8 ) , compared to Phosphate-buffered saline ( PBS ) injection ( nlarvae = 3/3 , Figure 1H , I ) . Confocal analysis confirmed the presence of a membrane-bound eGFP in cells displaying a typical myeloid morphology ( Figure 1J ) . To demonstrate that the Tg ( tnfa:eGFP-F ) line recapitulates transcriptional activation of tnfa , we performed a simultaneous detection of tnfa mRNA by in situ hybridization and GFP-F protein by immunofluorescence in amputated larvae 6 hpA . We observed a consistent overlap between tnfa and GFP-F signal in the fin ( nlarvae = 11/11 ) , showing the direct correlation of eGFP-F and tnfa transcriptional activation in the fin of the reporter line ( Figure 1K ) . In addition , we FACS-sorted GFP+ cells from wounded Tg ( TNFa:eGFP-F ) larvae 6 hpA and performed RT-qPCR to analyze tnfa expression . We observed a significant increase of tnfa mRNA level in eGFP+ cells as compared to eGFP− cells ( Figure 1L , M ) . All together these results indicate that the Tg ( TNFa:eGFP-F ) reporter line recapitulates transcriptional activation of tnfa . Then , with the ability to specifically track tnfa-expressing cells , we used Tg ( tnfa:eGFP-F ) fish to study macrophage activation by mating them with Tg ( mpeg1:mCherryF ) fish in which macrophages express farnesylated mCherry ( mCherryF ) under the control of the macrophage-specific mpeg1 promoter ( Ellett et al . , 2011; Nguyen-Chi et al . , 2014 ) . In intact Tg ( tnfa:eGFP-F/mpeg1:mCherryF ) larvae , no eGFP-F was observed in macrophages ( Figure 2A ) . We imaged double transgenic larvae Tg ( tnfa:eGFP-F/mpeg1:mCherryF ) using 4D confocal microscopy from 45 min post-amputation and found that macrophages were recruited to the wound from 1 hpA , some starting to express eGFP from 3 hpA ( Figure 2B , C and Video 1 ) . From 5 hpA already activated macrophages , that is , expressing tnfa arrived at the wound ( Video 1 and Figure 2C ) . Similarly , infection with a crimson-expressing E . coli in the muscle induced the expression of tnfa in phagocytes few hours following the infection ( Figure 2—figure supplement 1 , Video 2 ) . Imaging of the double transgenic larvae Tg ( tnfa:eGFP-F/mpeg1:mCherryF ) showed that tnfa-expressing phagocytes were mainly macrophages ( Figure 2—figure supplement 1 , Video 2 ) . These results show the dynamic macrophage activation in real-time in vivo including recruitment and rapid phenotypic change . During the revision of this paper , a similar result has been published ( Sanderson et al . , 2015 ) . 10 . 7554/eLife . 07288 . 003Figure 1 . The ( tnfa:eGFP-F ) reporter line recapitulates transcriptional activation of tnfa upon wound-induced inflammation and Escherichia coli infection . ( A–E ) Tumour necrosis factor alpha ( tnfa ) mRNA expression ( blue , arrowhead ) was detected by in situ hybridization using tnfa anti-sense probe: at 6 hpA in ( A ) intact ( control ) and ( B ) amputated fins from 3 dpf WT larvae , ( C ) in uninfected larvae ( 54 hpf , hours post-fertilization ) and ( D , E ) E . coli infected larvae ( 24 hpi , 54 hpf ) . Arrows show melanocytes ( black ) . ( E ) Imaging of tnfa mRNA expression in the muscle at higher magnification , asterisks show muscle fibres , scale bar in ( B ) = 100 μm and in ( E ) = 50 μm . ( F , G ) eGFP fluorescence ( green ) was analyzed by fluorescent microscopy in ( F ) intact ( control ) and ( G ) amputated Tg ( tnfa:eGFP-F ) fins at 6 hpA , dotted lines outline the caudal fin , scale bar = 100 μm and at 16 hpi in Tg ( tnfa:eGFP-F ) larvae injected with ( H ) PBS or ( I , J ) E . coli ( red ) in the muscle . Arrows show auto-fluorescent xanthophores . ( J ) Multi-scan confocal analysis of GFP expression in E . coli-infected Tg ( tnfa:eGFP-F ) larvae , scale bar = 20 μm . ( K ) tnfa mRNA and eGFP-F expressions were analyzed using microscopy at 6 hpA in amputated fins from 3 dpf Tg ( tnfa:eGFP-F ) larvae . Dotted lines delimit the caudal fin , arrowheads show overlapping signals , and arrows show the pigments . Scale bar = 100 μm . ( L ) Graphed data of representative fluorescence-activated flow cytometry analysis of eGFP+ cells in upon amputation . Tg ( tnfa:eGFP-F ) larvae were either kept intact ( control ) or amputated at 3 dpf , and cells were collected at 6 hr post-treatment . Green gates represent eGFP+ population and mean percentage of eGFP+ population ±s . e . m is indicated . ( M ) Relative expression of tnfa in eGFP- and GFP+ cells in amputated larvae . Real-time RT-PCR on separated cells using EF1a as a reference gene . Graph represents the mean value of three independent experiments ±s . e . m . p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 07288 . 00310 . 7554/eLife . 07288 . 004Figure 2 . Activation , morphology , and behaviour of TNF-α+ macrophages in ( tnfa:eGFP-F/mpeg1:mCherry-F ) transgenic larvae upon wound-induced inflammation . ( A ) eGFP-F ( green ) and mCherryF ( red ) fluorescence was analyzed by fluorescent microscopy in intact ( control ) and amputated Tg ( mpeg1:mCherryF/tnfa:eGFP-F ) fins at 6 hpA of 3 dpf larvae . Arrowheads show recruited macrophages that express tnfa , arrows show tnfa+ cells that are not macrophages , and asterisks show auto-fluorescent pigments . Dotted lines outline the caudal fin , scale bar = 100 μm . ( B ) Bright-field image of the wounded fin of a 3 dpf Tg ( mpeg1:mCherryF/tnfa:eGFP-F ) larva . Dotted red box shows the region imaged in C . ( C ) Representative time-lapse maximum projections show the activation of macrophages arriving at the wound in 3 dpf amputated Tg ( mpeg1:mCherryF/tnfa:eGFP-F ) . The time pA is shown on top right corner and indicated in hours and minutes , white lines outline the caudal fin . The transcriptional activation of tnfa ( green ) in recruited macrophage ( red , arrowhead ) was first observed from 3 hpA . Scale bar = 30 μm . White lines outline the caudal fin . ( D , E ) Maximum projections of confocal analysis of eGFP-F ( green ) and mCherryF ( red ) expressions in recruited macrophages at ( D ) 18 hpA and ( E ) 24 hpA in Tg ( mpeg1:mCherryF/tnfa:eGFP-F ) . tnfa+mpeg1+ macrophages exhibit a round and protrusive morphology , while tnfa−mpeg1+ macrophages exhibit a dendritic morphology . ( F ) Velocity of tnfa+mpeg1+ and tnfa−mpeg1+ macrophages ( N = 18 ) . ( G ) Frequency of macrophage–macrophage contacts and ( H ) time length of the contacts of tnfa−mpeg1+ and tnfa+mpeg1+ cells . Measurements were extracted from three independent videos of amputated Tg ( mpeg1:mCherryF/tnfa:eGFP-F ) , for contact frequency , N = 15 and for duration of the interaction , N = 11 macrophages . **p < 0 . 0001 . ( I ) Representative time-lapse maximum projections show the behaviour of tnfa+mpeg1+ macrophages , starting 19h20 pA during 42 min . Two macrophages ( green + red ) interact by cell–cell contact . These macrophages ( eGFP in grey ) remain attached up to 40 min . Scale bar 20 = μm . ( J ) Representative time-lapse maximum projections show the behaviour of tnfa−mpeg1+ macrophages , starting 25h12 pA during 52 min . Macrophages ( red ) barely establish cell–cell contact . Scale bar = 30 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07288 . 00410 . 7554/eLife . 07288 . 005Figure 2—figure supplement 1 . Activation of tnfa+ macrophages in ( tnfa:eGFP-F/mpeg1:mCherry-F ) transgenic larvae upon E . coli infection . ( A ) Diagram showing the site where Crimson E . coli or non-fluorescent E . coli ( blue ) were injected in the muscle of 3 dpf Tg ( mpeg1:mCherryF/tnfa:eGFP-F ) larvae . The red dotted box represents the region imaged by confocal microscopy . ( B ) Tg ( mpeg1:mCherryF/tnfa:eGFP-F ) larvae were infected with crimson-expressing E . coli ( blue ) at 3 dpf in the muscle and imaged from 30 min pi to 10 hr 30 min pi every 3 min 30 s . Representative time-lapse maximum projections show the expression of tnfa ( green ) induced in myeloid-like cells at the infection site from 3 hpA . The time pA is shown on top right corner , scale bar = 25 μm . ( C ) Maximum projections of confocal analysis of GFP-F ( green ) and mCherryF ( red ) expressions in Tg ( mpeg1:mCherryF/tnfa:eGFP-F ) . Larvae were previously infected with E . coli at 3 dpf in the muscle and imaged at 6 hpi . Arrowheads show macrophages expressing tnfa . Scale bar on left panels = 20 μm and on right panels = 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07288 . 00510 . 7554/eLife . 07288 . 006Video 1 . Transcriptional activation of tnfa in macrophages of ( tnfa:eGFP-F/mpeg1:mCherry-F ) transgenic larvae upon amputation . Representative time-lapse maximum projections show the transcriptional activation of tumour necrosis factor alpha ( tnfa ) in macrophages arriving at the wound in 3 dpf amputated Tg ( mpeg1:mCherryF/tnfa:eGFP-F ) . The time pA is shown on top right corner , white line outline the caudal fin . Scale bar = 30 μm . Image stacks were acquired every 3 min 30 s from 45 min pA to 7 hr 48 min pA at 2-μm intervals , 1024 × 512 pixel resolution using a confocal microscope TCSSP5 SP5 inverted equipped with a HCXPL APO 40×/1 . 25–0 . 75 oil objective ( Leica ) . Excitation wavelengths used were 488 nm for EGFP-F and 570 nm for mCherryF . DOI: http://dx . doi . org/10 . 7554/eLife . 07288 . 00610 . 7554/eLife . 07288 . 007Video 2 . Transcriptional activation of tnfa of ( tnfa:eGFP-F/mpeg1:mCherry-F ) transgenic larvae upon E . coli infection . Tg ( mpeg1:mCherryF/tnfa:eGFP-F ) larvae were infected with crimson-expressing E . coli ( blue ) at 3 dpf in the muscle and imaged from 30 min pi to 10 hr 30 min pi . Representative time-lapse maximum projections show the expression of tnfa ( green ) induced in myeloid-like cells at the infection site from 3 hpA . The time pA is shown on top right corner , scale bar = 25 μm . Image stacks were acquired every acquired 3 min 30 s at 2-μm intervals , 512 × 512 pixel resolution with a X2 zoom using a confocal microscope TCSSP5 SP5 inverted equipped with a APO 20× objective ( Leica ) . Excitation wavelengths used were 488 nm for EGFP-F and 580 nm for Crimson . To distinguish Crimson from mCherry , emission filter was adjusted from 630 to 750 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 07288 . 007 To test whether tnfa+ and tnfa− macrophages harboured different cellular characteristics , we first analyzed their morphology in fin-wounded Tg ( tnfa:eGFP-F/mpeg1:mCherryF ) larvae . tnfa+mpeg1+ cells displayed flattened and lobulated morphology ( Figure 2D ) , while tnfa−mpeg1+ were elongated and dendritic ( Figure 2E ) . As we observed that tnfa+mpeg1+ cells were predominant at the wound at 18 hpA and tnfa−mpeg1+ cells at 24 hpA ( data not shown ) , we imaged the behaviour of these macrophage populations in wounded fins from Tg ( tnfa:eGFP-F/mpeg1:mCherryF ) larvae at these time points . tnfa+mpeg1+ cells presented a lower velocity ( 0 . 32 μm/min ) than tnfa−mpeg1+ macrophages ( 1 . 09 μm/min , Figure 2F ) but a higher cell–cell contact frequency ( 0 . 036 VS 0 . 016 contacts/min ) with other macrophages ( Figure 2G , I , J and Videos 3 , 4 ) . Measurements of the duration of macrophage–macrophages contacts showed that these contacts lasted longer ( 48 . 6 min/contact ) than that of tnfa−mpeg1+ macrophages ( 15 . 9 min/contact , Figure 2H–J and Videos 3 , 4 ) . All together these data highlight different morphology and behaviour of macrophage phenotypes in live zebrafish suggesting the existence of macrophage subsets exhibiting different functions . 10 . 7554/eLife . 07288 . 008Video 3 . tnfa+ macrophage behaviour following amputation . Representative time-lapse maximum projections show the behaviour of tnfa+mpeg1+ macrophages in Tg ( mpeg1:mCherryF/tnfa:eGFP-F ) larvae following amputation , starting 19h20 pA during 42 min . Two macrophages tnfa+mpeg1+ interact by cell–cell contact . These macrophages ( GFP in grey ) remain attached up to 40 min . Scale bar = 20 μm . Image stacks were acquired every 3 min 30 s at 2 μm-intervals , 1024 × 512 pixel resolution using a confocal microscope TCSSP5 SP5 inverted equipped with a HCXPL APO 40×/1 . 25–0 . 75 oil objective ( Leica ) . Excitation wavelengths used were 488 nm for EGFP-F and 570 nm for mCherryF . DOI: http://dx . doi . org/10 . 7554/eLife . 07288 . 00810 . 7554/eLife . 07288 . 009Video 4 . tnfa− macrophage behaviour following amputation . Tg ( mpeg1:mCherryF/tnfa:eGFP-F ) caudal fins were amputated at 3 dpf . Representative time-lapse maximum projections show the behaviour of tnfa−mpeg1+ macrophages , starting 25h12 pA during 42 min . Macrophages ( red ) scarcely establish cell–cell contact . Scale bar = 30 μm . Image stacks were acquired every 4 min at 2 μm-intervals at 1024 × 512 pixel resolution using a confocal microscope TCSSP5 SP5 inverted equipped with a HCXPL APO 40×/1 . 25–0 . 75 oil objective ( Leica ) . Excitation wavelengths used were 488 nm for EGFP-F and 570 nm for mCherryF . DOI: http://dx . doi . org/10 . 7554/eLife . 07288 . 009 To quantify the respective frequency of tnfa+ macrophages ( mCherry+eGFP+ referred as dbl+ ) and tnfa− macrophages ( mCherry+eGFP− referred as mCh+ ) , we performed flow cytometry analysis on cells isolated from Tg ( tnfa:eGFP-F/mpeg1:mCherryF ) larvae at different time points following caudal fin amputation or E . coli . inoculation ( Figure 3A , B ) . While only 5 . 6% ± 0 . 9 ( s . e . m . ) dbl+ cells were detected in the mpeg1+ population of the intact larvae , a steady increase of the dbl+ population from 6 to 20 hpA ( up to 27 . 33 ± 0 . 2% ) was observed . This percentage decreased dramatically at 26 hpA to 8 . 75 ± 1% . In E . coli inoculation experiments , the frequency of dbl+ cells increased as soon as 3 hpi ( 55 . 60 ± 0 . 6% ) and remained stable until 26 hpi . These results demonstrate that wound-induced macrophage activation is transient compared to infection-induced macrophage activation . 10 . 7554/eLife . 07288 . 010Figure 3 . Isolation and molecular characterization of macrophage phenotypes . ( A ) Graphed data of representative fluorescence-activated flow cytometry analysis of tnfa+ and tnfa− macrophages upon inflammatory stimulations . Tg ( mpeg1:mCherryF/tnfa:eGFP-F ) larvae were either kept intact ( control ) , or amputated , or injected with PBS or with E . coli at 3 dpf , and cells were collected at 6 hr post-treatment . Red , green , and yellow gates represent mCherry+ , eGFP+ , and mCherry+eGFP+ populations , respectively . ( B ) Graph represents the kinetic of the frequency of mpeg1+tnfa+ macrophages in macrophage population ( mpeg1+ ) in three independent experiments following stimulation: amputation and E . coli infection ( E . coli ) at indicated time points . p < 0 . 05 vs 3 hpA , mean value of three experiments ±s . e . m . ( C ) Gating strategy to isolate control cells ( mCherry− eGFP− , neg ) , tnfa− macrophages ( mCherry+ eGFP− , mCh+ ) , tnfa+ macrophages ( mCherry+eGFP+ , dbl+ ) . ( D–H ) Relative expression of ( D ) mpeg1 , ( E ) tnfa , ( F ) tnfb , il1b , ( G ) il6 , ( H ) tgfb1 , ccr2 , and cxcr4b in cells neg , mCh+ , and dbl+ . Tg ( mpeg1:mCherryF/tnfa:eGFP-F ) were amputated at 3 dpf and cells were collected and separated at 6 hpA and 26 hpA . Real-time RT-PCR on separated cells using EF1a as a reference gene . Graph represents the mean value of five independent experiments ±s . e . m . Statistical significance between bars are indicated p < 0 . 05 , p < 0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 07288 . 010 To characterize at the molecular level tnfa− and tnfa+ macrophage populations during early and late phases of inflammation , we FACS-sorted dbl+ and mCh+ cells from Tg ( tnfa:eGFP-F/mpeg1:mCherryF ) tail-amputated larvae ( Figure 3C ) and analyzed them by qRT-PCR . In mammals , M1 and M2 macrophages are reported to be involved , respectively , in the initial phase of inflammation and in the resolution phase . Cell sorting was thus performed at 6 hpA and 26 hpA following caudal fin amputation since the kinetic analysis of macrophage subset activation ( Figure 3B ) suggested that these two time points correspond to initiation and resolution of inflammation , respectively . As expected , high levels of mpeg1 expression was observed in the mCh+ and dbl+ sorted cells at 6 and 26 hpA ( Figure 3D ) , and high levels of tnfa expression was detected in double-positive populations at 6 hpA ( Figure 3E ) . These observations demonstrated that fluorescence of these transgenes can be efficiently used to track and separate macrophage sub-populations . At 6 hpA , dbl+ macrophages expressed high levels of tnfb , il1b , and il6 compared to mCh+ macrophages ( Figure 3F , G ) that are well-known markers of M1 macrophages in mammals ( Mantovani et al . , 2002; Martinez et al . , 2006 ) . By contrast , mCh+ macrophages expressed low levels of these pro-inflammatory cytokines at both 6 and 26 hpA ( Figure 3F , G ) , but expressed high levels of tgfb1 , ccr2 , and cxcr4b ( Figure 3H ) , that are specifically expressed in mammalian M2 macrophages ( Mantovani et al . , 2002; Martinez et al . , 2006; Hao et al . , 2012; Beider et al . , 2014; Machado et al . , 2014 ) . Of note , neither Arginase 1 ( Arg1 ) , which is largely used as a M2 marker in mouse but not in human ( Chinetti-Gbaguidi and Staels , 2011; Pourcet and Pineda-Torra , 2013 ) , nor il10 ( data not shown ) , a known M2 marker in mammals ( Mantovani et al . , 2002 ) , was detected in zebrafish macrophages . Importantly , based on the stability of the eGFP in Tg ( tnfa:eGFP-F/mpeg1:mCherryF ) larvae allowing us to specifically track the behaviour and fate of pro-inflammatory macrophages , we found that the dbl+ pro-inflammatory macrophages changed their phenotype at 26 hpA . Indeed , dbl+ macrophages negative for M2 markers at 6 hpA , displayed at 26 hpA , in parallel to a significant decrease of tnfa , il1b , and il6 expression level , a significant increased expression level of ccr2 and cxcr4b ( Figure 3E , H ) . Of note , a tendency toward differential expression level was observed for tnfb and tgfb1 between 6 and 26 hpA . To go further and demonstrate that the same macrophages are present at the wound site during inflammation and its resolution , we generated the Tg ( mpeg1:GAL4/UAS:Kaede ) larvae to track macrophages exploiting the conversion of the native green fluorescence of Kaede into red fluorescence under UV light . Recruited macrophages were photoconverted 6 hpA and imaged at 26 hpA revealing that early recruited macrophages were still present at the wound area 20 hr later ( Figure 4—figure supplement 1 ) . Then , GFP+ macrophages were specifically tracked using time-lapse imaging of wounded Tg ( mpeg1:mCherryF/TNFa:GFP-F ) fins from 6 to 26 hpA . We show that initially recruited eGFP+ macrophages remain at the injury site and still express the GFP ( Figure 4A–C and Video 5 ) . The analysis of macrophage behaviour over time shows that among eGFP+ macrophages displaying an amyboid phenotype at the wound edge 6 hpA , 50% change toward a fibroblastic phenotype from 11 hpA when they moved distally ( Video 5 ) . All together these data show that pro-inflammatory macrophages underwent a phenotypic conversion toward an intermediate phenotype in which both M1 and M2 markers are expressed . In addition , this molecular characterization of macrophages in zebrafish reveals the conservation of macrophage subtypes between zebrafish and human . 10 . 7554/eLife . 07288 . 011Figure 4 . M1-like macrophages convert their phenotype toward M2-like phenotype in the wounded fin . ( A ) Diagram showing the site where caudal fin was transected ( dotted red line ) in 3 dpf Tg ( mpeg1:mCherryF/tnfa:eGFP-F ) larvae . The black dotted box represents the region imaged by confocal microscopy . ( B ) Representative time-lapse maximum projections of 3 dpf Tg ( mpeg1:mCherryF/tnfa:eGFP-F ) amputated fins showing the fate of tnfa+ macrophages ( magenta + green ) at the indicated times pA ( hours:minutes ) from 6 hpA to 26 hpA . White lines delimit the caudal fin . Scale bar = 30 μm . ( C ) Tracking of tnfa+ macrophages from 6 to 26 hpA . The distinct colours of the lines correspond to the distinct macrophages that were indicated with an arrowhead in B . ( D ) Diagram representing macrophage activation and polarization in zebrafish . Unpolarized macrophages ( mpeg1+ ) are mobilized and recruited to the wound following fin amputation . They are activated and polarized toward a M1-like phenotype ( pro-inflammatory ) few hours following fin amputation . After 24 hpA , in response to changes in environmental cues , the same macrophages progressively change their phenotype toward intermediate phenotypes and maybe fully polarized M2-like phenotype ( non-inflammatory ) . Main markers of macrophage subtypes are indicated and resemble those found in human ( tnfa/b indicates tumour necrosis factor alpha; il1b , interleukin 1-beta; il6 , interleukin 6; tgfb1 , tumour growth factor beta 1; ccr2 , c–c chemokine receptor type 2; cxcr4b , chemokine ( C-X-C motif ) receptor 4b ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07288 . 01110 . 7554/eLife . 07288 . 012Figure 4—figure supplement 1 . Recruited macrophages remain in the region of tissue injury at 26 hpA . ( A ) Diagram showing the site where caudal fin was transected ( dotted red line ) in 3 dpf Tg ( mpeg1:GAL4/UAS:Kaede ) larvae . The black dotted box represents the region imaged by confocal microscopy . ( B ) Representative maximum projections of confocal analysis of Kaede ( green and red ) expression in recruited macrophages at 6 hpA in Tg ( mpeg1:GAL4/UAS:Kaede ) fins . Images were acquired before photoconversion at 6 hpA ( left panel ) , immediately after photoconversion at 6 hpA ( middle panel ) , and at 26 hpA ( right panel ) . The blue dashed boxes represent the region that was scanned using UV laser allowing photoconversion of few macrophages at the wound leaving them with red fluorescence rather than green . Arrowheads show the traced and photoconverted macrophages at 26 hpA and white dotted line outline the amputated fin . ( ncells = 13 , nlarvae = 5 ) Scale bar = 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 07288 . 01210 . 7554/eLife . 07288 . 013Video 5 . Recruited GFP+ macrophages persist in the region of tissue injury at 26 hpA . Tg ( mpeg1:mCherryF/tnfa:eGFP-F ) caudal fins were amputated at 3 dpf . Representative time-lapse maximum projections show the movements of tnfa+mpeg1+ macrophages ( green + magenta ) , starting 6 hpA to 26 hpA . Coloured lines correspond to eGFP+ macrophage tracking that stay in the wounded fin and still express eGFP at 26 hpA . Scale bar = 30 μm . Image stacks were acquired every 4 min 40 s at 3 . 5-μm intervals at 512 × 512 pixel resolution using a confocal microscope TCSSP5 SP5 inverted equipped with a HCXPL APO 40×/1 . 25–0 . 75 oil objective ( Leica ) . Excitation wavelengths used were 488 nm for EGFP-F and 570 nm for mCherryF . DOI: http://dx . doi . org/10 . 7554/eLife . 07288 . 013 In conclusion , we identified macrophage subsets in zebrafish and described their behaviour and fate during a process of inflammation ( Figure 4D ) . Live imaging of transparent transgenic zebrafish larvae allowed the first real-time visualization of macrophage activation and polarization . In parallel , a molecular analysis of macrophage sub-populations highlights the evolutionary conservation of macrophages from fish to mammals . We propose that in response to wounding zebrafish , unpolarized macrophages are recruited to the inflammation site and adopt a M1-like phenotype . Subsequently , they progressively convert their functional phenotype from M1-like to M2-like in response to progressive inflammatory microenvironment changes within the tissue ( Figure 4D ) . Live imaging of the new transgenic line we generated opens new avenues to study in real time in live vertebrates the full spectrum of macrophage activation , polarization , and functions . All animal experiments described in the present study were conducted at the University of Montpellier according to European Union guidelines for handling of laboratory animals ( http://ec . europa . eu/environment/chemicals/lab_animals/home_en . htm ) and were approved by the Direction Sanitaire et Vétérinaire de l'Hérault and Comité d'Ethique pour l'Expérimentation Animale under reference CEEA-LR-13007 . Fish and embryo maintenance , staging , and husbandry were as previously described ( Nguyen-Chi et al . , 2014 ) . Experiments were performed using the AB zebrafish strain ( ZIRC ) and the transgenic line Tg ( mpeg1:mCherryF ) to visualize macrophages . For the photoconversion experiments , a cross of Tg ( mpeg1:Gal4 ) gl25 and Tg ( UAS:kaede ) rk8 lines was used , using breeders selected for progeny with negligible silencing of the UAS transgene . The TNFa promoter ( Gene ID: 405785 ) was amplified from zebrafish genomic DNA using primers zTNFaP4 ( CCCGCATGCTCCACGTCTCC ) and zTNFaE11N ( TTATAGCGGCCGCCCGACTCTCAAGCTTCA ) . The resulting fragment was phosphorylated using T4PNK , digested by NotI and cloned in a farnesylated eGFP ( eGFP-F ) derivative of pBSKI2 ( Thermes et al . , 2002 ) . The resulting plasmid ( pI2promTNFa:eGFP-F ) harbours a 3 . 8-kb fragment of the zebrafish tnfa promoter , including part of the first coding exon . It uses the endogenous ATG codon of tnfa to drive the translation of eGFP-F . The expressed eGFP-F harbours the first 7 amino acids of zebrafish TNFa at its N-terminus ( MKLESRA ) . The expression cassette is flanked by two I-SceI sites . pI2promTNFa:eGFP-F was co-injected in fertilized eggs with the enzyme I-SceI ( New England Biolabs , France ) . Developing embryos were injected with non-pathogenic E . coli at 3 dpf ( days post-fertilization ) , and those that developed a specific green fluorescence were raised as putative founders . The offspring of putative founders was tested the same way in order to establish the stable transgenic line . Low expression of eGFP in the pharynx at 3 dpf was used to check larvae for the presence of the transgene . Caudal fin amputation was performed on 3 dpf larvae as described in Pase et al . ( 2012 ) . The caudal fin was transected with a sterile scalpel , posterior to muscle and notochord under anaesthesia with 0 . 016% Tricaine ( ethyl 3-aminobenzoate , Sigma Aldrich , France ) in zebrafish water . Larvae were inoculated at 3 dpf by 2 . 103 CFU E . coli K12 bacteria harbouring either DsRed ( van der Sar et al . , 2003 ) or Crimson ( Clontech , France ) expression plasmid or no plasmid . Imaging was performed as previously described ( Nguyen-Chi et al . , 2014 ) using a confocal TCS SP5 inverted microscope with a HCXPL APO 40×/1 . 25–0 . 75 oil objective ( Leica , France ) . Image stacks for time-lapse videos were acquired every 3–5 min , typically spanning 30–60 μm at 2-μm intervals , 1024 × 512 or 512 × 512 pixel resolution . The 4D files generated from time-lapse acquisitions were processed using Image J . They were compressed into maximum intensity projections and cropped . Brightness , contrast , and colour levels were adjusted for maximal visibility . Velocity of macrophages was measured using Manual Tracking Image J plugin . Frequency of macrophage–macrophage interaction and duration of interactions were measured manually on stack images . For tracking of macrophages , eGFP-F+ mCherryF+ cells from Tg ( mpeg1:mCherryF/tnfa:eGFP-F ) wounded fins were tracked using time-lapse image series from 6 hpA ( hours post-amputation ) to 26 hpA . 300 Tg ( tnfa:eGFP-F/mpeg1:mCherryF ) larvae were either amputated or infected as described above , then crushed on a 70-µm cell strainer ( Falcon , France ) . Isolated cells were washed in PBS/2 mM ethylenediaminetetraacetic acid ( EDTA ) /2% Foetal Calf Serum ( FCS ) , filtered through a 40-µm cell strainer , and counted . Counting of mCherry+eGFP− and mCherry+eGFP+ cells was performed on LSRFortessa ( BD Bioscience , France ) , and data analyzed using the Flowjo software ( Tree start , Ashland , Or , USA ) . Sorting was done using FACS ARIA ( BD Bioscience , France ) and collected in 50% FCS/50% Leibovitz L-15 medium ( 21083-027 , Gibco , France ) on ice . To isolate total RNA , cells were lysed in QIAzol Lysis Reagent ( Qiagen , France ) and RNA extracted using miRNeasy mini kit ( Qiagen-21704 , France ) . 20 ng of total RNA was reverse transcribed using High Capacity RNA Reverse Transcription kit ( Applied Biosystems , France ) . QPCR were performed on a LightCycler 480 system ( Roche , France ) , following manufacturer's instructions ( SYBR Green format , Roche Applied Science , Meylan , France ) and using primers in Supplementary file 1: denaturation 15 s at 95°C , annealing 10 s at 64°C , and elongation 20 s at 72°C . Expression levels were determined with the LightCycler analysis software ( version-3 . 5 ) from 5 independent experiments . The relative amount of a given mRNA was calculated by using the formulae 2−∆Ct with ef1a as reference . Significance testing for Figures 1M , 3D–H was done using Mann–Whitney unpaired t-test , one-tail and Figure 2F–H using Mann–Whitney unpaired t-test , two-tails using GraphPad Prism 6 Software . p < 0 . 05 , p < 0 . 01 , **p < 0 . 0001 . A tnfa probe was amplified from total cDNA by PCR using tnfa . 55 and tnfa . 58 primers ( Supplementary file 1 ) and cloned in plasmid pCRII-TOPO . Digoxigenin ( DIG ) -labelled ( Roche , France ) sense and anti-sense RNA probes were in vitro transcribed ( Biolabs , France ) . In situ hybridizations on whole-mount embryos were as previously described ( Nguyen-Chi et al . , 2012 ) . For simultaneous detection of eGFP-F proteins and tnfa mRNA by immuno-detection and in situ hybridization , fixed and rehydrated Tg ( tnfa:eGFP-F ) larvae were permeabilised in ice in 100% ethanol for 5 min , then in a mixture of 50% Xylene-50% ethanol for 1 hr and in 80% acetone for 10 min at −20°C as described in Nagaso et al . ( 2001 ) . After washes in PBS-0 . 1% Tween , larvae were post-fixed in 4% paraformaldehyde ( PFA ) for 20 min . Subsequent steps of hybridization , washes , and staining with NBT-BCIP ( Roche , France ) were as previously described in Nguyen-Chi et al . ( 2012 ) . Next , unspecific-binding sites were saturated in PBS-1% bovin serum albumin ( BSA ) -1% lamb serum-10% Goat serum and larvae incubated 3 days with an anti-GFP antibody ( MBL , 1/500 ) . After extensive washes , larvae were incubated with a goat anti-rabbit antibody . Stained embryos were imaged using a MVX10 Olympus microscope with MVPLAPO 1× objective and XC50 camera and using a Zeiss Axioimager with a Zeiss 40× Plan-Apo 1 . 3 oil objective . Tg ( mpeg1:GAL4/UAS:Kaede ) embryos were raised to 3 dpf in the dark , and caudal fin was transected as described above . At 6 hpA , larvae were mounted in 1% low-melting point agarose . A 405-nm Laser Cube 405-50C on a confocal TCS SP5 inverted microscope with a HCXPL APO 40×/1 . 25–0 . 75 oil objective ( Leica ) was used to photoconvert the Kaede-labelled cells using 6% laser power scanning for 60 s ( optimized before the experiments; data not shown ) . Fins were imaged before and after the photoconversion ( at 6 and 26 hpA ) in the green and red channels .	Inflammation plays an important role in helping the body to heal wounds and fight off certain diseases . Immune cells called macrophages—which are perhaps best known for their ability to engulf and digest microbes and cell debris—help to control inflammation . In mammals , different types of macrophage exist; the most functionally extreme of which are the M1 macrophages that stimulate inflammation and M2 macrophages that reduce the inflammatory response . Macrophages acquire different abilities through a process called polarization , which is controlled by signals produced by a macrophage's environment . Polarization has been well investigated in human and mouse cells grown in the laboratory , but less is understood about how this process occurs in live animals . Nguyen Chi , Laplace-Builhe et al . investigated whether zebrafish larvae ( which are naturally transparent ) could form an experimental model in which to investigate macrophage polarization in living animals . Zebrafish were first genetically engineered to produce two fluorescent proteins: one that marks macrophages and one that marks M1 macrophages . These fluorescent proteins allow the movement and polarization of macrophages to be tracked in real time in living larvae using a technique called confocal microscopy . Nguyen Chi , Laplace-Builhe et al . also isolated macrophage cells from these zebrafish at different times during the inflammatory process to identify which macrophage subtypes form and when . The results show that unpolarized macrophages move to the sites of inflammation ( caused by wounds or bacterial infection ) , where they become polarized into M1 cells . Over time , these M1 macrophages progressively convert into an M2-like macrophage subtype , presumably to help clear up the inflammation . Furthermore , Nguyen Chi , Laplace-Builhe et al . show that the M1 and M2 macrophage subtypes in zebrafish are similar to those found in mammals . Therefore , genetically engineered zebrafish larvae are likely to prove useful for studying macrophage activity and polarization in living animals .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "immunology", "and", "inflammation" ]	2015	Identification of polarized macrophage subsets in zebrafish
Listeners locate potential mates using species-specific vocal signals . As tetrapods transitioned from water to land , lungs replaced gills , allowing expiration to drive sound production . Some frogs then returned to water . Here we explore how air-driven sound production changed upon re-entry to preserve essential acoustic information on species identity in the secondarily aquatic frog genus Xenopus . We filmed movements of cartilage and muscles during evoked sound production in isolated larynges . Results refute the current theory for Xenopus vocalization , cavitation , and favor instead sound production by mechanical excitation of laryngeal resonance modes following rapid separation of laryngeal arytenoid discs . Resulting frequency resonance modes ( dyads ) are intrinsic to the larynx rather than due to neuromuscular control . Dyads are a distinctive acoustic signature . While their component frequencies overlap across species , their ratio is shared within each Xenopus clade providing information on species identity that could facilitate both conspecific localization and ancient species divergence . Editorial note: This article has been through an editorial process in which the authors decide how to respond to the issues raised during peer review . The Reviewing Editor's assessment is that all the issues have been addressed ( see decision letter ) . In the transition from water to land in early tetrapods , lungs replaced gills for respiration ( Daeschler et al . , 2006 ) . Many current tetrapods use air movement to empower specialized vocal organs such as the larynx of frogs and mammals and the syrinx of birds ( Elemans et al . , 2015 ) . The resulting sounds are shaped by a combination of vibrating elements and cavity resonances to voice different acoustic qualities that convey sex , age , species , emotional state and even intent ( Hall et al . , 2013 ) . While voice provides essential information for social interactions , we know surprisingly little about how vertebrate vocal organs create these complex acoustic features . In particular , when an ancestral tetrapod leaves the land and vocalizes underwater without air movement , how are communication sounds produced and then shaped to maintain essential social information , and how do they diversify during speciation ? Frogs in the secondarily aquatic genus Xenopus ( Evans et al . , 2015 ) present an informative system for addressing these questions . In Xenopus , social communication is dominated by vocal signaling ( Kelley et al . , 2017 ) . Males in each species produce distinctive advertisement calls underwater whose acoustic features inform species identity ( Evans et al . , 2015 ) . These calls consist of a series of sound pulses that form species-typical temporal patterns and characteristically include two dominant frequencies ( DFs ) ( Hall et al . , 2013; Tobias et al . , 2011 ) . The sound pulses that comprise Xenopus calls are produced in the larynx , ( Tobias and Kelley , 1987 ) a vocal organ interposed between the nasal and buccal cavities and the lungs ( Figure 1A ) . Vocal folds are absent ( Ridewood , 1897 ) and a separate glottis gates air flow to and from the lungs ( Brett and Shelton , 1979 ) . The larynx consists of a cricoid frame or ‘box’ of hyaline cartilage flanked bilaterally by bipennate muscles . These insert anteriorly , via a tendon , onto paired , closely apposed arytenoid cartilage discs ( Figure 1B; Figure 1—figure supplement 1 ) whose medial faces are coated by mucopolysaccharide secreted by adjacent cells ( Yager , 1992 ) . The discs are suspended in elastic tissue composed of elastic cartilage and elastin fibrils ( Figure 1—figure supplement 1 ) ( Yager , 1992 ) . Electrical stimulation of laryngeal muscles or nerves results in species-specific sound pulses , both in situ and ex vivo ( Yager , 1992; Tobias and Kelley , 1987 ) . Sound is thus produced without air flow or vocal folds . In X . borealis , separations of the paired arytenoid discs accompany sound pulses , ( Yager , 1992 ) but how disc motion results in sound production has not been resolved . An unusual mechanism – implosion of air bubbles or cavitation ( Yager , 1992 ) - is the currently accepted ( Irisarri et al . , 2011; Ladich and Winkler , 2017 ) hypothesis for underwater laryngeal sound production in Xenopus . In this scenario , the high velocity separation of the arytenoid discs causes formation of bubbles that then implode and produce sounds . Cavitation bubbles are known to produce hydrodynamic propeller noise ( Carlton , 2012 ) and the ‘snaps’ of some species of shrimp ( Versluis et al . , 2000 ) . However , a priori cavitation - creating "a bubble between the discs at a pressure below ambient … that … implodes as air rushes into the cleft at high speeds , producing a click" ( Yager , 1992 ) - seems an unlikely cause of sound production in Xenopus . The small film of fluid between the arytenoid discs should allow neither high velocity flow nor bubble formation , and bubbles have not yet been observed . Additionally , cavitation bubble collapse produces a high amplitude pressure pulse ( ~50–100 kPa ) , several orders of magnitude louder than the radiated sound pressure of Xenopus sound pulses . Finally , the duration of pressure transients produced by the collapse of cavitation bubbles are in the microsecond range , rather than the millisecond range of Xenopus sound pulses . To empirically test the cavitation hypothesis , we filmed isolated X . laevis larynges during sound pulse production evoked by stimulation of the laryngeal nerve ( Tobias and Kelley , 1987 ) . As reported previously , ( Yager , 1992 ) disc movements accompanied sound production ( Figure 1C , E , F; Video 1 ) . To track the position of the arytenoid discs , we placed 40–80 µm carbon microspheres over the discs ( Figure 1B , C ) and computed disc position at subpixel resolution by interpolating the 2D intensity correlations of the spheres with each image ( See Materials and methods , Figure 1D ) . This approach revealed disc position at speeds of up to 44 , 000 fps and allowed us to calculate disc velocity and acceleration profile in relation to sound pressure ( Figure 1E–H ) . Nerve stimulation first produces isometric contraction of the bipennate muscles during which the arytenoid discs remain apposed . In favorable preparations , a fluid layer could then be observed retreating from the medial surface of the discs with increasing isometric force . This observation suggests that the discs are kept together by a capillary binding force through a liquid bridge . When bilaterally exerted muscle force overcomes the binding force , the liquid bridge ruptures ( Figure 1F ) and the discs separate rapidly with mean gap peak acceleration of 13 . 5 ± 9 . 1 g ( N = 3; range: 6 . 4–23 . 8 g , where 1 g = 9 . 82 m/s2 ) and mean peak velocity of 50 . 3 ± 21 . 9 mm/s ( N = 3; range: 35 . 5–75 . 4 mm/s ) ( see examples: Figure 1E; Figure 2B ) . Disc peak deceleration ( 10 . 6 ± 5 . 2 g; N = 3; range: 6 . 5–16 . 4 g ) occurs only 0 . 57 ± 0 . 10 ms after peak acceleration . In this short time , the gap between the discs enlarges to 18 . 6 ± 5 . 5 µm ( range: 14 . 1–24 . 8 µm ) , a value that is 27 . 5 ± 8 . 0% ( range: 22 . 8–35 . 6% ) of the maximum gap width of 72 . 6 ± 34 . 5 µm ( N = 3; range: 39 . 6–108 . 4 µm ) . Peak acceleration and velocity precede sound onsets by 0 . 85 ± 0 . 33 and 0 . 51 ± 0 . 33 ms respectively ( Figure 1E–H ) . The interval between disc separation and subsequent sound radiation reveals that these events are clearly associated . The delay ( Figure 1F ) between disc acceleration and velocity shows less variation ( 0 . 34 ± 0 . 06 ms , N = 3 ) indicating that sound onset timing after disc separation , rather than disc kinematics , is more variable between preparations . At 40 Hz nerve stimulation rates ( within natural call sound pulse rates ( Tobias et al . , 2011 ) ) , the first few stimuli do not result in sound pulse production ( Figure 2A ) . This corroborates previous results that male laryngeal neuromuscular synapses are ‘weak , ’ i . e . , require facilitation to release sufficient neurotransmitter to generate a muscle action potential and contraction ( Tobias and Kelley , 1987; Ruel et al . , 1997 ) . Subsequently , over multiple stimulations , peak-to-peak ( ptp ) sound pressure increases linearly with both peak velocity and acceleration and reaches a maximum received level of ~180 mPa ptp ( at 44 mm ) in all preparations ( Figure 2B ) . Below a minimal peak velocity of 28 ± 12 mm/s and minimal peak acceleration of 4 . 6 ± 3 . 1 g , no sound is detected , suggesting a threshold disc velocity or acceleration required to produce sound . When water was introduced via the glottis into the supradisc space , no sound was produced , corroborating earlier observations in X . borealis ( Yager , 1992 ) . Of disc gap width , peak velocity and peak acceleration , only disc peak velocity does not reach threshold when the liquid bridge holding the discs together is disrupted ( Figure 2—figure supplement 1 ) . This observation supports the hypothesis that a threshold disc peak velocity is required for sound production . In summary , three sets of observations do not support the cavitation hypothesis: ( 1 ) we do not observe bubble formation , ( 2 ) sound pressures are six orders of magnitude too low ( mPa vs kPa ) , and ( 3 ) the onset of sound production is three orders of magnitude too slow ( ms vs µs ) . We have previously reported ( Tobias and Kelley , 1987 ) that sound pulses produced ex vivo in male X . laevis larynges include species-specific frequencies also present in actual calls . To determine whether spectral features of calls reflect species-typical laryngeal features , we first recorded male advertisement calls from representative Xenopus species . In the three clades of this sub-genus - L ( which includes X . laevis ) , M ( which includes X . borealis ) and A ( which includes X . andrei and X . amieti ) ( Evans et al . , 2015 ) - each repeated sound pulse in the male advertisement call has two simultaneously produced frequency bands: a higher dominant frequency ( DF2 ) and a lower frequency ( DF1 ) ( Tobias et al . , 2011 ) , termed dyads ( Figure 3A , C , D ) . All but one species produce advertisement calls made up of harmonic dyads in which DF1 and DF2 are related by a small-integer ratio . The exception is X . allofraseri in which pulses contain harmonic stacks . There is a broad range for both the lower and the higher dominant frequency across Xenopus ( Figure 3C , D ) . Either DF1 or DF2 can be shared by different species . However , in contrast to individual frequencies , the ratio of DF2/DF1 is specific to , and highly conserved within , each clade ( Figure 3E ) . The mean ratio ( Table 1 ) for A clade species is 1 . 33 ( ±0 . 004 ) , for L clade species is 1 . 23 ( ±0 . 020 ) , and for M clade species is 2 . 02 ( ±0 . 032 ) . Exceptions include X . wittei in the A clade ( 1 . 19 , not 1 . 34 ) and X . laevis South Africa in the L clade ( 1 . 14 , not 1 . 24 ) . Thus , species in different clades can share either DF1 or DF2 but not the distinctive ratio . To determine how dyads are produced , we first confirmed that our recordings of sound pulses were free of possible acoustic artifacts produced by interactions with the recording tank . Using a hydrophone and laser Doppler vibrometry , we obtained simultaneous recordings of sounds and body vibrations from a calling male under the same conditions in which we obtained the data in Figure 3 . The same dyad was present in both sounds and vibrations from a vocalizing male ( Figure 4A ) . Thus , the dyads reported here are produced by the frog . To determine whether dyads are produced solely by the larynx ( without contributions from other organs ) , we recorded sound and vibrations produced by isolated larynges in response to nerve stimulation ( Figure 4B ) . Isolated larynges produced the same dyads as the intact animal . Thus , the generation of dyads is intrinsic to the larynx and is not influenced by extra-laryngeal tissues . To examine a potential role for the internal air spaces , we reduced the volume of the lumen in isolated X . laevis male larynges by inserting a large bead or replaced air with helium ( as in a previous study; Yager , 1992 ) . We also placed weights on dorsal surface of the larynx . None of these manipulations affected DF1 , DF2 or the DF2/DF1 ratio . These observations indicate that the dyads do not reflect the volume of the internal chambers or the mass of the cricoid cartilage . However , the ability of tissues containing elastic cartilage , such as the pinna of the ear , to deform and reform rapidly suggested that this tissue - also present in the larynx -might be essential to producing dyads . The interior of the larynx ( Figure 4C ) includes a central , air-filled chamber ( C ) separated from smaller , lateral chambers ( L ) by elastic cartilage septa ( Figure 4C , inset ) . We isolated the larynx and drilled a small hole into the middle of the dorsal hyaline cartilage through which a pin was used to puncture the elastic cartilage on both sides . After puncture , the elastic cartilage was no longer structurally intact ( Figure 4—figure supplement 1 ) . In the three species examined - X . boumbaensis ( A clade; n = 6 ) , X . victorianus ( L clade , n = 2 ) and X . laevis South Africa ( L clade- DF2/DF1 exception; n = 4 ) - the hole in the hyaline cartilage by itself ( Figure 4—figure supplement 1A ) did not affect the frequencies produced by motor nerve stimulation ( Figure 4—figure supplement 1B ) ) . However , puncturing the elastic cartilage ( Figure 4—figure supplement 1C ) either abolished the DF peaks by shifting the two narrowly tuned bands to a single , intermediate , broader band frequency ( 5/12 larynges; Figure 4D ) , abolished one peak and broadened the other ( 2/12 ) , abolished both peaks ( 4/12 ) or shifted peaks to a different ratio ( 1/12 ) . In X . borealis , ‘opening’ the cricoid box by removing a rectangular portion of the ventral laryngeal wall detunes the larynx ( Yager , 1992 ) . This detuning could also have been due to disruption of elastic cartilages . Because the elastic cartilage partitions the internal air chambers of the larynx , the puncture created a single air space . Our data do not support the prevailing ( Yager , 1992; Irisarri et al . , 2011; Ladich and Winkler , 2017 ) cavitation hypothesis for sound production in Xenopus . Two alternative mechanisms that could account for the association between disc movement and sound production are: 1 ) acoustic excitation by a rapid pressure drop between the discs or 2 ) disc movement-associated mechanical excitation of the larynx . Sound pressure reduction between the discs may produce a propagating sound pressure wave , exciting air cavity resonances within the larynx . This mechanism should result in a bipole sound source with strong directional radiation pattern ( Larsen and Wahlberg , 2017 ) . Because of impedance mismatch between the air cavities and cartilages of the larynx , however , this mechanism would produce a poor , low intensity sound . In addition , replacing air with helium should alter the frequency distribution of cavity resonances ( Rand and Dudley , 1993 ) , an effect not observed in this study , nor previously ( Yager , 1992 ) . Alternatively , separation of the arytenoid discs might mechanically excite vibration of laryngeal elements . This mechanism would result in a monopole sound source – the entire larynx - with a more omnidirectional radiation pattern , effectively coupled to the medium , producing a more intense sound . The sound pressure produced by such a vibrating monopole structure depends on its space and time averaged velocity ( Hambric and Fahnline , 2007 ) , which is consistent with our observations of a linear correlation between disc velocity and sound pressure , and the minimal velocity required for sound production . This mechanism is also consistent with previous experiments in which splitting 'the elastic sac surrounding the discs' ( Yager , 1992 ) prevented sound pulse production . We thus favor the second explanation and propose that disc movements specifically excite vibration of the elastic tissues surrounding the discs ( illustrated in Figure 1—figure supplement 1 ) . As key features of the Xenopus larynx - including lack of vocal folds and modification of the laryngeal box and cartilages - are shared across Pipid species ( Ridewood , 1897; Ridewood , 1900 ) , this proposed mechanism of underwater sound production may also be shared . In Xenopus , water is prevented from entering the larynx during underwater calling by inhibition of glottal motor neurons ( Zornik and Kelley , 2007 ) , thus ensuring the attainment of the disc peak velocity values identified here as required for sound production . The sound-protection afforded would not be required in another pipid , Hymenochirus merlini , that has reverted to calling in air ( Irisarri et al . , 2011 ) , presumably through an open glottis . We predict that H . merlini calling is also powered by disc separation rather than air flow . Our experimental results support the hypothesis that arytenoid disc movements subsequently excite two natural vibratory resonance frequencies of the larynx itself . These harmonic dyads require intact elastic cartilage septa that separate the central laryngeal lumen from the lateral laryngeal chambers . Both species-specific individual DFs and the clade-specific dyad ratio are thus intrinsic to the larynx rather than the result of laryngeal or respiratory muscle modulation by neural circuitry . Which , as yet unidentified , characteristics of laryngeal tissue geometry and properties result in species-specific DFs and their ratios remain to be determined , but are likely to reflect a common tuning mechanism in descendants of ancestral Xenopus species ( Sassoon et al . , 1986; Baur et al . , 2008 ) . Three mechanisms have been identified for producing and shaping vertebrate laryngeal vocalizations: the myoelastic aerodynamic theory ( MEAD ) , active muscle contraction ( ACM ) and intralaryngeal aerodynamic whistles . The MEAD mechanism ( Titze , 1980 ) explains the physical basis for sounds produced by isolated larynges of mammals and terrestrial frogs ( Suthers et al . , 2006 ) as well as by syringes in birds ( Elemans et al . , 2015 ) . The ACM mechanism requires motor neuron-driven contraction of intrinsic laryngeal muscles to produce , for example , purring in cats ( Remmers and Gautier , 1972 ) . Intralaryngeal aerodynamic whistles produce ultrasound in mice and probably all murine rodents ( Mahrt et al . , 2016; Roberts , 1975 ) . All of these mechanisms require air flow to produce vocalizations and thus none of them explain how Xenopus call . The return to water in ancestral Xenopus was instead accompanied by a novel mechanism for laryngeal sound production: disc movement-induced excitation of laryngeally intrinsic resonance modes - dyads - shaped spectrally by material properties of elastic cartilage septa . Thus , the evolutionary change that allowed sound to be produced underwater without airflow in Pipids is not only responsible for production of sound pulses , but also their spectral features . Species-specific , rhythmic activity patterns of laryngeal motor neurons drive the precise temporal pattern of laryngeal muscle contractions responsible for the arytenoid disc separations that produce vocalizations ( Leininger and Kelley , 2013; Barkan et al . , 2017; Barkan et al . , 2018 ) . In terrestrial frogs , as in other vocal vertebrates , acoustic features of male advertisement calls contain information on species identity ( Gerhardt and Huber , 2002 ) . This is also true for Xenopus; other call types - such as the male release call - vary little between species ( Tobias et al . , 2014 ) . Information on species identity can serve to reduce interspecific mating and costs associated with hybrid offspring , including male sterility , that lead to restricted gene flow and speciation ( Lemmon and Lemmon , 2010 ) . Across Xenopus , temporal patterns of the advertisement call are homoplasious: the same pattern - for example a click-type call – recurs in genetically distant species ( Tobias et al . , 2011; Roberts , 1975 ) . While temporal information can be ambiguous due to homoplasy , our data suggest that spectral information in advertisement calls – constrained by the morphology of the larynx – is more phylogenetically informative ( Gingras et al . , 2013 ) . However , Xenopus evolution has also been shaped by multiple rounds of inter-specific hybridization resulting in genomic introgression and the numerous highly polyploid species of the phylogeny , particularly A clade species ( Evans et al . , 2015 ) . Rapid oviposition once eggs are ovulated places a premium on locating a male . When different species share the same pond , a female mating with a male from the same clade is more likely to produce viable and fertile offspring . The peripheral auditory system of females is tuned to their species' own dyad: DF1 , DF2 and the DF2/DF1 ratio ( Hall et al . , 2016 ) . Species-specific complementarity between vocal production and perception should reinforce the divergence of populations during speciation by limiting gene flow . The acoustic advantage to a gravid female of locating the most genetically-compatible calling male using the clade-specific common harmonic vocal signature thus may drive co-evolution of the vocal organ in the male and auditory perception in the female . Subjects for this study were all sexually mature , male , clawed frogs from the sub-genus Xenopus . We either obtained frogs from commercial suppliers ( Avifauna , Xenopus Express , Xenopus One , or Nasco ) or from our Xenopus colony at Columbia University ( species and populations: X . pygmaeus , X . ruwensoriensis , X . amieti , X . boumbaensis , X . allofraseri , X . andrei , X . itombwensis , X . wittei , X . vestitus , X . lenduensis , X . largeni , X . gilli , X . poweri , X . laevis Nigeria , X . laevis South Africa , X . victorianus , X . petersii , X . laevis Malawi , X . borealis and X . muelleri; see ( Tobias et al . , 2011 ) for details on geographic locales; species nomenclature as revised ( Evans et al . , 2015 ) . ) Frogs were housed in 2–5 L of water in polycarbonate tanks , under a 12–12 light-dark cycle , fed frog brittle ( Nasco; Ft . Atkinson , WI , USA ) and had their water changed twice per week . All animals were housed and handled in accordance with the guidelines established by the Danish Animal Experiments Inspectorate ( Copenhagen , Denmark ) and the Columbia University Animal Care and Use Committee . We used high speed films of tissue movements in the isolated larynx preparation ( Tobias and Kelley , 1987 ) of 5 adult male X . laevis to test the cavitation hypothesis . The animals were euthanized and their vocal organ and attached lungs were removed . The air-filled larynx was submerged in physiological saline in a 66 mm Petri dish . After isolation , larynges were pinned , dorsal side up , via extra-laryngeal cartilages , to a Sylgard coated recording dish submerged in oxygenated Ringers solution . The bilateral , freed motor nerves were drawn into suction electrodes for stimulation ( WPI Linear stimulus isolator model A395R-C ) . All isolated larynges produced sound pulses . Sound was recorded with a 1/2-inch pressure microphone-pre-amplifier assembly ( model 46AD , G . R . A . S . , Denmark ) and amplified ( model 12AQ , G . R . A . S . , Denmark ) . The microphone and recording chain sensitivity was measured before each experiment ( sound calibrator model 42AB , G . R . A . S . , Denmark ) . The microphone was placed at 22–24 mm away from the mounted larynx . Because the signal-to-noise ratio of the hydrophone was lower than the microphone and the timing of sound events did not differ , we used the microphone signal for further analysis . Microphone , hydrophone and stimulation timing signals were low-pass filtered at 10 kHz , ( custom-built filter , ThorLabs , Germany ) and digitized at 30 kHz ( USB 6259 , 16 bit , National Instruments , Austin , Texas ) . Larynges were imaged with a 12 bit high-speed camera ( MotionPro-X4 , 12 bit CMOS sensor , Integrated Design Tools , Inc . ) mounted on a stereomicroscope ( M165-FC , Leica Microsystems ) . The preparation was back-lighted to visualize the arytenoid discs by a plasma light source ( HPLS200 , Thorlabs , Germany ) through liquid light guides and reflected of a 45° angled silver coated prism ( MRA series , Thorlabs ) to absorb heat . To track the position of landmarks , we placed 40–80 µm diameter carbon spheres on the surface of the larynx , muscles and arytenoid discs as illustrated in Figure 1 . The position of spheres was tracked at subpixel precision by interpolating the 2D intensity cross-correlations of the same sphere in an initial frame to each movie image ( Figure 1B–D ) . Velocity and acceleration of the spheres were calculated by differentiation of their position . All control and analysis software was written in MATLAB . In all five preparations , we filmed the larynx in toto following varying rates of nerve stimulation . In three preparations we obtained sufficiently high contrast images of the arytenoid discs at high imaging frame rates of 10 , 00–44 , 000 fps to allow automated position extraction during stimulation of the bilateral motor nerves at 40 Hz for 50 cycles . Data acquisition on the NI board and camera system was synchronized by a 1 ms TTL pulse . The camera was triggered at the positive rise of this 1 ms TTL pulse . The camera’s specifications allow shutter speed as short as 1 µs . During earlier synchronization tests ( Elemans et al . , 2015 ) , we determined that the trigger accuracy was below the duration of one frame ( maximally 21 µs ) and thus well below the relevant time scales investigated here . Arytenoid gap width was defined as distance moved between the two markers from their resting position and perpendicular to the midline . Minimum and maximum acceleration of gap width were calculated per stimulus . Sound was bandpass filtered between 1–4 kHz ( 3th order butterworth filter implemented with zero phase-shift; filtfilt algorithm ) . The noise floor was defined as three times the standard deviation of a 67 ms background recording prior to each stimulation experiment . However , because sound energy did not fully dissipate in the experimental chamber between consecutive nerve stimulations , especially after 30–40 cycles , we used a threshold of 0 . 01 Pa to determine sound onset per stimulus . The first detectable sound pulse typically occurred after 2–3 stimulations . This is consistent with our earlier work showing that male laryngeal neuromuscular synapses are ‘weak , ’ i . e . , require facilitation to release sufficient neurotransmitter to generate a muscle action potential and contraction ( Ruel et al . , 1997 ) . We used linear regressions - including only measurements above sound threshold – to calculate the minimal disc velocity and acceleration associated with sound generation . For in vivo recordings , frogs were placed in a 75 L aquarium . To promote vocal behavior , males were injected with human chorionic gonadotropin ( hCG; Sigma: 50–200 IU depending on body size ) one day prior to and on the day of recording . Males were then paired with a conspecific , sexually unreceptive female in a glass aquarium ( 60 × 15 × 30 . 5 cm , L × W × H; water depth = 23 cm; 20◦ C ) . A hydrophone ( High Tech , Gulfport , MI , USA; output sensitivity −164 . 5 dB at 1 V/µPa , frequency sensitivity 0 . 015–10 kHz; or Cornell Bioacoustics , output sensitivity −163 dB at 1 V/µPa ) was used to record calls to a Marantz digital recorder CD or flash card ( CDR300 , Marantz , Mahwah , NJ , USA; 44 . 1 kHz sampling rate ) or on a computer ( Macintosh ) via a Lexicon A/D converter . To measure values for DF1 and DF2 , three non-consecutive advertisement calls ( the smallest vocal unit as described in ( Tobias et al . , 2011 ) ) were analyzed from 3 males of each species . Dominant frequencies were calculated from fast Fourier transforms ( FFT ) with maximum Q values ( peak frequency/maximum frequency 6 dB below peak frequency - minimum frequency 6 dB below peak frequency; Table 1 ) . The initial attack segment of each sound pulse was not included in the analysis because it is more broadband than the sustained portion of the call . The values shown in Table 1 are the mean of individual means for all calls by species recorded . Advertisement calls of single males were recorded in aquaria with a hydrophone ( H2a , Aquarian Audio Products; Anacortes , WA , USA ) and vibration velocities were recorded simultaneously with a portable laser ( PDV 100 laser , Polytec Inc . ; Irvine , CA , USA ) directed at the ventral surface of the singing frog . We recorded from one each of X . laevis . South Africa , X . borealis , X . muelleri , X . new tetraploid , and X . boumbaensis . We then isolated the larynx as described above . To access the elastic cartilages ( Figure 4—figure supplement 1C ) , we drilled a small hole in the dorsal surface of the larynx ( Figure 4—figure supplement 1A ) and then sealed it with a small piece of Parafilm . To ensure that the hole had no effect , sound and laser recordings were obtained before and after this procedure ( Figure 4—figure supplement 1B ) . The Parafilm was then removed and a 30 g needle used to puncture the elastic cartilage on both sides , after which the Parafilm was replaced . At the end of the experiment , the larynx was split saggitally ( 'butterflied' ) and the disruption of elastic cartilage was confirmed by visual inspection ( Figure 4—figure supplement 1C , illustrating intact vs punctured elastic cartilage ) . Sound and laser recordings were digitized ( PreSonus Audio box; Baton Rouge , LA , USA ) and stored on a Macintosh computer .	The voice is a unique characteristic that we use to identify one another – including someone's sex , age and mood . We speak by using air flow to vibrate our vocal folds , commonly known as vocal cords . The land-living ancestors of the African clawed frog Xenopus also used breath and vocal cords to communicate , but they returned to aquatic life 180 million years ago and had to evolve a different way to create sounds . Today’s Xenopus live in water and use a new mechanism that lets them sing for hours underwater without coming up to breathe . Males from each major group of Xenopus species produce courtship songs with harmonic intervals corresponding to an octave , a perfect fourth , or a major or minor third . Today's Xenopus species do not have any vocal cords . Instead , they have an elaborate set of vocal components: the muscles of the larynx contract paired , movable rods that end in discs . For the past 40 years , it was thought that these frogs create sounds by collapsing small air bubbles between the discs , similar to snapping shrimp . But such bubbles have never been observed , and exactly how these frogs manage to create sounds underwater has been a mystery . Here , Kwong-Brown et al . filmed the larynx as it was stimulated to produce sounds and discovered that the rapid separation of the discs excites the larynx and the surrounding tissues to create the harmonic frequencies . Then , to determine how the frog creates its harmonic intervals , Kwong-Brown et al . tried to ‘detune’ the larynx . In a series of experiments , they placed weights on the surface of the larynx , drilled a hole in the cartilage and filled it with helium , or introduced small glass beads . None of these attempts had any effect . However , rupturing the elastic cartilages within the larynx – which separate its internal cavity into three chambers – disrupted the harmonic intervals . This new way of creating underwater sounds helped to maintain the quality of the frog’s voice and may explain how Xenopus can shape its songs to convey crucial information to others , such as identifying species , sex and social intent .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology", "research", "communication" ]	2019	The return to water in ancestral Xenopus was accompanied by a novel mechanism for producing and shaping vocal signals
Clustering of receptors associated with immunoreceptor tyrosine-based activation motifs ( ITAMs ) initiates the macrophage antimicrobial response . ITAM receptors engage Src-family tyrosine kinases ( SFKs ) to initiate phagocytosis and macrophage activation . Macrophages also encounter nonpathogenic molecules that cluster receptors weakly and must tune their sensitivity to avoid inappropriate responses . To investigate this response threshold , we compared signaling in the presence and absence of receptor clustering using a small-molecule inhibitor of Csk , which increased SFK activation and produced robust membrane-proximal signaling . Surprisingly , receptor-independent SFK activation led to a downstream signaling blockade associated with rapid degradation of the SFK LynA . Inflammatory priming of macrophages upregulated LynA and promoted receptor-independent signaling . In contrast , clustering the hemi-ITAM receptor Dectin-1 induced signaling that did not require LynA or inflammatory priming . Together , the basal-state signaling checkpoint regulated by LynA expression and degradation and the signaling reorganization initiated by receptor clustering allow cells to discriminate optimally between pathogens and nonpathogens . Macrophages are myeloid-derived hematopoietic cells that guard against infection and orchestrate a broad immune response to pathogens . Bridging the innate and adaptive immune systems , they survey tissues for signs of pathogen invasion and present antigen to T cells . When activated , they engulf and destroy foreign organisms and secrete cytokines ( Mosser and Edwards , 2008; Murray and Wynn , 2011; Wynn et al . , 2013 ) . While macrophage activity promotes an effective immune response to tissue distress , inappropriate activation of macrophages can cause tissue injury . Accumulation of activated macrophages is linked to diseases such as rheumatoid arthritis , type 1 diabetes , and obesity ( Murray and Wynn , 2011; Wynn et al . , 2013 ) . Macrophage stimulation through immunoreceptor tyrosine-based activation motif ( ITAM ) - and hemi-ITAM-coupled receptors initiates pathogen destruction via phagocytosis and the production of reactive oxygen species , which require activated macrophages to be physically close to the pathogen . To prevent nonproductive activation and tissue damage , macrophages must differentiate between receptor ligands displayed during a pathogen encounter ( e . g . , a β-glucan-presenting fungal cell wall or an IgG-opsonized cell ) vs . environmental noise ( e . g . , glycans shed from fungi or other cell debris ) . One mechanism by which cells make this distinction is by using receptor clustering to sense the size of a potential stimulus . Large-diameter stimuli that are more likely to be associated with pathogenic particles nucleate large-scale receptor clusters , forming a signaling complex called a phagocytic synapse . In contrast , small-diameter stimuli induce only small-scale or transient receptor clusters ( Goodridge et al . , 2011 ) . Hereafter , we will refer to ligands that cluster ITAM/hemi-ITAM receptors efficiently as ‘strong stimuli’ and ligands that cluster receptors poorly as ‘weak stimuli’ . In mast cells , a related myeloid subtype , treatment with a high dose of weak stimulus can produce robust receptor phosphorylation without stable receptor clustering . Interestingly , the resulting membrane-proximal signaling patterns , as well as the associated immunopathologies , are qualitatively different than those produced in response to a strong stimulus ( Carroll-Portillo et al . , 2010; Suzuki et al . , 2014 ) , which indicates that myeloid cells can distinguish between strong and weak stimuli beyond simply sensing the degree of receptor phosphorylation . Receptor engagement initiates interactions with an active pool of Src-family tyrosine kinases ( SFKs ) , which phosphorylate intracellular ITAMs within the receptor , the tyrosine kinase Syk , cytoskeleton-associated proteins , and other membrane-proximal signaling proteins ( Lowell , 2010; Goodridge et al . , 2011 ) . This membrane-proximal signaling initiates downstream signaling , including the phospholipase Cγ-mitogen-activated protein kinase ( PLCγ-MAPK ) and phosphoinositide 3-kinase-protein kinase B/Akt ( PI3K-Akt ) pathways . The signaling cascades initiated by the SFKs in response to strong stimuli then enact a transcriptional program of macrophage activation leading to microbicidal activity ( Lowell , 2010; Byeon et al . , 2012 ) . To some extent , macrophages can modulate their responsiveness to activating stimuli according to their environment . For example , inflammatory cytokines secreted by other immune cells trigger a process termed inflammatory priming , a transcriptional program that increases macrophage reactivity . Priming is known to upregulate the expression of proteins involved in signal detection , pathogen killing , and stimulation of other immune cells ( Platanias , 2005 ) . However , it is not clear how inflammation might modulate macrophage responses to weak stimuli by changing the extent of receptor clustering necessary for downstream signaling . It is also unclear whether strong macrophage stimuli are defined by their ability to robustly activate the SFKs or whether receptor clustering itself relays qualitative information that indicates a genuine pathogen encounter . Clustered receptors are immobilized within the cell membrane , participate in costimulatory interactions , and organize the signaling complex spatially , integrating SFK-dependent and SFK-independent steps into a productive signaling response ( Sohn et al . , 2008; Bethani et al . , 2010; Dustin and Groves , 2012 ) . Until now , it has not been possible to study the contributions of SFK activity to macrophage activation independently of the integrated effects of receptor clustering . To investigate the role of SFK activation in macrophage signaling independently of ligand-induced receptor clustering , we developed an experimental system to activate the SFKs directly by inhibiting their negative regulator , the tyrosine kinase Csk . Csk phosphorylates the inhibitory-tail tyrosine of the SFKs , promoting an autoinhibited state in which the SH2 domain of the SFK interacts with the phosphorylated inhibitory tail and constrains the kinase domain in an inactive conformation ( Brown and Cooper , 1996; Okada , 2012 ) . The phosphatases CD45 and CD148 dephosphorylate the inhibitory-tail tyrosine of the SFK , releasing this autoinhibition and enabling subsequent activation-loop autophosphorylation and full activation ( Chow and Veillette , 1995; Zhu et al . , 2011 ) . The continuous competition of Csk and CD45/CD148 sets the steady state of SFK inhibitory-tail phosphorylation in the cell . Inhibiting Csk activity disrupts this steady state , allowing CD45 and CD148 initiate bulk SFK activation unopposed ( Schoenborn et al . , 2011 ) . We previously described a Csk variant ( analog-sensitive Csk , ‘CskAS’ ) that is sensitized to inhibition by a 3-iodobenzyl analog of the kinase inhibitor PP1 ( 3-IB-PP1 ) . The bulkiness of 3-IB-PP1 prevents its interaction with WT Csk and other WT kinases ( Okuzumi et al . , 2010; Schoenborn et al . , 2011; Tan et al . , 2014 ) . We have also described a transgenic mouse that expresses CskAS but not WT Csk and have studied the effects of CskAS inhibition in T cells ( Schoenborn et al . , 2011; Tan et al . , 2014 ) . Macrophages express a distinct subset of Src family members ( Lyn , Hck , and Fgr ) ( Lowell , 2004 ) , and , unlike naive T cells , macrophages express the ITAM-proximal kinase Syk ( Wang et al . , 2010 ) and SFK-dependent negative-regulatory phosphatases associated with immunoreceptor tyrosine-based inhibitory motifs ( ITIMs ) ( Veillette et al . , 2002; Lowell , 2004 ) . This distinct membrane-proximal signaling environment likely modifies the signaling program initiated by macrophage receptors in service of cell-specific functions . We derived macrophages from the bone marrow of CskAS mice and compared SFK signaling without receptor clustering ( by inhibiting CskAS with 3-IB-PP1 ) to SFK signaling induced by receptor clustering ( by ligating the hemi-ITAM receptor Dectin-1 with depleted zymosan ) . Receptor-independent SFK activation by 3-IB-PP1 induced robust membrane-proximal signaling but no downstream signaling through the MAPKs or Akt . We determined that this signaling blockade was caused by rapid degradation of the SFK LynA , which resulted in a loss of function that could not be compensated for by the other SFKs . We were able to rescue downstream signaling by priming the macrophages , which led to the upregulation of LynA . Receptor clustering enabled the participation of the other SFKs in the activation of downstream MAPK , Akt , and calcium signaling independently of LynA . From the data presented in this article , we propose a model to explain how macrophages are prevented from responding to weak stimuli , how inflammation increases macrophage sensitivity to weak stimuli , and how receptor clustering rewires SFK signaling to enable macrophage activation . In this study , we examined signaling in BMDMs induced by activating the SFKs in the absence of receptor clustering ( by inhibiting CskAS with 3-IB-PP1 ) and compared this to signaling induced by receptor clustering ( mediated by the Dectin-1 ligand zymosandep ) . Treatment with 3-IB-PP1 induced robust membrane-proximal signaling , but signal propagation was blocked downstream of PLCγ and PI3K , with no apparent activation of Erk , JNK , Akt , or NFAT . We determined that propagation of receptor-independent signaling induced by 3-IB-PP1 requires the SFK LynA , but we observed that LynA protein was almost completely degraded within the first three minutes after 3-IB-PP1 addition . This blockade was overcome by priming macrophages with IFN-γ or GM-CSF before 3-IB-PP1 treatment , which led to increased expression of LynA , thus buffering it against total degradation and leading to a more sustained presence of LynA during the first minutes of 3-IB-PP1 treatment . The other macrophage SFKs , Hck and Fgr , were also activated in response to 3-IB-PP1 treatment but were not capable of compensating for the lack of LynA activity after LynA degradation or in Lyn-deficient cells . In contrast , receptor clustering by zymosandep induced signaling through the MAPKs and Akt independently of LynA function and inflammatory priming . 3-IB-PP1 induces SFK activation without large-scale receptor clustering or formation of a phagocytic synapse , a quality shared with weak physiological stimuli . In the experiments discussed here , we treated cells with a high dose of 3-IB-PP1 to induce robust SFK activation and unmask signaling that would otherwise be undetectable with weak stimulation . When we titrate 3-IB-PP1 to lower doses , the signaling pattern is unchanged , just generally weaker in strength . Therefore , we believe that the strong SFK activation in response to 3-IB-PP1 is an advantage of this experimental system and allows the comparison of signaling in the absence of receptor clustering , as in a weak physiological stimulus , to signaling in response to zymosandep , which mimics a strong physiological stimulus ( Goodridge et al . , 2011 ) . In this context , differences in signaling between 3-IB-PP1- and zymosandep-treated cells reveal how receptor clustering relays signaling information . Importantly , we found that Erk , JNK , Akt , and NFAT activation patterns do not only depend on the strength of SFK activation and membrane-proximal signaling , but instead arise from the different roles of LynA and the other SFKs in the presence and absence of induced receptor clustering . We conclude , therefore , that large-scale receptor clustering does qualitatively change the chain of events necessary for signaling to occur independently of general SFK activation strength . We propose that the responsiveness of macrophages to weak stimuli is regulated by a signaling checkpoint in which degradation of LynA occurs when weak or low-valency ligands engage surface receptors; by contrast , LynA expression is maintained but not required when strong or high-valency ligands induce large-scale receptor clustering . As a result , signaling induced in the absence of significant receptor clustering is blocked at the level of Erk , JNK , Akt , and NFAT , while signaling induced by high-valency receptor clustering proceeds through these pathways . However , when cells are primed with IFN-γ , they upregulate expression of LynA sufficiently that even weak or low-valency ligands can produce downstream Erk , JNK , Akt , and NFAT signaling . While priming sensitizes macrophages to small amounts of high-valency pathogen ligands , it will also increase the chance to signaling to host-derived or nonpathogenic low-valency ligands that may lead to immunopathology . Our finding that LynA promotes signaling in response to 3-IB-PP1 was initially surprising because Lyn-deficient myeloid cells have defects in negative regulatory signaling ( Harder et al . , 2001 , 2004; Lamagna et al . , 2013 ) . Lyn phosphorylates ITIMs and their associated phosphatases , and so Lyn deficiency leads to hyperactivation of cell signaling ( Scapini et al . , 2009; Ingley , 2012 ) . However , such Lyn-deficient cells lack both LynA and LynB . When mast cells derived from Lyn−/− mice are reconstituted either with LynA or LynB , LynA expression leads to more robust PLCγ phosphorylation and better association of PLCγ with the adaptor protein LAT in response to FcϵRI ligation . In contrast , LynB expression drives the formation of a LynB-SHIP1 complex and impairs calcium and degranulation responses ( Alvarez-Errico et al . , 2010 ) . These observations are consistent with our conclusion that in macrophages , LynA is the primary mediator of downstream signaling in response to 3-IB-PP1 . The persisting activity of LynB after LynA degradation is likely to activate negative feedback responses and ensure deterioration of membrane-proximal signaling . It is likely , however , that the loss of some positive function of LynA leads to the primary blockade in signaling because Lyn-deficient macrophages ( lacking both LynA and LynB ) lose all capacity to signal through Erk and Akt in response to 3-IB-PP1 . In addition , LynB , like LynA , is upregulated during priming and fails to suppress LynA-mediated signaling . Therefore , we conclude that the effects of LynA activity are dominant or orthogonal to any negative-regulatory effects of LynB . We do not yet know how LynA uniquely promotes downstream signaling , but we speculate that some activity of LynA may protect PLCγ from inhibition by Dok-3 . Studies in B cells have shown that a complex of Dok-3 , SHIP1 , and Grb2 associates with Slp76 , Btk , and PLCγ and negatively regulates PLCγ , Ras/MAPK , and Akt function ( Honma et al . , 2006; Stork et al . , 2007; Losing et al . , 2012 ) . Two phosphorylation sites of Dok-3 ( Y378 and Y399 ) are not required for SHIP1-binding or inhibitory function ( Lemay et al . , 2000; Robson et al . , 2004 ) , and we speculate that these sites could sense LynA activity and block Dok-3 function . We have detected general tyrosine phosphorylation of Dok-3 and its association with SHIP1 in response to 3-IB-PP1 ( Figure 6—figure supplement 1 ) , and the role of Dok-3 in signaling will be the subject of future studies . Our observations diverge from the standard model of how SFK activation is reversed , namely through dephosphorylation of the activation loop by PTPN22/Pep or other phosphatases and subsequent phosphorylation of the inhibitory tail by Csk ( Brown and Cooper , 1996; Wu et al . , 2006; Zikherman et al . , 2010; Negro et al . , 2012; Okada , 2012 ) . However , previous work has shown that activated SFKs can phosphorylate the E3 ubiquitin ligase c-Cbl , which then targets them for degradation ( Rao et al . , 2002 ) . In fact , Lyn ubiquitination in mast cells has been observed within 15–30 s of FcϵRI activation ( Kyo et al . , 2003 ) . What surprised us was that bulk protein degradation appeared to account for all loss of activated Lyn after 3-IB-PP1 treatment . This mechanism is reminiscent of the regulation of receptor tyrosine kinases by Cbl ( Thien and Langdon , 2001 ) and internalization/degradation processes for deactivating the T-cell receptor ( Liu et al . , 2000 ) . This observation has important implications for how macrophages might respond to repeated encounters with weak stimuli . The LynA checkpoint would not only shut signaling down but also render macrophages refractory to restimulation by another weak stimulus , at least until redistribution at the membrane or new synthesis replenishes local reserves of LynA . We have observed one form of this desensitization during 3-IB-PP1 wash-out and subsequent stimulation with the growth factor M-CSF , but in vivo this process likely occurs in a highly spatially localized manner . This degradation and desensitization could prevent the accumulation of hyperactivated macrophages at sites rich in weak stimuli but lacking genuine pathogenic invaders . Activation of macrophage subsets is a key factor in the progression and severity of autoimmune diseases , cancer cell growth and differentiation , the success of cancer immunotherapies , and induction and maintenance of obesity ( Wynn et al . , 2013 ) . Moreover , alterations in macrophage sensitivity through defects in membrane-proximal negative regulators have already been linked directly to disease . For example , loss of negative-regulatory receptor tyrosine kinases Tyro3 , Axl , and Mer exacerbates lupus-like disease , rheumatoid arthritis , and inflammatory bowel disease through enhanced Toll-like receptor signaling ( Rothlin et al . , 2007; Lemke and Rothlin , 2008; Rothlin and Lemke , 2010 ) . Loss of either SHP-1 or SHIP1 , both negative regulators of ITAM signaling , causes severe myeloproliferation , skin abscesses , and lung failure ( Van Zant and Shultz , 1989; Shultz et al . , 1993; Tsui et al . , 1993; Helgason et al . , 1998; Abram et al . , 2013 ) . The LynA checkpoint constitutes an additional system by which macrophages sense inflammation and set basal sensitivity accordingly . Understanding the ways that macrophages become hypersensitive to stimulation , including the bypass of the LynA checkpoint , will provide insights into how we might limit this process and reduce the contribution of macrophage signaling to disease pathology . All mice used in these experiments were derived from the strain C57/BL6 . CskAS mice ( Tan et al . , 2014 ) are heterozygous for a BAC transgenic CskAS allele on a Csk-null background . Hck−/− mice ( Lowell et al . , 1994 ) and Lyn−/− mice ( Chan et al . , 1997 ) were used for antibody characterization and breeding . Lyn−/−CskAS mice were generated by crossing CskAS and Lyn−/− mice . All animals were backcrossed to C57/BL6 for 15 generations . BMDMs were prepared using standard methods ( e . g . , as described in Zhu et al . , 2008 ) . Bone marrow was extracted from femura/tibiae of mice . After hypotonic lysis of erythrocytes , BMDMs were derived on untreated plastic plates ( BD Falcon , Bedford , MA ) by culturing in Dulbecco's Modified Eagle Medium ( Corning Cellgro , Manassas , VA ) containing approximately 10% heat-inactivated fetal calf serum ( Omega Scientific , Tarzana , CA ) , 0 . 11 mg/ml sodium pyruvate ( UCSF Cell Culture Facility ) , 2 mM penicillin/streptomycin/L-glutamine ( Sigma-Aldrich , St . Louis , MO ) , and 10% CMG-12-14-cell-conditioned medium as a source of M-CSF ( Takeshita et al . , 2000 ) . After 6 or 7 days , cells were resuspended in enzyme-free ethylenediaminetetraacetic acid ( EDTA ) buffer and replated in untreated 6-well plates ( BD Falcon ) at 1 M cells per well in unconditioned medium ±25 U/ml IFN-γ ( Peprotech , Rocky Hill , NJ ) or 10 ng/ml GM-CSF ( eBioscience , San Diego , CA; Goodridge et al . , 2009 ) . BMDMs were resuspended and stained with PE-Cy7-conjugated anti-CD11c ( eBioscience clone N418 , 25-0114 ) , FITC-conjugated anti-F4/80 ( eBioscience clone BM8 , 11-4801 ) , and PE-conjugated CD11b ( BD Pharmingen , San Jose , CA , clone M1/70 553311 ) . BD and eBioscience antibodies to CD45 were used as single-stained controls . Data were collected on a BD LSRFortessa flow cytometer running FACSDiva software and analyzed in FlowJo ( TreeStar , Ashland , OR ) . Figures were made in Adobe Creative Suite ( San Jose , CA ) . Zymosan ( Sigma ) was depleted of TLR2 agonist as described previously ( Underhill , 2003 ) . Briefly , intact zymosan suspended in phosphate-buffered saline ( PBS ) was subjected to five 15-min cycles of boiling and sonication followed by pelleting and resuspension in fresh PBS . The sample was then boiled in 10 M NaOH for 1 hr , washed , resuspended in PBS , and counted . Stocks of 1000 M zymosandep particles per ml were stored at −20°C and thawed , briefly sonicated , pelleted , and resuspended before use . Stimuli prepared in Roswell Park Memorial Institute- ( RPMI- ) 1640 medium were added to adherent BMDMs at 37°C . In experiments with zymosandep , all stimuli were applied by pulse spinning at 37°C . BMDMs were stimulated with 10 zymosandep particles per cell ( Underhill , 2003 ) , 10 μM 3-IB-PP1 CskAS inhibitor ( Okuzumi et al . , 2009; Tan et al . , 2014 ) , 2–50 ng/ml recombinant mouse M-CSF ( eBioscience ) , 1 μM BAY 61-3606 Syk inhibitor ( Calbiochem; Yamamoto et al . , 2003 ) , 20 μM PP2 SFK inhibitor ( Calbiochem EMD Millipore , Billerica , MA; Hanke et al . , 1996 ) or the actin-remodeling agents Cytochalasin D ( 10 µM ) , Latrunculin A ( 0 . 5 µM ) , or Jasplakinolide ( 1 µM ) . Reactions were stopped by placing the plate on ice , aspirating the stimulus , adding nonreducing SDS sample buffer , scraping cells off the plate , incubating 5 min at 37°C , adding 50 mM dithiothreitol ( DTT ) , sonicating on a Diagenode Biorupter ( Denvillle , NJ ) , and boiling for 15 min . Samples were separated for immunoblot analysis by sodium dodecyl sulfate polyacrylamide gel electrophoresis ( SDS-PAGE ) ( Novex NuPAGE Thermo Fisher Scientific , Grand Island , NY ) . Stimulated BMDMs were lysed in 1% Lauryl Maltoside Lysis Buffer containing 150 mM NaCl , 0 . 01% sodium azide , and 10 mM Tris , pH 7 . 6 with 2 mM NaVO4 , 0 . 01 mg/ml Aprotinin , 0 . 01 M NaF , 0 . 01 mg/ml Leupeptin , 0 . 01 mg/ml Pepstatin A , 2 mM phenylmethanesulfonyl fluoride ( PMSF ) , and 0 . 4 mM EDTA . After scraping the plates , cells and detergent were incubated 30 min on ice . The lysate was then cleared by ultracentrifugation for 15 min at 50 , 000 rpm at 4°C in a Beckman ( Pasadena , CA ) TLA120 . 2 rotor . The lysates were precleared for 30 min at 4°C with Protein G Sepharose beads ( Life Technologies , Carlsbad , CA ) and normal rabbit or goat serum as appropriate ( Jackson ImmunoResearch , West Grove , PA ) . Protein G-Sepharose beads were covalently conjugated to Lyn or Hck antibodies ( see table below ) using dimethyl pimelimidate ( Sigma ) . Antibody-bound beads were added to the lysate and mixed 1 . 5 hr at 4°C to immunoprecipitate Hck or Lyn . Finally , the samples were applied to micro bio-spin chromatography columns ( Bio-Rad , Hercules , CA ) , washed , and eluted with SDS Sample Buffer . After transfer from SDS-PAGE to Immobilon P membrane ( EMD Millipore ) , blots were blocked with 3% bovine serum albumin ( BSA ) in 25 mM Tris , pH 8 . 0; 125 mM NaCl; and 0 . 02% NaN3 . Antibodies were applied in solutions containing 20 mM Tris , pH 8 . 0; 125 mM NaCl; and 0 . 05% Tween-20 ( plus 2% BSA for primary or 0 . 5% powered milk for secondary antibody solution ) . High-stringency washes after antibody binding were in similar Tris-buffered saline and Tween-20 ( TBST ) buffer with 200 mM NaCl . Antibodies were obtained from Cell Signaling Technology ( Danvers , MA ) , Santa Cruz ( Dallas , TX ) , ProMab Biotechnologies ( Richmond , CA ) , Promega ( Madison , WI ) , Life Technologies , and Sigma-Aldrich ) . Primary staining was performed with the following antibodies:AntibodySource-catalog no . ID/cloneErk1Santa Cruz-93-GC-16-GErk1/2Santa Cruz-93 + Santa Cruz-154C-16 + C-14FgrProMab-20318/Santa Cruz-503386G2 ( PM ) /M-60 ( SC ) HckSanta Cruz-1428M-28JNKCell Signaling-9252–LynACell Signaling-2796/Lowell LabC13F9 ( CS ) /7478 . 5 ( Lowell ) LynA + BSanta Cruz-1544NFAT1Cell Signaling-5861D43B1pAktS473Cell Signaling-4058244F9p-c-CblY700Cell Signaling-8869D16D7pErk1/2T202/Y204Cell Signaling-4377197G2pJNK1/2T183/Y185Promega-V7931–pLynY507Cell Signaling-2731–pHckY527Life-44-912Anti-pSFKnegpPI3Kp85-Y458/p55-Y199Cell Signaling-4228–pPLCγ1Y783Life-44-696 G–pPLCγ2Y759Cell Signaling-3874–pPLCγ2Y1217Cell Signaling-3871–pSFKY416Cell Signaling-4058–pSHIP1Y1020Cell Signaling-3941–pSHP-1Y564Cell Signaling-8849D11G5pSirpαY452Ken Swanson Lab10-1989pSykY352Cell Signaling-271765E4pTyrosineWeiss Lab + Millipore-05-9474G10 + pY20SykWeiss Lab5F5SHIP1Santa Cruz-8425P1C1VinculinSigma-Aldrich-V9264hVIN-1 Horseradish peroxidase- ( HRP ) conjugated secondary antibodies were from Southern Biotech ( Birmingham , AL ) , and blots were developed using Thermo Scientific SuperSignal Femto reagent . Blots were visualized on a Kodak Imagestation . Some blots were reprobed after HRP inactivation by azide treatment and freezing . Where applicable , immunoblots were analyzed quantitatively by performing densitometry with ImageJ software ( Bethesda , MD ) . Briefly , images were background-subtracted , and strips of the blot corresponding to each band were demarcated and analyzed for each time point/gel lane . Error bars reflect independent analysis of blots from at least three independent experiments . Figures were prepared in Graphpad Prism ( La Jolla , CA ) .	Macrophages are white blood cells that protect the body from infection by bacteria and other microbes . Proteins and sugars on the microbe bind to receptor proteins on the surface of the macrophage , which triggers the macrophage to engulf the cell and destroy it . Macrophages also release molecules that are toxic to the microbe and activate other immune responses in the body . It is vital that macrophages can tell the difference between normal body cells and microbes because if macrophages are activated at the wrong time , they can damage tissues and cause inflammatory diseases . When the receptor proteins are activated by contact with a microbe , they interact with a family of proteins called SFKs , which in turn stimulate communication systems called signaling pathways inside the macrophages that activate their immune responses . However , these responses are only triggered if many receptors are activated and clustered together . Thus , macrophages are able to identify whether a cell is a normal part of the body or is a foreign invader by sensing the number of receptors that cluster together as they bind to the cell . However , it is not clear if it is the clustering itself that signals a genuine encounter with a microbe , or whether this information comes from the strength of the response by the SFK proteins . Here , Freedman et al . addressed this question by developing a new experimental system that allows SFKs to be directly activated in macrophages in the absence of receptor clusters . The experiments used macrophages obtained from genetically engineered mice and found that the direct activation of SFK proteins alone does not fully activate signaling pathways in the macrophages . A signaling blockade occurs due to the rapid destruction of an SFK protein called LynA . When the macrophages were exposed to molecules that are signals of inflammation in the body , they produced more LynA . This allowed these macrophages to be activated even without the formation of receptor clusters by interactions with pathogens . Freedman et al . 's findings reveal that LynA acts as a checkpoint that primes macrophages to respond more aggressively when the body is under attack . Future work will be aimed at understanding how LynA is destroyed and how the detection of real pathogens overcomes the checkpoint .	[ "Abstract", "Introduction", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "immunology", "and", "inflammation" ]	2015	LynA regulates an inflammation-sensitive signaling checkpoint in macrophages
Mitochondrial division , essential for survival in mammals , is enhanced by an inter-organellar process involving ER tubules encircling and constricting mitochondria . The force for constriction is thought to involve actin polymerization by the ER-anchored isoform of the formin protein inverted formin 2 ( INF2 ) . Unknown is the mechanism triggering INF2-mediated actin polymerization at ER-mitochondria intersections . We show that a novel isoform of the formin-binding , actin-nucleating protein Spire , Spire1C , localizes to mitochondria and directly links mitochondria to the actin cytoskeleton and the ER . Spire1C binds INF2 and promotes actin assembly on mitochondrial surfaces . Disrupting either Spire1C actin- or formin-binding activities reduces mitochondrial constriction and division . We propose Spire1C cooperates with INF2 to regulate actin assembly at ER-mitochondrial contacts . Simulations support this model's feasibility and demonstrate polymerizing actin filaments can induce mitochondrial constriction . Thus , Spire1C is optimally positioned to serve as a molecular hub that links mitochondria to actin and the ER for regulation of mitochondrial division . Mitochondrial division is a complex process that is essential for survival in mammals ( Nunnari and Suomalainen , 2012; Archer , 2014 ) and is facilitated by the actin cytoskeleton ( De Vos et al . , 2005; DuBoff et al . , 2012; Korobova et al . , 2013 , 2014; Hatch et al . , 2014; Li et al . , 2015 ) . Two distinct steps define mitochondrial division - an initial constriction of mitochondrial membranes , followed by final membrane scission ( Friedman et al . , 2011; Korobova et al . , 2013; Murley et al . , 2013; Korobova et al . , 2014 ) . Scission is mediated by the dynamin-related protein , Drp1 , which self-assembles on the surface of the mitochondrial outer membrane into helices that drive final mitochondrial division ( Lackner and Nunnari , 2009; Archer , 2014 ) . The initial constriction step narrows the mitochondrial tube diameter , which is necessary for Drp1 helix assembly ( Labrousse et al . , 1999; Yoon et al . , 2001; Legesse-Miller et al . , 2003; Ingerman et al . , 2005; Friedman et al . , 2011; Mears et al . , 2011; Murley et al . , 2013 ) . This step is independent of Drp1 and occurs at ER-mitochondria intersection zones where ER tubules associate with and wrap around the mitochondrial outer membrane along the plane of mitochondrial division ( Friedman et al . , 2011; Korobova et al . , 2013; Murley et al . , 2013; Korobova et al . , 2014 ) . Along these zones , actin filaments polymerize , providing the force needed for constriction that floppy ER tubules lack ( Korobova et al . , 2013 , 2014; Hatch et al . , 2014 ) . Inverted formin 2 ( INF2 ) is a formin family protein that promotes actin filament polymerization in a regulated fashion ( Korobova et al . , 2013 , 2014 ) . An ER-anchored splice isoform of INF2 ( usually referred to as INF2-CAAX ) ( Chhabra et al . , 2009; Korobova et al . , 2013 ) has been shown to facilitate mitochondrial constriction and division via its actin polymerization activity ( Korobova et al . , 2013 ) . Given INF2 , but not actin assembly , is localized throughout the ER , how INF2-mediated actin assembly is specifically triggered at ER-mitochondria intersections to ensure mitochondrial division remains an open central question . Spire proteins are membrane-binding actin-nucleators that interact with and regulate formin proteins ( Bosch et al . , 2007; Quinlan et al . , 2007; Pechlivanis et al . , 2009; Kerkhoff , 2011; Pfender et al . , 2011; Schuh , 2011; Vizcarra et al . , 2011; Zeth et al . , 2011; Quinlan , 2013; Montaville et al . , 2014 ) . Synergistic promotion of actin assembly by Spire and formin proteins has been implicated in driving a variety of processes ranging from vesicle trafficking to DNA repair in the nucleus ( Pfender et al . , 2011; Schuh , 2011; Montaville et al . , 2014; Belin et al . , 2015 ) . In these systems , membrane-associated Spire proteins nucleate actin filaments , which are then further polymerized by formin proteins , ultimately leading to actin-dependent translocation . Given these characteristics of Spire proteins , we set out to investigate whether any Spire proteins could be involved in helping promote INF2- and actin-dependent constriction and division of mitochondria at ER-mitochondria association zones . Vertebrates have two known Spire genes , Spire1 and Spire2 . Each Spire protein contains highly conserved domains with specific capabilities , including: four actin-monomer binding WH2 domains necessary for nucleating actin filaments; an mFYVE domain that binds to intracellular membranes and facilitates oligomerization ( Kerkhoff , 2011; Dietrich et al . , 2013 ) ; and an N-terminal KIND domain that serves to bind to and regulate formin proteins ( Bosch et al . , 2007; Quinlan et al . , 2007; Pechlivanis et al . , 2009; Kerkhoff , 2011; Pfender et al . , 2011; Vizcarra et al . , 2011; Zeth et al . , 2011; Quinlan , 2013; Montaville et al . , 2014 ) . Although no Spire proteins have been shown previously to localize to mitochondria or the ER ( Dietrich et al . , 2013 ) , here we report a previously uncharacterized alternate splice-isoform of Spire1 ( named Spire1C ) that localizes to mitochondria , promotes actin assembly on mitochondrial surfaces , and interacts with ER-anchored INF2 to regulate mitochondrial constriction and division . Our results support a model where Spire1C and INF2 coordinately drive actin- and ER-dependent mitochondrial division . They also reveal that Spire1C directly links mitochondria to both the actin cytoskeleton and the ER . We identified and characterized a novel alternate splice-isoform of Spire1 that contains KIND and WH2 domains common to all Spire proteins , as well as a previously uncharacterized unique 58 amino acid alternate exon sequence ( ExonC ) ( Figure 1A and see ‘Materials and methods’ and Figure 1—figure supplements 1–3 for details on Spire1C cloning , probe generation , and sequence information ) . Because of its alternative ExonC sequence , we named the Spire1 isoform Spire1C ( Figure 1A ) . After determining that Spire1C mRNA was present in multiple mouse tissues ( Figure 1—figure supplement 3 ) , we examined its presence and conservation among species . Using the UCSC Genome Browser , we searched for additional DNA or mRNA sequences that contain Spire1C ( Kent et al . , 2002 ) . When compared to mouse , we found striking identity in ExonC for rat , rabbit , human , dog , elephant , opossum , platypus , and chicken . It is not found in the annotated zebrafish sequence . Compared to our amplified mouse sequence of 58 amino acids , human ExonC differs by 2 residues ( 96% identical ) , platypus by 8 residues ( 86% identical ) , and chicken by 12 residues ( 79% identical ) . Sequence homology is strongest among mammals , but chicken maintains conservation that is unlikely to be solely due to chance . Conservation among species extends beyond the coding exon and into the upstream and downstream intronic regions of the gene , stretching ∼150 bases in the 3′ direction ( data not shown ) . These conserved extensions are likely involved in splicing regulation of the exon . The high level of conservation within and surrounding ExonC suggests that its role is indispensible for the health of the organism . 10 . 7554/eLife . 08828 . 003Figure 1 . The Spire1 alternate exon ExonC is necessary and sufficient for localization to mitochondria . ( A ) Full length Spire1C domain structure: The number ranges indicate the amino acid regions of the conserved domains probed in this study . ( B ) Spire1C localizes to mitochondria . myc-Spire1C: U2OS cells cotransfected with myc-Spire1C and mitoRFP show robust localization of myc-Spire1C to mitochondria . α-ExonC: U2OS cells stained with an antibody raised against ExonC ( α-ExonC ) and expressing mitoRFP show endogenous Spire1C labeling on mitochondria . GFP-ExonC: U2OS cells cotransfected with GFP-ExonC and mitoRFP show robust targeting of GFP-ExonC to mitochondria . myc-Spire1ΔC: U2OS cells cotransfected with myc-Spire1ΔC and mitoRFP show no specific targeting of myc-Spire1ΔC to mitochondria . All cells were fixed and primary antibodies were counterstained with Alexa-488 secondary antibody before imaging with confocal fluorescence microscopy . Scale bars: 10 μm . Inserts are magnifications of the boxed regions . DOI: http://dx . doi . org/10 . 7554/eLife . 08828 . 00310 . 7554/eLife . 08828 . 004Figure 1—figure supplement 1 . Construction of the complete Spire1C protein sequence as explained in detail in the ‘Materials and methods’ . Spire1C gene amplified from mouse brain cDNA . ( A ) Initial construct based on NM_194355 was later combined with the sequence based on AK129296 to form full-length Spire1C protein of 802 amino acids with all known exons . ( B ) Spire1C diagram showing both characterized ( red , blue , yellow , and orange ) and uncharacterized ( purple , green , and black ) domains ( indicated in the legend on the bottom ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08828 . 00410 . 7554/eLife . 08828 . 005Figure 1—figure supplement 2 . Constructs used to probe Spire1 function . ( A ) Spire1C constructs used in this study . Residue numbers are for the 802 a . a . full-length mouse Spire1C protein . ( B ) Coomassie-stained gels of purified proteins used for antibody production ( pET constructs ) and affinity purification ( GST constructs ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08828 . 00510 . 7554/eLife . 08828 . 006Figure 1—figure supplement 3 . Spire1C contains a previously uncharacterized alternate exon of 58 amino acids . ( A ) Spire1 ExonC was found to be present in the tissues listed . *PCR from cDNA did not yield conclusive results , but the ExonC-containing gene was originally amplified from mouse brain cDNA . ( B ) Sequencing results from TOPO pCR2 . 1 vector using the M13 reverse primer . ExonC sequence is shown in red . ( C ) Translated protein sequence from red portion in C . ( D ) Representative immunoblots comparing commercial antibodies to Spire1 used to detect endogenous Spire1 in HeLa and PC12 cell lysates , 25 μg per lane . ( E ) Affinity purified Spire1 C-terminal antibody and ExonC antisera ( both generated in this study ) tested on HeLa and PC12 cell lysates , as in ( A ) . Strikingly different protein bands are observed for the same cell type with different antibodies and between cell types with the same antibody . Molecular weight protein markers are the same size for each blot , though the positions vary slightly ( as indicated ) . ( F ) The Phyre server was used to predict any secondary structural elements in ExonC . Phyre compiles structure predictions from three different algorithms ( psipred , jnet , and sspro ) to form a consensus prediction and probability for each structural element . Probability ranges from 1–10 , with 10 being the most probable . Yellow amino acids are small/polar , green are hydrophobic , red are charged , and purple are aromatic + cysteine . DOI: http://dx . doi . org/10 . 7554/eLife . 08828 . 006 When cells were transfected with a myc-tagged Spire1C construct , the protein showed extensive co-distribution with the mitochondrial marker mitoRFP ( Figure 1B , myc-Spire1C ) . A polyclonal antibody generated against a peptide containing the unique 58 amino acids in ExonC ( Figure 1B and Figure 1—figure supplements 1–3 ) also showed extensive mitochondria-specific labeling within cells ( Figure 1B , α−ExonC ) ( see Figure 1—figure supplements 2 , 3 and ‘Materials and methods’ for additional information on α-ExonC ) . Testing the role of ExonC in mitochondrial targeting of Spire1C , we found that a GFP-tagged ExonC fusion protein robustly localized to mitochondria in expressing cells ( Figure 1B , GFP-ExonC ) . By contrast , a myc-tagged Spire1C construct lacking ExonC never showed specific mitochondrial localization ( Figure 1B , myc-Spire1ΔC ) . These results suggest that Spire1C is an endogenously expressed mitochondria-associated protein that targets to mitochondria via its ExonC domain . To examine whether Spire1C distributes on the surface or interior of mitochondria we compared the distribution of GFP-ExonC ( marking Spire1C ) with mitoRFP ( marking the mitochondrial matrix ) using structured illumination microscopy ( SIM ) , which gives a twofold resolution improvement over conventional confocal imaging ( Allen et al . , 2014 ) . GFP-ExonC labeling on mitochondria surrounded that of mitoRFP labeling ( Figure 2A ) , suggesting Spire1C localizes to the periphery of mitochondria , most likely on the mitochondrial outer membrane . 10 . 7554/eLife . 08828 . 007Figure 2 . Spire1C localizes to the mitochondrial outer membrane with its formin and actin binding domains facing the cytoplasm . ( A ) ExonC localizes to the peripheral region of mitochondria . Left: structured iIllumination microscopy ( SIM ) image of a U2OS cell transfected with GFP-ExonC and mitoRFP reveals localization of GFP-ExonC to the periphery of mitochondria . Right: Magnification of boxed region on left . The insert on the lower right corner is a fluorescence intensity linescan of the rectangular boxed region indicating the inversely related profiles of GFP-ExonC vs mitoRFP . Scale bar: 10 μm . ( B ) Illustration of the principle of the fluorescence protease protection ( FPP ) assay performed on mitochondrial outer membrane ( MOM ) proteins . If a fluorescent protein tag faces the cytoplasm ( green circle ) , it is degraded by trypsin and its fluorescence is depleted . If the protein tag faces the interior of the mitochondria ( blue triangle ) , it is protected from trypsin in the cytoplasm , and thus its fluorescence remains after trypsin addition . An intermembrane space ( IMS ) marker , OMI-mCherry , serves as a control to verify that the mitochondrial outer membrane has not been permeabilized by the digitonin treatment , and furthermore to confirm that trypsin is not degrading proteins in the IMS . ( C ) Schematic of the constructs used in our FPP assays . ( D ) Cells cotransfected with OMI-mCherry and GFP-Spire1C ( N-terminal GFP tag , left ) or GFP-ExonC ( N-terminal GFP-tag , middle ) were treated with 20 μM digitonin and 4 mM trypsin . In both cases OMI-mCherry fluorescence remained , whereas the N-terminal GFP tags were mostly depleted within 60 s after trypsin treatment . In contrast , in cells transfected with ExonC-GFP ( C-terminus GFP tag ) , GFP fluorescence remains unchanged after 60 s of trypsin treatment . Scale bar: 10 μm . ( E ) GFP-Spire1C laterally diffuses along the mitochondrial outer membrane . A small region of a mitochondrion labeled with GFP-Spire1C was photobleached ( white boxed region ) . The rapid , directional recovery from the unbleached region into the bleached region ( see also Figure 2—figure supplement 2 ) suggests GFP-Spire1C is stably associated with and laterally diffuses along the mitochondrial outer membrane . ( F ) GFP-Spire1C does not readily exchange with the cytoplasm or neighboring mitochondria since photobleaching of an entire mitochondrion resulted in very low fluorescence recovery over the same period of time as in ( E ) . Scale bar: 1 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08828 . 00710 . 7554/eLife . 08828 . 008Figure 2—figure supplement 1 . GFP-Spire1C laterally diffuses on the mitochondrial outer membrane . ( A ) After partial bleaching of mitochondria labeled with GFP-Spire1C , rapid fluorescence recovery occurs in a directional fashion . The fluorescence intensity in the region closest to the unbleached region ( green rectangle ) increases more rapidly than the fluorescence intensity in the region farther away ( red rectangle ) from the unbleached region . Scale bar: 2 μm . ( B ) Graph of the fluorescence intensity of the green and red regions in ( A ) shows the differential rate of recovery over time for each region . DOI: http://dx . doi . org/10 . 7554/eLife . 08828 . 008 To confirm Spire1C's mitochondrial outer membrane localization , we employed a fluorescence protease protection ( FPP ) assay ( Lorenz et al . , 2006 ) , which can determine a protein's membrane topology ( Figure 2B ) . GFP was fused to the N-terminus of Spire1C ( GFP-Spire1C ) , the N-terminus of ExonC ( GFP-ExonC ) , or to the C-terminus of ExonC ( ExonC-GFP ) ( Figure 2C ) . The constructs were then co-expressed in cells with OMI-mCherry , a mitochondrial intermembrane space ( IMS ) protein ( Muñoz-Pinedo et al . , 2006 ) . Thereafter , the plasma membrane of the cells was gently permeabilized with digitonin , followed by treatment with trypsin to extinguish cytoplasmic GFP fluorescence . Because mitochondrial membranes are not permeabilized by digitonin treatment , we reasoned that only if the fluorescent tag from these constructs faced the cytoplasm would their fluorescence be abolished by the trypsin . Both GFP-Spire1C and GFP-ExonC lost nearly all their fluorescence within 60 seconds of trypsin treatment . By contrast , fluorescence from ExonC-GFP was protected , similar to that seen for co-expressed OMI-mCherry , which as a mitochondrial IMS protein should be insensitive to trypsin ( Muñoz-Pinedo et al . , 2006 ) ( Figure 2D; Videos 1–3 ) . These results suggest that the WH2 and KIND domains of Spire1C face the cytoplasm ( since both reside N-terminal to ExonC ) ( Figure 2C ) , a topology where they could participate in formin-binding and actin-nucleation activity . The data further suggest that the C-terminus of ExonC is not exposed to the cytoplasm . This raises the possibility that ExonC is embedded in the outer membrane , either as a transmembrane or hairpin protein . 10 . 7554/eLife . 08828 . 009Video 1 . A U2OS cell coexpressing GFP-Spire1C ( N-terminus tag ) and OMI-mCherry displays rapid loss of GFP fluorescence signal after the addition of 10 μM digitonin and 4 mM trypsin , whereas mCherry fluorescence persists , indicating that trypsin is degrading the GFP tag on the N-terminus of Spire1C in the cytoplasm , but not OMI-mCherry , which resides in the mitochondria intermembrane space ( IMS ) . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08828 . 00910 . 7554/eLife . 08828 . 010Video 2 . A U2OS cell coexpressing GFP-ExonC ( N-terminus tag ) and OMI-mCherry displays rapid loss of GFP fluorescence signal after the addition of 10 μM digitonin and 4 mM trypsin , whereas mCherry fluorescence persists , indicating that trypsin is degrading the GFP tag on the N-terminus of Spire1C in the cytoplasm , but not OMI-mCherry , which resides in the mitochondria IMS . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08828 . 01010 . 7554/eLife . 08828 . 011Video 3 . A U2OS cell coexpressing ExonC-GFP ( C-terminus tag ) and OMI-mCherry displays no loss of GFP fluorescence signal after the addition of 10 μM digitonin and 4 mM trypsin , indicating the GFP tag on the C-terminus of ExonC is protected within the mitochondrial lumen . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08828 . 011 In support of this notion , transmembrane domain prediction software ( Claros and von Heijne , 1994 ) indicated that residues 26–46 within ExonC form an α-helix characteristic of prokaryotic transmembrane domains . Secondary structure prediction software PHYRE ( Kelley and Sternberg , 2009 ) also predicted a second α-helix within ExonC ( Figure 1—figure supplement 3 ) . Interestingly , when we expressed a full-length Spire1C construct with GFP fused to the C-terminus ( Spire1C-GFP ) , the protein remained mostly cytoplasmic and no longer properly localized to mitochondria ( data not shown ) . The protein also rapidly escaped cells treated with digitonin , suggesting it was not able to target efficiently to mitochondria . Taken together , our data suggests that Spire1C is a mitochondrial outer membrane protein , possibly with a hairpin conformation given ExonC's two predicted α-helix domains ( Figure 1—figure supplement 3 ) . We next performed photobleaching experiments to investigate the dynamics of Spire1C's association with mitochondrial membranes . Photobleaching of a portion of a mitochondrial element that expressed GFP-Spire1C resulted in rapid recovery of fluorescence , with replenishment arising first in regions close to the bleach site and later at regions further away ( Figure 2E and Figure 2—figure supplement 1 ) , similar to that seen for GFP-tagged proteins that feely diffuse along membranes ( Cole et al . , 1996 ) . In contrast , very little recovery during the same time period occurred when an entire mitochondrial element expressing GFP-Spire1C was photobleached ( Figure 2F ) . Thus , Spire1C appears to diffuse laterally along mitochondrial membranes , rather than rapidly bind and dissociate from the membrane , further supporting the idea of Spire1C being a mitochondrial outer membrane protein . Given that Spire proteins can nucleate actin via their highly conserved WH2-repeat domain ( Quinlan et al . , 2005; Loomis et al . , 2006; Salles et al . , 2009 ) , we next investigated whether overexpression of Spire1C promotes actin assembly on mitochondria . In non-transfected cells , only low levels of actin co-localized with mitochondria , without any apparent specificity ( Figure 3 and Figure 3—figure supplement 1 , control ) . Upon Spire1C overexpression , however , actin accumulated to high levels specifically on mitochondria ( Figure 3 and Figure 3—figure supplement 1 , Spire1C overexpression ) . This actin enrichment on mitochondrial surfaces was not dependent on Spire's formin-binding KIND domain , since overexpression of Spire1C lacking the KIND domain ( Spire1CΔKIND ) still induced actin enrichment on mitochondrial surfaces ( Figure 3 , Spire1CΔKIND overexpression ) . In contrast , actin enrichment on mitochondria was muted upon overexpression of Spire1C mWH2 ( Figure 3 and Figure 3—figure supplement 1 , Spire1C mWH2 overexpression ) , which contains mutations in its WH2 domains that block Spire-mediated actin nucleation ( Loomis et al . , 2006; Salles et al . , 2009 ) ( see ‘Materials and methods’ ) . These results demonstrated that Spire1C promotes actin assembly on mitochondria , most likely through Spire1C's ability to nucleate actin filaments . 10 . 7554/eLife . 08828 . 012Figure 3 . Spire1C promotes actin assembly on mitochondrial surfaces . Overexpression of Spire1C causes actin accumulation on mitochondria . Control: SIM image of a Cos7 cell expressing mitoEmerald and stained with phalloidin-568 to visualize actin shows low amounts of overlap between mitochondria and actin ( Mander's: 0 . 43 ± 0 . 020 , ncells = 26 ) . Spire1C overexpression: A Cos7 cell expressing mitoEmerald and overexpressing myc-Spire1C and stained with phalloidin-568 reveals significantly increased actin accumulation on mitochondria ( Mander's: 0 . 64 ± 0 . 066 , ncells = 19 , p < 0 . 05 ) compared to control cells . GFP-Spire1CΔKIND: A Cos7 cell overexpressing the formin-binding deficient GFP-Spire1CΔKIND stained with phalloidin-568 reveals significant accumulation of actin on mitochondria compared to control cells ( Mander's: 0 . 55 ± 0 . 017 , ncells = 18 , p < 0 . 05 ) . A Cos7 cell overexpressing GFP-Spire1C mWH2 displays no increased accumulation of actin ( Mander's: 0 . 41 ± 0 . 040 , ncells = 15 , p = 0 . 32 ) compared to control cells . Scale bar: 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08828 . 01210 . 7554/eLife . 08828 . 013Figure 3—figure supplement 1 . Spire1C promotes actin assembly near mitochondria in a WH2-dependent fashion . control ( α-ExonC ) : SIM image of a U2OS cell stained with phalloidin-488 and α-Spire1C to visualize endogenous Spire1C localization relative to actin shows very low amounts of overlap between mitochondria and actin . myc-Spire1C ( α-myc ) : A U2OS cell overexpressing myc-Spire1C , stained with anti-myc and phalloidin-488 reveals increased actin accumulation on mitochondria compared to non-transfected cells . myc-Spire1C mWH2 ( α-myc ) : A U2OS cell overexpressing the actin-nucleation deficient myc-Spire1C mWH2 labeled with anti-myc and phalloidin-488 reveals muted accumulation of actin on mitochondria compared to wild-type myc-Spire1C overexpression . Scale bars: 15 μm . The yellow-boxed regions are magnified underneath each image . All cells were counterstained with Alexa-568 secondary antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 08828 . 013 Given that actin assembly has been shown to play an important role in regulating mitochondrial fission ( De Vos et al . , 2005; DuBoff et al . , 2012; Korobova et al . , 2013 , 2014; Hatch et al . , 2014; Li et al . , 2015 ) , the observation that Spire1C localizes to mitochondria and promotes actin assembly on mitochondrial surfaces raised the possibility that Spire1C plays a role in mitochondrial fission . To test this directly , we assessed the effect of Spire1C overexpression , mutation and depletion on mitochondrial morphology , length , and fission . While all cells in all conditions in this study displayed a combination of fragmented and tubular mitochondria , overexpression of Spire1C resulted in a significant shift towards fragmented mitochondria compared to control cells ( Figure 4A , +Spire1C ) . In contrast , overexpression of Spire1C mWH2 or Spire1CΔKIND resulted in a shift towards tubular mitochondria ( Figure 4A ) . Depletion of Spire1C using shRNA also resulted in a shift towards tubular mitochondria ( Figure 4B ) . To quantify these changes , we measured mitochondrial lengths in each of these conditions , and found that Spire1C overexpression resulted in shorter mitochondria on average , whereas mutation or depletion of Spire1C resulted in longer mitochondria ( Figure 4C , left graph , and Figure 4—figure supplement 1 ) , resembling the effect of disrupting mitochondrial fission due to the depletion of Drp1 or INF2 from cells ( Bleazard et al . , 1999; Labrousse et al . , 1999; Wakabayashi et al . , 2009; Korobova et al . , 2013 , 2014; Hatch et al . , 2014 ) . To determine whether these morphological changes were a result of altered mitochondrial fission dynamics , we counted the number of mitochondrial fission events in each of these conditions . We found that fission events increased in frequency when cells were overexpressing Spire1C , whereas overexpression of Spire1C mWH2 or Spire1CΔKIND decreased the frequency of fission events ( Figure 4C , right graph ) . Similarly , shRNA-mediated knockdown of Spire1C decreased the frequency of fission events ( Figure 4C , right graph ) . Taken together , our results show Spire1C promotes mitochondrial fission via Spire1C's actin-nucleating and formin-binding capabilities . 10 . 7554/eLife . 08828 . 014Figure 4 . Spire1C promotes mitochondrial fission via its formin-binding KIND and actin-nucleating WH2 domains . ( A ) U2OS cells overexpressing Spire1C display shorter mitochondria ( second panel , 2 . 2 ± 0 . 5 μm , nmitochondria = 211 , ncells = 14 , p < 0 . 0001 ) , whereas cells overexpressing Spire1C mWH2 ( third panel , 6 . 2 ± 1 . 52 μm , nmitochondria = 332 , ncells = 15 , p < 0 . 0001 ) or Spire1CΔKIND ( fourth panel , 9 . 0 ± 1 . 50 μm , nmitochondria = 232 , ncells = 16 , p < 0 . 0001 ) display longer , more tubulated mitochondria compared to control cells ( first panel , 3 . 57 ± 0 . 45 μm , nmitochondria = 322 , ncells = 34 ) . Scale bar: 10 μm . ( B ) Cells transfected with mitoEmerald ( and neighboring non-transfected cells ) stained with α-ExonC ( upper row ) showed robust colocalization of mitoEmerald and α-ExonC , with a mixture of tubulated and fragmented mitochondria . Cells cotransfected with Spire1 shRNA and mitoEmerald with no detectable α-ExonC labeling ( lower row ) display long , tubulated mitochondria ( 6 . 1 ± 1 . 26 μm , nmitochondria = 222 , ncells = 17 ) . All primary antibodies were counterstained with Alexa-568 secondary antibody . Scale bar: 15 μm . ( C ) Left: Average mitochondrial lengths for control cells and cells overexpressing Spire1C , Spire1C mWH2 , Spire1 shRNA or Spire1CΔKIND . Right: Average number of mitochondrial fission events in one cell in a timespan of 10 min for control ( ncells = 17 ) , Spire1C overexpressing ( ncells = 10 , p < 0 . 0001 ) , Spire1C mWH2 overexpressing ( ncells = 12 , p < 0 . 0001 ) , Spire1 knockdown ( ncells = 25 , p < 0 . 0001 ) and Spire1CΔKIND overexpressing ( ncells = 22 , p < 0 . 0001 ) cells . At least 3 separate experiments were performed for all conditions . Error bars represent standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 08828 . 01410 . 7554/eLife . 08828 . 015Figure 4—figure supplement 1 . Distribution of mitochondrial lengths measured in each condition . Histogram showing distribution of frequencies of mitochondrial lengths for each of the conditions shown in the bar graphs in Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 08828 . 015 Since ER tubules have been implicated in mitochondrial constriction and division ( Friedman et al . , 2011; Korobova et al . , 2013; Murley et al . , 2013; Korobova et al . , 2014 ) , we investigated whether Spire1C influences ER-mitochondria association . Upon overexpression of Spire1C or Spire1C mWH2 , we observed no significant change in ER-mitochondrial overlap or crossover sites of ER tubules and mitochondria compared to control cells ( Figure 5A , B ) . However , in cells overexpressing Spire1CΔKIND , we observed a significant decrease in ER-mitochondria overlap , as well as a decrease in the number of ER tubules crossing over mitochondria ( Figure 5A , B ) . This suggested that the KIND domain of Spire1C might play a role in regulating the extent of ER-mitochondria intersections within cells , and that Spire1CΔKIND-mediated disruption of mitochondrial fission ( see Figure 4C , right panel ) could be due to a reduction in ER-mediated mitochondrial constriction in these cells ( Friedman et al . , 2011; Korobova et al . , 2013; Murley et al . , 2013 ) . 10 . 7554/eLife . 08828 . 016Figure 5 . Spire1CΔKIND overexpression reduces the amount of ER-mitochondria overlap . ( A ) Confocal images of cells expressing Ii33-mCherry and overexpressing GFP-Spire1C ( second row ) or GFP-Spire1C mWH2 ( third row ) , but not GFP-Spire1CΔKIND ( fourth row ) , display significant overlap of mitochondria with ER , similar to control cells expressing mitoEmerald . The images on the right-hand side show a magnified view of the boxed region in the merge image , with overlapping pixels in displayed in white . Scale bar: 15 μm . ( B ) Bar graph representing the average number of ER-mitochondria intersections per cell . We were able to resolve an average of 14 . 3 ± 3 . 5 intersections in control cells ( ncells = 12 ) . GFP-Spire1C overexpressing cells had 14 . 7 ± 1 . 93 ( ncells = 15 ) ER-mitochondria intersections per cell . GFP-Spire1C mWH2 expressing cells had an average of 14 . 7 ± 1 . 62 ( ncells = 11 ) ER-mitochondria intersections per cell . Spire1CΔKIND expressing cells had 6 . 8 ± 1 . 33 ( ncells = 14 , p < 0 . 05 ) ER-mitochondria intersections per cell . DOI: http://dx . doi . org/10 . 7554/eLife . 08828 . 016 One way the KIND domain of Spire1C could affect ER-mediated mitochondrial division is by binding to ER-anchored INF2 . To test this possibility , we performed in vitro GST pull-down assays . We found that the C-terminal half of INF2 ( INF2-CT ) , but not the N-terminal half ( INF2-NT ) , associated with a GST-tagged Spire1C KIND domain , but not GST alone ( Figure 6A ) . Addition of the N-terminal half of INF2 ( INF2-NT ) inhibited the interaction between Spire1C KIND and INF2-CT in our GST pulldown assays ( Figure 6A; last well ) , suggesting that INF2's ability to self-interact ( Chhabra and Higgs , 2006; Ramabhadran et al . , 2012 , 2013 ) can regulate its association with Spire1C . We further confirmed the interaction between Spire1C's KIND domain and INF2 using fluorescence anisotropy ( Ramabhadran et al . , 2013 ) ( Figure 6B ) , which showed a specific interaction between Spire1C's KIND domain and INF2 . Taken together , these data demonstrate that the Spire1C KIND domain directly binds to INF2 . 10 . 7554/eLife . 08828 . 017Figure 6 . Spire1C and inverted formin 2 ( INF2 ) directly interact and work together to regulate mitochondrial fission . ( A ) INF2-CT directly binds to Spire1-KIND in vitro . GST pull-down assays in actin polymerization buffer , containing combinations of the following: 20 μM GST or GST-Spire1 KIND bound to glutathione-sepharose beads; 1 μM INF2-CT; and 10 μM INF2-NT . Co-incubation of the GST-Spire1 KIND domain pulls down INF2-CT ( third to last lane ) , but not INF2-NT ( second to last lane ) . INF2-CT pulldown is inhibited by the addition of the INF2-NT ( last lane ) . STDS lane represents 0 . 2 μM INF2-CT . ( B ) Fluorescence anisotropy binding curve of purified INF2-CT ( 20 nM ) labeled with tetramethylrhodamine succinimide mixed with varying concentrations of Spire1 KIND or bovine serum albumin reveals a direct interaction between Spire1 KIND and INF2-CT . ( C ) Cells overexpressing a constitutively active INF2 mutant ( INF2 A149 alone , ncells = 16 , nmitochondria = 232 ) display very short , fragmented mitochondria compared to control cells ( p < 0 . 0001 ) . Cells overexpressing A149 and Spire1C ( +Spire1C , ncells = 14 , nmitochondria = 461 ) or Spire1C mWH2 ( +Spire1C mWH2 , ncells = 20 , nmitochondria = 377 ) similarly display very short mitochondria . In contrast , cells overexpressing A149 and Spire1CΔKIND display longer , more tubulated mitochondria ( +Spire1CΔKIND , ncells = 18 , nmitochondria = 379 , p < 0 . 0001 ) . ( D ) Cells overexpressing Spire1C and treated with scrambled siRNA display shorter , more fragmented mitochondria ( +scramble siRNA , ncells = 20 , nmitochondria = 434 , p < 0 . 05 ) . In contrast , cells overexpressing Spire1C and treated with INF2 siRNA display significantly longer , more tubulated mitochondria ( Spire1C + INF2 siRNA , ncells = 22 , nmitochondria = 627 , p < 0 . 0001 ) . Scale bars: 5 μm . ( E ) Bar graph displaying average mitochondria lengths for each of the conditions in this figure . Error bars represent standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 08828 . 017 We next tested whether disrupting Spire1C's interaction with INF2 inhibits mitochondrial fission . To test this hypothesis , we first asked whether removing the KIND domain from Spire1C blocks the increase in mitochondrial division associated with overexpressing a constitutively active INF2 mutant , INF2 A149 ( Korobova et al . , 2013 , 2014 ) . Consistent with previous reports ( Korobova et al . , 2013 , 2014 ) , we found that overexpressing INF2 A149 resulted in significant shortening of mitochondria ( Figure 6C , E; A149 alone ) . Similarly , co-overexpression of INF2 A149 and Spire1C or INF2 A149 and Spire1C mWH2 resulted in very short , fragmented mitochondria ( Figure 6C , E +Spire1C or +Spire1C mWH2 ) . By contrast , overexpression of INF2 A149 and Spire1CΔKIND significantly disrupted INF2 A149-mediated mitochondrial fragmentation ( Figure 6C , +Spire1CΔKIND ) . These results suggest that while Spire1C actin-nucleating activity may be unnecessary for mitochondrial fission when INF2 is constitutively active , INF2 binding to the Spire1C KIND domain is necessary for INF2 to maximally induce mitochondrial fission . In further confirmation of the hypothesis that Spire1C and INF2 jointly work to drive mitochondrial fission , we found that siRNA-mediated knockdown of INF2 disrupted Spire1C-mediated upregulation of mitochondrial fission ( Figure 6D ) . Taken together , the results suggest that Spire1C interacts with INF2 via Spire1C's KIND domain , and that this interaction promotes mitochondrial fission . Mitochondrial fission would be co-dependent on Spire1C and INF2 if Spire1C's interaction with INF2 drives ER-mediated mitochondrial constriction . To test this hypothesis , we employed confocal fluorescence imaging of ER and mitochondria to examine mitochondrial constriction sites in cells co-expressing the ER marker Ii33-mCherry and different variants of Spire1C . Mitochondria constriction was visible at sites of ER-mitochondria crossover slightly more frequently in cells overexpressing Spire1C compared to cells not overexpressing the construct ( Figure 7A , B , Spire1C , see arrows for ER-mediated mitochondrial constrictions , and arrowheads for ER-mitochondria intersections that didn't result in constrictions ) . Notably , cells overexpressing Spire1C mWH2 displayed significantly fewer mitochondrial constrictions at ER-mitochondria intersections ( Figure 7A , B , Spire1C mWH2 ) . Similarly , cells overexpressing Spire1CΔKIND showed a decrease in the frequency of constrictions at ER-mitochondria intersections ( Figure 7A , B , Spire1CΔKIND ) . RNA-mediated Spire1C knockdown also resulted in decreased constrictions ( Figure 7A , B , Spire1C shRNA ) . Finally , overexpressing Spire1C in cells treated with INF2 siRNA showed a significant reduction in mitochondrial constrictions ( Figure 7A , B , INF2 siRNA + Spire1C ) . Taken together , our results suggest that Spire1C and INF2 work together to promote mitochondrial division by driving ER-mediated mitochondrial constriction , and that this process is dependent on Spire1C's ability to nucleate actin filaments on mitochondrial surfaces , as well as the ability for Spire1C and INF2 to interact via the Spire1C KIND domain . 10 . 7554/eLife . 08828 . 018Figure 7 . Spire1C overexpression enhances mitochondrial constriction via its WH2 and KIND domains in cooperation with INF2 . ( A ) Representative confocal images of U2OS cells expressing Ii33-mCherry in order to visualize ER tubules crossing over mitochondria in cells expressing mitoEmerald ( first row ) or overexpressing GFP-Spire1C ( second row ) , GFP-Spire1C mWH2 ( third row ) , GFP-Spire1CΔKIND ( fourth row ) , or GFP-Spire1C while treated with INF2 siRNA ( fifth row ) . Arrows indicate ER-mitochondria intersection points associated with mitochondrial constriction . Arrowheads indicate ER-mitochondria intersections not resulting in mitochondrial constriction . Scale bar: 1 μm . ( B ) Bar graphs representing the average percentage of ER-mitochondria intersections associated with mitochondrial constriction for each construct used . In cells expressing mitoEmerald , 55 . 2 ± 5 . 5% ( nintersections = 172 , ncells = 12 ) of ER-mitochondria intersections appeared to result in mitochondrial constriction . In GFP-Spire1C overexpressing cells , 71 . 5 ± 4 . 5% ( nintersections = 221 , ncells = 15 , p < 0 . 05 ) of ER-mitochondria intersections resulted in mitochondrial constriction . In GFP-Spire1C mWH2 overexpressing cells , 24 . 1 ± 2 . 4% ( nintersections = 162 , ncells = 11 , p < 0 . 01 ) of ER-mitochondria intersections appeared to result in mitochondrial constriction . In Spire1C knockdown cells , 37 . 2 ± 5 . 7% ( nintersections = 123 , ncells = 11 , p < 0 . 001 ) of ER-mitochondria intersections appeared to result in mitochondrial constriction . In GFP-Spire1CΔKIND overexpressing cells , 29 . 5 ± 4 . 7% ( nintersections = 95 , ncells = 14 , p < 0 . 01 ) of ER-mitochondria intersections appeared to result in mitochondrial constriction . In GFP-Spire1C overexpressing cells treated with INF2 siRNA , 27 . 5 ± 5 . 3% ( nintersections = 178 , ncells = 16 , p < 0 . 01 ) of ER-mitochondria intersections appeared to result in mitochondrial constriction . DOI: http://dx . doi . org/10 . 7554/eLife . 08828 . 018 Given the above experimental results , we used in silico simulations to test a potential model in which forces mediated by the actin cytoskeleton induce mitochondrial constriction . In this scenario , actin filament polymerization within the gap between the ER tubule surrounding the mitochondria and the mitochondrial outer membrane ( Figure 8A ) results in localized pressure that drives mitochondrial constriction to diameters required for Drp1 helix formation ( Korobova et al . , 2013 ) . This pressure could originate either from forces exerted by actin polymerization against the mitochondrial outer membrane ( Korobova et al . , 2013 ) , or by myosin-II dimer mediated contraction of actin filaments lying between the ER and mitochondrial membranes ( Hatch et al . , 2014; Korobova et al . , 2014 ) , or by a concerted action of these two complementary mechanisms . 10 . 7554/eLife . 08828 . 019Figure 8 . Putative model for how mitochondrial Spire1C and ER-anchored INF2 could mediate mitochondrial constriction via actin filament assembly . ( A ) Spire1C:actin complexes on mitochondria associate with INF2 on the ER . Actin filaments nucleated by Spire1C are elongated by the actin polymerization activity of INF2 . The actin filament elongation activity exerts pressure on the mitochondrial outer membrane , thereby driving constriction of the latter . Tethering complexes may play a role in maintaining association between ER and mitochondrial membranes . Myosin-II dimers and the related contractile actin ring , which may also be involved in mitochondrial constriction , are not shown for simplicity . ( B ) Computational results showing mitochondrial shapes resulting from deformation by constricting pressure P developed by the actin polymerization and/or actin contractile based mechanisms ( see also Figure 8—figure supplement 1 , Figure 8—source data 1 , and ‘Materials and methods’ for more information ) . The mitochondrial constriction site was modeled as a tubular membrane of about 680 nm length and with initial radius R = 230 nm . The dark blue strip in the middle represents the 50 nm wide zone of the pressure application . The images correspond to 3° of the mitochondria constriction characterized by cross-sectional radii r in the narrowest place of 145 nm , 110 nm and 65 nm . The corresponding values of the pressure P , the required numbers of the polymerizing actin filaments , Nf , and the required tensions in the actin contractile ring , γm , are presented in Figure 8—figure supplement 1 and Figure 8—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 08828 . 01910 . 7554/eLife . 08828 . 020Figure 8—source data 1 . Specific values of the system parameters and the computational results for the three specific extents of mitochondrial constriction presented in Figure 8 , Figure 8—figure supplement 1 , and discussed in the main text . DOI: http://dx . doi . org/10 . 7554/eLife . 08828 . 02010 . 7554/eLife . 08828 . 021Figure 8—figure supplement 1 . Computational results of simulations of mitochondrial constriction mediated by actin polymerization and actin constriction mechanisms . ( A ) The cross-sectional radius in the narrowest place of the mitochondria shape , r , as a function of the pressure , P , exerted on the limited region in the middle of the constriction region ( see Figure 8 of the main text ) . The radius , r , and the pressure , P , are presented in the universal dimensionless forms , r/R , and PR3/2κ , where R is the initial ( preceding the deformation ) mitochondrial radius and κ is the membrane bending modulus . The dashed lines indicate the specific deformations presented in Figure 8 of the main text . ( B ) The number of the actin filaments , Nf , and the tension in the actin contractile ring , γm , providing the pressure as functions of the resulting mitochondria deformation . The deformation is quantified by r/R ( see ( A ) for definition ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08828 . 021 To substantiate this , we created a simulation of mitochondrial constriction in response to a localized pressure generated by the above-mentioned mechanisms ( Figure 8B , Figure 8—figure supplement 1 , and Figure 8—source data 1 ) . We modeled the constriction site of the mitochondrial outer membrane as a membrane tubule whose resistance to deformations is characterized by a bending modulus of 8 × 10−20 Joules , typical for a lipid bilayer ( Helfrich , 1973 ) . The pressure deforming the membrane tubule was applied in the middle of the constriction zone along a strip of 50-nm thickness , corresponding to that of a typical ER tubule , while the computed shapes of the mitochondria constriction region corresponded to those of three different constriction events imaged with electron tomography ( Friedman et al . , 2011 ) ( Figure 8B ) . The computed pressure values required for generation of these 3 degrees of mitochondrial constriction ( Figure 8—figure supplement 1 and Figure 8—source data 1 ) enabled us to calculate the numbers of polymerizing actin filaments , Nf , or the tension , γm , which has to be developed within the actin contractile system . Assuming that the force developed by one polymerizing actin filament is about 1 pN ( Footer et al . , 2007 ) , the estimated filament number , Nf , varies between 10–20 . The actomyosin tension values , γm , range from 2 to 3 pN . The obtained estimations for both Nf and γm are perfectly reasonable physiologically , which supports the feasibility of the suggested mechanisms . Thus , our results and model are fully consistent with previous studies suggesting that tightly regulated actin assembly at ER-mitochondria intersection sites facilitates mitochondrial membrane scission by Drp1 ( Friedman et al . , 2011; Korobova et al . , 2013 , 2014; Murley et al . , 2013; Hatch et al . , 2014; Li et al . , 2015 ) . A key event in the mitochondrion's life cycle is its division into distinct mitochondrial elements . Prior work studying this process demonstrated that division occurs at sites where ER wraps around mitochondria ( Friedman et al . , 2011; Murley et al . , 2013 ) , with the ER providing a platform for actin polymerization mediated by the ER-anchored formin INF2 ( Korobova et al . , 2013 , 2014; Hatch et al . , 2014 ) . This actin meshwork is proposed to provide the force that drives mitochondrial constriction prior to Drp1-mediated mitochondrial division . Missing from this picture has been a molecular player that regulates INF2-mediated actin polymerization , ensuring that polymerization occurs specifically at ER-mitochondria contact sites to drive mitochondrial constriction and division . Here , we demonstrate that Spire1C , a novel mitochondrial outer membrane protein , can serve this role by both binding to INF2 as well as by acting as an actin-nucleator . Spire proteins are membrane-binding actin-nucleators that interact with and regulate formin proteins ( Bosch et al . , 2007; Quinlan et al . , 2007; Pechlivanis et al . , 2009; Pfender et al . , 2011; Schuh , 2011; Vizcarra et al . , 2011; Quinlan , 2013 ) . Given this , Spire proteins are potential candidates for regulating the actin polymerization activity of INF2 on mitochondrial membranes . In searching for such a protein , we identified a specific isoform of Spire1 , Spire1C , which resides on mitochondria and interacts with INF2 . Spire1C is distinct from other Spire proteins in having mitochondrial outer membrane localization . This localization is a result of Spire1C's unique ExonC domain , which serves as a mitochondria-targeting sequence . Spire1C undergoes lateral diffusion on the mitochondrial outer membrane , and is oriented with its formin-binding and actin-nucleating domains facing the cytoplasm . Spire1C promotes actin assembly on mitochondrial outer membranes; when Spire1C is overexpressed a massive buildup of actin around mitochondria is observed . The actin buildup is dependent on Spire1C's actin-nucleating WH2 domain , but not its formin-binding KIND domain . Therefore , Spire1C's canonical actin-nucleating domain drives actin accumulation on mitochondria independently of its interactions with formin proteins . Given Spire1C's ability to assemble actin filaments on mitochondrial membranes , we examined whether modulating Spire1C's activity could affect mitochondrial lengths or their division rates , and we found that it can . Overexpressing Spire1C increases mitochondrial division rates while depleting Spire1C has the opposite effect , causing mitochondria to become highly elongated . Because the increased fission seen in cells overexpressing Spire1C depends not only on Spire1C's actin-nucleating WH2 domain but also its formin-binding KIND domain , we reasoned that Spire1C-mediated mitochondrial fission also depended on formin proteins . Given INF2's previously established role as a formin protein involved in ER-mediated mitochondrial constriction and division ( Korobova et al . , 2013 , 2014; Hatch et al . , 2014 ) , we hypothesized that Spire1C could be working together with INF2 . We postulated that Spire1C could be promoting mitochondrial division by interacting with ER-anchored INF2 , in order to enable mitochondria to come into close proximity with the ER so that actin-nucleation by Spire1C could enhance actin assembly mediated by INF2 . Supporting this possibility , we found in GST pulldown and fluorescence anisotropy assays that the Spire1C KIND domain directly binds to INF2 . In cells overexpressing Spire1C lacking its KIND domain-mediated INF2-binding activity ( Spire1CΔKIND ) or its actin-nucleating activity ( Spire1C mWH2 ) , ER-mitochondria associations leading to mitochondrial constrictions were significantly decreased . Moreover , overexpressing Spire1CΔKIND prevented mitochondria from dividing in cells expressing a constitutively active INF2 mutant ( INF2 A149 ) that normally induces dramatic mitochondrial fission ( Korobova et al . , 2013 , 2014 ) . Finally , Spire1C overexpression in cells lacking INF2 failed to induce mitochondrial fission . All these observations suggest a model in which mitochondrial Spire1C and ER-anchored INF2 conspire to mediate mitochondrial constriction via actin filament assembly . In this scheme , Spire1C:actin complexes on mitochondria associate with INF2 on the ER , acting together with other ER-mitochondria tethering complexes ( Rowland and Voeltz , 2012 ) to draw the two organelles together . This results in the ER wrapping around the mitochondria . Once this occurs , actin filaments nucleated by Spire1C are elongated by the actin polymerization activity of INF2 , similar to the ‘rocket launcher’ mechanism shown for other actin nucleating and formin proteins ( Breitsprecher et al . , 2012 ) . Because INF2 can both polymerize and sever actin filaments , a complex meshwork of actin grows between the ER and mitochondria , which may be further impacted by myosin-II dimer recruitment ( Hatch et al . , 2014; Korobova et al . , 2014 ) as well as perhaps other actin-regulatory proteins such as cofilin or Arp2/3 ( Derivery et al . , 2009; Liu et al . , 2009; Li et al . , 2015 ) . The growing actin meshwork between the ER and mitochondria then exerts pressure on the mitochondrial outer membrane causing its constriction . Our computational modeling of this process confirmed that polymerizing actin filaments from this meshwork has sufficient force to bend and constrict the mitochondrial membrane once the filaments abut the mitochondria surface . Clearly , further work is needed to clarify the mechanism by which Spire1C and INF2 facilitate mitochondrial division . First , a better understanding of how Spire1C and INF2 interact with and regulate one another's activities during mitochondrial constriction is required . The relatively low affinity between Spire1C's KIND domain and INF2 detected in our assays , if applicable to living cells , is consistent with Spire1C:INF2 dissociation once INF2 begins to elongate actin filaments . Second , as several proteins are known to tether ER-mitochondrial membranes ( Rowland and Voeltz , 2012 ) , the role of these proteins in promoting or disrupting Spire1C's interaction with INF2 also needs to be studied . Such interactions may underlie differences in the mode of mitochondrial division seen in cells undergoing apoptosis , mitophagy , mitosis , or in response to toxins such as LLO from Listeria ( Chan , 2012; Hoppins and Nunnari , 2012; Youle and van der Bliek , 2012; Stavru et al . , 2013 ) . Finally , the precise organization of the actin meshwork responsible for constricting mitochondria needs to be characterized at higher resolution . This will help determine whether the actin meshwork constricts mitochondria by myosin-mediated contraction , by elongating filaments pushing , or by a combination of both . While we have focused on Spire1C's role in mitochondrial constriction , the establishment of Spire1C as a mitochondrial outer membrane protein suggests that Spire1C is optimally positioned to serve as a molecular hub that links mitochondrial dynamics to the actin cytoskeleton as well as to the ER . While our appreciation of the role of actin in mitochondrial division is rapidly growing ( De Vos et al . , 2005; Korobova et al . , 2013 , 2014; Hatch et al . , 2014; Li et al . , 2015 ) , there are other important functions for actin in mitochondrial dynamics , such as mitochondrial motility in neurons ( Hollenbeck and Saxton , 2005; Pathak et al . , 2010 ) , mitochondrial partitioning prior to cell division in fibroblasts ( Quintero et al . , 2009; Rohn et al . , 2014 ) , and perhaps also the partitioning of mitochondrial DNA ( Boldogh et al . , 2003 , 2004; Reyes et al . , 2011 ) . This list is almost certainly not exhaustive; there may yet be other known and unknown roles for the actin cytoskeleton in mitochondrial biology , and vice versa . Interestingly , Spire1C directly interacts with the tail domain of myosin Va ( data not shown ) , an actin-binding motor protein that has been shown to be involved in both mitochondrial and ER movement in neurons ( Wagner et al . , 2011 ) . In other cellular systems myosin Vb , Rab11a , and Spire proteins cooperate to drive actin-based vesicle movements and dynamics ( Schuh , 2011; Montaville et al . , 2014 ) —perhaps similar mechanisms exist for mitochondrial movements . Along these lines , it is interesting to note that Rab11a has also been implicated in mitochondrial dynamics ( Landry et al . , 2014 ) —exploring these findings in the context of Spire1C function may provide new insight towards mitochondrial movements and dynamics , and perhaps the relationship between actin-dependent motility and actin-dependent fission . Finally , the recent discovery of a role for the ER in mediating endosomal constriction and division raises the possibility that endosomal isoforms of Spire ( Kerkhoff , 2006; Liu et al . , 2009 ) are playing a similar role in promoting ER/actin/INF2-mediated endosomal fission . In fact , results from previous studies suggest that overexpression of the endosomal Spire2 protein lacking its KIND domain may result in endosome elongation ( Dietrich et al . , 2013 ) , which would be analogous to what we have observed for Spire1CΔKIND overexpression and mitochondria . In conclusion , our identification and characterization of Spire1C as an ER- and actin-binding mitochondrial outer membrane protein opens the door for novel avenues towards understanding the regulation of myriad roles of actin , mitochondria , and the ER in cellular function and disease ( Rappold et al . , 2014 ) . U2OS and Cos-7 cells were purchased from ATCC ( Manassas , VA ) . U2OS cells stably expressing GFP-INF2 was described in Chhabra et al . ( 2009 ) . All cells were grown in DMEM ( Invitrogen , Carlsbad , CA ) with 10% fetal bovine serum . For imaging , fibronectin coated coverslips ranging between 168 and 172 μm ( for fixed cell imaging ) or #1 . 5 LabTek chambers ( for live cell imaging ) were incubated with 10 μg/ml of fibronectin in PBS at 37°C for 30 min prior to plating the cells . Transient transfections were performed using FuGene 6 ( Promega , Madison , WI ) according to the manufacturer's recommendations . For overexpression experiments , 1 μg of DNA per coverslip was used . For minimal perturbation while imaging ER and mitochondria , 50 ng of DNA was used as described in Friedman et al . ( 2011 ) . For siRNA transfections , cells were treated as in Korobova et al . ( 2013 ) . Briefly , U2OS cells stably expressing GFP-INF2 ( Chhabra et al . , 2009 ) were plated on 6-well plates , treated with 63 pg of siRNA per well , and analyzed 72 hr post-transfection . siRNA or shRNA-mediated knockdown was confirmed by loss of GFP-INF2 or GFP-Spire1C fluorescence . Ii33-mCherry was a generous gift from P Satpute ( National Institutes of Health , Bethesda , MD ) . MitoEmerald and mitoRFP were gifts from A Rambold ( National Institutes of Health , Bethesda , MD ) . Spire1C was amplified from mouse brain cDNA using the sequence of NM_194355 as a reference . We expected to obtain a sequence yielding protein corresponding to GI 37595748 , however , it contained an additional 58 residues ( ExonC ) . Thorough examination of all available mammalian Spire1 isoforms revealed at least 3 alternatively-spliced exons , which we refer to as exons A ( majority of KIND domain ) , B ( protein sequence AVRPLSMSHSFDLS ) , and C ( protein sequence VPRITGVWPRTPFRPLFSTIQTASLLSSHPFEAAMFGVAGAMYYLFERAFTSRWKPSK ) . To obtain a full-length Spire1C construct , we amplified the mouse Spire1 gene AK129296 , which contains the full KIND domain through the first 3 WH2 domains , along with NM_194355 , which contains a partial KIND domain and ExonC without Exon B . A series of amplifications of partial gene sequences was then performed to obtain versions of mouse Spire1 that were ± each of exons A , B , and C ( Figure 1—figure supplement 1 ) . Each variant of the Spire1 gene was cloned into the AscI and PacI sites of modified pCS2+ vectors containing epitope tag sequences adjacent to the multiple cloning site , creating Spire1C constructs with either N-terminal 6x-myc or fluorescent protein tags . For the FPP assay and knockdown experiments , human Spire1C ORF XM_005258122 ( acquired from Genscript USA Inc . , Piscataway Township , NJ ) was cloned into the pEGFP-C1/pmApple-C1 and pEGFP-N1/pmApple-N1 vectors ( Clontech ) using the XhoI-BamHI and NheI-AgeI restriction sites , respectively . A nucleation-deficient mutant of Spire1C ( Spire1C mWH2 ) was generated by utilizing internal PstI and AfeI restriction sites in the Spire1C gene . A sequence of 822 nucleotides of the Spire1C gene between internal PstI and AfeI sites was synthesized ( Genscript USA Inc . ) that contained alanines in place of the key hydrophobic residues required for nucleation in all four WH2 domains ( Quinlan et al . , 2005; Loomis et al . , 2006; Quinlan et al . , 2007 ) . Insertion of the alanine-mutated WH2 domains was confirmed with DNA sequencing . All primers used are shown below along with the WH2 mutant insertion . Primers for spire1 gene amplification and plasmid constructionPrimer 1 GCGCGGCGCGCCATGGAACTGCATACATTTCTGACCAAAATTAAGAGPrimer 2 GCGCTTAATTAATCAGATCTCGTTGATAGTCCGTTCTGAAGPrimer 3 GAGCAGGCGCGCCATGGCCAATACCGTGGAGGCTGPrimer 4 GGCGTTAATTAATCTAGTCTGCTCCGTCTAATTTCTTCPrimer 5 GACGGCGCGCCATGGCGCAGCCCTCCAGPrimer 6 GGCGTTAATTAATCTAGTCTGCTCCGTCTAATTTCTTCPrimer 7 CCATGTGCTCCAGGAAGAAGCCPrimer 8 CTGCCTTCCAAGCCATACTCTACTCTAC All primers are 5′ to 3′ . AscI or PacI restriction sites are underlined . Oligos flanking ExonC were designed to amplify the spire1C gene . A mouse tissue cDNA panel ( Clontech , Mountain View , CA ) was used as a template for amplification using Primers 7 and 8 . Amplified DNA containing ExonC was ∼400 bp , while DNA lacking this exon was ∼200 bp . Multiple oligomer sets were utilized to eliminate non-specific amplification while capturing as many on-target amplifications as possible . PCR reactions were run on a 2% agaorse gel , and bands of the appropriate size were excised from the gel and purified using a QIAquick gel extraction kit ( Qiagen , Germantown , MD ) . Purified DNA was cloned into the pCR II-TOPO vector using the TOPO-TA cloning kit ( Invitrogen ) . Sequence analysis was used to confirm the sequence of the amplified and inserted DNA . The Spire1C gene was amplified from mouse cDNA as described above . Vectors used for protein purification include a modified avidin-6x his-MBP-TEV-3x FLAG-Precision construct under the P1 promoter , as well as a modified pGEX vector for N-terminal GST fusion proteins . All vector backbones were gifts of Dr . Aaron Straight and are described in Figure 1—figure supplement 2 . ExonC and the C-terminal 50 residues of mouse Spire1 were cloned into the avi-his-MBP-TEV-3xFLAG-precision vector described above or a modified pGEX vector ( for N-terminal GST fusion proteins ) using standard techniques ( Figure 1—figure supplement 2 ) . Avi-his-MBP-TEV-3xFLAG-precision constructs were expressed and purified as described above with the following modifications . For avi-his-MBP-TEV-3xFLAG-precision constructs , protein was purified over Ni-NTA resin , and the eluate was further purified on an S-200 gel filtration column , followed by a HiTrap Q column to remove any contaminating DNA . Protein samples were sent to Cocalico Biologicals and injected into rabbits to produce antisera . GST fusion proteins were used for affinity column construction for affinity purification of antibodies . These proteins were purified by using single colonies of transformed Rosetta ( DE3 ) cells to inoculate 400 ml Terrific Broth ( TB; Invitrogen ) cultures containing 100 μg/ml carbenicillin , 34 μg/ml chloramphenicol , which was grown overnight at 37°C . This culture was diluted into 2 l of fresh TB with antibiotics and grown to an O . D . of 0 . 8–0 . 9 , at which time it was moved to 23°C . After 1 hr at 23°C , cultures were induced with 0 . 5 mM isopropyl β-D-1-thiogalactopyranoside for 3–4 hr and harvested as described for purification of avi-his-MBP-TEV-3xFLAG-Precision proteins above . Cells were thawed in lysis buffer ( 50 mM Tris , 1 M NaCl , 1 mM EDTA , 1 mM DTT , and protease inhibitor cocktail , pH 7 . 8 ) , sonicated , and lysates were centrifuged for 30 min at 125 , 000×g 4°C . Supernatant was applied to hydrated glutathione resin ( 2 ml bed volume per liter culture ) , protein bound for 1 hr at 4°C , and resin was washed extensively with lysis buffer . Protein was eluted with elution buffer ( 50 mM Tris , 150 mM NaCl , 1 mM EDTA , 20 mM reduced glutathione , 1 mM DTT , and protease inhibitor cocktail , pH 7 . 8 ) and loaded onto a HiTrapQ anion exchange column . A salt gradient of 150 mM to 1 M was used for protein elution . Spire1 affinity columns were made using GST-fusion proteins following the method of Finan et al . ( 2011 ) . Proteins were coupled to Affi-Gel 10 by washing with 5 resin/column vol ( CV ) of 0 . 2 M glycine , pH 2 and quickly equilibrating with PBS . Antisera was filtered through a 0 . 2 μm filter and flowed over the column continuously overnight . The column was washed with 50 CV wash buffer ( PBS with 500 mM NaCl and 0 . 1% Tween 20 ) , followed by 2 . 5 CV 0 . 2× PBS . Antibody was eluted with 1 CV 0 . 2 M glycine , pH 2 directly into 1 M Tris , pH 8 . 5 to neutralize the solution . Concentration of elution fractions was checked on a Nanodrop spectrophotometer using the IgG setting . The column was washed with 20 CV PBS , the antisera was re-filtered , and the purification process was repeated to isolate additional antibody . Fractions with an O . D . > 0 . 2 were pooled , dialyzed into PBS containing 50% glycerol , and stored at −20°C . Antisera to ExonC yielded no IgG after affinity purification . Instead , whole antisera were used for subsequent experiments to probe ExonC function and localization . Two rabbit polyclonal antibodies described above and three commercially available antibodies were used for examining expression patterns of Spire1 protein in various cell types . The antibodies discussed are: ( 1 ) Rabbit polyclonal anti-Spire1 C-term ( affinity-purified ) , ( 2 ) Rabbit polyclonal anti-Spire1 ExonC ( whole antisera ) , ( 3 ) Sigma mouse monoclonal anti-Spire1 , ( 4 ) Abcam ( Cambridge , UK ) mouse monoclonal anti Spire1 , and ( 5 ) Abnova ( Taipei , Taiwan ) rabbit antisera to Spire1 . Notably , all of the commercially-available antibodies were targeted to residues 482–584 of the Spire1 isoform lacking ExonC ( NP_064533 ) , and thus could only detect non-ExonC containing isoforms . Western blots were performed with 5–50 μg cell lysate and antibodies/antisera was tested at various concentrations , temperatures , and lengths of time for best conditions . Optimized conditions for all antibodies used in this work are described below . HRP-conjugated goat anti-rabbit secondary antibody was used at 1:20 , 000 in all cases . SourceSpeciesTypeTarget regionIB dilutionTimeTemperatureIF dilutionSigmamousemonoclonal482–584 of NP_064533; ‘last 100 a . a . ’; flanks ( but does not contain ) z′1:20001 hr23°Cn/a–O/N4°Cn/aAbcammousemonoclonal1:1000O/N4°Cn/aAbnovarabbitantisera1:1000O/N4°Cn/aThis studyrabbitpolyclonal , affinity purifiedlast 51 residues1:2000O/N4°C1 to 1001:30001 hr23°C–This studyrabbitantiseraExonC1:1000O/N4°C1 to 500 Spire-KIND ( amino acids 1–234 ) was expressed as a GST fusion protein in bacteria , and purified on glutathione-sepharose ( GE Biosciences , Buckinghamshire , UK ) followed by Superdex200 gel filtration ( GE Biosciences ) of the glutathione-eluted GST-fusion protein . GST-KIND or GST alone was re-bound to glutathione-sepharose in binding buffer ( 50 mM KCl , 1 mM MgCl2 , 1 mM EGTA , 10 mM Hepes-HCl pH 7 . 4 , 1 mM DTT , 0 . 02% thesit ( Sigma , St . Louis , MO ) , 10 μg/ml aprotinin , 2 μg/ml leupeptin , 0 . 5 mM benzamidine ) . INF2-CT ( amino acids 469–1249 , containing FH1 , FH2 and C-terminal regions ) and INF2-NT ( amino acids 1–420 , containing DID and dimerization region ) were purified as described ( Ramabhadran et al . , 2012 ) . Proteins were mixed at 20 μM GST protein , 1 μM INF2-CT and 10 μM INF2-NT in binding buffer and incubated overnight , then quickly washed once in binding buffer . Proteins in glutathione sepharose-bound pellet were resolved by SDS-PAGE . INF2 C-term ( human CAAX variant , amino acids 941–1249 ) was expressed in bacteria , purified and labeled on its N-terminal amine with tetramethylrhodamine succinimide as described ( Ramabhadran et al . , 2013 ) . Labeled INF2-C-term ( 20 nM ) was mixed with varying concentrations of Spire-KIND or bovine serum albumin ( BSA ) in 10 mM Hepes pH 7 . 4 , 50 mM KCl , 1 mM MgCl2 , 1 mM EGTA , 1 mM DTT , 0 . 5 mM Thesit detergent ( nonaethylene glycol monododecyl ether ) at 23°C for 1 hr before reading fluorescence anisotropy in an M-1000 fluorescence plate reader ( Tecan Inc ) at 530 nm excitation and 585 nm emission . Cells were washed in phosphate buffered saline ( PBS; pH 7 . 4 ) then fixed with 4% paraformaldehyde for 30 min . Cells were then permeabilized with 0 . 1% Triton X-100 for 30 min before being blocked overnight with 4% BSA at 4°C . The next day , cells were incubated with primary antibody for 2 hr , rinsed three times with PBS for 10 min each , then incubated with secondary antibodies ( Invitrogen ) for 1 hr , rinsed three times with PBS for 10 min each , then counterstained with phalloidin ( Invitrogen ) for 30 min , then rinsed with PBS three times , then mounted using ProLong Gold antifade reagent . Confocal images were acquired with an Apochromat 63× 1 . 4 NA objective lens ( Carl Zeiss , Jena , Germany ) on a Marinas spinning disk confocal imaging system ( Intelligent Imaging Innovations , Denver , CO ) using an EM charge-coupled device camera ( Evolve; Photometrics , Tucson , AZ ) , or a 100× Apo TIRF 1 . 49 NA objective ( Nikon Instruments , Tokyo , Japan ) on a Yokogawa CSU-X1 spinning disk system using an EM charge-coupled device camera ( Evolve; Photometrics ) . Cells were imaged in HEPES-buffered growth media . Confocal z-stacks were taken using 200 nm steps . Images were deconvoluted using Slidebook 6 . Individual 16-bit tiff image files were exported , then processed using ImageJ . GFP-Spire1C photobleaching experiments presented in Figure 2 and Figure 2—figure supplement 1 were carried out on a Marianas spinning disc confocal microscope equipped with a Mosaic Digital Illumination System . The laser power entering the Mosaic was 9 mW . Image acquisition and photobleaching of GFP-Spire1C ( including region selection and 405 laser exposure control ) were carried out using Slidebook 6 software . SIM imaging of fixed cells was performed using an ELYRA SIM ( Carl Zeiss ) with an Apochromat 63× 1 . 4 NA oil objective lens . Five angles of the excitation grid with five phases each were acquired for each channel and each z-plane , which were spaced at 110 nm each . SIM processing was performed using the SIM module of the Zeiss Zen software package . 16-bit grayscale tiffs were subsequently exported to ImageJ for quantification and processing into rendered colored images . Channels in maximum projection images were aligned in the xy-plane using maximum projection images of fluorescent beads . All image analysis and processing was performed using ImageJ . Mitochondria lengths were measured manually by first setting the scale according to pixel size , drawing a line along the length of the mitochondria , then using ImageJ's ‘measure’ function . Colocalization analysis and rendering was performed using the colocalization plugin included in the MacBiophotonics ImageJ plugin bundle ( http://rsb . info . nih . gov/ij/plugins/mbf/index . html ) . When calculating Pearson's values , the mitochondria channel was used as a mask for colocalization . ER-mitochondria intersection sites were visually identified as regions where ER tubules could be clearly visualized crossing mitochondria—these regions were always in the periphery of the cell , significantly restricting the total number of intersections that could be reliably identified . Mitochondria constriction sites were visually identified as regions defined by relative narrowing of mitochondria diameter or reduced fluorescence . Magnifications of boxed regions were generated using ImageJ . Color images of merged 16-bit tiffs exported from the microscope were generated using the ImageJ ‘merge channels’ function . Statistical analysis was performed using Excel ( Microsoft , Redmond , Washington ) . p-values were determined using the unpaired Student's t-test or ANOVA , as appropriate . The deformed shapes of the membrane tubule representing the constriction site of the mitochondrial outer membrane were determined by minimizing the energy of the membrane bending upon the condition of a given pressure P acting on a limited region in the middle of the tube ( Figure 8B ) . The value of the bending energy , FB , was determined by ( 1 ) FB= ∫12 κ J2 dA , where κ = 8 × 10−20 Joule is the lipid bilayer bending modulus , J is the local total curvature of the membrane surface changing along the membrane surface and equal at each point to the sum of the local principal curvatures ( Helfrich , 1973; Spivak , 1979 ) . The integration in Equation 1 is performed over the whole surface of the deformed tubule . The boundary conditions for the energy minimization consisted in the requirements that at the tubule left and right edges ( i ) the tubule cross-sectional radius , r , remains equal to its initial ( preceding the deformation ) value r = R , and ( ii ) the tubule profile remains parallel to the tubule axis . While the tubule length L = 680 nm was required to remain constant during the deformation , the tubule surface area was free to change . This means that the membrane lateral tension was taken to be zero , which guaranteed that the membrane bending energy was the sole contribution to the membrane elastic energy . The energy minimization and the shape determination for each pressure value were performed using Brakke's ‘Surface Evolver’ program ( Brakke , 1992 ) .	Mitochondria are structures within cells that provide the energy to power many biological processes that are essential for complex life . These structures are also highly dynamic and go through cycles of fission ( in which a single mitochondrion splits in two ) and fusion ( in which two mitochondria merge into one ) . These processes both maintain the correct number of mitochondria in a cell and remove damaged ones , and defects in either can result in many diseases . Previous research had shown that mitochondria are in close contact with another cellular structure called the endoplasmic reticulum . The points of contact mark the sites where mitochondria undergo fission , as small tubes of the endoplasmic reticulum wrap around , and then constrict , to split a mitochondrion . Other recent work revealed that a protein called INF2 is anchored on the endoplasmic reticulum where it promotes mitochondrial constriction . This protein builds actin subunits into long filaments that provide the force for constriction . However , it was not clear how INF2 became active , and whether there are proteins on mitochondria that interact with INF2 or actin . Manor , Bartholomew et al . have now used a combination of microscopy-based methods and biochemical analysis to discover that a mitochondrial protein called Spire1C performs all of these roles . Spire1C is found on the outer membrane of mitochondria; it interacts with INF2 to drive the formation of actin filaments that constrict mitochondria . These results suggest that Spire1C bridges the endoplasmic reticulum with the network of actin filaments . Further experiments then showed that increasing Spire1C levels in cells resulted in the mitochondria becoming fragmented due to increased constriction . On the other hand , depleting Spire1C had the opposite effect and caused mitochondria to become unusually elongated . Following on from this work , the next challenge is to see if Spire1C is used differently or similarly in the different processes that involve mitochondrial fission .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2015	A mitochondria-anchored isoform of the actin-nucleating spire protein regulates mitochondrial division
The vertebrate gut microbiota have been implicated in the metabolism of xenobiotic compounds , motivating studies of microbe-driven metabolism of clinically important drugs . Here , we studied interactions between the microbiota and indomethacin , a nonsteroidal anti-inflammatory drug ( NSAID ) that inhibits cyclooxygenases ( COX ) -1 and -2 . Indomethacin was tested in both acute and chronic exposure models in mice at clinically relevant doses , which suppressed production of COX-1- and COX-2-derived prostaglandins and caused small intestinal ( SI ) damage . Deep sequencing analysis showed that indomethacin exposure was associated with alterations in the structure of the intestinal microbiota in both dosing models . Perturbation of the intestinal microbiome by antibiotic treatment altered indomethacin pharmacokinetics and pharmacodynamics , which is probably the result of reduced bacterial β-glucuronidase activity . Humans show considerable inter-individual differences in their microbiota and their responses to indomethacin — thus , the drug-microbe interactions described here provide candidate mediators of individualized drug responses . The composition of the intestinal microbiota is relatively stable in adult humans , but the taxa present differ considerably among individuals ( Gill et al . , 2006; Arumugam et al . , 2011; Wu et al . , 2011 ) . The gut microbiota is influenced by host genetics ( Turnbaugh et al . , 2009; Benson et al . , 2010 ) , ageing ( Biagi et al . , 2010; Agans et al . , 2011 ) , the use of antibiotics ( De La Cochetière et al . , 2005; Dethlefsen et al . , 2008; Jernberg et al . , 2010; Dollive et al . , 2013 ) , lifestyle ( Annalisa et al . , 2014 ) , diet ( Wu et al . , 2011; Zoetendal and de Vos , 2014; Wu et al . , 2014 ) , time of day ( Thaiss et al . , 2014; Zarrinpar et al . , 2014; Liang et al . , 2015 ) , pet ownership ( Song et al . , 2013 ) and concomitant disease ( Zhao , 2013; Wu et al . , 2013 ) . Bacterial communities in the intestine help maintain mucosal structure ( Stappenbeck et al . , 2002; McDermott and Huffnagle , 2014 ) , defend against pathogens ( Littman and Pamer , 2011 ) , and metabolize dietary constituents such as fiber ( Sekirov et al . , 2010 ) , peptides , proteins , ( Farthing , 2004 ) and xenobiotics ( Zheng et al . , 2011; Nicholson et al . , 2012; Haiser et al . , 2013 ) . The intestinal microbiota contain ~3 . 3 million microbial genes , including genes encoding xenobiotics biodegradation and metabolism pathways ( Qin et al . , 2010 ) . These bacteria are implicated in biotransformation of over 30 approved drugs by direct or indirect mechanisms ( Okuda et al . , 1998; Sousa et al . , 2008; Clayton et al . , 2009; Haiser et al . , 2013 ) . For example , bacterially generated p-cresol competes with the widely used analgesic acetaminophen for O-sulfonation ( Clayton et al . , 2009 ) , and digoxin is directly inactivated by the gut Actinobacterium Eggerthella lenta ( Haiser et al . , 2013; 2014 ) . Nonsteroidal anti-inflammatory drugs ( NSAIDs ) suppress prostanoid production by inhibiting the cyclooxygenase ( COX ) -1 and -2 enzymes . NSAIDs are widely used for relief of pain and inflammation ( Green , 2001 ) . A limitation of these drugs is their association with adverse gastrointestinal ( GI ) complications ( Graumlich , 2001 ) . Coincidental disruption of both COX enzymes , such as is achieved at therapeutic doses of indomethacin in humans , is necessary to evoke GI lesions in experimental animals ( Brodie et al . , 1970; Stewart et al . , 1980 ) . However , germ-free ( Robert and Asano , 1977 ) and antibiotic-treated ( Koga et al . , 1999 ) rats are resistant to indomethacin-induced intestinal lesions , suggesting a role for the microbiota . Limited information is available as to the impact of NSAIDs on the composition of microbiome: indomethacin is reported to increase intestinal Enterococcus faecalis and decrease segmented filamentous bacteria ( SFB ) ( Dalby et al . , 2006 ) , while DuP 697 , a COX-2 inhibitor , increases the abundance of Gram-negative rods in rats ( Kinouchi et al . , 1998 ) . However , whether indomethacin induces compositional changes in intestinal microbiota and whether these changes are involved in indomethacin enteropathy remains unknown . Here , we investigate interactions between indomethacin and the intestinal microbiota . Deep sequencing of longitudinal samples provided evidence that indomethacin affects the composition of the gut microbiota following both acute and chronic exposure . Indomethacin undergoes enterohepatic recirculation ( Harman et al . , 1964 ) — it is glucuronidated in the liver by UDP-glucuronosyltransferases ( UGTs ) and the glucuronide is delivered to the SI with bile acids where it is de-conjugated and reabsorbed . Previously , a specific inhibitor of bacterial β-glucuronidase was reported to reduce GI damage inflicted by the anticancer drug CPT-11 ( Roberts et al . , 2013 ) and by several NSAIDs , including diclofenac , indomethacin and ketoprofen ( Saitta et al . , 2014 ) . Here , we provide direct pharmacokinetic evidence documenting the influence of the intestinal microbiota on indomethacin metabolism via de-glucuronidation of its metabolites during enterohepatic recirculation . Given that inter- and intra-individual variation in the intestinal microbiota is high in humans , these results suggest a possible role for the intestinal microbiota in diversification of human responses to NSAIDs . To assess the effect of indomethacin on the intestinal microbiota , we first analyzed the composition of the luminal and tissue-associated microbiota in mice prior to drug exposure ( Figure 1 ) . Eight anatomical sites were analyzed — the small and large intestines were analyzed at proximal , middle , and distal sites , and cecum and feces were also compared . For each of the intestinal sites , luminal contents and mucosa were compared . We purified DNA from tissue or feces and used 16S rRNA gene sequencing and community analysis implemented using the QIIME pipeline ( Caporaso et al . , 2010b ) to characterize geographic differences . The microbiota were compared between GI sites using UniFrac ( Lozupone and Knight , 2005 ) , and visualized using Principal Coordinate Analyses ( PCoA ) ( Caporaso et al . , 2010b ) of unweighted UniFrac distances ( describing the bacterial lineages presented in samples ) and weighted UniFrac distances ( describing the proportions of bacterial lineages in samples ) . 10 . 7554/eLife . 08973 . 003Figure 1 . Geographic heterogeneity of basal intestinal microbiota composition along the intestine in mice . Bacterial communities colonized in the mouse intestine were profiled using 16S rRNA gene sequencing and analyzed using QIIME ( Caporaso et al . , 2010b ) . ( A ) Principal coordinates analysis ( PCoA ) of unweighted ( left ) and weighted ( right ) UniFrac values ( Lozupone et al . , 2011 ) , depicting the comparison of microbial communities from luminal content ( round ) , mucosal tissue ( triangle ) , or feces ( square ) . The base line microbiota compositions along the intestine are heterogeneous at anatomical sites . Each point represents a sample , and each sample is colored according to the habitat sites in the intestine . N=17–20 . ( B ) Heat map of the microbiota composition in luminal content ( upper ) and mucosal tissue ( lower ) along the intestine . Each column represents sample , and each row represents one phylum . The proportions of phyla are indicated by the color code to the right . Anatomical sites of the intestine are indicated at the bottom . N=17–20 . DOI: http://dx . doi . org/10 . 7554/eLife . 08973 . 003 The composition of the intestinal microbiota varied considerably by anatomical site ( Figure 1A ) . Most samples were dominated by Bacteriodetes and Firmcutes . In the lumen , the cecum was dominated by Firmicutes ( 58 . 64% ± 3 . 49% ) , while Bacteroidetes were more abundant in the large intestine ( LI ) ( 50 . 51% ± 4 . 03% at proximal , 67 . 91% ± 2 . 72% at middle , 72 . 14% ± 2 . 25% at distal LI ) . ( Figure 1B , upper ) . In the mucosa , Firmicutes were dominant in SI ( 52 . 40% ± 6 . 56% at proximal , 48 . 34% ± 5 . 65% at middle , 45 . 49% ± 6 . 61% at distal SI ) , cecum ( 60 . 48% 2 . 76% ) , and proximal LI ( 85 . 70% ± 2 . 00% ) ( Figure 1B , lower ) . Proteobacteria were abundant in the proximal SI ( 14 . 0% ± 4 . 27% in the lumen , 22 . 31% ± 6 . 37% in the mucosa ) and distal LI ( 11 . 34% ± 2 . 72% ) , though results were more heterogeneous than at other sites , probably in part reflecting low bacterial biomass in the starting material ( Figure 2A ) . LI lumen and fecal compositions exhibited considerable similarity ( Figure 1A , p<0 . 001 for both weighted and unweighted UniFrac distance , ADONIS test ) . The composition of the microbiota in the luminal content differed from that at the mucosal surface throughout the intestine ( p<0 . 001 for both weighted and unweighted UniFrac distance , ADONIS test ) , as indicated by the separation of luminal content samples and mucosal tissue samples ( Figure 1A ) . 10 . 7554/eLife . 08973 . 004Figure 2 . Indomethacin induces changes in microbial composition along the intestine in mice . Bacterial load in samples were inferred from 16S rRNA gene quantitative PCR . Bacterial communities colonized in the mouse intestine were profiled using 16S rRNA gene sequencing and analyzed using QIIME ( Caporaso et al . , 2010b ) . ( A ) 16S rRNA gene copies per gram of luminal contents ( left ) and mucosal tissues ( right ) at anatomical sites along the intestine in indomethacin ( red ) , PEG400 ( blue ) , and untreated ( black ) groups . Microbial loads at anatomical sites along the intestine are barely different between PEG400 and indomethacin groups , although PEG400 causes changes by itself . *p<0 . 01 by multiple t test comparing PEG400 versus indomethacin groups , FDR corrected . #p<0 . 05 , ###p<0 . 001 , ####p<0 . 0001 by multiple t test comparing untreated versus PEG400 groups , FDR corrected . N=20/group . Mean ± S . E . M . shown . SI , small intestine; Ce , cecum; LI , large intestine . P , proximal; M , middle; D , distal . Observed Species ( B ) and Shannon Index ( C ) are used to estimate richness and diversity of microbial communities in luminal content ( left ) and mucosal tissue ( right ) at anatomical sites along the intestine in indomethacin ( red ) , PEG400 ( blue ) , and untreated ( black ) groups . Indomethacin altered microbial diversity in the distal intestine , although PEG400 also causes changes in the distal intestine by itself . p<0 . 05 , p<0 . 01 , p<0 . 001 by multiple t test comparing PEG400 versus indomethacin groups , FDR corrected . #p<0 . 05 , ###p<0 . 001 , ####p<0 . 0001 by multiple t test comparing untreated versus PEG400 groups , FDR corrected . N=20/group . Relative abundance of Peptococcaceae ( D ) and Erysipelotrichaceae ( E ) at anatomical sites along the intestine are significantly elevated in indomethacin ( red ) group than in PEG400 ( blue ) and untreated ( black ) group in both luminal content and mucosal tissues of the distal gut . p<0 . 05 , **p<0 . 001 by QIIME analysis , FDR corrected . Mean ± S . E . M . shown . SI , small intestine; Ce , cecum; LI , large intestine . P , proximal; M , middle; D , distal . DOI: http://dx . doi . org/10 . 7554/eLife . 08973 . 00410 . 7554/eLife . 08973 . 005Figure 2—figure supplement 1 . Indomethacin induces small intestinal damage in C57BL/6 mice . Representative sections of small intestinal injuries 24 hr after 10 mg/kg indomethacin treatment , including macroscopic views ( left ) and hematoxylin and eosin ( H&E ) staining ( right ) . Macroscopically identified lesion areas were cut out for histopathology analysis by H&E staining at the center of the area . Red rectangle outlines the mucosal erosion ( A ) and ulcerations ( B-D ) observed . DOI: http://dx . doi . org/10 . 7554/eLife . 08973 . 00510 . 7554/eLife . 08973 . 006Figure 2—figure supplement 2 . Inhibitory effects of acute indomethacin treatment on COX-1 and COX-2 in C57BL/6 mice . Mice were administered by gavage with or without 10 mg/kg indomethacin ( red ) or PEG400 ( blue ) and urine were collected for the analysis of prostanoid metabolites . PGD-M ( A ) , PGE-M ( B ) , PGI-M ( C ) , and Tx-M ( D ) are reduced in indomethacin-treated mice . N=6/group . p<0 . 05 , p<0 . 01 , p<0 . 001 , *p<0 . 0001 by Mann-Whitney test , multiplicity adjusted . Mean ± S . E . M . shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08973 . 00610 . 7554/eLife . 08973 . 007Figure 2—figure supplement 3 . C57BL/6 mice are systemically and locally exposed to indomethacin . Mice were administered by gavage 10 mg/kg indomethacin in PEG400 ( red ) or PEG400 alone ( blue ) . Urine , feces , plasma , and intestines were collected from mice at 6 hr after drug administration . Indomethacin concentrations were measured in samples and corrected by sample weight . Indomethacin is detected along the intestine in both luminal content ( A ) and mucosal tissue ( B ) in mice of indomethacin ( red ) group , but not in those of PEG400 ( blue ) or untreated ( black ) groups . In feces ( C ) , urine ( D ) , and plasma ( E ) , indomethacin is also detected in mice of indomethacin ( red ) group , but not in those of PEG400 ( blue ) or untreated ( black ) groups . N=10/group . Mean ± S . E . M . shown . SI , small intestine; Ce , cecum; LI , large intestine . P , proximal; M , middle; D , distal . DOI: http://dx . doi . org/10 . 7554/eLife . 08973 . 00710 . 7554/eLife . 08973 . 008Figure 2—figure supplement 4 . Inhibitory effects of chronic indomethacin treatment on COX-1 and COX-2 in C57BL/6 mice . Mice were receiving control diet ( black ) or indomethacin diet ( 20 ppm , red ) for 7 days and urine were collected for the analysis of prostanoid metabolites . PGD-M ( A ) , PGE-M ( B ) , PGI-M ( C ) , and Tx-M ( D ) are reduced in indomethacin-treated mice . N=10/group . *p<0 . 0001 by Mann-Whitney test , multiplicity adjusted . Mean ± S . E . M . shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08973 . 00810 . 7554/eLife . 08973 . 009Figure 2—figure supplement 5 . Chronic indomethacin treatment induces changes in microbial composition along the intestine in mice . Bacterial communities colonized in the mouse intestine were profiled using 16S rRNA gene sequencing and analyzed using QIIME ( Caporaso et al . , 2010b ) . Observed Species ( A ) and Shannon Index ( B ) are used to estimate richness and diversity of microbial communities in luminal content ( left ) and mucosal tissue ( right ) at anatomical sites along the intestine in indomethacin ( red ) and control ( black ) groups . Indomethacin altered microbial diversity in the cecum lumen . p<0 . 05 by one-tailed Mann-Whitney test . N=9–10/group . Relative abundance of Peptococcaceae ( C ) and Erysipelotrichaceae ( D ) at anatomical sites along the intestine are significantly elevated in indomethacin ( red ) group than in control ( black ) group in both luminal content and mucosal tissues of the distal gut . p<0 . 05 , p<0 . 01 by one-tailed Mann-Whitney test . Mean ± S . E . M . shown . SI , small intestine; Ce , cecum; LI , large intestine . P , proximal; M , middle; D , distal . DOI: http://dx . doi . org/10 . 7554/eLife . 08973 . 009 A single dose of 10 mg/kg indomethacin was introduced into mice by gavage to test for effects on the microbiota . This acute dose is clinically relevant and is thus widely used in animal models ( Fukumoto et al . , 2011; Tanigawa et al . , 2013 ) . Mucosal erosion and ulcerations were observed in SI 24 hr after indomethacin treatment ( Figure 2—figure supplement 1 ) but not with vehicle control ( PEG400 ) . To analyze the compositional changes in microbial communities before the appearance of indomethacin-induced lesions , we analyzed animals after 6-hr of treatment . Urinary prostanoid metabolites ( Figure 2—figure supplement 2 ) were suppressed , predominantly reflective of COX-1 ( PGD-M and Tx-M ) and COX-2 ( PGI-M and PGE-M ) inhibition . Indomethacin was detected in plasma and urine as well as in luminal contents and mucosal tissues along the intestine ( Figure 2—figure supplement 3 ) , suggesting local and systemic exposure to the drug . Sixty mice were analyzed to characterize microbial responses to indomethacin exposure . Mice were randomly divided into three equal groups , receiving ( i ) 10 mg/kg indomethacin in PEG400 vehicle; ( ii ) PEG400 only and ( iii ) an untreated group . Examination of the bacterial biomasses using 16S rRNA gene qPCR revealed no effect of indomethacin , although a vehicle effect was evident ( Figure 2A ) . Mice treated with PEG400 showed a decrease in luminal biomasses and an increase in mucosal tissue biomasses in the distal end of the LI , compared to untreated mice . Similarly in humans , treatment with Golytely , which contains PEG 3350 , has been associated with changes in the mucosal-associated microbiota in colon ( Harrell et al . , 2012 ) . Microbial community structure along the intestine was quantified for observed species , which reflects the richness by measuring the number of operational taxonomic units ( OTUs ) and the Shannon Index , which indicates the diversity by taking account of both the number of OTUs and the evenness of distribution of reads among the OTU categories . Comparison between the indomethacin and PEG400 groups revealed changes primarily in the LI . Indomethacin caused an increase in richness ( Figure 2B ) in the middle ( p<0 . 01 , FDR corrected ) and distal LI luminal content ( p<0 . 01 , FDR corrected ) , as well as in feces ( p<0 . 01 , FDR corrected ) , without influencing the mucosal tissues . Diversity ( Figure 2C ) was decreased in the luminal content of the cecum ( p<0 . 01 , FDR corrected ) and in the mucosal tissue of proximal LI ( p<0 . 001 , FDR corrected ) , while it was increased in feces ( p<0 . 05 , FDR corrected ) . PEG400 alone increased microbial diversity in the distal LI . The abundance of some bacterial lineages was also affected by indomethacin . Peptococcaceae expanded in the luminal content of cecum ( 0 . 45% ± 0 . 07% ) , the proximal LI ( 0 . 38% ± 0 . 06% ) and distal LI ( 0 . 67% ± 0 . 06% ) , as well as in mucosal tissues of the cecum ( 0 . 55% ± 0 . 06% ) and the proximal LI ( 0 . 38% ± 0 . 06% ) ( Figure 2D ) . Erysipelotrichaceae expanded in the mucosal tissues of cecum ( 0 . 17% ± 0 . 06% ) and middle LI ( 0 . 12% ± 0 . 05% ) , yet were less affected in the luminal content ( Figure 2E ) . Separation of bacterial communities between indomethacin- or PEG400-treated mice was most evident in the LI mucosal tissues ( p=0 . 004 for proximal LI , p=0 . 009 for middle LI , p=0 . 009 for distal LI; weighted UniFrac distance ) . Since indomethacin is also chronically used in humans , we sought to understand the effects of chronic drug exposure in the mouse model . We thus introduced indomethacin in the diet , bypassing vehicle effects , in a second age and gender matched mouse cohort . Twenty mice were randomly divided into two groups , receiving control diet or an indomethacin-supplemented diet ( 20 ppm ) , administered for 7 days , and were then sacrificed one day later . This dose was selected based on the previous work ( Chiu et al . , 2000; Fjære et al . , 2014; Leibowitz et al . , 2014 ) and to ensure tolerability . Indomethacin significantly suppressed urinary prostanoid metabolites ( Figure 2—figure supplement 4 ) , suggesting COX-1 ( PGD-M and Tx-M ) and COX-2 ( PGI-M and PGE-M ) inhibition . The suppression was to a similar extent as was observed for the 6-hr treatment ( Figure 2—figure supplement 2 ) . Chronic indomethacin treatment was associated with a decrease in the Shannon Index in the luminal content of cecum . Richness and diversity were not affected at other GI sites ( Figure 2—figure supplement 5A , B ) . Several compositional changes detected in the acute treatment study were reproduced after chronic treatment . Peptococcaceae increased in relative abundance as in the acute study – in the chronic study this lineage expanded in the luminal content of cecum ( 0 . 28% ± 0 . 08% ) , proximal LI ( 0 . 35% ± 0 . 09% ) and distal LI ( 0 . 43% ± 0 . 13% ) , as well as in mucosal tissues of cecum ( 0 . 10% ± 0 . 02% ) and proximal LI ( 0 . 11% ± 0 . 02% ) ( Figure 2—figure supplement 5C ) . Erysipelotrichaceae , which also expanded in the acute treatment , expanded in the chronic treatment in mucosal tissues of cecum ( 0 . 21% ± 0 . 07% ) ( Figure 2—figure supplement 5D ) . Thus , both acute and chronic dosing affected the microbiota . A single oral dose of indomethacin induced alterations in microbial diversity in the distal intestine and caused compositional changes along the intestine , with only slight effects on microbial biomasses . Chronic indomethacin treatment exhibited some of the same effects on microbial composition for both the lineages affected and directions of change . Collection of fecal pellets from the same mouse before and after indomethacin treatment allowed analysis of within-individual compositional changes over time ( Figure 3A ) . We detected significant clustering between 0 hr and 6 hr microbial communities in the indomethacin-treated group ( p<0 . 01 , ADONIS test ) and PEG400 group ( p<0 . 05 , ADONIS test ) , but not in untreated group ( p>0 . 5 , ADONIS test ) . However , drug treatment explains more of the observed variation in the indomethacin group ( R2 = 0 . 22 , ADONIS test ) than in PEG400 group ( R2 = 0 . 05 , ADONIS test ) , indicating indomethacin had a greater effect in modulating fecal microbiota composition than PEG400 . The influence of indomethacin was not due to changes in microbial biomasses , since there were no significant changes in 16S rRNA gene copy number between PEG400 and indomethacin groups , as measured by qPCR ( Figure 3B ) . There may be a vehicle-induced decrease in microbial biomass , likely due to its purgative effect . Indomethacin also induced an increase in the Shannon Index ( Figure 3C , right ) , without influencing observed species in the fecal microbiota , suggestive of an increase in evenness associated with drug exposure . 10 . 7554/eLife . 08973 . 010Figure 3 . Indomethacin induces longitudinal changes in fecal microbiota composition . Microbiota composition in fecal pellets before ( 0 hr ) and after ( 6 hr ) treatment with or without indomethacin or PEG400 is analyzed by 16S rRNA gene profiling , including sequencing and quantitative PCR . ( A ) Principal coordinates analysis ( PCoA ) of weighted UniFrac values ( Lozupone et al . , 2011 ) , comparing the fecal microbial communities at 0 hr ( black ) versus 6 hr ( blue ) of untreated ( left ) , PEG400 ( middle ) , and indomethacin ( right ) groups . Each point represents a sample . Fecal microbial communities at 0 hr and 6 hr are not separated in untreated group ( p>0 . 5 ) , and significantly clustered in PEG400 group ( p<0 . 5 ) and in indomethacin group ( p<0 . 01 ) . Clustering was analyzed by ADONIS test . ( B ) 16S rRNA gene copies per gram of feces at 0 hr and 6 hr ( left ) , and Fold changes ( right ) in indomethacin ( red ) , PEG400 ( blue ) , and untreated ( black ) groups . Both PEG400 and indomethacin groups have lower bacterial loads at 6 hr , whereas these are no between-group differences at 0 hr or 6 hr . *p<0 . 0001 by Mann-Whitney test comparing 0 hr versus 6 hr . N=20/group . Mean ± S . E . M . shown . ( C ) Both Observed Species ( left ) and Shannon Index ( right ) are increased at 6 hr in indomethacin-treated mice , while unchanged in Untreated and PEG400 groups . p<0 . 01 by multiple t test , FDR corrected . N=19–20/group . Mean ± S . E . M . shown . ( D ) The relative abundance of Bacteroidetes ( left ) is decreased and that of Firmicutes ( right ) is increased at 6 hr ( blue ) after indomethacin treatment . *p<0 . 01 by multiple t test , FDR corrected . N=19–20/group . ( E ) Indomethacin induced a decrease in the relative abundance of S24-7 ( family ) , and increases in those of Ruminococcus , Lachnospiraceae sp . , Lachnospiraceae sp . , rc4-4 , and Anaeroplasma at 6 hr ( blue ) . p<0 . 05 , p<0 . 01 , p<0 . 001 by QIIME analysis , FDR corrected . N=19–20/group . Mean ± S . E . M . shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08973 . 01010 . 7554/eLife . 08973 . 011Figure 3—figure supplement 1 . Longitudinal effects of acute indomethacin treatment in fecal microbiota composition . Microbiota composition in fecal pellets before ( 0 hr , black ) and after ( 6 hr , blue ) treatment with or without indomethacin or PEG400 is analyzed by 16S rRNA gene profiling . ( A ) The relative abundance of Clostridiales sp . is increased in both PEG400 and indomethacin groups at 6 hr . ( B ) The relative abundance of Ruminococcaceae sp . is decreased in the PEG400 group but increased in the indomethacin group at 6 hr . ( C ) PEG400 induced an increase of Lactobacillus ( left ) and a decrease of Oscillospira ( right ) , whereas there is no change in the untreated or indomethacin treated groups . p<0 . 05 , p<0 . 01 , p<0 . 001 by QIIME analysis , FDR corrected . N=19–20/group . Mean ± S . E . M . shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08973 . 01110 . 7554/eLife . 08973 . 012Figure 3—figure supplement 2 . Longitudinal effects of chronic indomethacin treatment in fecal microbiota composition . Fecal microbiota composition before ( day 0 ) and after ( day 8 ) treatment in mice receiving control or indomethacin diet is analyzed by 16S rRNA gene profiling . ( A ) Observed species ( left ) and Shannon Index ( right ) showed no significant difference after indomethacin treatment in both control and indomethacin groups . ( B ) The relative abundance of Bacteroidetes ( left ) and Firmicutes ( right ) showed no significant difference after indomethacin treatment in both control and indomethacin groups . Mann-Whitney test . N=9–10/group . Mean ± S . E . M . shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08973 . 01210 . 7554/eLife . 08973 . 013Figure 3—figure supplement 3 . Longitudinal effects of chronic indomethacin treatment on genera abundance in fecal microbiota . Fecal microbiota composition before ( day 0 ) and after ( day 8 ) treatment in mice receiving control or indomethacin diet is analyzed by 16S rRNA gene profiling . Indomethacin induced increases in the relative abundance of Ruminococcus and Anaeroplasma at day 8 . p<0 . 05 , p<0 . 01 by two-tailed Mann-Whitney test . N=9–10/group . Mean ± S . E . M . shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08973 . 013 A phylum-level shift was evident in indomethacin-treated mice ( Figure 3D ) , with significantly decreased Bacteroidetes ( 73 . 84% ± 3 . 39% at 0 hr versus 56 . 02% ± 2 . 04% at 6 hr , p<0 . 01 after FDR correction ) and increased Firmicutes ( 24 . 95% ± 3 . 25% at 0 hr versus 41 . 97% ± 2 . 01% at 6 hr , p<0 . 01 after FDR correction ) . These trends were also detectable at lower taxonomic levels ( Figure 3E ) , including decreased S24-7 spp . ( Bacteroidetes ) , and increased Ruminococcus , Lachnospiraceae , and rc4-4 ( Firmicutes ) , and Anaeroplasma ( Tenericutes ) . PEG400 treatment increased Clostridiales spp . and Lactobacillus , while decreased Ruminococcaceae spp . and Oscillospira ( Figure 3—figure supplement 1 ) , consistent with the clustering observed in the PCoA plots ( Figure 3A ) . However , indomethacin may counteract the effect of PEG400 , leading to the increase of Ruminococcaceae spp . , or unchanged Lactobacillus and Oscillospira . The chronic indomethacin treatment introduced in the diet had diverse effects on the fecal microbiota . Richness , diversity , and the relative abundance of Bacteroidetes and Firmicutes were not significantly affected ( Figure 3—figure supplement 2A , B ) . However , changes at the genus level including expansion of Ruminococcus ( 0 . 49% ± 0 . 09% on day 0 versus 0 . 80% ± 0 . 13% on day 8 , p<0 . 05 ) and Anearoplasma ( 0 . 24% ± 0 . 05% on day 0 versus 0 . 72% ± 0 . 20% on day 8 , p<0 . 01 ) ( Figure 3—figure supplement 3 ) , matching effects seen in the acute dosing study . To investigate the impact of intestinal microbiota on the metabolism of indomethacin , we used antibiotics to deplete the microbiota , then compared metabolism in treated and control mice . Mice were treated with either control ( water ) or an antibiotic cocktail ( 1 g/L neomycin and 0 . 5 g/L vancomycin ) for 5 days . Fecal pellets were collected daily . The 16S rRNA gene copy numbers in feces were reduced by five orders of magnitude after antibiotic treatment , and this was maintained for up to 2 days after treatment cessation ( Figure 4A ) . Body weight , food intake , and water intake were not affected by antibiotic treatment over the time-course studied ( Figure 4—figure supplement 1A–C ) . Microbial diversity analysis revealed a significant decrease , starting at day 4 , with recovery still incomplete by day 7 ( Figure 4B ) . After 5 days of antibiotic treatment , mice showed a significantly shifted composition of the fecal microbiota , with a reduction in Bacteroidetes and Firmicutes , and a concomitant expansion of Proteobacteria ( Figure 4—figure supplement 1D ) . 10 . 7554/eLife . 08973 . 014Figure 4 . Microbiota-depletion with antibiotics alters the pharmacokinetics of indomethacin in mice . Mice were subjected to control water ( Con ) or antibiotic cocktail ( Abx , neomycin and vancomycin ) for 5 days ( blue-shaded area ) . Upon the cessation of 5-day treatment , mice were administered by gavage with 10 mg/kg indomethacin . Plasma was collected sequentially for pharmacokinetic analysis . Fecal microbiota compositions over time were analyzed using 16S rRNA gene profiling . ( A ) Longitudinal analysis of 16S rRNA gene copies per gram of feces reveals a significant reduction in microbial load in Abx group ( red ) . ( B ) Longitudinal analysis of observed species reveals decreased microbial richness in Abx group ( red ) . p<0 . 01 , **p<0 . 001 by multiple t test , FDR corrected . N=4–6/group . Mean ± S . E . M . shown . In antibiotic-treated mice ( red ) , indomethacin has increased oral clearance ( C ) and elimination rate constant ( Kel ) ( D ) , as well as decreased area under the curve ( AUCtotal ) ( E ) , half-life ( t1/2 ) ( F ) , and apparent volume of distribution ( Vd ) ( G ) . p<0 . 05 , *p<0 . 01 by Mann-Whitney test . N=6/group . Mean ± S . E . M . shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08973 . 01410 . 7554/eLife . 08973 . 015Figure 4—figure supplement 1 . Antibiotic-treatment causes compositional changes in intestinal microbiota without affecting body weight , food intake , and water intake in C57BL/6 mice . Mice were subjected to antibiotic water ( Abx , neomycin and vancomycin ) or control water ( Con ) for 5 days ( blue-shaded ) and the body weight , food intake , and water intake were monitored daily . Fecal pellets were collected for the analysis of microbiota composition . Body weight ( A ) , Food intake ( B ) , and Water intake ( C ) are not affected by antibiotic treatment . Note: on day 5 , animal cages were changed by facility staff and the movement caused water loss . N=6/group . Mean ± S . E . M . shown . ( D ) Heat map of the longitudinal analysis of bacterial lineages detected in feces of control ( light green ) or antibiotic treated ( dark green ) mice before ( day 0 ) and after ( Dday 5 , 6 , and 7 ) treatment . Each column represents one individual mouse of the time and treatment group indicated . Microbial composition is shifted in the Abx group but is stable in Con group . The proportions of bacterial lineages are indicated by the color code to the right . DOI: http://dx . doi . org/10 . 7554/eLife . 08973 . 015 Mice treated with the antibiotic cocktail or control were administered 10 mg/kg indomethacin by gavage on day 5 followed by sequential blood sampling over 48 hr . In the antibiotic-treated mice , the oral clearance of indomethacin was increased by 19 . 6% ( Figure 4C ) , and the elimination rate constant Kel was increased by 55 . 2% ( Figure 4D ) , indicating an increased elimination of indomethacin . The total area-under-the-curve ( AUCtotal ) of indomethacin , which is a measurement of total drug exposure , was decreased by 16 . 8% in antibiotic-treated mice ( Figure 4E ) . The half-life ( t1/2 ) of indomethacin was decreased by 37 . 5% ( Figure 4F ) , and the apparent volume of distribution ( Vd ) of indomethacin was decreased by 46 . 1% ( Figure 4G ) in antibiotic-treated mice . The variances of AUCtotal , t1/2 and Vd of indomethacin were significantly smaller in antibiotic-treated mice than in control mice ( p=0 . 01 for AUCtotal , p=0 . 04 for t1/2 , and p=0 . 006 for Vd , F test ) , suggesting intestinal bacteria as one of the sources of inter-mouse variation in response to indomethacin . A second set of age and gender matched mice treated with or without the antibiotic cocktail were administered 10 mg/kg indomethacin by gavage on day 5 , and urine and feces were collected for the following 24 hr . Detection of indomethacin and indomethacin-glucuronide was confirmed by incubating samples with or without β-glucuronidase . As shown in the representative spectra ( Figure 5—figure supplement 1A ) , after incubation with β-glucuronidase , the peak of indomethacin glucuronide was diminished and that of indomethacin greatly increased . This change was detected in each of the samples studied ( Figure 5—figure supplement 1B , C ) . To evaluate enzyme activity , we compared the ratio of indomethacin-glucuronide to indomethacin in mice pretreated with or without the antibiotic cocktail . In urine , the ratio was significantly higher in antibiotic-treated mice for the first 12 hr following indomethacin administration ( 93 . 3% higher at 4 hr , 55 . 6% higher at 8 hr , 43 . 4% at 12 hr; Figure 5B ) . In feces , indomethacin-glucuronide was barely detectable in control mice , yet was readily detected in antibiotic-treated mice ( Figure 5C ) . Indomethacin suppressed urinary prostanoid metabolites irrespective of treatment with the antibiotic cocktail ( Figure 5D ) . In control mice , indomethacin reduced these metabolites in a time-dependent fashion . A similar pattern was evident for PGD-M and PGI-M in antibiotic-treated mice . However , concentrations of the most abundant prostanoids , PGE-M and Tx-M , were suppressed to a lesser degree by indomethacin , and their concentration started to recover faster in antibiotic-treated mice compared to control mice . Evidently β-glucuronidase-catalyzed de-glucuronidation was impaired due to antibiotic-treatment , partially suppressing the inhibitory effect of indomethacin on COX enzymes . Thus , the intestinal microbiota influences the disposition and efficacy of indomethacin in the host , at least in part by regulating its de-glucuronidation and reabsorption from the intestine . 10 . 7554/eLife . 08973 . 016Figure 5 . Metabolism and efficacy of indomethacin in antibiotic-treated mice are altered . Upon the cessation of 5-day treatment with antibiotic cocktail ( Abx , neomycin and vancomycin ) or control water ( Con ) , mice were administered by gavage with 10 mg/kg indomethacin . Urine and feces were collected at indicated time for metabolic analysis . ( A ) Chemical structures of indomethacin ( left ) and indomethacin glucuronide ( right ) . Enzyme catalyzing the glucuronidation is UDP-glucuronosyltransferase ( UGT ) , and the one catalyzing the de-glucuronidation is β-glucuronidase . The ratio of indomethacin-glucuronide to indomethacin in urine ( B ) and feces ( C ) are higher in Abx group ( red ) than in Con group ( blue ) . p<0 . 05 , *p<0 . 01 by Mann-Whitney test . N=6/group . Mean ± S . E . M . shown . ( D ) Urinary prostanoid metabolites were analyzed with LC/MS and values are corrected by creatinine . In Con mice ( blue ) , all metabolites were reduced time-dependently . In Abx mice ( red ) PGD-M and PGI-M remained suppressed 24 hr after indomethacin , whereas PGE-M and Tx-M concentrations recovered more quickly . Two-way ANOVA revealed significant effect of time in PGD-M ( p=0 . 001 ) and PGI-M ( p=0 . 0004 ) , and significant antibiotic effect of PGE-M ( p<0 . 0001 ) and Tx-M ( p=0 . 0002 ) . In Abx mice , PGE-M was higher mice at 24 hr , and Tx-M was higher at 4 hr and 24 hr . N=6/group . p<0 . 05 , **p<0 . 01 by multiple comparison test , adjusted . Mean ± S . E . M . shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08973 . 01610 . 7554/eLife . 08973 . 017Figure 5—figure supplement 1 . β-glucuronidase catalyzes de-glucuronidation reaction . Mice were administered by gavage 10 mg/kg indomethacin in PEG400 after 5 days of antibiotic treatment . Urine and feces were collected at indicated times for the analysis of glucuronidation by in vitro by incubation with or without β-glucuronidase . ( A ) Representative spectra of LC/MS measurements of indomethacin and its metabolites . The peak denoting Acyl-b-D-glucuronide Indomethacin ( indomethacin glucuronide ) is larger without β-glucuronidase ( left ) than with β-glucuronidase ( right ) . The peak denoting indomethacin is smaller without β-glucuronidase ( left ) than with β-glucuronidase ( right ) . In urine ( B ) and feces ( C ) samples of control ( upper ) and antibiotic ( lower ) groups , the proportions of indomethacin glucuronide are smaller with β-glucuronidase added . Similarly , the proportions of indomethacin are larger with β-glucuronidase added . Each graph shows the longitudinal changes in one mouse . N=6/group . DOI: http://dx . doi . org/10 . 7554/eLife . 08973 . 017 Here , we investigated interactions between the intestinal microbiota and the NSAID indomethacin . We documented a bidirectional interaction — indomethacin altered the composition of the intestinal microbiota , and the intestinal microbiota altered indomethacin metabolism . The presence of intestinal bacteria boosted the circulating concentrations of indomethacin , which resulted in measurable changes in prostaglandin metabolism . Apparently , bacterial-encoded de-gluconuridation enzymes deconjugated the indomethacin-gluconuride catabolic intermediate to allow indomethacin recycling . We showed that only a single oral dose of indomethacin was sufficient to perturb the intestinal microbiota , specifically within the cecum , LI and feces . Drug-induced effects were less evident in the SI , possibly attributable to higher inter-individual variance in composition ( Figure 1A ) which limits the ability to detect the impact of interventions . The selected dose of indomethacin resulted in systemic drug exposure , including detection in luminal contents , along the mucosa of the SI and LI , and in plasma , urine and feces . Reflective of its mechanism of analgesic and anti-inflammatory action , this dose of indomethacin suppressed endogenous biosynthesis of prostaglandins derived primarily from COX-1 ( PGD2 and Tx ) and COX-2 ( PGI2 and PGE2 ) ( McAdam et al . , 1999; Ricciotti and FitzGerald , 2011 ) , as reflected by urinary excretion of their major metabolites . It also resulted in intestinal damage , reminiscent of the upper and lower GI complications induced by NSAIDs in humans ( Allison et al . , 1992; Smale et al . , 2001; Sostres et al . , 2013 ) . The alterations in the intestinal microbiota induced by indomethacin — specifically expansion of pro-inflammatory bacteria — may have functional consequences . For example , indomethacin induces expansion of Erysipelotrichaceae in LI mucosa . This Gram-positive family of Firmicutes has been associated with parenteral nutrition-induced liver injury ( Harris et al . , 2014 ) , obesity ( Zhang et al . , 2009 ) , colorectal cancer ( Zhu et al . , 2014; Chen et al . , 2012 ) , and Crohn’s disease ( Kaakoush et al . , 2015 ) . Indomethacin also induced pro-inflammatory shifts in the composition of fecal microbiota , for example , a significantly increased ratio of Firmicutes to Bacteroidetes . This shift has been previously reported in genetically obese mice ( Ley et al . , 2005; Turnbaugh et al . , 2006 ) , obese children ( Bervoets et al . , 2013 ) , and obese adults ( Ley et al . , 2006 ) . A decrease of S24-7 , a family of Bacteroidetes , such as induced here by indomethacin , has been observed in a mouse model of colorectal cancer ( Liang et al . , 2014 ) and in mice with high fat diet-induced obesity ( Evans et al . , 2014 ) . Lachnospiraceae , also increased following indomethacin administration , have been associated with lupus ( Zhang et al . , 2014 ) , drug-induced liver toxicity ( Xu et al . , 2015 ) and the development of obesity and diabetes in genetically susceptible mice ( Kameyama and Itoh , 2014 ) . While the importance of many of these observations remains to be established , they do suggest possible mechanisms of adverse health consequences . The observed influence of PEG400 , the vehicle used in our acute dosing study , on intestinal microbiota was not surprising . PEG used in bowel preparation in humans has been reported to have an effect on microbial diversity and composition ( Harrell et al . , 2012 ) . PEG binds water and prevents absorption in the large intestine , and so may hydrate the LI microbiome or flush upstream bacteria to downstream sites . However , our comparative study design allowed us to detect indomethacin effects independently of vehicle effects . Despite the PEG effect , a number of compositional changes quantified in the acute study were reproduced in the chronic treatment study . Expansion of Peptococcaceae and Erysipelotrichaceae in the GI tract and Ruminococcus and Anearoplasma in feces were seen in both studies . Whether and how these strains contribute to indomethacin-induced GI toxicity is unclear and warrants further study . Several changes in the microbiota observed in the acute study were not detected in the chronic study , which could be due to their involvement in the initial response only , or due to gradual recovery of the intestinal microbiota during chronic drug exposure . Systemic exposure to indomethacin was also reduced to insure tolerability in the chronic dosing study . The influence of indomethacin on the LI microbiota may be clinically important , since indomethacin causes complications in the lower GI tract with a similar incidence as in the upper GI tract ( Sostres et al . , 2013 ) . The patients receiving indomethacin were reported to show increased large intestinal permeability ( Suenaert et al . , 2003 ) , colonic ulceration and bleeding ( Oren and Ligumsky , 1994 ) , multiple colonic perforations ( Loh et al . , 2011 ) , and hemorrhage ( Langman et al . , 1985 ) . Compositional changes in LI microbiota associated with indomethacin administration may be involved in inflammatory processes directly in the lower GI tract , and might also affect the upper GI tract indirectly . For example , metabolic products produced by the LI microbiota may modify the local environment or enter the circulation , hence changing inter-bacterial interactions or host physiology at other sites including the SI . Here , we provide pharmacokinetic evidence that indomethacin metabolism is influenced by intestinal bacteria — specifically , antibiotic suppression of intestinal bacteria significantly reduced the level of its de-glucuronidation . In the absence of bacterial de-gluconuronidation , indomethacin reabsorption into the circulation was reduced , resulting in increased elimination , a shortened half-life and reduced drug exposure . Concomitantly , indomethacin-induced suppression of PGE2 and Tx was attenuated in microbiota-depleted mice , suggesting a reduction in indomethacin efficacy resulting from the loss of intestinal bacteria . The reduction in drug exposure due to antibiotic treatment may also explain the attenuated enteropathy associated with indomethacin in rats pretreated with antibiotics ( Koga et al . , 1999 ) , and further support previous reports from Boelsterli and colleagues ( Saitta et al . , 2014 ) , which reported that a small molecule inhibitor of bacterial β-glucuronidase were protective against NSAID-induced ulcerations in small intestine . Indomethacin shows considerable inter-individual variation in pharmacokinetics , efficacy and risk of GI complications ( Brune , 1985; 1987 ) that is not explained by human genetic variation ( Nakajima et al . , 1998; Martin et al . , 2001 ) . Our finding that depletion of intestinal bacteria significantly reduced inter-mouse variability of half-life , volume of distribution , and drug exposure suggests bacteria-mediated metabolism as a source of variation in drug response . Given that multiple human intestinal bacteria encode β-glucuronidase genes ( Dabek et al . , 2008; Gloux et al . , 2011 ) and that the intestinal microbiota are variable amongst individuals ( Wu et al . , 2011; Arumugam et al . , 2011 ) , differences in bacteria-mediated metabolism provide a reasonable explanation for inter-individual differences in indomethacin pharmacokinetics ( Brune , 1985; 1987 ) . The pharmacokinetics of orally dosed indomethacin shows circadian variation both in humans and in rats , which may also reflect a contribution from the microbiota . A prolonged apparent half-life of indomethacin was observed in patients dosed in the evening compared to those dosed in the morning or at noon ( Guissou et al . , 1983 ) , accompanied by fewer undesirable effects ( Levi et al . , 1985 ) . We and others have shown that the intestinal microbial load and composition varies during the day-night cycle ( Thaiss et al . , 2014; Zarrinpar et al . , 2014; Liang et al . , 2015 ) , including strains bearing β-glucuronidase activity . Taken together with our findings here , intra-individual variation of the intestinal microbiota during the 24-hr light-dark cycle may contribute to indomethacin chronopharmacology . In summary , a single oral dose of indomethacin elicited changes in composition and diversity of the microbiota . Reciprocally , the intestinal microbiota influenced indomethacin metabolism and its effectiveness as a systemic prostaglandin inhibitor . These results suggest that a dynamic interplay with the intestinal microbiome may contribute to adverse effects and variability in efficacy of indomethacin and perhaps other drugs . All C57BL/6 mice were purchased from the Jackson Laboratory and housed in our animal facility for at least 2 weeks before the performance of experiments . Male mice 10–14 weeks of age were used for all experiments . All animals were fed at libitum with regular chow diet ( 5010 , LabDiet , St . Louis , MO , USA ) for the course of study . Mice were kept under 12-hr light/12-hr dark ( LD ) cycle , with lights on at 7 am and off at 7 pm . Experimental protocols were reviewed and approved by the Institute for Animal Care and Use Committee at the University of Pennsylvania . All chemicals used were purchased from Sigma-Aldrich ( St . Louis , MO , USA ) unless otherwise stated . Bacterial DNA was isolated from samples ( fecal pellets , luminal contents , and mucosal tissues ) using PSP Spin Stool DNA Plus Kit ( Stratec , Berlin , Germany ) with a slight modification . Briefly , samples were thawed on ice and transferred to Lysing Matrix E tubes ( MP Biomedicals , Solon , OH , USA ) with 1400 μl of stool stabilizer from the PSP kit . They were then disrupted using the TissueLyser II ( Qiagen , Valencia , CA , USA ) for 6 min at 30 Hz . Samples were then heated at 95°C for 15 min , cooled on ice for 1 min , and spun down at 13 , 400 g for 1 min . The supernatant was then transferred to the PSP InviAdsorb tubes and the rest of the protocol for the PSP Spin Stool DNA Plus was followed according to the manufacturer's instructions . To maximize the extraction efficiency , each sample underwent two rounds of elution . Extracted DNA was quantified using NanoDrop 1000 ( Thermo Scientific , Wilmington , DE , USA ) and stored at −20°C for future use . Every DNA extraction included a negative extraction control in which water was used instead of fecal pellets . All controls went through the same DNA extraction process as well as following amplification and sequencing processes . Quantification of 16S rRNA gene was performed by real-time PCR using the Taqman method in triplicate reactions with 10 ng of DNA per reaction . Degenerate bacterial 16S rRNA gene-specific primers were used for amplification and their sequences were as follows: forward primer , 5’-AGAGTTTGATCCTGGCTCAG-3; reverse primer , 5’-CTGCTGCCTYCCGTA-3’; probe: 5' - /56-FAM/TAA +CA+C ATG +CA+A GT+C GA/3BHQ_1/ - 3'; + precedes the position of LNA base . Quantitative PCR was performed on a 7900HT Real-Time PCR System ( Applied Biosystems , Grand Island , NY , USA ) . Thermocycling was performed as follows: initiation at 95°C for 5 min followed by 40 cycles of 94°C × 30 s , 50°C × 30 s , and 72°C × 30 s . Signals were collected during the elongation step at 72°C . A standard curve prepared from a near full length clone of Escherichi coli 16S inserted into a Topo Vector was used for normalization for each run of real-time PCR . The V1-V2 region has performed well in reconstruction experiments and been used extensively previously in studies of the intestinal microbiome ( Liu et al . , 2007; Chakravorty et al . , 2007; Lozupone et al . , 2007; Turnbaugh et al . , 2009; Wu et al . , 2010; Song et al . , 2013 ) , and so was chosen here . A total of 100 ng of DNA was amplified with barcoded primers annealing to the V1-V2 region of the 16S rRNA gene using AccuPrime Taq DNA Polymerase System with Buffer 2 ( Life Technologies , Grand Island , NY , USA ) . PCR reactions were performed on a thermocycler using the following conditions: initiation at 95°C for 5 min followed by 20 cycles of 95°C × 30 s , 56°C × 30 s , and 72°C × 1 min 30 s , then a final extension step at 72°C for 8 min . The amplicons from each DNA sample , which was amplified in quadruplicate , were pooled and purified with Agencourt AMPure XP beads ( Beckman Coulter , Beverly , MA , USA ) following the manufacturer’s instructions . Purified amplicon DNA samples were then sequenced using the 454/Roche GS FLX Titanium chemistry ( 454 Life Sciences , Branford , CT , USA ) . All novel sequence data were deposited at NCBI’s Sequence Read Archive under Accession Numbers SRP 059293 and SRP 068846 . Sequence data were processed with QIIME v 1 . 8 . 0 ( Caporaso et al . , 2010b ) using default parameters . Firstly , samples with less than 200 counts were removed from further analysis . Sequences were clustered into operational taxonomic units ( OTUs ) at 97% similarity and then assigned taxonomy using the uclust consensus taxonomy classifier . Sequences were aligned using PyNAST ( Caporaso et al . , 2010a ) and a phylogenetic tree was constructed using FastTree ( Price et al . , 2009 ) . Weighted and unweighted UniFrac ( Lozupone and Knight , 2005 ) distances were calculated for each pair of samples for the assessment of community similarity and generation of principal coordinate analysis ( PCoA ) plots . Taxonomic composition and alpha diversity were generated for each sample . To compare bacterial abundances across sample groups , group_significance . py was used with default parameters . To estimate the functional profile for each microbiota sample , the reads were analyzed with Phylogenetic Investigation of Communities by Reconstruction of Unobserved States ( PICRUSt ) version 1 . 0 . 0 ( Langille et al . , 2013 ) following the instructions . Predicted metagenomes were collapsed into KEGG ( Kyoto Encyclopedia of Genes and Genomes ) pathways ( Ogata et al . , 1999 ) and analyzed with STAMP ( Parks and Beiko , 2010 ) . Plasma indomethacin concentrations at 1 , 2 , 4 , 6 , 8 , 24 , 30 , 48 hr post administration were plotted against time to generate the 'plasma indomethacin concentration versus time curve' . With this curve , the area under the curve ( AUCtotal ) was calculated according to the trapezoidal rule and the elimination rate constant ( Kel ) was obtained as the slope value . The half-life ( t1/2 ) was calculated ast1/2=ln2/kel . The apparent volume of distribution Vd was calculated as Vd=dose/C0 . C0 was extrapolated using the plasma drug concentration versus time curve . The oral clearance Cl was calculated as Cl=dose/AUCtotal . Histology of the injured SI was analyzed as described ( Imaoka et al . , 2010 ) . Briefly , SIs were removed and perfused with phosphate buffered saline ( PBS ) . Tissues were opened along the antimesenteric attachment and pinned down for macroscopic examination . The injured segments of the small intestine were trimmed , fixed overnight in 4% ( vol/vol ) paraformaldehyde at 4°C , washed with PBS , and dehydrated with ethanol before embedding in paraffin . The sections were cut and stained with hematoxylin and eosin ( H&E ) staining . To calculate the precise relative amount of indomethacin metabolite , standard curves were prepared in mouse urine for indomethacin , and acyl-β-D-glucuronide Indomethacin ( Santa Cruz Biotechnology , Dallas , TX , USA ) . Individual stock solutions of each compound ( 100 ng/µl ) were prepared in ACN and stored at −80°C . Working solutions were prepared by mixing equal amounts of corresponding stock solutions and performing serial dilutions with ACN . Seven-point calibration samples ( 0 , 0 . 032 , 0 . 16 , 0 . 8 , 4 , 20 and 100 ng/µl ) for indomethacin and its metabolite were prepared . One large urine sample was obtained from mice without exposure to indomethacin . For each sample , 40 µl of ( 1 ng/µl ) d4-Indomethacin ( Santa Cruz Biotechnology , Dallas , TexasTX , USA ) , 10 µl calibration standards were added to 20 mouse urine . The calibration curves were also prepared with β-Glucuronidase hydrolysis . The samples were extracted by SPE before LC/MS . Indomethacin and its metabolites were measured using a TSQ Quantum Ultra triple quadrupole mass spectrometer ( Thermo Scientific , Wilmington , DE , USA ) equipped with an ESI ion source . The Mass Spectrometer was connected to a Thermo Scientific Accela HPLC Systems and equipped with a PAL auto sampler and thermocontroller ( set at 4°C ) . The CSH C18 Column ( 2 . 1 mm Xx 150 mm , 130Å , 1 . 7 µm , Waters ) was used at a constant 40°C . The mobile phase ( A ) ( 90% H2O/10% ( B ) , 0 . 2% acetic acid ) and mobile ( B ) ( 90% ACN/10% methanol ) was used at a flow rate of 350 μl/min with a binary gradient ( 0–12 min , 10–50% B; 12–12 . 5 min , 50–100% B; 12 . 5–16 min , 100% B; 16 . 2–20 min , 10% B ) . Mass spectrometry was performed in negative mode . The transition for Indomethacin and d4-Indomethacin are 355 . 9>311 . 9 and 359 . 9>315 . 9 , respectively . The transition for acyl-β-D-glucuronide Indomethacin is 533 . 1>193 . 3 . Both Q1 and Q3 were operated at 0 . 7 m/z FWHM . Peak area ratios of target analytes to d4-Indomethacin internal standards were calculated by Xcalibur Quan software . The data were fitted to the calibration curves to calculate the precise relative amount of indomethacin metabolites . Prostanoid metabolites were measured using a Waters Acquity UPLC system comprising a binary pump , an autosampler , and a Xevo TQ-S triple quadrupole mass spectrometer equipped with an electrospray ionization source ( Waters , Milford , MA , USA ) . Chromatographic separation was performed on a Waters UPLC CSH C18 column ( 2 . 1 mm x 150 mm , 130 Å , 1 . 7 µm ) . The UPLC mobile phases consisted of ( A ) ( 95%H2O/5% ( B ) , pH=5 . 7 ) and ( B ) 95%ACN/5% methanol . The initial gradient began with 0% B . Mobile phase B increased linearly to 10% at 17 min , to 10 . 5% at 17 . 5 min , to 11 . 5% at 32 min , to 20% at 35 min , to 43% at 43 min , to 100% at 43 . 5 min , and finally go back to 0% at 45 . 5 min . A 0 . 35 ml/min flow rate was used throughout the UPLC gradient . The autosampler temperature was set at 4°C and the UPLC column was heated at 50°C . The MS was operated under negative ion mode at MRM mode . The transitions were monitored as previously described ( Song et al . , 2007 ) . Briefly , systemic production of PGI2 , PGE2 , PGD2 , and TxB2 was determined by quantifying their major urinary metabolites: 2 , 3-dinor 6-keto PGF1α ( PGI-M ) ; 7-hydroxy-5 , 11-diketotetranorprostane-1 , 16-dioic acid ( PGE-M ) ; 11 , 15-dioxo-9α-hydroxy-2 , 3 , 4 , 5-tetranorprostan-1 , 20-dioic acid ( tetranor PGD-M ) ; and 2 , 3-dinor TxB2 ( Tx-M ) , respectively . esults were normalized with creatinine ( Oxford Biomedical Research , Rochester Hills , MI , USA ) . Peak areas were obtained using MassLynx software ( Waters ) . Statistical analyses were performed using PRISM or QIIME ( Caporaso et al . , 2010b ) . Mann-Whitney test , Wilcoxon test , multiple t test , or QIIME analysis were conducted as indicated in figure legend . All data were expressed as means ± SEM .	The bacteria that inhabit the digestive tract do more than just help to break down food . Scientists are increasingly discovering that having a healthy and diverse community of gut bacteria is essential to overall health . Changes in these communities may increase the likelihood of harmful inflammation and diseases like obesity . Medications may alter the gut bacteria; for example , antibiotics used to treat infections may wipe out beneficial bacteria in the digestive tract . Other drugs like nonsteroidal anti-inflammatory drugs ( NSAID ) , which are sold over-the-counter to treat headaches and by prescription to treat pain , are known to damage the lining of the digestive tract . However , it was not clear how this affects the communities of bacteria in the gut . Liang et al . have now used genome-sequencing tools to determine which types and quantities of bacteria are normally present at various points along the digestive tract of healthy mice . The mice were then treated with an NSAID called indomethacin , which shifted the composition of intestinal bacteria towards a pro-inflammatory structure . Liang et al . then treated mice with antibiotics before giving them indomethacin to reduce the overall number of gut bacteria . These bacteria-depleted mice showed altered metabolism of indomethacin , and reduced blood levels of the drug , as evident in the production of fatty molecules called prostaglandins in the body . This is because the interactions between the gut bacteria and indomethacin modifies the inhibitory effect of indomethacin on enzymes known as COX-1 and COX-2 . This may explain why some drugs work better in some people than others , as different people have different bacteria in their guts . In the future , Liang et al . aim to investigate whether the communities of gut bacteria would be differently influenced by specifically inhibiting the action of either COX-1 or COX-2 , given that drugs that inhibit COX-2 cause fewer gastrointestinal complications . There are also plans to explore whether alterations in the communities of gut bacteria are a driver or a passenger in gastrointestinal ailments , following ingestion of indomethacin . Previous work has shown the influence of the host molecular clock on gut bacteria . Therefore , Liang et al . will also ask if taking indomethacin at different times of day might improve how well the drug works and cause fewer side effects in animal models and eventually in humans .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2015	Bidirectional interactions between indomethacin and the murine intestinal microbiota
Astrocytes respond to and regulate neuronal activity , yet their role in mammalian behavior remains incompletely understood . Especially unclear is whether , and if so how , astrocyte activity regulates contextual fear memory , the dysregulation of which leads to pathological fear-related disorders . We generated GFAP-ChR2-EYFP rats to allow the specific activation of astrocytes in vivo by optogenetics . We found that after memory acquisition within a temporal window , astrocyte activation disrupted memory consolidation and persistently decreased contextual but not cued fear memory accompanied by reduced fear-related anxiety behavior . In vivo microdialysis experiments showed astrocyte photoactivation increased extracellular ATP and adenosine concentrations . Intracerebral blockade of adenosine A1 receptors ( A1Rs ) reversed the attenuation of fear memory . Furthermore , intracerebral or intraperitoneal injection of A1R agonist mimicked the effects of astrocyte activation . Therefore , our findings provide a deeper understanding of the astrocyte-mediated regulation of fear memory and suggest a new and important therapeutic strategy against pathological fear-related disorders . Astrocytes are the most abundant glial cells in the central nervous system ( Allen and Barres , 2009; Bushong et al . , 2002 ) and are recognized for their classical supportive , metabolic , and protective roles ( Allaman et al . , 2011; Oliveira et al . , 2015 ) . In addition , increasing evidence has shown that they are actively involved in modulating synaptic transmission and plasticity ( Allen and Eroglu , 2017; Chen et al . , 2019; Weiss et al . , 2019; Zhang et al . , 2003 ) . Astrocytes respond to neuronal activity with a transient increase in the cytosolic Ca2+ concentration , as a result triggering the release of gliotransmitters , which in turn causes feedback regulation of neuronal activity and synaptic transmission ( Araque et al . , 2014; Zhang et al . , 2003 ) . However , most of these studies were performed at the synaptic or cellular level , and the roles of astrocytes in mammalian behavior remain incompletely understood . Memory is the biological process of retaining and retrieving what we learn over time , and it is crucial for survival . However , remembering traumatic fearful events can be maladaptive , leading to both inappropriate behavioral responses and grave physical or psychological harm ( Izquierdo et al . , 2016 ) . In humans , this can lead to various psychiatric disorders including post-traumatic stress disorder ( PTSD ) , panic disorder , phobias , and depression ( Maren et al . , 2013; Parsons and Ressler , 2013 ) . The estimated lifetime prevalence of fear- and stress-related disorders is close to 29% ( Kessler et al . , 2005 ) . Yet limiting pathological fear is a considerable challenge since fear memories are rapidly acquired and temporally enduring . Fear extinction , such as exposure therapy , is a fundamental behavioral method to reduce fear and anxiety in humans . However , exposure therapy is a context-dependent learning process that does not erase the initial fear memory ( Parsons and Ressler , 2013; Tovote et al . , 2015 ) . Fear memory could spontaneously recover or renew when patients are exposed to contexts similar or identical to those in which they first experienced trauma ( Britton et al . , 2011; Maren , 2011; Tovote et al . , 2015 ) . Therefore , it is urgent to find new therapeutic strategies for treating these disorders . Memory consolidation is a process by which newly acquired information is gradually stabilized by molecular and cellular processes after initial training ( Mednick et al . , 2011 ) . New memories are labile and vulnerable to disruption during early consolidation ( Abraham and Williams , 2008; Dudai , 2004; Izquierdo et al . , 1999 ) . Therefore , a feasible and more effective strategy for treating pathological fear would be to prevent fear memory consolidation soon after a traumatic event . Context and cue processing are two major components of fear learning and memory . Moreover , context processing is essential for understanding the meaning of cues in a particular context . The dysregulation of contextual fear processing may lead to pathological fear-related disorders such as PTSD , phobias , panic disorder and depression ( Maren et al . , 2013; Parsons and Ressler , 2013 ) . The hippocampus is thought to be critical in the formation of contextual fear memory ( Bast et al . , 2003; Besnard et al . , 2020; Daumas et al . , 2005; Maren , 2001; Maren et al . , 2013 ) . Hippocampal pyramidal neurons and interneurons have received much attention related to the contextual fear memory process in the mammalian brain ( Lamsa and Lau , 2019; Roy et al . , 2017; Xia et al . , 2017; Zhu et al . , 2014 ) . In vivo animal and human studies have found dynamic morphological and molecular changes in astrocytes during hippocampus-based contextual or spatial memory processes ( Choi et al . , 2016; Sagi et al . , 2012 ) , indicating the functional involvement of astrocytes in memory processes . However , as to whether and how astrocyte activity regulates contextual fear memory remains unclear . It has been demonstrated that channelrhodopsin-2 ( ChR2 ) expression is nontoxic , safe , stable , and functional ( Aravanis et al . , 2007; Cardin et al . , 2010; Deisseroth , 2011; Doroudchi et al . , 2011; Zhang et al . , 2006 ) . At present , ChR2 is widely used to explore the role of glia in regulating rodent behavior and circuits by precisely manipulating their Ca2+ signaling ( Bang et al . , 2016; Figueiredo et al . , 2011; Gourine et al . , 2010; Nam et al . , 2016; Perea et al . , 2014; Yamashita et al . , 2014 ) . Our previous work showed that the [Ca2+]i elevation induced by ChR2 in astrocytes is entirely from the extracellular space ( Yang et al . , 2015 ) . Due to its proximity to the plasma membrane where exocytosis occurs , transmembrane Ca2+ influx may be more efficient in inducing gliotransmitter release than Ca2+ release from intracellular stores ( Chen et al . , 2013; Tan et al . , 2017; Yang et al . , 2015 ) . Rats are well-adapted to the natural environment and are generally considered to perform well in learning and memory tasks ( Crawley , 2007 ) . The rat brain is relatively large , so injury from optical fibers is relatively small , thus photostimulation can be delivered specifically to the hippocampus . To clarify the exact role of astrocytes in fear memory and fear-related anxiety , we generated transgenic rats with astrocyte-specific ChR2 expression . We found that the optogenetic activation of astrocytes in CA1 within a critical time window after fear conditioning disrupted memory consolidation and persistently reduced contextual fear memory and fear-related anxiety . Conversely , reducing astrocyte Ca2+ activity increased fear memory . Notably , our data revealed that the gliotransmitter adenosine and adenosine A1 receptors ( A1Rs ) were responsible for the fear memory attenuation and anxiolytic effect . Furthermore , intraperitoneal ( i . p . ) injection of the A1Rs agonist 2-chloro-N6-cyclopentyladenosine ( CCPA ) within the defined time window also decreased contextual fear memory and fear-related anxiety . Therefore , our findings demonstrate that astrocytes participate in the regulation of contextual fear memory through purinergic signaling . This provides a deeper understanding of the astrocyte-mediated regulation of fear memory and suggests an important therapeutic strategy against pathological fear-related disorders . ChR2 induces calcium elevation in astrocytes and it has been used as a tool for astrocyte activation ( Chen et al . , 2013; Gourine et al . , 2010; Tan et al . , 2017; Yang et al . , 2015 ) . To specifically manipulate the activity of astrocytes in the brain , we generated ChR2 knock-in rats ( GFAP-ChR2-EYFP ) with astrocyte-specific promoter-glial fibrillary acidic protein ( GFAP ) . Co-staining of GFAP in EYFP-positive cells in the hippocampus verified the specific expression of ChR2 in astrocytes . We observed that 96 . 7% of the EYFP-positive astrocytes were GFAP-positive ( Figure 1A and B ) . Furthermore , 92 . 5% of the GFAP-positive astrocytes in CA1 expressed EYFP ( Figure 1A and C ) . We did not detect an EYFP signal in CA1 neurons ( Figure 1A ) . We further confirmed these results with neuron and astrocyte co-culture . EYFP was only expressed in astrocytes but not neurons in vitro ( Figure 1—figure supplement 1A ) . In addition , we also confirmed the specific expression of ChR2 in different brain areas such as the motor cortex , lateral posterior thalamic nucleus , and dorsomedial hypothalamic nucleus ( Figure 1—figure supplement 1B–D ) . To establish the functional response of astrocytes to optogenetic manipulation , Ca2+ imaging and whole-cell recording were performed in rat hippocampal slices . After loading astrocytes with a Ca2+ dye ( Rhod-2 AM ) , we observed that blue light stimulation increased the Ca2+ level in astrocytes expressing ChR2 ( Figure 1D and E; Figure 1—video 1 ) . Membrane depolarization was recorded in astrocytes that expressed ChR2-EYFP after light stimulation . We did not detect changes in the membrane potential in neurons with the same stimulation in these transgenic rats ( Figure 1F ) . Astrocytes were further confirmed with different electrophysiological features; they showed a linear current-voltage relationship under voltage-clamp mode , and no action potentials were detected even when they were depolarized to 0 mV ( Figure 1G and H; Figure 1—figure supplement 1E and F ) . To test the role of astrocytes in fear memory , we bilaterally implanted optical fibers into GFAP-ChR2-EYFP rats . To more comprehensively understand and to infer links with clinical symptoms ( McCarthy et al . , 2017 ) , we examined the effects of astrocyte activation on fear memory in rats of both sexes . The fiber location in the hippocampus was confirmed after completing the experiments ( Figure 2A and B ) . Photostimulation ( 473 nm , 10 Hz , 30 s on/30 s off , 15 min ) in the stratum radiatum of dorsal CA1 followed the Pavlovian fear conditioning paradigm in which an initial tone stimulus ( conditioned stimulus , CS , 30 s ) co-terminated with a scrambled foot shock ( unconditioned stimulus , US , 2 s ) ( Figure 2C; Figure 2—figure supplement 1A ) . Contextual and cued fear memory were measured on day 2 after fear acquisition ( Figure 2C; Figure 2—figure supplement 1A ) . The photostimulated rats showed no differences in freezing levels during fear conditioning compared with controls ( sham operation ) ( Figure 2D; Figure 2—figure supplement 1B ) . Contextual fear memory ( Figure 2E; Figure 2—figure supplement 1C ) but not cued fear memory ( Figure 2F; Figure 2—figure supplement 1D ) was significantly decreased in the photostimulated rats . To further differentiate the role of astrocytes in memory consolidation and retrieval , rats were given photostimulation only during the recall test and showed freezing similar to controls ( Figure 2—figure supplement 2A–C ) . These findings suggest that astrocyte photoactivation impairs memory consolidation , but not memory retrieval . To exclude the effect of light stimulation alone on freezing levels , littermate wild-type ( WT ) rats underwent the same protocol of photostimulation immediately after fear conditioning , and displayed freezing levels similar to sham-operated GFAP-ChR2-EYFP rats tested on day 2 ( Figure 2—figure supplement 2D–F ) . We further tested the effects of longer photostimulation on fear memory attenuation . GFAP-ChR2-EYFP rats were fear conditioned as above ( Figure 2—figure supplement 3A ) and received repeated photostimulation ( 15 min , four times at 15-min intervals ) , which successfully attenuated their fear memory to levels comparable to those obtained after a single 15-min photostimulation ( Figure 2—figure supplement 3B and C ) . Astrocyte Ca2+ signaling is attenuated by expression of the Ca2+-extruder PMCA2w/b ( plasma membrane Ca2+ ATPase isoform two splice variant w/b ) in hippocampal astrocytes ( Yu et al . , 2020; Yu et al . , 2018 ) . To investigate whether astrocyte activity is required for fear memory attenuation under normal conditions , we used an adeno-associated virus ( AAV5 ) expressing human PMCA2w/b ( hPMCA2w/b ) with an astrocyte-specific GfaABC1D promoter to reduce hippocampal astrocyte Ca2+ signaling ( Figure 2G–I ) . In the control group , we microinjected tdTomato instead of hPMCA2w/b ( Figure 2—figure supplement 4A and B ) . We ran behavioral tests 21 days after viral injection , and found that control and hPMCA2w/b rats did not differ in distance moved and center zone exploration time in the open-field test ( OFT ) ( Figure 2—figure supplement 4C and D ) . To assess whether fear memory was affected by reducing hippocampal astrocyte Ca2+ signaling , we used the Pavlovian fear conditioning paradigm again , and the contextual and cued fear memory were measured on day 2 after fear acquisition ( Figure 2C ) . We found that control and hPMCA2w/b rats had comparable learning curves for fear conditioning ( Figure 2—figure supplement 4E ) . Interestingly , contextual fear memory but not cued fear memory ( Figure 2J and K ) was significantly increased in hPMCA2w/b rats . These results demonstrated that hippocampal astrocyte activity is indeed required for contextual fear memory . Severe traumatic stress leads to anxiety-like emotional responses , such as PTSD ( Britton et al . , 2011; Parsons and Ressler , 2013 ) . To test the effect of astrocyte activation on fear-related anxiety-like behavior , we performed photostimulation immediately after fear conditioning . Thirty minutes after the cued fear test on day 2 , we assessed the anxiety-like behavior of rats in the open field ( Figure 2C ) . The total moving distance ( Figure 2L and M ) and center zone exploration time ( Figure 2L and N ) of the fear-conditioned rats were significantly reduced compared with those of the controls ( without fear conditioning ) , indicating that anxiety-like behavior was enhanced . Notably , photostimulation immediately after fear conditioning significantly rescued the decreased moving distance ( Figure 2L and M ) and center zone exploration time ( Figure 2L and N ) , indicating that astrocytic activation exerts an anxiolytic effect . We then assessed whether the motor performance was directly affected during photostimulation using the OFT . We used 3-min epochs as the paradigm for light OFF–ON–OFF . Photostimulation of astrocytes in CA1 did not affect the distance moved and center zone exploration time ( Figure 2O–Q ) . To determine whether there is a time-window of astrocyte activation after training for the disruption of fear memory consolidation , rats were photostimulated for 15 min in CA1 1–3 hr after fear conditioning ( Figure 3A ) , and the contextual fear memory was assessed 24 hr later . Control and photostimulated rats had comparable learning curves for fear conditioning ( Figure 3B ) . Photostimulation of astrocytes 1 hr , but not 2 hr or 3 hr ( Figure 3C ) after fear conditioning resulted in a significantly decreased fear response , indicating that astrocyte activation within a critical time window after traumatic events is more efficient for disrupting memory consolidation and reducing contextual fear memory . To measure whether the contextual fear memory attenuation induced by astrocyte activation is long-lasting , we assessed it on days 2 , 3 , 8 , 16 , and 26 after fear acquisition ( Figure 3D ) . Compared with the controls ( sham operation ) , the rats receiving photostimulation immediately after fear acquisition showed a persistent decrease of contextual fear memory lasting up to 26 days ( Figure 3E ) . On day 27 , some of the photostimulated rats were re-trained in a new conditioning chamber and exhibited a learning curve comparable to controls conditioned on day 1 ( Figure 3F ) . The other photostimulated rats ( conditioned ) performed the object location recognition ( OLR ) task and the percentages of time spent exploring the objects were similar to controls in the sample phase ( Figure 3H ) , indicating that astrocyte activation does not affect new learning . Furthermore , retrained rats had normal memory retention on day 28 during contextual fear test compared with controls on day 2 ( Figure 3G ) . In the test phase of the OLR on day 28 , the conditioned rats spent more time exploring the object in a novel location relative to the object in a familiar location ( Figure 3I ) and the discrimination index was similar to controls ( Figure 3J ) . All these data suggested that astrocyte activation does not affect new memory formation . These results reveal that the astrocyte activation reliably and persistently attenuates temporally coupled fear memory . Previous studies in our lab ( Chen et al . , 2013; Tan et al . , 2017 ) and those of other investigators Pascual et al . , 2005 found that the ATP derived from astrocytes degrades to adenosine and leads to the suppression of nearby synapses . To determine whether the photostimulation of astrocytes triggered increased ATP and adenosine concentrations in CA1 of GFAP-ChR2-EYFP rats , we performed in vivo microdialysis experiments ( Figure 4A ) . Following the collection of samples of baseline dialysate , we initiated the second phase of microdialysis starting with 15 min of photostimulation . The ATP and adenosine concentrations in the interstitial fluid were significantly higher ( ATP: 1st , 0 . 74 ± 0 . 26 nM , 2nd , 1 . 98 ± 0 . 38 nM; adenosine: 1st , 2 . 88 ± 0 . 17 μM , 2nd , 3 . 42 ± 0 . 11 μM ) in the photostimulation phase ( Figure 4B ) , demonstrating that astrocyte activation induces ATP release in the hippocampus . To determine whether the fear memory attenuation induced by astrocyte activation was mediated by ATP or its degradation product , adenosine , we implanted infusion cannulae bilaterally into dorsal CA1 ( Figure 4C ) . ATP-γ-S , the specific P2Y receptor agonist 2- ( methylthio ) adenosine 5'-diphosphate ( MesADP ) , or vehicle were separately injected into the dorsal CA1 bilaterally after fear conditioning , then we assessed contextual fear memory on day 2 ( Figure 4D ) . The vehicle and drug-treated rats had comparable learning curves for fear conditioning ( Figure 4E–G ) . There was no difference in contextual fear memory ( Figure 4H ) . Further , agonists of adenosine receptors were bilaterally delivered to the dorsal CA1 immediately after fear conditioning . The rats given the non-hydrolyzable adenosine analog 5′- ( N-ethylcarboxamido ) adenosine ( NECA ) and the specific A1R agonist CCPA displayed significantly decreased contextual fear memory on day 2 compared with vehicle-treated rats ( Figure 4I ) , while the specific adenosine A2A receptor ( A2AR ) agonist CGS 21680 hydrochloride had no effect ( Figure 4J ) . NECA and CCPA in the dorsal CA1 did not affect spontaneous locomotor activity or center zone exploration time in the OFT ( Figure 4—figure supplement 1A–E ) . These results showed that the activation of A1Rs after memory acquisition mimicked the effects of astrocyte activation on contextual fear memory , indicating that A1Rs mediate the contextual fear memory attenuation . To further confirm that adenosine participates in the astrocyte activation-induced attenuation of contextual fear memory , we separately injected into CA1 the ectonucleotidase inhibitor ARL 67156 trisodium salt hydrate , which prevents ATP from converting into adenosine , the specific A1R antagonist 8-cyclopentyl-1 , 3-dimethylxanthine ( CPT ) , or the specific A2AR antagonist SCH 58261 . Vehicle and drug treatment were paired with photostimulation of astrocytes in CA1 immediately after fear conditioning , and contextual fear memory was examined on day 2 ( Figure 5A and B ) . Control rats ( without photostimulation or pharmacological treatment ) , photostimulated rats ( without pharmacological treatment ) , and photostimulated rats paired with vehicle or drug treatment had comparable learning curves for fear conditioning ( Figure 5C–E ) . Compared to controls , astrocyte activation alone or paired with vehicle significantly decreased contextual fear memory to a comparable level ( Figure 5F–H ) . Astrocyte activation paired with ARL 67156 reversed the freezing level to that of controls , showing that the attenuation effect of astrocyte photostimulation on fear memory was almost totally blocked by ARL 67156 ( Figure 5F ) . These results indicate that the fear memory attenuation induced by astrocyte activation is abrogated with diminished levels of adenosine . CPT treatment also dramatically blocked the attenuation of fear memory induced by astrocyte photostimulation ( Figure 5G ) , indicating that astrocyte activation attenuates fear memory in an A1R-dependent manner . Notably , CPT in CA1 did not affect the spontaneous locomotor activity and center zone exploration time in the OFT ( Figure 5—figure supplement 1A–C ) . The specific A2AR antagonist SCH 58261 failed to block the attenuation of fear memory induced by astrocyte activation ( Figure 5H ) . Altogether , these results suggest that A1Rs are necessary for the contextual fear memory attenuation induced by astrocyte activation . To further determine the therapeutic applicability of CCPA for fear memory attenuation , we tested whether the i . p . injection of CCPA decreases fear memory . Rats were injected with vehicle or CCPA immediately after fear conditioning and the contextual fear memory was assessed on days 2 and 3 ( Figure 6A ) . Rats with vehicle and CCPA injection had comparable learning curves for fear conditioning ( Figure 6—figure supplement 1A–C ) . Rats injected with vehicle or CCPA ( 0 . 03 mg/kg ) did not differ in contextual fear memory tested on days 2 and 3 ( Figure 6B ) . However , rats injected with higher doses of CCPA ( 0 . 1 and 0 . 3 mg/kg ) showed significantly attenuated contextual fear memory tested on days 2 and 3 , compared to rats with vehicle injection ( Figure 6C and D ) . To determine whether there is also a time-window of CCPA injection after training for the disruption of fear memory consolidation , rats were injected with vehicle or CCPA ( 0 . 1 mg/kg ) 0 . 5–2 hr after fear conditioning , and the contextual fear memory was assessed on days 2 and 3 ( Figure 6E ) . Rats with vehicle and CCPA injection had comparable learning curves for fear conditioning ( Figure 6—figure supplement 1D–F ) . CCPA injection 0 . 5 hr and 1 hr ( Figure 6F and G ) , but not 2 hr ( Figure 6H ) after fear conditioning significantly reduced the fear response . These results show that CCPA injection within a critical time-window after a traumatic experience disrupts memory consolidation and reduces contextual fear memory . To test whether CCPA affects fear-related anxiety-like behavior , rats were injected with vehicle or CCPA ( 0 . 1 mg/kg , i . p . ) immediately after fear conditioning , and tested in the open field 3 hr or 24 hr later . Compared with controls ( without fear conditioning ) , the fear-conditioned rats with vehicle injection showed a decreased moving distance and center zone exploration time in the OFT , while CCPA injection significantly rescued the decreased moving distance and center zone exploration time ( Figure 6I–N ) . Notably , CCPA injection at 0 . 1 mg/kg did not affect the motor performance and center zone exploration time of unconditioned rats in the OFT ( Figure 6—figure supplement 2A–C ) . CCPA injection at the higher concentrations of 0 . 3 mg/kg decreased the moving distance in the OFT ( Figure 6—figure supplement 2A and B ) . However , the motor performance recovered 5 hr after higher doses CCPA injection ( Figure 6—figure supplement 2D ) . Accumulating evidence suggest that astrocyte activity is crucial for synaptic regulation and plasticity , which are considered to be involved in learning and memory ( Chen et al . , 2013; Martin-Fernandez et al . , 2017; Tan et al . , 2017; Yang et al . , 2003; Zhang et al . , 2003 ) . Previous studies have reported the essential role of astrocytes in memory processing ( Gao et al . , 2016; Robin et al . , 2018; Suzuki et al . , 2011; Vignoli et al . , 2016 ) . However , it remains unclear whether and how astrocyte activity regulates contextual fear memory . We provide evidence for the first time that astrocyte activation by precise optogenetic stimulation within a defined time window ( 1 hr ) disrupts memory consolidation and dramatically reduces contextual fear memory and the related anxiety . The fear memory attenuation is robust and persistent , while learning capacity remains intact . On the contrary , reducing Ca2+ activity in astrocytes increased fear memory . Our further results reveal that astrocyte activation increases the extracellular ATP and adenosine concentrations , and the fear memory attenuation and anxiolytic effect are mediated by A1R activation . Thus , our findings provide a deeper understanding of astrocyte-mediated regulation of fear memory and the complexity of the functional consequences of astrocyte regulation . The hippocampus is essential for the formation of contextual fear memory ( Bast et al . , 2003; Daumas et al . , 2005; Maren , 2001; Maren et al . , 2013 ) , and hippocampal lesions selectively affect contextual fear memory but not cued fear memory ( Kim and Fanselow , 1992; Phillips and LeDoux , 1992 ) . Context plays a central role in understanding the meaning of cues , abstracting the surrounding information and anticipating the future in a particular context . Disorders in contextual processing may lead to fear-related diseases , including PTSD , panic disorder , phobias and depression ( Maren et al . , 2013; Parsons and Ressler , 2013 ) . PTSD may be the most representative of context-processing diseases , given that its core features involve recurrent and intrusive recollections ( flashbacks ) of an experienced traumatic event ( Rosenbaum , 2004 ) . Finding new mechanisms underlying contextual memory processing is thus critical in developing strategies for relieving sufferers from pathological fear . Memory consolidation is a molecular and cellular processes by which newly-acquired information is gradually stabilized by strengthening synaptic connections after the initial training ( Dudai , 2004 ) . It has been proposed that long-term potentiation ( LTP ) , a form of synaptic plasticity , is an intrinsic property of this consolidation ( Abraham and Williams , 2008; Kandel et al . , 2014; Yang et al . , 2018 ) . Once LTP or memory has passed into the protein synthesis-dependent phase , LTP is highly resistant to disruption , while memory becomes more difficult to erase and is said to have been consolidated ( Abraham and Williams , 2008; Lu et al . , 2018; Mednick et al . , 2011 ) . So if the time between learning and interference is sufficiently long for the processes associated with LTP to have occurred , the memory trace would be less vulnerable . Our present study showed that photoactivation of astrocytes in CA1 within the early memory consolidation phase ( 1 hr ) but not beyond it ( 2 hr , 3 hr or during retrieval ) after a fear stimulus produced a long-lasting attenuation of contextual fear memory and an anxiolytic effect . This finding is in accordance with reports that new memories are sensitive to interference within a short time ( 1 hr ) after learning , but not after a relatively long interval ( 6 hr ) during which the memory trace becomes consolidated ( Mednick et al . , 2011 ) . More importantly , our results suggest a new strategy of astrocyte-based disruption of memory consolidation , which leads to persistent fear memory attenuation with attenuated spontaneous recovery of fear memory . Astrocytes respond to neuronal activity and release different neuroactive molecules , among which ATP ( adenosine ) , glutamate , D-serine , and gamma-aminobutyric acid are the major gliotransmitters identified as regulators of synaptic transmission ( Araque et al . , 2014 ) . Recent study showed that activating astrocytes by chemogenetic or optogenetic recruitment of their Gq-coupled signaling before or during memory acquisition induces N-methyl-D-aspartate-dependent long-term potentiation in CA1 through the release of D-serine and enhances memory acquisition ( Adamsky et al . , 2018 ) . However , another study showed that chemogenetic activation of astrocytes before memory retrieval reduces the expression of an acquired cued fear response , which is mediated by inhibiting excitatory synapses from the basolateral amygdala via A1R activation and enhancing inhibitory synapses from the lateral subdivision of the central amygdala via A2A activation ( Martin-Fernandez et al . , 2017 ) . These contradictory results indicate that different patterns of stimulation may induce astrocytes to release different substances which then activate their associated receptors on nearby neurons , leading to different forms of synaptic regulation ( Covelo and Araque , 2018 ) . Previous work in our lab established that the gliotransmitters ATP and adenosine but not glutamate and D-serine are involved in astrocyte-mediated synaptic suppression ( Chen et al . , 2013; Tan et al . , 2017; Zhang et al . , 2003 ) . Astrocyte-derived ATP increases the activity of cholecystokinin-expressing interneurons through the activation of P2Y1 receptors and decreases pyramidal neuron activity through the activation of A1Rs in CA1 , resulting in the downregulation of the whole network activity in the hippocampus ( Tan et al . , 2017 ) . Similarly , a recent publication showed that adenosine degraded from the astrocyte-derived ATP upregulates the synaptic inhibition of pyramidal neurons by somatostatin-expressing interneurons via A1R activation ( Matos et al . , 2018 ) . The activation of A1Rs inhibits adenylate cyclase , which causes a decrease in the second messenger cAMP , inhibits voltage-gated Ca2+ channels and activates G-protein-coupled inwardly rectifying K+ channels through Gi/o βγ-subunits , and decreases the excitability of pyramidal neurons ( Burnstock et al . , 2011; Fields and Burnstock , 2006; Tan et al . , 2017; Wetherington and Lambert , 2002 ) . Previous studies demonstrate that neurons with relatively high intrinsic excitability promote memory integration ( via co-allocation to overlapping engrams ) , while decreased excitability promotes memory separation ( Josselyn , 2010; Josselyn and Frankland , 2018 ) . Our in vivo microdialysis experiments showed that photostimulation of astrocytes led to significant increase in extracellular ATP and adenosine concentrations . Both an ectonucleotidase inhibitor ( which prevents ATP from converting to adenosine ) and a specific A1R antagonist dramatically reversed astrocyte activation-induced fear memory attenuation . But a specific A2AR antagonist did not have this effect . Furthermore , A1R agonist application within the critical time window mimicked the effects of astrocyte activation on fear memory and the anxiolytic effect . Therefore , we demonstrate that the astrocyte-mediated attenuation of contextual fear memory and the related anxiety found in our experimental conditions depends on A1Rs , which could act as potential new molecular targets for the treatment of fear-related disorders . Together , our results identify a functional role of astrocytes in contextual fear memory and the related anxiety and reveal that astrocytes are essential elements in fear memory processing through purinergic signaling in the hippocampus . Consequently , our findings provide a deeper understanding of astrocyte-mediated regulation of fear memory and suggest a new and important therapeutic strategy against pathological fear-related disorders . An attempt to use this strategy in clinic could be very promising . GFAP-ChR2-EYFP rats ( Sprague-Dawley background ) were generated at the Institute of Neuroscience , Chinese Academy of Sciences . To generate GFAP-ChR2-EYFP knock-in rats , we designed the single guide ( sg ) RNA near the stop codon in the last exon of the GFAP gene , and constructed a donor plasmid containing the ChR2-EYFP sequence . The plasmid was used as a template to repair the double-strand break by homologous recombination . Super-ovulated female Sprague-Dawley rats were mated to Sprague-Dawley males , and fertilized embryos were collected from the oviducts . Cas9 mRNA , sgRNAs , and donor were mixed and injected into the cytoplasm of fertilized eggs using a Narishige IM300 microinjector and the zygotes were cultured for several hours . Thereafter , 20–25 embryos were transferred into the oviducts of pseudopregnant Sprague-Dawley rats . The genotypes of mutant mice were determined by PCR of genomic DNA extracted from the tail . Experiments were conducted on 2- to 4-month-old male and female GFAP-ChR2-EYFP rats and WT Sprague-Dawley rats . The rats were housed with food and water available ad libitum in a temperature-controlled room with a 12 hr light/dark cycle ( lights on at 07:00 ) . Rats were singly housed after surgery . All experimental procedures were approved by the Animal Advisory Committee at Zhejiang University ( 2019–2# ) and were performed in strict accordance with the National Institutes of Health Guidelines for the Care and Use of Laboratory Animals ( 2006–398# ) . All surgeries were performed under sodium pentobarbital anesthesia , and every effort was made to minimize suffering . Rats were deeply anesthetized with pentobarbital sodium ( 1% , wt/vol ) and placed on a stereotaxic frame ( Kopf , USA ) . Body temperature was kept stable throughout the procedure with a heating pad . A scalp incision was made with eye scissors . The skull was exposed and perforated with a stereotaxic drill at the target region . For optogenetic fiber implantation , two optical fibers ( core diameter 200 µm , NA 0 . 37; Newdoon , China ) were bilaterally implanted into CA1 ( AP , –3 . 75 mm; ML , ±2 . 46 mm; DV , –2 . 63 mm relative to bregma ) for optogenetic manipulations . The optical fibers were connected to a laser source using a fiber optic patch cord ( Newdoon , China ) . The intensity of laser stimulation was measured before each experiment at the tip of the optical fiber via a laser power meter ( LP1 , Sanwa , Japan ) . For pharmacological experiments , two guide cannulae ( RWD Life Science , China ) were bilaterally implanted into CA1 at the above coordinates for drug infusion . For photostimulation coupled with pharmacological experiments , an optical fiber ( AP , –3 . 75 mm; ML , 1 . 85 mm; DV , –3 . 10 mm relative to bregma , at a 12° angle ) and an infusion cannula ( AP , –3 . 75 mm; ML , 3 . 01 mm; DV , –1 . 76 mm relative to bregma , at a 15° angle ) were unilaterally implanted into CA1 for simultaneous photostimulation and drug infusion . The optical fibers and cannulae were fixed to the skull with dental cement . After the implantation surgery , rats were allowed to recover for 7 days before behavioral tests . After experiments , the positions of the optical fibers and cannulae were verified histologically . Rats with incorrect positioning of optical fibers or cannulae were excluded . pZac2 . 1-GfaABC1D-mCherry-hPMCA2w/b was from Baljit Khakh Lab Plasmids in Addgene ( Addgene plasmid # 111568 ) . AAV5 GfaABC1D mCherry-hPMCA2w/b ( 7 . 0 × 1012 viral genomes/ml ) and AAV5 GfaABC1D tdTomato ( 6 . 0 × 1012 viral genomes/ml ) were constructed by Vigene Biosciences ( Shandong , China ) . Rats were anesthetized with pentobarbital sodium ( 1% , wt/vol ) and placed on a stereotaxic frame ( Kopf , USA ) . Viruses were injected bilaterally into the dorsal CA1 ( 450 nl , 50 nl/min ) via a microsyringe . The needle was left in place for 15–20 min after the end of infusion to allow diffusion of the virus . After injection , mice were allowed 3 weeks of recovery and then performed behavioral tests . The fear-conditioning task was carried out in a 25 × 25 × 25 cm conditioning chamber ( Panlab Harvard Apparatus , Spain ) placed inside a sound-protected box . This task consisted of three phases: fear conditioning ( training ) , testing for contextual fear ( a hippocampus-dependent test ) ( Kim and Fanselow , 1992 ) , and cued fear ( a hippocampus-independent test ) ( Phillips and LeDoux , 1992 ) . Rats were handled for 3 days before training was begun . On day 1 , rats were placed into a fear conditioning chamber with a grid floor capable of delivering foot shocks , after the chamber was cleaned with 70% alcohol . The baseline freezing level was measured during a 2-min exploration period prior to the first conditioned stimulus ( CS ) . Rats were then exposed to a 30 s tone ( CS , 2 kHz , 85 dB ) that co-terminated with a 2 s scrambled foot shock ( unconditioned stimulus , US , 0 . 6 mA ) . A total of three tone-shock pairings were delivered with an inter-tone interval of 60 s . The rats remained in the conditioning chamber for a 90 s consolidation period following the last US . The conditioning phase lasted for 7 min , and all the processes were carried out in a relatively dark chamber . Twenty-four hours after conditioning , the rats were placed in the conditioning chamber for a 5-min contextual test . The cued test was carried out in a relatively bright chamber that had a context and smell different from the conditioning chamber ( cleaned with 1% acetic acid ) . In the cued test , rats received a CS recall test ( 3 presentations of the 30 s CS alone with a 30-s inter-tone interval ) . The rats were considered to freeze if no movement was detected for 2 s . In the contextual test , the freezing level was calculated as the percentage of freezing time during 5 min in context . In fear-conditioning and the cued test , the freezing level was calculated as the percentage of freezing time during three presentations of the CS . The data were automatically recorded using commercial software ( FREEZING , Panlab Harvard Apparatus , Spain ) . When assessing the effect of astrocyte activation on fear memory , the rats received photostimulation for 15 min ( 473 nm , 10 Hz , 20 ms pulses , 1–3 mW at the fiber tip , 30 s light on , 30 s light off ) immediately after fear conditioning or with an interval of 1 , 2 , and 3 hr . Sham operation was defined as surgery including the implantation of optical fibers and cannulae but without photostimulation . For pharmacological manipulations , drugs or vehicle were delivered by intracerebral or i . p . injection immediately after fear conditioning . The OFT is a classical test to measure anxiety-like behavior in rodents ( Cai et al . , 2018 ) . Rats were placed in the corner of a black open field arena ( 100 × 100 × 40 cm ) at the start of the experiment and allowed to freely explore for 5 min . The center of the open field was defined as the central 50% of the arena . The locomotor activity of the rats in the open field was video-recorded and analyzed with automatic behavioral tracking software ( ANY-maze , Stoelting Co . , USA ) . The open field chamber was cleaned with 70% alcohol between animals . In the optogenetic studies , the total test time of 9 min was divided into three consecutive 3-min epochs consisting of stimulation off , stimulation on , and stimulation off periods ( OFF-ON-OFF ) . The total distance was defined as the distance moved in a 3- or 5 min OFT . The center zone time was defined as the exploration time in the center area of the open field during a 3- or 5-min test . Rats were habituated to a square testing arena ( 100 cm ×100 cm × 40 cm ) for 10 min per day for 3 consecutive days . After habituation , the OLR task was divided into a sample phase and a test phase , each lasting 5 min . In the sample phase , each rat was placed in the arena , exposed to two identical objects , and then returned to its home cage . After a 24 hr delay , each rat was returned to the arena for the test phase when it was exposed to the same objects as in the sample phase except that one of these objects ( displaced object , DO ) was moved to a novel location in the arena . The other ( undisplaced object , UDO ) remained in the original location . The time spent in exploring each object was recorded . Exploration was defined as touching the object with the nose or directing the nose to the object at a distance of no more than 2 cm . The percentage of time spent exploring an object was defined as: ( time at novel or familiar ) / ( time at novel + time at familiar ) . The discrimination index was calculated as: ( time at novel − time at familiar ) / ( time at novel + time at familiar ) , where novel refers to the object in a novel location , and familiar refers to the other object . Microdialysis was conducted on awake , freely moving rats as previously described ( Nam et al . , 2016 ) with modifications . For optical stimulation coupled with simultaneous microdialysis , adult rats were implanted stereotaxically with an optical fiber ( AP , –3 . 75 mm; ML , 1 . 85 mm; DV , –3 . 10 mm relative to bregma , at a 12° angle ) and a microdialysis probe guide cannula ( AP , –3 . 75 mm; ML , 3 . 01 mm; DV , –1 . 76 mm relative to bregma , at a 15° angle; CMA 12 , CMA Microdialysis AB , Sweden ) . Microdialysis sampling was started after 6 days of recovery from surgery . On the day of sampling , a microdialysis probe ( CMA 12 Elite , membrane length 2 mm , CMA Microdialysis AB , Sweden ) was inserted into CA1 through the probe cannula 3 hr before the start of microdialysis . The probe was connected to a syringe pump ( CMA 402 ) with polyethylene tubing and perfused continuously with artificial cerebrospinal fluid ( aCSF ) at a constant flow rate of 1 . 5 μl/min . After pre-washing ( 30 min ) and recovery ( 1 hr ) , 40 min samples of baseline dialysate were collected by an 820 microsampler ( Univentor , Malta ) as the first microdialysis . Then the second microdialysis for 40 min began with 15 min photostimulation . Extracellular fluid was collected in plastic vials in the presence of the ectonuclease inhibitor ARL67156 ( 100 μM final concentration ) . All the dialysis samples were stored at –80°C for later analysis . A previously described procedure for ATP measurement ( Zhang et al . , 2003 ) with some modifications was used . In brief , the extracellular ATP concentration in the samples was quantified with a bioluminescent assay kit ( FLAA , Sigma-Aldrich ) . A calibration curve was generated with standard ATP samples and the luminescence of the dialysis medium was measured as the background ATP level . A 50 µl sample was added to 50 µl of ATP assay mix containing luciferase-luciferin buffer . The luminescence was measured by a luminometer ( Varioskan Flash , Thermo Scientific , USA ) according to the manufacturer’s instructions . The concentration of extracellular adenosine in the samples was quantified with an adenosine assay kit ( Fluorometric , K327-100 , BioVision ) according to the protocol . Adenosine was measured using adenosine deaminase followed by a multi-step enzymatic approach resulting in the generation of an intermediate that reacts with the adenosine probe to form a fluorescent product . The fluorescent product was measured at excitation/emission = 535/587 nm . ATP-γ-S , NECA , CCPA , CPT and ARL67156 trisodium salt hydrate were from Sigma-Aldrich; MesADP , SCH58261 and CGS 21680 hydrochloride were from Tocris . NECA , CCPA , CPT , SCH58261 , and CGS 21680 were made up to stock solution in dimethyl sulfoxide ( DMSO ) and then diluted to their final concentrations in sterile 0 . 9% saline . ATP-γ-S , MesADP , and ARL67156 were dissolved in sterile 0 . 9% saline and diluted to their final concentrations in sterile 0 . 9% saline . Intracerebral drug delivery was through previously-implanted infusion cannulae . On the day of the experiment , the internal cannulae that protruded 2 mm beyond the ends of the guide cannulae were inserted , and drugs ( 1 μl/side ) were infused bilaterally . The vehicle control groups were given an equivalent amount of DMSO dissolved in sterile 0 . 9% saline or equivalent sterile 0 . 9% saline . For i . p . injection , rats were given CCPA ( 0 . 03 , 0 . 1 , 0 . 3 , or 1 mg/kg body weight ) or the appropriate vehicle . All rats were anesthetized with sodium pentobarbital and then perfused transcardially with 0 . 9% NaCl followed by 4% paraformaldehyde ( PFA , wt/vol ) dissolved in phosphate-buffered saline ( PBS , pH 7 . 4 ) . The brains were removed and postfixed in 4% PFA at 4°C overnight , then cryoprotected in 30% sucrose ( wt/vol ) for 3–4 days at 4°C . Coronal sections ( 40 μm ) were cut on a microtome ( CM 1950 , Leica , Germany ) and stored in PBS at 4°C for further use . For immunostaining , each section was treated with 0 . 5% Triton X-100 ( vol/vol ) for 10 min . After washing with PBS , the sections were blocked in 10% bovine serum albumin ( BSA , wt/vol ) with 5% donkey serum ( wt/vol ) for 1 . 5 hr at room temperature and then incubated with primary antibody ( rabbit anti-GFAP 1:500 , Millipore; mouse anti-NeuN , 1:400 , Millipore; Rabbit anti-DsRed 1:500 , Clontech ) diluted in 5% BSA ( wt/vol ) at 4°C for 24 hr . The sections were then washed three times ( 10 min each ) in PBS , followed by incubation with secondary antibody ( 1:1000 Alexa Fluor 568 anti-rabbit , Invitrogen; 1:1000 Alexa Fluor 647 anti-mouse , Invitrogen; 1:1000 Alexa Fluor 488 donkey anti-mouse , Invitrogen ) for 2 hr at room temperature . They were then incubated for 5 min with DAPI and rinsed three times with PBS . Finally , the sections were mounted on microscope slides and coverslipped . Fluorescence images were acquired with an Olympus FV-1200 ( Japan ) confocal microscope . The acute brain slices were prepared following a previously described protocol ( Ting et al . , 2018 ) . Briefly , GFAP-ChR2-EYFP rats ( 3–4 weeks postnatal ) were anesthetized with pentobarbital sodium , then perfused transcardially with cold ( 2–4°C ) oxygenated ( 95% O2/5% CO2 ) N-methyl-D-glucamine ( NMDG ) - 4- ( 2-hydroxyethyl ) −1-piperazineethanesulfonic acid ( HEPES ) aCSF before decapitation . Then the whole brain was removed rapidly into cold oxygenated aCSF containing ( in mM ) : 92 NMDG , 2 . 5 KCl , 1 . 25 NaH2PO4 , 30 NaHCO3 , 20 HEPES , 25 glucose , 2 thiourea , 5 Na-ascorbate , 3 Na-pyruvate , 0 . 5 CaCl2·4H2O , and 10 MgSO4·7H2O . The pH of the NMDG–HEPES aCSF was titrated to pH 7 . 3–7 . 4 with concentrated HCl . After the brain was swiftly dissected , transverse slices ( 300 μm ) were cut on a vibratome ( VT1200s , Leica , Germany ) and transferred into a recovery chamber with NMDG-HEPES aCSF . Then , the slices were transferred into HEPES-aCSF ( in mM ) : 92 NaCl , 2 . 5 KCl , 1 . 25 NaH2PO4 , 30 NaHCO3 , 20 HEPES , 25 glucose , 2 thiourea , 5 Na-ascorbate , 3 Na-pyruvate , 2 CaCl2·2H2O , and 2 MgSO4·7H2O ( pH 7 . 3–7 . 4 ) at room temperature and allowed to recover for at least 1 hr . The recording aCSF contained ( in mM ) : 119 NaCl , 2 . 5 KCl , 1 . 25 NaH2PO4 , 24 NaHCO3 , 12 . 5 glucose , 2 CaCl2·4H2O , and 2 MgSO4·7H2O ( pH 7 . 3–7 . 4 ) continuously bubbled with 95% O2/5% CO2 . For recording , an individual slice was transferred to a submerged recording chamber and continuously perfused with the above aCSF ( 3 . 0 ml/min ) at 26°C . The slice was visualized under a microscope ( BX51WI , Olympus , Japan ) using infrared differential interference contrast optics . For whole-cell patch clamp recording from hippocampal slices , the patch electrodes were made from borosilicate glass capillaries ( B-120-69-15 , Sutter Instruments , USA ) and had resistances in the 3–5 MΩ range . The internal solution contained ( in mM ) : 130 K-gluconate , 4 KCl , 10 HEPES , 4 MgATP , 0 . 3 Na2GTP , 10 Na2-phosphocreatine , and 0 . 3 EGTA . Recordings were made with an Axon 700B patch-clamp amplifier and 1320A interface ( Axon Instruments , USA ) . The signals were filtered at 2 kHz using amplifier circuitry , sampled at 10 kHz , and analyzed using Clampex 9 . 0 ( Axon Instruments ) . Photostimulation was delivered by 473 nm solid-state laser diodes , and light pulses were generated with a custom-built high-speed shutter; the power density of the blue light was 1–3 mW/mm2 . Blue light was delivered to the slices through a thin quartz fiber ( 200 μm diameter , custom made ) . Ca2+ imaging in hippocampal slices was performed using a confocal laser scanning microscope ( Olympus FV-1200 , Japan ) . Astrocytes were bulk loaded in slices with Rhod-2 AM ( 20 μM , Invitrogen ) . The fluorescence intensity was measured at an excitation wavelength of 550 nm and emission wavelength of 580 nm . Ca2+ signals were calculated as the relative change in fluorescence ( ΔF/F ) , where F is the fluorescence intensity before photostimulation and ΔF is the change in fluorescence after photostimulation . Primary hippocampal cultures were prepared as described previously ( Kaech and Banker , 2006 ) with some modifications . Embryonic day 18 rat hippocampi were dissected and dissociated with 0 . 125% trypsin . Cells were re-suspended in Dulbecco’s modified Eagle’s medium containing 10% fetal bovine serum and 10% Ham’s F-12 ( all from Gibco ) at a cell density of 20 , 000 mL−1 . The neurons were plated on a layer of astrocytes and maintained at 37°C in a 5% CO2 incubator . The culture medium was changed every 2–3 days . The co-cultured cells were used for fluorescence image acquisition . All statistics were calculated with GraphPad Prism ( Version 6 . 01 ) . One-way or two-way repeated measures ANOVA followed by the Bonferroni or Newman-Keuls post hoc test and standard two-tailed paired or unpaired t-tests were used as indicated in the figure legends . Normal distribution was determined by D’Agostino-Pearson , Shapiro-Wilk , and Kolmogorov-Smirnov normality tests . Animals were randomly assigned to treatment groups . Data are presented as the mean ± SEM . Statistical significance was set at p<0 . 05 .	Memory is the record of what we learn over time and is essential to our survival . But not all memories are helpful . Repeatedly recalling a traumatic event – such as an assault – can be harmful . About 1 in 3 people who experience severe trauma go on to develop post-traumatic stress disorder ( PTSD ) , in which they re-live the traumatic event in the form of flashbacks and nightmares . Others develop panic disorder , phobias or depression . Preventing this chain of events is challenging because fear memories form rapidly and last a long time . Current treatments involve re-exposing individuals to the traumatic event . This could be real-life exposure in the case of a phobia . Or it could involve visualizing the event , in the case of PTSD . Controlled re-exposure can help individuals learn new coping strategies . But it does not erase the initial fear memory . A better approach might be to take advantage of the fact that new memories are unstable . To form a long-lasting memory trace , newly acquired information must go through a process called consolidation to stabilize it . This process takes place in an area of the brain called the hippocampus . If consolidation does not occur , new memory traces can fade away . Li , Li et al . now show that preventing consolidation in the rat brain stops the animals from forming lasting memories of a stressful event , namely a foot shock . In the study , the rats first learned to associate a foot shock with a tone . This training took place inside a specific chamber . After learning the association , the rats began to freeze – a sign of fear – whenever they entered the chamber . This happened even if the tone was not played . But Li , Li et al . showed that they could reduce this fear response by activating cells in the hippocampus known as astrocytes , shortly after the learning episode . Activating the astrocytes made them release a substance called adenosine . Molecules of adenosine then bound to and activated proteins called adenosine A1 receptors . Administering a drug that activated these receptors directly had the same effect as activating the astrocytes themselves . This suggests that drugs of this type could one day help patients with fear-related disorders such as PTSD and phobias . For this to become a reality , new studies must test different drugs and find the best ways of administering them . After testing in animal models , the next step will be preliminary clinical trials in people .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2020	Activation of astrocytes in hippocampus decreases fear memory through adenosine A1 receptors
The Atlantic herring is one of the most abundant vertebrates on earth but its nucleotide diversity is moderate ( π = 0 . 3% ) , only three-fold higher than in human . Here , we present a pedigree-based estimation of the mutation rate in this species . Based on whole-genome sequencing of four parents and 12 offspring , the estimated mutation rate is 2 . 0 × 10-9 per base per generation . We observed a high degree of parental mosaicism indicating that a large fraction of these de novo mutations occurred during early germ cell development . The estimated mutation rate – the lowest among vertebrates analyzed to date – partially explains the discrepancy between the rather low nucleotide diversity in herring and its huge census population size . But a species like the herring will never reach its expected nucleotide diversity because of fluctuations in population size over the millions of years it takes to build up high nucleotide diversity . Empirical observations of nucleotide diversity in different species show that the variation is often much smaller than would be expected from simple population genetic models ( Leffler et al . , 2012 ) . The Atlantic herring ( Clupea harengus ) is a good example of the paradox , since , in spite of an enormous census population size about 1012 ( Supplementary file 1 ) , its nucleotide diversity ( π = 0 . 3% ) ( Martinez Barrio et al . , 2016 ) is middle-of-the-road when compared to terrestrial mammals , e . g . 0 . 1% for humans ( Hernandez et al . , 2011 ) and 0 . 9% for European rabbits ( Carneiro et al . , 2014 ) with much smaller census populations . A large census population does not necessarily mean that the long term effective population size ( Ne ) is large but the extremely low genetic differentiation at selectively neutral loci between geographically distant populations strongly suggests that current Ne must be high and genetic drift very low in the Atlantic herring ( Martinez Barrio et al . , 2016 ) . Before the NextGenerationSequencing-era , mutation rates were estimated by comparative genomics , by relating sequence differences to fossil record-dated estimates of species divergence times , or by tracking changes at specific loci in experimental studies . However , since species divergence is hard to date and the use of a small subset of loci can introduce bias , these methods have limited accuracy ( Drake et al . , 1998 ) . More recently , affordable whole genome sequencing has facilitated two approaches to estimate mutation rates: mutation accumulation lines and parent-offspring comparisons . The mutation accumulation approach , where an inbred line is maintained for a number of generations and the mutation rate is measured by counting up differences between the first and last generation , has the advantage of scalability , since it is possible to increase the number of mutation events observed by including more generations . On the other hand , the approach requires an organism that can be reproduced as viable inbred lines , and it is difficult to fully eliminate purifying selection against deleterious new mutations . The parent-offspring approach , which relies on using high coverage whole-genome sequencing to detect differences between parents and their offspring , alleviates the cultivation related issues , and has thus become the preferred method for estimating the mutation rate in non-model organisms . The trade-off is that the total number of mutation events per progeny will typically be small . Currently , the number of studies using any of the methods outlined above remains small , and the available data is somewhat biased towards unicellular organisms ( Dettman et al . , 2016; Dillon et al . , 2015; Farlow et al . , 2015; Lee et al . , 2012; Ness et al . , 2012; Zhu et al . , 2014 ) , insects ( Keightley et al . , 2014 , 2015 , 2009 ) and mammals ( Harland et al . , 2016; Kong et al . , 2012; Uchimura et al . , 2015; Venn et al . , 2014 ) , while including a single plant ( Ossowski et al . , 2010 ) and one bird ( Smeds et al . , 2016 ) . In all , this leaves large sections of the tree of life essentially unexplored . This is problematic for drawing general conclusions about the relationship between neutral diversity , effective population size and mutation rate , which is a topic of considerable interest in population genetics ( Leffler et al . , 2012; Lynch , 2010 ) . In this study , to our knowledge the first of its kind in a teleost , we estimate the genome-wide point mutation rate in Atlantic herring . The Atlantic herring was chosen due to its suitability as a population genetic model system; it is one of the most abundant vertebrate species on earth with external reproduction involving large numbers of gametes per reproducing adult . In essence , these properties make the Atlantic herring one of the best approximations of a randomly mating , infinite size population among vertebrates . In addition , there exists a high-quality draft genome assembly ( Martinez Barrio et al . , 2016 ) , which is a pre-requisite for a study of this kind . We have employed the parent-offspring approach , and base our measurement on two families , each containing two parents and six offspring . We here estimate the spontaneous mutation rate to be 2 . 0 × 10−9 per site per generation in the Atlantic herring , six-fold lower than the rate in humans and the lowest rate reported so far for a vertebrate . We have generated two-generation experimental pedigrees for spring-spawning Atlantic and Baltic herring ( classified as a subspecies of the Atlantic herring by Linnaeus ( 1761 ) ) , each comprising the two parents and six offspring ( Table 1 ) . We performed whole-genome sequencing of these two pedigrees using genomic DNA isolated from muscle tissue . As detection of de novo mutations requires high sequence coverage , we sequenced each individual to ~45–71 x ( Table 1 ) , in line with the procedures used in previous studies ( Keightley et al . , 2015; Kong et al . , 2012 ) . The sequences were aligned to the recently published Atlantic herring genome ( Martinez Barrio et al . , 2016 ) . A total of 5 . 3 ( Atlantic ) and 5 . 2 ( Baltic ) million raw SNPs were detected in each pedigree , respectively , using GATK ( see Materials and methods ) ( McKenna et al . , 2010 ) . 10 . 7554/eLife . 23907 . 003Table 1 . Summary of the pedigrees used for whole-genome sequencing . DOI: http://dx . doi . org/10 . 7554/eLife . 23907 . 003NoIDPedigreeSequencing depth ( x ) De novo mutationsPedigree 1 , Atlantic herring1AM8Father65 . 7N . A . 2AF8Mother70 . 2N . A . 3AA1Offspring65 . 614AA2Offspring70 . 925AA3Offspring47 . 206AA4Offspring66 . 937AA5Offspring64 . 248AA6Offspring61 . 21Pedigree 2 , Baltic herring9BM19Father71 . 8N . A . 10BF21Mother65 . 1N . A . 11BB1Offspring74 . 5212BB2Offspring61 . 6113BB3Offspring75 . 0014BB4Offspring69 . 9215BB5Offspring60 . 6216BB6Offspring62 . 61N . A . = Not available . Detection of de novo mutations with high confidence requires a careful examination of raw variant calls and application of highly stringent filtering criteria . Using a standard genotype-calling pipeline will typically lead to the great majority of novel sequence variants detected being false positives . Screening of provisional candidate mutations in a single offspring indicated that this was the case , as many candidates could not be verified using Sanger sequencing . Hence , in order to minimize the frequency of false positives by the de novo calls using only the GATK variant caller , we separately performed variant calling using SAMTOOLS ( Li et al . , 2009 ) and only selected novel mutations detected by both variant callers ( Figure 1 ) . In addition , we applied strict filtering criteria in order to remove variants detected due to sequencing and alignment errors . We excluded variant calls from genomic regions with low mappability ( see Materials and methods ) and repetitive regions detected by Repeat Masker ( Smit et al . , 2013 ) . Furthermore , we defined the cut-off parameters for sequence depth , SNP and genotype quality-related statistics using the set of SNPs that were fixed for different alleles in both parents and thus heterozygous in all offspring ( Figure 1 , Materials and methods ) . As this strict filtering could lead to failure to detect some fraction of true heterozygotes , we estimated the false negative rate of our pipeline by calling SNPs in each individual offspring separately , in order to eliminate bias stemming from shared SNPs present in multiple individuals being called with higher power . For this analysis we used 116 , 910 polymorphic sites where the parents were homozygous for different alleles in the joint genotype calling . The expectation is that these sites are heterozygous in all offspring , but that information did not influence SNP calling . By separating the individuals , we mimicked the situation for de novo mutations , which are typically not shared . Using the same pipeline as for the de novo detection , the average detection rate of such heterozygous positions across all offspring was 94 . 1% , yielding a false negative rate of 5 . 9% . As an alternative way of estimating the false negative rate , we used a simulation procedure where we generated mutated reads for 1000 positions within callable regions . Each site in each offspring had its frequency of mutated reads determined by a sample from the observed frequency distribution of called heterozygous sites in the original data set ( see Materials and methods ) . Across all offspring , we found an overall frequency of 2 . 7% false negative calls , while roughly 9% of sites failed to generate a call ( Supplementary file 2 ) . Overall , the two methods used are in agreement . However , for the purpose of the final calculations we will use the empirical estimate of 5 . 9% , which includes both incorrect and failed calls , as it is derived directly from the real data set . The choice has minor effects on the estimated mutation rate , as using the simulated value would result in the final rate being approximately 5% higher . 10 . 7554/eLife . 23907 . 004Figure 1 . Flowchart describing the de novo mutation-calling pipeline . A schematic illustration of the steps used in calling and filtering the candidate mutations . DOI: http://dx . doi . org/10 . 7554/eLife . 23907 . 00410 . 7554/eLife . 23907 . 005Figure 1—figure supplement 1 . Sanger sequencing chromatograms of the de novo mutations . Chromatograms from the identified target offspring and its parents for each region containing a candidate de novo mutation . DOI: http://dx . doi . org/10 . 7554/eLife . 23907 . 005 This stringent filtering procedure identified a total of 17 candidate de novo mutations , nine in the Atlantic pedigree and eight in the Baltic one ( Tables 1 and 2 ) . Two of the 17 de novo mutations were each found in two different offspring from the same pedigree . 10 . 7554/eLife . 23907 . 006Table 2 . Summary of the de novo mutations identified in Atlantic herring . DOI: http://dx . doi . org/10 . 7554/eLife . 23907 . 006SNP positionMutationScaffold:positionIdRefVarFreq†Origin‡Type§Region1157:174 , 127AA4TA1/50 ( - ) MTVIntergenic153:2 , 684 , 380AA2TG9/50 ( 18% ) PTVIntronic241:7 , 752 , 158AA5CA5/50 ( 10% ) MTVIntergenic4:5 , 098 , 858AA5TC2/50 ( 4% ) MTSIntronic481:1 , 927 , 799AA4 , AA5CA6/50 ( 12% ) PTV3' UTR61:815 , 077AA4AT3/50 ( 6% ) N . A . TVIntergenic62:613 , 919AA1 , AA6CA6/50 ( 12% ) MTVIntergenic729:1 , 499 , 224AA2CT4/50 ( 8% ) MTSIntronic887:195 , 946AA5GA1/50 ( - ) PTSIntronic10:1 , 443 , 002BB4CT1/46 ( - ) PTSIntronic151:267 , 875BB5AT1/46 ( - ) PTVExonic177:1 , 045 , 894BB1AG1/46 ( - ) PTSIntronic194:478 , 776BB6AG1/46 ( - ) N . A . TSIntronic246:1 , 890 , 479BB4TC1/46 ( - ) PTSIntergenic257:380 , 993BB2GA1/46 ( - ) MTSIntergenic26:2 , 976 , 192BB1TC2/46 ( 4% ) PTSIntronic37:1 , 374 , 669BB5GA1/46 ( - ) MTSIntronicSame mutation detected in two progeny . †Number of siblings carrying the de novo mutation; - the frequency of transmission was only estimated when two or more progeny with the de novo mutation was detected . ‡M:Maternal , P:Paternal , N . A . = Not available§TV = Transversion , TS = Transition We performed Sanger sequencing of the genomic regions around each of these putative de novo mutations in all parents and the 12 offspring ( Figure 1—figure supplement 1 ) . This confirmed that all 17 putative de novo mutation events were genuine and all the peak ratios of two alleles were close to 1:1 consistent with germ-line mutations . Thus , we did not observe any false positives in this study . In order to estimate the transmission frequencies of our detected de novo mutations , we measured the rate of transfer of the de novo mutations in a larger set of offspring ( n = 46 and 50 per family ) , in order to infer when during the formation of the parental germ line the mutation occurred ( Table 2 ) . For eight out of seventeen de novo mutations we observed more than one sibling carrying exactly the same mutation ( Table 2 ) . The range of occurrences for the de novo mutations was one to nine among the 50 offspring . Even the maximum of the observed transfer rates ( 18% for scaffold 153: 2 , 684 , 380 T>G ) was significantly lower than the 50% expected for a fixed mutation ( p=1 . 4×10−3 , Fisher’s exact test ) . About half of the de novo mutations were present in two or more offspring , indicating that they occurred during early germ cell divisions . Assuming that the number of cell divisions from zygote to mature sperm or egg is similar in Atlantic herring to the one in mammalian species , we can conclude from a recent simulation study ( Harland et al . , 2016 ) that it would be highly unlikely to observe such a high rate of parental mosaicism unless a large fraction of the de novo mutations occurred during early germ cell divisions . Further , the incidence of parental mosaicism differed significantly between the two families included in this study ( Table 2; p=0 . 01 , Fisher’s exact test ) . The finding that the same mutation was observed in two or more siblings for eight of the putative de novo mutations confirms that these must be germ-line mutations and not somatic mutations . We also explored if the 17 germ line de novo mutations had a paternal or maternal origin . For 14 of the de novo mutations , we could detect an additional segregating site within the same Illumina sequencing read ( length 125 bp ) or mate-pair read that spanned the respective de novo mutation and was uniquely associated with either parent . In these cases , the parental origin could be directly inferred . We were able to infer the parental origins of one additional de novo mutation by PCR cloning and sequencing . Out of the 15 mutations for which their parental origin was determined , there was no significant difference between paternal ( eight ) and maternal ( seven ) mutations ( Table 2 ) . A paternal bias in the origin of de novo mutations has been shown in mammals , such as human ( ratio = 3 . 9 ) ( Kong et al . , 2012 ) and chimpanzee ( ratio = 5 . 5 ) ( Venn et al . , 2014 ) , where the main reason is thought to be the larger number of cell divisions during spermatogenesis than during oogenesis ( Crow , 2000 ) . While the numbers are small , a binomial test against the human ratio indicates that the gender bias in herring , if it exists at all , is significantly weaker than in humans ( p=0 . 004 ) . In herring , both sexes produce large numbers of gametes and males only produce sperm during the spawning season ( a few months per year ) . Furthermore , the high degree of parental mosaicism indicates that a large fraction of the de novo mutations reported here must have occurred during early germ cell development when we do not expect a strong gender effect . These circumstances offer a reasonable explanation to the balanced parental origin of de novo mutations in the Atlantic herring . Among the 17 de novo mutations , there were 10 transitions and seven transversions , yielding a transition/transversion ratio of 1 . 4 , which can be compared with a genome-wide ratio of 1 . 1 for previously reported SNPs ( Martinez Barrio et al . , 2016 ) . An overrepresentation of transitions is expected , and the observed ratio falls in the range found in previous de novo mutation studies . For example , Kong et al . identified 3344 transitions out of 4933 events ( ratio = 2 . 1 ) in humans ( Kong et al . , 2012 ) , while Keightley et al . found five out of nine events ( ratio = 1 . 25 ) in the tropical butterfly Heliconius melpomene ( Keightley et al . , 2015 ) . In humans and other mammals there is a well-established excess of CpG>TpG mutations ( Kong et al . , 2012 ) . CpG methylations also occur in teleosts ( Rai et al . , 2010 ) , but in our dataset only 1 out of 17 de novo mutations was of this type . This frequency ( 6% ) is below , but not significantly different from the frequency reported for human ( 19% ) ( Kong et al . , 2012 ) ( binomial test , p=0 . 06 ) . There were six mutations located in intergenic regions , nine intronic mutations , one 3’ UTR mutation and one exonic mutation . In all , this is a distribution that does not deviate significantly from random expectation , given the composition of the genome after mappability filtering ( p=0 . 65 , Fisher’s exact test ) . We identified nine and eight de novo mutations in the Atlantic herring and the Baltic herring pedigrees , respectively . Since we had 12 progeny in total and two of the mutations were detected twice among the sequenced progeny , our estimate of the de novo mutation rate is 0 . 79 ( 19/24 ) . After strict filtering of genomic regions with low mappability and repetitive sequences , we had ~442 Mb of sequence available for variant screening . Based on the distribution of read coverage in a random subset of the genome , we estimated that 2 . 6% of this region have insufficient depth for successful SNP calling , giving us a final callable region of 442 × 0 . 974 = 431 Mb ( representing ~51% of the genome ) . The mutation rate per site per generation can thus be estimated as 19/ ( 2 × 12×431 x 106 ) =1 . 8×10−9 ( 95% CI = 1 . 1–2 . 7 × 10−9 , assuming that the mutations are Poisson distributed ) . If we correct for the estimated false negative rate ( 5 . 9% ) we obtain: 2 . 0 × 10−9 ( 95% CI = 1 . 1–2 . 9 × 10−9 ) . Based on historical sampling of several herring stocks , we estimated the minimum generation time of Atlantic herring before the onset of large-scale commercial fishing to be approximately six years ( Supplementary file 3 ) . Using this historical generation time , the mutation rate per site per year in the Atlantic herring was estimated at 3 . 3 × 10−10 ( 95% CI = 1 . 9×10−10 – 4 . 8 × 10−10 ) . This study provides new insights regarding factors affecting the mutation rate and levels of nucleotide diversity in vertebrates . Our finding of a high degree of parental mosaicism for the detected de novo mutations is consistent with several recent studies indicating that the early cleavage cell divisions in the germ-line are particularly mutation-prone ( Harland et al . , 2016; Rahbari et al . , 2016; Ségurel et al . , 2014 ) . A high rate of de novo mutations at early germ-cell divisions has also been reported for Drosophila ( Gao et al . , 2014 ) . The estimated mutation rate ( μ = 2 . 0×10−9 ) for the Atlantic herring is the lowest for a vertebrate species to date ( Table 3 ) ; about six-fold lower than in humans . It should be noted that this number reflects the rate in the callable fraction of the genome , which by definition does not contain repeat regions . Thus , the true genomic average could be somewhat higher , as replication of repetitive regions tends to be more error-prone , but the decreased calling power in those regions makes diversity hard to estimate in an unbiased fashion . However , these issues are not unique to the Atlantic herring , similar caveats apply to estimates of mutation rates in other species as well , and the results should thus be comparable across species . In this study we surveyed about 51% of the current genome assembly for the Atlantic herring and we used our previously published population data ( Martinez Barrio et al . , 2016 ) to estimate the nucleotide diversity in the parts of the genome that were included and excluded in the current study to address the concern that we may have underestimated the mutation rate because the rate is higher in the part that was excluded . This analysis showed that the nucleotide diversities in the excluded and included parts were almost identical ( π = 0 . 00318 and π = 0 . 00304 , respectively ) . In conclusion , this analysis does not indicate a major difference in mutation rates between the two parts of the genome . 10 . 7554/eLife . 23907 . 007Table 3 . Summary of mutation rates measured to date . DOI: http://dx . doi . org/10 . 7554/eLife . 23907 . 007SpeciesTaxonomic groupμMethodGenome size ( Mb ) Ne† Pseudomonas aeruginosaBacteria7 . 9 × 10−11MA16 . 32 . 1 × 108Burkholderia cenocepaciaBacteria1 . 3 × 10−10MA28 . 12 . 5 × 108Escherichia coliBacteria2 . 2 × 10−10MA34 . 61 . 6 × 108Chlamydomonas reinhardtiiUnicellular eukaryotes2 . 1 × 10−10MA41207 . 8 × 107Saccharomyces cerevisiaeUnicellular eukaryotes1 . 7 × 10−10MA512 . 21 . 2 × 107Schizosaccharomyces pombeUnicellular eukaryotes2 . 1 × 10−10MA612 . 61 . 4 × 107Arabidopsis thalianaPlants7 . 1 × 10−9MA71192 . 8 × 105Pristionchus pacificusInvertebrates2 . 0 × 10−9MA81331 . 8 × 106Caenorhabditis elegansInvertebrates1 . 5 × 10−9MA91005 . 2 × 105Caenorhabditis briggsaeInvertebrates1 . 3 × 10−9MA91082 . 7 × 105Drosophila melanogasterInvertebrates3 . 2 × 10−9MA10 PO111441 . 4 × 106Heliconius melpomeneInvertebrates2 . 9 × 10−9PO122742 . 1 × 106Daphnia pulexInvertebrates5 . 7 × 10−9MA132508 . 2 × 105Atlantic herring ( Clupea harengus ) Teleosts2 . 0 × 10−9PO8504 . 0 × 105Collared flycatcher ( Ficedula albicollis ) Birds4 . 6 × 10−9PO1411182 . 0 × 105Mouse ( Mus musculus ) Mammals5 . 4 × 10−9MA15 , 1628081 . 8 × 105Cattle ( Bos taurus ) Mammals9 . 7 × 10−9PO1727253 . 7 × 104Chimpanzee ( Pan troglodytes ) Mammals1 . 2 × 10−8PO1832312 . 9 × 104Human ( Homo sapiens ) Mammals1 . 2 × 10−8PO1932362 . 4 × 104MA = Mutation Accumulation , PO = Parent-Offspring . The values are from the following sources: 1 . Dettman et al . ( 2016 ) ; 2 . Dillon et al . ( 2015 ) ; 3 . Lee et al . ( 2012 ) ; 4 . Ness et al . ( 2012 ) ; 5 . Zhu et al . ( 2014 ) ; 6 . Farlow et al . ( 2015 ) ; 7 . Ossowski et al . ( 2010 ) ; 8 . Weller et al . ( 2014 ) ; 9 . Denver et al . ( 2012 ) ; 10 . Keightley et al . ( 2009 ) ; 11 . Keightley et al . ( 2014 ) ; 12 . Keightley et al . ( 2015 ) ; 13 . Keith et al . ( 2016 ) ; 14 . Smeds et al . ( 2016 ) ; 15 . Lindsay et al . ( 2016 ) ; 16 . Uchimura et al . ( 2015 ) ; 17 . Harland et al . ( 2016 ) ; 18 . Venn et al . ( 2014 ) ; 19 . Kong et al . ( 2012 ) . †Ne is calculated as π/4μ . The underlying π estimates are all from Lynch et al . ( 2016 ) except for herring ( present study ) , collared flycatcher ( Ellegren et al . , 2012 ) and cattle ( Daetwyler et al . , 2014 ) . By combining the now estimated mutation rate with the neutral diversity level ( π = 0 . 0032 ) found by Martinez Barrio et al . ( 2016 ) and the expected relationship between nucleotide diversity , the mutation rate and effective population size ( Ne ) for selectively neutral alleles ( π = 4 Ne μ ) , we obtain an estimated Ne of 4 × 105 . While this number is larger than for most terrestrial animal species , it is still much lower than the census population size of the herring , about 1012 ( Supplementary file 1 ) . There are several factors that may contribute to this discrepancy , but demographic history stands out as the most prominent factor . Using coalescent analysis and allele frequency distributions , Martinez Barrio et al . ( 2016 ) showed that the herring population is expanding from a previous bottleneck . Since the diversity-based estimate of effective population size can be considered as an average over time this bottleneck still have a major impact on the current nucleotide diversity . Population genetics theory implies that it will take 4Ne generations before populations reach their genetic equilibrium ( Kimura and Ohta , 1973 ) . We have estimated the generation interval to approximately six years in this study ( Supplementary file 3 ) and a conservative estimate of the current ( not long-term ) Ne is 107 , which appears reasonable since we estimated long-term Ne at 4 × 105 and we have evidence for population expansion ( e . g . excess of rare alleles ( Martinez Barrio et al . , 2016 ) ) . These figures indicate that it will take about 240 million years before the herring populations reach genetic equilibrium . Thus , it is obvious that a species with a large population size like the herring and a relatively long generation interval will never reach genetic equilibrium . Background selection ( the elimination of deleterious alleles ) and selective sweeps will also lead to reductions in nucleotide diversity at linked neutral sites ( Gillespie , 2000 , 2001 ) . Furthermore , highly efficient purifying selection decreases the fraction of the genome that appears as selectively neutral ( Ohta , 1973 ) which is also expected to lead to a slightly reduced nucleotide diversity . The fact that the observed mutation rate is unusually low in the Atlantic herring is of interest in relation to the drift-barrier hypothesis ( Lynch et al . , 2016 ) , which predicts that the purging of slightly deleterious mutations affecting the mutation rate is particularly effective in species that have a very large effective population size , large fecundity and close to random mating , conditions which the Atlantic herring meets ( Table 3 ) . However , since the population size of the Atlantic herring appears to have fluctuated over time ( Martinez Barrio et al . , 2016 ) , it remains unclear exactly how powerful selection has been in a time-averaged perspective , which means the support for the drift-barrier hypothesis is not unconditional . Additionally , the low body temperature of a marine fish may also slow down the metabolic rate which has been suggested to decrease the mutation rate ( Martin and Palumbi , 1993 ) . In a recently released study , Malinsky et al . ( 2017 ) used three trios representing three species of Lake Malawi cichlids and estimated the overall mutation rate to 3 . 5 × 10−9 compared with 2 . 0 × 10−9 for the Atlantic herring . However , these fish both have a lower estimated effective population size than herring ( Malinsky et al . , 2017 ) and live in warmer waters . In conclusion , there is still a need to compare our data with mutation rates from additional species , with lower populations sizes but similar body temperatures , before we can draw firm conclusions about the relationship between population size and mutation rate . According to simple , ideal-case population genetic models there should be a positive relationship between nucleotide diversity and population size , so that a population at mutation-drift balance has a nucleotide diversity of 4Nμ . However , as outlined above , this expectation is disrupted by population size fluctuations over time and selective forces . In practice , population sizes are only weakly , if at all , correlated with nucleotide diversity ( Leffler et al . , 2012 ) . Our finding that the inherent mutation rate is approximately six times lower in Atlantic herring than in humans indicates that differences in intrinsic mutation rate is also an important factor when comparing nucleotide diversities among species . In the case of the Atlantic herring , the low mutation rate , the demographic history and efficient positive and negative selection , all contribute to explaining the apparent disparity between nucleotide diversity and the census population size in the Atlantic herring . Two full-sib families were generated by crossing wild-caught Atlantic herring from Bergen ( Norway ) and Baltic herring from Hästskär ( Sweden ) . For each family , six offspring from a total of 50 progeny were selected for sequencing together with the two parents . Our aim was to determine the mutation rate to its order of magnitude and one to two significant digits . Thus , a samples size of 12 progeny was expected to result in about 100 detectable novel mutations based on previously known vertebrate mutation rates and the size of the genomic regions we could use to detect mutations . Genomic DNA was isolated from muscle tissue using Qiagen DNeasy Blood and Tissue kit . DNA libraries were constructed using the TruSeq PCR-free kit . All individuals were sequenced on Illumina HiSeq2500 machines , using 2 × 125 bp paired reads to a sequencing depth of ~47–71X ( Table 1 ) . The short reads were aligned to the Clupea harengus reference genome ( Martinez Barrio et al . , 2016 ) using BWA v0 . 6 . 2 ( Li and Durbin , 2009 ) with default parameters . The data were then filtered based on mappability , calculated using GEM ( Derrien et al . , 2012 ) , within the reference assembly , so that only positions with mappability 1 that were also inside 1 kb windows with average mappability >0 . 95 were included in the downstream analysis; 442 Mb ( 52% ) of genome sequence passed this filtering step . The sequence data have been deposited in the SRA archive ( PRJNA356817 ) . Sequence alignments from the previous step were used for calling variants using two separate tools; GATK v3 . 3 . 0 ( McKenna et al . , 2010 ) and SAMTOOLS v . 1 . 19 ( Li et al . , 2009 ) . We used GATK HaplotypeCaller with default parameters that performs simultaneous calling of SNP and Indels via local de novo assembly of haplotypes ( see GATK manual for details ) . We ran HaplotypeCaller separately for each individual to generate intermediate genomic VCF ( Danecek et al . , 2011 ) files ( gVCF ) . Afterwards , we used the GenotypeGVCFs module in GATK to merge gVCF records from each individual ( altogether 12 from the two pedigrees ) using the multi-sample joint aggregation step that combines all records , generate correct genotype likelihood , re-genotype the newly merged record and re-annotate each of the called variants and thereby generate a VCF file . For SAMTOOLS , we used the standard multi-sample SNP calling pipeline ( Li et al . , 2009 ) using the ‘mpileup’ module for calling raw variants . Once we got the raw variant calls , we filtered small insertions and deletions and only used SNPs for downstream analysis . Furthermore , we also removed SNPs that had missing genotypes in one or both parents , as these SNPs were not informative . Afterwards , we extracted a subset of SNPs where parents were homozygous for different allele and all six offspring were heterozygous ( the genotype calls were considered heterozygous in offspring if the minor allele frequency was >25% ) . The SNP quality annotations in this set of ‘known’ heterozygous offspring were used as proxy to consider the quality parameter of true SNPs in the dataset . We extracted various SNP quality annotations recorded in the VCF file like total read depth , mapping quality , mapping quality rank sum , base quality , base quality rank sum , read position rank sum , quality by depth , genotype quality , allele depth ( see GATK manual for details on these parameters ) and examined their distributions in the subset of our known heterozygous offspring . As these quality parameters were close to being normally distributed , we used the threshold of mean ±2 x standard deviation for each of these quality estimates as the standard cut-offs for our in-house SNP filtering pipeline to filter raw SNPs in our entire dataset ( Figure 1 ) . From the filtered SNP dataset generated in the previous step , we further selected those sites where both parents were homozyogous for the reference allele and at least one offspring carried the variant allele in the heterozygous state . These two sets of raw novel mutations in offspring independently called by GATK and SAMTOOLS were then intersected and the sites that were detected by both variant callers were considered as our true de novo mutations among the progeny . PCR amplification and Sanger sequencing of both strands verified all candidate mutations . We inferred the parental origin of the de novo mutations based on flanking SNP alleles that could be verified by Sanger sequencing and only have been transferred from one of the parents . The parental origin of fourteen de novo mutations could be directly deduced from SNP alleles segregating between the two parents present on the same short Illumina read and mate-pair read as the de novo mutation ( at least 5 reads ) . The parental origin of one additional de novo mutations was determined via cloning PCR fragments and sequencing; we sequenced at least 7 independent clones for each de novo mutation . Firstly , we estimated the false negative rate by performing genotype calls at those nucleotide positions where the parents were fixed for different alleles . The genotype calls for progeny were done without using the information for parents to mimic the detection of de novo mutations . Secondly , we also used simulation to estimate the false negative rate . From the previously determined callable fraction of the genome , we selected approximately 1000 sites without any existing polymorphism for each offspring and then introduced de novo mutations . Then , we aligned the new reads and called SNPs using the pipeline described in Figure 1 . Finally , we compared the SNP calls with expected genotypes based on the mutated sites and calculated the false negative rate . The generation length of populations with overlapping generations is equal to the mean age of parents ( Hill , 1979 ) . Following Miller and Kapuscinski ( 1997 ) , this was approximated as the mean age of spawners ( age-specific number of fish multiplied by the age-specific proportion of reproductive fish ) weighted by age-specific mean weights . In our analyses we used age-specific weights as proxy for age-specific fecundity , since in Atlantic herring weights and fecundity are strongly and nearly linearly correlated ( Arula et al . , 2012; Oskarsson and Taggart , 2006 ) . We estimated the generation time for the herring stocks with data starting shortly after the end of the World War II , a period characterized by still low commercial exploitation which started to increase after the early 1960s . The stocks were the North Sea/Skagerrak/Kattegat/English Channel , the Celtic Sea , the West of Scotland/West of Ireland , the Irish Sea and the Norwegian spring spawning herring . Data on age-specific abundance , maturity and mean weight were extracted from stock assessment reports ( ICES , 2015 , ICES , 2016 ) . The generation time was very similar for almost all the stocks in the first available period after the World War II , characterized by low exploitation , i . e . in 1947-1965 . During this period , the generation time declined between ~6 years in late 1940s ( corresponding to the lowest exploitation ) and ~5 years in 1965 , decreasing further in successive years . No data were available for the period before 1947 when the generation time was likely to have been higher . The Norwegian spring spawning herring showed a higher generation time than the other stocks , oscillating around 10 years in the 1950s . We therefore consider the generation time of 6 years as a minimum estimate for Atlantic herring under no or moderate exploitation .	Evolution by natural selection favours the survival of individuals that are well suited to their environment . This process depends on genetic differences between individuals that make some more able to survive than others . These genetic differences are the result of mutations in DNA of germ-line cells , that is , the cells that produce egg cells and sperm . These mutations mean that new offspring always have a few small differences in some of the genes they inherited from each of their parents . DNA contains strings of molecules known as bases . These act as individual “letters” in the genetic code of an individual . Rapid sequencing of DNA to find out the order of these bases makes it possible to study the rate of mutations within a species . This provides a way to measure how different an individual is from its parents and , by extension , the potential of the species to diversify and adapt to different environments . There are over a trillion Atlantic herring in the Atlantic Ocean , so this fish is an ideal model to study the effects of germ-line mutations on genetic diversity . In 2016 , a group of researchers reported that there is relatively little genetic diversity across Atlantic herring . Given the large population , this suggested that the mutation rate in this species may be low . Feng , Pettersson , Lamichhaney et al . – who were also involved with the earlier work – sequenced the DNA of two families of Atlantic herring raised in captivity to calculate the rate of germ-line mutations in this species . The results showed that , on average , two changes occur per one billion letters in the genetic code in each generation . That is one to two new mutations per egg cell or sperm . This is the lowest mutation rate yet recorded in any animal with a backbone and is around six times lower than the mutation rate in humans . Whilst the low mutation rate in Atlantic herring means there are few differences between individual fish , the extremely large number of these fish on the planet still means that there is enough diversity across the population to allow the species to adapt to changing conditions . This work is important for conservation as it highlights the great variation in potential genetic diversity across species . Future work will need to examine why the mutation rate in Atlantic herring is so low and compare it more widely to mutation rates in other species .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology" ]	2017	Moderate nucleotide diversity in the Atlantic herring is associated with a low mutation rate
Corrective responses to limb disturbances are surprisingly complex , but the neural basis of these goal-directed responses is poorly understood . Here we show that somatosensory feedback is transmitted to many sensory and motor cortical regions within 25 ms of a mechanical disturbance applied to the monkey’s arm . When limb feedback was salient to an ongoing motor action ( task engagement ) , neurons in parietal area 5 immediately ( ~25 ms ) increased their response to limb disturbances , whereas neurons in other regions did not alter their response until 15 to 40 ms later . In contrast , initiation of a motor action elicited by a limb disturbance ( target selection ) altered neural responses in primary motor cortex ~65 ms after the limb disturbance , and then in dorsal premotor cortex , with no effect in parietal regions until 150 ms post-perturbation . Our findings highlight broad parietofrontal circuits that provide the neural substrate for goal-directed corrections , an essential aspect of highly skilled motor behaviors . The motor system is capable of performing a wide range of skilled motor behaviors , from pouring tea in a cup to catching a ball while running . Optimal feedback control ( OFC ) has become an influential theory for interpreting voluntary motor control ( Todorov and Jordan , 2002; Scott , 2004; 2012 ) . OFC includes state estimation ( i . e . position and velocity of the body segments ) based on sensory and internally generated feedback . It also includes a control policy that uses these state estimates to generate motor commands to muscles that generate movement to attain a behavioral goal . Importantly , feedback gains within the control policy are selected based on the behavioral goal . As feedback is an essential component of OFC , an important approach to probe the properties of the motor system is to use small mechanical loads applied to the limb to disturb the motor system and observe how it responds to attain different behavioral goals ( Scott , 2004 ) . This approach has been used to show how muscle responses are modified in under 100ms due to behavioral factors such as task instruction ( Hammond et al . , 1956; Rothwell et al . , 1980; Pruszynski et al . , 2008; Shemmell et al . , 2009 ) , the properties of the spatial target and selection of alternate goals ( Yang et al . , 2011; Nashed et al . , 2012; Selen et al . , 2012 ) , location of obstacles in the environment ( Nashed et al . , 2012; 2014 ) , mechanical properties of the limb and environment ( Kurtzer et al . , 2008; Shemmell et al . , 2010; Cluff and Scott , 2013; Weiler et al . , 2015 ) , and timing constraints or task urgency ( Crevecoeur et al . , 2013; Omrani et al . , 2013; Cluff and Scott , 2015 ) . The neural basis of this task-dependent feedback processing is , however , largely unexplored . Frontoparietal circuits are known to play an important role in voluntary control , but the focus for the last 30 years has been on motor planning and the initiation of motor actions ( Kalaska et al . , 1997; Andersen and Cui , 2009 ) . While the mathematical details of OFC are not implemented at the neural level , the OFC framework creates an important dichotomy between state estimation and the control policy . From this perspective , perturbation-related activity in a brain region that is not influenced by the behavioral goal would be associated with the former , whereas perturbation-related activity that is modified by the behavioral goal , would be associated with the latter . There are sporadic observations that several sensory and motor cortical areas respond to mechanical loads applied to the limb ( Tanji et al . , 1980; Chapman et al . , 1984; Crammond et al . , 1986; Boudreau et al . , 2001; Weber and He , 2004; London and Miller , 2012; Spieser et al . , 2013 ) . However , task-dependent changes in neural activity have only been examined in primary sensory ( S1 ) and motor cortices ( M1 ) ( Evarts and Tanji , 1974; Wolpaw , 1980; Omrani et al . , 2014; Pruszynski et al . , 2014 ) . Thus , the extent to which other cortical regions are involved in generating these task-dependent feedback responses ( i . e . part of the control policy ) remains largely unexplored . Here we use mechanical disturbances applied to the arm of non-human primates ( NHPs ) under multiple behavioral conditions to reveal task-dependent feedback processing across frontoparietal circuits . Our first experiment examined the timing and magnitude of neural responses in a number of frontal and parietal cortical regions elicited by mechanical loads applied to the forelimb as monkeys maintained their hands at a central target ( Posture Task , Figure 1A , Herter et al . , 2007 ) . The activity of 611 neurons was recorded in 5 different cortical regions associated with voluntary motor control ( A5:posterior parietal area 5 , A2:primary somatosensory area 2 , S1:primary somatosensory area 1 & 3 , M1:primary motor cortex and PMd:dorsal premotor cortex , See Materials and methods ) . We found many neurons in each cortical region displayed significant perturbation-related activity within 100 ms of loads being applied in their preferred torque direction ( Figure 1D , refer to Table 1 for details on the number of neurons recorded in each area and neurons responsive to perturbations ) . 10 . 7554/eLife . 13141 . 003Figure 1 . Behavioral tasks and perturbation responses across cortical areas . Each task varied in how the monkeys were instructed to respond to perturbations applied to the limb: correct for the disturbance in the Posture Task ( A ) , not required to respond in the Movie Task ( B ) , move to a spatial target in the IN/OUT Task ( C ) . ( D ) Perturbation response , in the cell’s preferred torque direction , in the Posture Task of a neuron in A5 and in M1 . Tick marks denote action potentials , each row representing a separate trial . ( E ) Population responses in each cortical area ( Mean±2SEM ) . Arrows depict when population signal surpassed threshold ( baseline + 3SD ) for >20 ms . Pre-perturbation baseline activity ( -100:0 ms pre-perturbation , horizontal insets ) in each area is stated below each corresponding population response . Scale bars denote 20 sp/s . Population activity between 60–250 ms ( denoted by thick horizontal line ) is compressed for visualization purposes . The 'American Pie' picture is reproduced with permission from Universal Studios . DOI: http://dx . doi . org/10 . 7554/eLife . 13141 . 003© 1999 , Universal Pictures , All Rights Reserved1999Universal PicturesFigure 1 part B photo ( from the film American Pie ) is reproduced with permission from Universal Studios . 10 . 7554/eLife . 13141 . 004Figure 1—figure supplement 1 . Locations of recorded neurons from each session in Monkey P . ( A5:cyan , S1:green , M1:navy blue , PMd:red ) and the locations of the inserted pins ( black dots ) . We applied up to 0 . 5 mm random jitter on each data point for visualization purposes to avoid multiple sessions obscuring each other . DOI: http://dx . doi . org/10 . 7554/eLife . 13141 . 00410 . 7554/eLife . 13141 . 005Table 1 . Number of neurons in each area recorded in each monkey and neurons with significant change in activity ( two-sample t-test , p<0 . 05 ) in response to perturbation or across tasks . DOI: http://dx . doi . org/10 . 7554/eLife . 13141 . 005A5A2S1M1PMd#cells in Monkey P7004014750#cells in Monkey X000900#cells in Monkey A555503569#cells with significant perturbation response ( /recorded neurons ) in the Posture Task87/12547/5533/40213/27264/119#cells with significant perturbation response recorded in both the Posture & Movie Tasks65/8736/4321/26129/16054/100#cells with significant change in activity between Posture & Movie Tasks ( /significant perturbation response ) 22/654/366/2174/12917/54#cells with significant reduction in activity in Movie Task ( /significant task effect ) 20/223/45/665/7415/17#cells with significant perturbation response ( /recorded neurons ) in the IN-OUT Task35/4121/2518/1883/9258/87#cells with significant change in activity in the IN/OUT task ( /significant perturbation response ) 9/350/212/1831/8316/58#cells with significant increase in activity in OUT target ( /significant task effect ) 4/9NA1/224/317/16 Although neurons in each cortical region responded to the applied perturbations , the magnitude of the perturbation-related responses varied across cortical regions ( one way ANOVA , p<0 . 0001 , F ( df=4 , error=439 ) =10 . 62 , See Table 2 for details ) . A post-hoc analysis revealed that the response in S1 was significantly larger than responses in all other areas ( p<0 . 001 , Tukey's least significant difference ( LSD ) procedure ) . M1 activity was also significantly larger than that of A5 and PMd ( p=0 . 002 and 0 . 0009 respectively ) . The perturbation response was not significantly different across other areas ( p>0 . 07 in all other comparisons ) . 10 . 7554/eLife . 13141 . 006Table 2 . Neural activity [mean ± SD] in each area , 50-100 ms post perturbation minus baseline , in response to the perturbation and across tasks . sp/s: spikes per second . DOI: http://dx . doi . org/10 . 7554/eLife . 13141 . 006A5A2S1M1PMdPerturbation response30 . 3 ± 27 . 4 sp/s38 . 7 ± 26 . 1 sp/s68 . 5 ± 50 . 1 sp/s43 . 3 ± 27 sp/s26 . 9 ± 31 . 6 sp/sDifferential activity across areas ( Movie ) 11 . 2 ± 12 . 8 sp/s5 . 6 ± 16 . 9 sp/s8 . 2 ± 20 . 8 sp/s14 . 4 ± 19 sp/s8 . 3 ± 18 . 3 sp/sChange ratio across areas ( Movie ) 36%15%9%36%40%Differential activity across areas ( IN/OUT ) -0 . 3 ± 6 . 3 sp/s-2 . 2 ± 10 . 2 sp/s-1 . 1 ± 6 sp/s7 . 4 ± 16 . 5 sp/s3 . 2 ± 14 . 2 sp/s We explored how trial-by-trial changes in perturbation-related activity correlated with variations in the timing of the corrective response . Neural activity ( 50 to 100 ms post-perturbation ) for each trial in the neuron’s preferred torque direction was compared to the time of joint reversal ( max distance from start position in joint space before returning to central target ) . We found the activity of 22/213 M1 neurons correlated with kinematic changes in the postural perturbation task ( p<0 . 05 threshold ) . This low number partially reflects our study design in which only a small number of trials were collected per load condition ( n=10 ) . Across the M1 population , the median correlation was -0 . 11 ( p=0 . 002 in comparison with 0 using a Wilcoxon signed rank test ) . This low , although significant correlation is not surprising given that correlations between proximal arm muscle activity and motor corrections during similar time epochs are in the -0 . 2 or -0 . 3 range ( Crevecoeur et al . , 2013 ) . Perturbation-related activity in the other cortical regions was not significantly correlated with motor corrections although the data samples were smaller than that for M1 ( median correlation coefficient , A5: -0 . 014 , A2: -0 . 001 , S1: -0 . 11 , PMd: 0 . 03 , p>0 . 35 in all areas in comparison with 0 using a Wilcoxon signed rank test ) . Figure 1E displays population signals of the cells with a significant perturbation response for each cortical region . Perturbation onset time for each cortical area was calculated based on the time point when the population signal crossed 3SD above baseline activity ( and remained above for at least 20 ms ) , highlighting that perturbation-related activity arrived quickly in all cortical regions ( onset time , A5:21 ms , A2:18 ms , S1:17 ms , M1:22 ms , PMd:25 ms ) . Comparisons of perturbation-related activity to baseline activity for calculating the onset time is not sensitive to sample size , but assumes variance remains constant throughout time ( see Materials and methods ) . As an alternate method , we used a one-sample running t-test to compare when the population signal was different from baseline activity ( 1 ms steps ) . We identified the first point in time that the evoked population activity became significantly different from 0 , and remained significant for at least 20 ms ( Figure 1E , the first point in time the lower edge of the shaded area depicting 2SE rises above the baseline level ) . Response onset times , using this technique were generally similar to those observed using the 3SD technique ( A5:25 ms , A2:25 ms , S1:19 ms , M1:24 ms , PMd:34 ms ) . Thus , sensory feedback from the limb is rapidly transmitted throughout sensory and motor cortical regions . We used a bootstrap technique to identify the likely order of onset times for perturbation responses and task-dependency across cortical areas . To do so , we resampled ( with replacement ) cells in each population and then calculated the response onset and rank-order of each cortical area . We performed this procedure 10 , 000 times , and calculated the proportion of times each cortical area assumed a rank across the resampled data . The left panel in Figure 2A shows the percentage activity onset times in each area were ranked from 1st to 5th . This analysis revealed that primary somatosensory cortex was the first to respond ( with S1 and A2 ranking 1st 95% of the time-S1:77% & A2:18% ) , premotor cortex was the last to respond ( ranking last , 92% ) , and A5 and M1 responded in between these extremes . 10 . 7554/eLife . 13141 . 007Figure 2 . Response onset across tasks . ( A ) Different areas were ranked based on their activity/task onset times across 10000 random iterations of data in each area . Proportion of times each area assumed a rank is plotted for each task . ( B ) Absolute and ( C ) cumulative distributions of change in activity across targets ( in the IN/OUT task ) in each area . DOI: http://dx . doi . org/10 . 7554/eLife . 13141 . 007 We next examined whether perturbation responses were altered when the animal was not engaged in a limb motor task . Perturbation responses during the Posture Task were compared with those observed when the monkey was sitting quietly watching a movie and not required to respond to the perturbation ( Movie Task , Figure 1B , Omrani et al . , 2014 ) . In the Movie Task , the monkey watched a movie displayed on the virtual reality display as the robot moved the monkey’s unseen hand to the central target . At a random time point , a perturbation was applied to the limb ( same load conditions as used in the Posture Task ) . Hand and joint motions were similar across tasks for the first 100 ms . Importantly , corrective movements , and corresponding long-latency muscle stretch responses in the Movie Task , were greatly diminished as compared to the Posture Task ( average muscle response to the perturbation started 34 ms post-perturbation and differentiated across tasks 45 ms post-perturbation , Omrani et al . , 2014 ) . Activity of 416 neurons was recorded in the Posture and Movie Tasks , with 305 displaying significant perturbation responses in the Posture Task ( refer to Table 1 for details on the number of neurons responsive to perturbation in each area ) . Perturbation-related activity was commonly modulated between the Posture and Movie Tasks ( Figure 3A , refer to Table 1 for details on the number of neurons ) . In most cases , the response was smaller in the Movie Task ( refer to Table 1 for details on the number of neurons ) . 10 . 7554/eLife . 13141 . 008Figure 3 . Perturbation responses compared across the Posture and Movie Tasks . ( A ) Perturbation response across the Posture ( green ) and Movie ( red ) tasks ( same neurons as in Figure 1D ) . ( B ) Population signal and ( C ) differential signal across tasks ( Posture - Movie ) in each cortical area . Each scale bar represents 20 sp/s . DOI: http://dx . doi . org/10 . 7554/eLife . 13141 . 008 The magnitude of the perturbation response was significantly different across tasks and areas ( mixed ANOVA with task as within-subject and area as between-subject variables , p<0 . 0001 , F ( 1 , 300 ) =62 . 7 for the task effect and p=0 . 04 , F ( 4 , 300 ) =2 . 5 for the interaction of task and area ) . The magnitude of the reduction was significant in A5 , M1 and PMd ( p<0 . 01 , one sample t-test ) , and marginally not significant in S1 and A2 ( p=0 . 086 & 0 . 052 respectively ) . In order to evaluate the task effect directly , we compared the differential signal across tasks using a One-way ANOVA ( with area as fixed variable and cells as random variables , p=0 . 04 , F ( 4 , 300 ) =2 . 5 ) . Decreases in activity across tasks were only significant in A5 , M1 and PMd ( one sample t-test , p<0 . 0001 in A5 and M1 , p=0 . 0016 in PMd and p=0 . 052 & 0 . 086 in A2 and S1 respectively , refer to Table 2 for details on the differential activity in each area across tasks ) . A post-hoc analysis ( LSD ) revealed that M1 differential response was significantly bigger than the differential response in A2 ( p=0 . 009 ) and PMd ( p=0 . 03 ) . No other significant differences were found among other areas ( p>0 . 13 in all other comparisons ) . We also quantified the relative change in the perturbation response in each area ( i . e . evoked response in the Posture minus that in the Movie tasks , divided by the evoked response in the Posture Task , refer to Table 2 for details on the relative change in each area across tasks ) and found it to be significantly different across areas ( One way ANOVA with area as fixed variable and cells as random variables , p=0 . 019 , F ( 4 , 300 ) =3 ) . A post-hoc analysis ( LSD ) revealed that the relative changes in the perturbation response were significantly smaller in S1 and A2 than the change ratio in other areas ( relative to A2 , A5:p=0 . 036 , M1:0 . 024 and PMd:0 . 016 , and relative to S1 , A5: p=0 . 025 , M1:0 . 019 , PMd:0 . 012 ) . The change ratio was not significantly different between A5 , M1 and PMd ( relative to A5 , M1:p=0 . 95 , PMd:0 . 65 , also p=0 . 57 in M1-PMd comparison ) or between A2 and S1 ( p=0 . 64 ) . Of particular importance is that the timing of the change in the perturbation response varied across the cortical areas . The population signal for A5 was reduced in the Movie as compared to the Posture Task at 23 ms , effectively at the same time as the onset of the initial perturbation response in this cortical region ( Figure 3B , C , differential signal >3SD of baseline ) . In contrast , other cortical areas only displayed a significant reduction in the population signal at ~40 ms or later after the applied load ( M1: 38 ms , PMd: 42 ms , S1: 40 ms and A2: 70 ms ) . The running t-test was also used to identify when the population activity was different between tasks . In the Posture-Movie tasks comparison , we identified the first point in time that the differential signal ( activity in the Posture Task – activity in the Movie Task ) became significantly different from 0 , and remained significant for at least 20 ms ( Figure 3C , the first point in time the lower edge of the shaded area depicting 2SE rises above baseline ) . Response differentiation times were 25 ms , 45 ms and 52 ms for A5 , M1 and PMd , respectively . As shown in Figure 3C , the 2SE shaded area never rises ( and stays ) above zero in S1 and A2 , hence no response differentiation time was detected for these areas . The failure to identify these onset times likely reflects the influence of sample size on the running t-test . Finally , our bootstrap analysis identified the most likely order of response differentiation to be A5 ( ranking 1st 89% , all other regions each less than 5% , middle panel Figure 2A ) , then M1 , PMd , S1 and last A2 ( ranking last 73% ) . Somatosensory feedback also permits rapid transition from one motor task to another ( Hammond et al . , 1956; Rothwell et al . , 1980; Johansson and Flanagan , 2009 ) . Our last experiment quantified how perturbation responses were altered across sensory and motor cortices when the load instructed the monkey to move to a second spatial target . Loads in this target selection task either pushed the hand into the spatial target ( IN ) or away from it ( OUT ) , eliciting a larger corrective response in the latter condition ( Figure 1C , Pruszynski et al . , 2014 ) . Note that in this condition , the monkey should always be engaged in the task but produce different magnitudes of response for each target . Differences in muscle responses between the IN and OUT targets begin ~85 ms after the applied load ( See Materials and methods ) . The onset of this differential response is slower than our previous study ( ~70 ms , Pruszynski et al . , 2014 ) , which likely reflects that that study used a background load to prime the muscles , whereas the present study did not . Activity of 263 neurons was recorded in the IN and OUT tasks; with 215 displaying significant perturbation responses in the OUT target Task ( refer to Table 1 for details on the number of neurons responsive to perturbation in each area ) . Perturbation responses in M1 and PMd were commonly altered between the IN and OUT targets ( Figure 4 , refer to Table 1 for details on the number of neurons ) . Altered responses were also observed in A5 , but rarely in somatosensory cortex . Significant increases in activity in the OUT target were predominantly observed in M1 , but not in A5 and PMd ( refer to Table 1 for details on the number of neurons ) . 10 . 7554/eLife . 13141 . 009Figure 4 . Perturbation responses compared across different target positions . ( A ) Perturbation response for the OUT ( navy blue ) versus IN ( cyan ) targets ( same neurons as in Figure 1D ) . ( B ) Population signal and ( C ) average differential signal across tasks ( OUT-IN ) in each cortical area . Each scale bar represents 20 sp/s . DOI: http://dx . doi . org/10 . 7554/eLife . 13141 . 009 The perturbation magnitude was not significantly different across targets but was significantly different across areas ( mixed ANOVA with target as within-subject and area as between-subject variables , p=0 . 2 , F ( 1 , 211 ) =1 . 64 for the target effect and p=0 . 005 , F ( 4 , 211 ) =3 . 85 for the interaction of target and area ) . We also directly compared the differential signal across targets ( OUT-IN 50–100 ms post perturbation ) using a one-way ANOVA ( with area as the fixed variable and cells as random variables , p=0 . 005 , F ( 4 , 211 ) =3 . 85 ) . M1 was the only area , which had a significant difference in its activity across the two targets in the 50–100 ms period ( one sample t-test , p=0 . 0001 in M1 , p=0 . 08 in PMd and p>0 . 3 in A5 , A2 and S1 , refer to Table 2 for details on the differential activity in each area across tasks ) . A post-hoc analysis ( LSD ) revealed that M1 differential response between IN and OUT targets was significantly larger than in A5 ( p=0 . 005 ) , A2 ( p=0 . 004 ) and S1 ( p=0 . 017 ) but not different than that of PMd ( p=0 . 08 ) . The differential activity was not significantly different across other areas ( p>0 . 1 in all other comparisons ) . We examined whether the absence of any significant target effect in A5 and PMd was due to the fact that half the cells were increasing and half were decreasing their activity between IN and OUT targets ( see Pruszynski et al . , 2014 ) . We therefore compared the absolute change in activity across targets ( Figure 2B , absolute activity change in A5:4 . 6 sp/s ± 4 . 2 sp/s , A2:5 . 4 sp/s ± 8 . 9 sp/s , S1:4 . 2 sp/s ± 4 . 3 sp/s , M1:12 . 8 sp/s ± 12 . 8 sp/s , PMd:10 sp/s ± 10 . 5 sp/s ) . The absolute change in magnitude was significantly different across areas ( One way ANOVA , p<0 . 001 , F ( 4 , 211 ) =5 . 92 ) . A post-hoc analysis ( LSD ) revealed that the absolute change in activity in M1 was significantly larger than in A5 ( p<0 . 001 ) , A2 ( p=0 . 004 ) and S1 ( p=0 . 002 ) but not different than in PMd ( p=0 . 13 ) . The absolute change in activity was also bigger in PMd compared to the absolute change in A5 ( p=0 . 016 ) and S1 ( p=0 . 04 ) . We also examined the cumulative distributions of change in activity for each area ( Figure 2C ) . A two-sample Kolmogorov-Smirnov test revealed that the distribution was significantly different in M1 compared to A5 ( p<0 . 001 ) , A2 ( p=0 . 001 ) , S1 ( p=0 . 004 ) , and in PMd compared to A5 ( p=0 . 05 ) . All other comparisons were not significant ( p>0 . 1 across all areas ) . The timing of the change in the perturbation-response for the IN and OUT targets also varied across cortical areas . Differences in the population signals for the IN and OUT targets were first observed in M1 , 66 ms after perturbation onset , and then in PMd at 98 ms ( Figure 4B , C , differential signal >3SD of baseline ) . In contrast , population signals in S1 and A5 did not show any difference across targets until ~150 ms post-perturbation , and the differential signal in A2 never passed threshold in this task . For the IN-OUT target comparison , the baseline activity was already significantly different in M1 & PMd ( M1:2 . 4 sp/s ± 7 . 2 sp/s , PMd:3 . 6 sp/s ± 12 . 5 sp/s , one sample t-test , p=0 . 009 & 0 . 029 in M1 & PMd respectively ) . Thus , in using the running t-test technique , we compared the differential activity relative to its baseline difference rather than 0 . We identified the first point in time the differential activity was significantly different from baseline , and remained significant for at least 20 ms ( Figure 4C , the first point in time the lower edge of the shaded area depicting 2SE rises above the baseline line ) . With this technique , response differentiation times were 72 ms , 180 ms and 160 ms for M1 , PMd and A5 respectively . As observed ( Figure 4C ) , the 2SE shaded area never rises ( and stays ) above the zero line in S1 and A2 , and hence no response differentiation time was detected for these areas . The fact that neurons could increase or decrease activity between the IN versus OUT target could impact the onset time for observing differences in the population signals associated with each target ( See also Heming et al . , 2016 ) . To rule this out , we reversed the differential sign for cells that significantly decreased their activity in the OUT versus IN target when calculating the difference in the population signals ( this means we used IN-OUT in these cells rather than OUT-IN ) . Differences in the population signals tended to be slightly earlier ( 3SD technique , M1:58 ms , PMd:78 ms , A5:157 ms , S1:173 ms and no differentiation time for A2 ) , but the order of onset times remained the same with only M1 and PMd showing significant onset times before 100 ms . Finally , the bootstrap technique identified the most likely order of response differentiation to be M1 first ( 86% , all others regions each less than 5% , except for A2 which was 8% , right panel in Figure 2A ) , PMd second ( 58% ) , followed by S1 and A5 ( similar ratio of 35% ) and finally A2 ( 55% ) . Figure 5 provides an overview of the main results on how perturbation-related activity is transmitted across sensory and motor cortical regions , and how this spatiotemporal pattern of activity is altered by behavioral context . The top panel highlights perturbation responses when the monkey is not rewarded for responding to the mechanical load , termed the 'default response' . In this case , limb feedback is rapidly transmitted across the cortex with the greatest and earliest activity in S1 quickly followed by responses in adjacent cortical regions . The middle panel displays how the perturbation response changes between the Movie and Posture Tasks , termed 'task engagement' . In this case , the perturbation response increases immediately in A5 and then in motor cortical regions with minimal effect in primary somatosensory cortex . Finally , the bottom panel displays how perturbation responses are altered when they cue the generation of a goal-directed movement , termed 'target selection' . In this case , perturbation-responses first increase in M1 , and then PMd , with minimal effect in sensory areas until 150 ms . 10 . 7554/eLife . 13141 . 010Figure 5 . Context-dependent patterns across sensorimotor cortex -Default response ( top panel ) , is represented by activity patterns in the Movie Task . Task engagement ( middle panel ) , is represented by the differential signal between the Movie and Posture Tasks . Target selection ( bottom panel ) , is represented by differential signal between the OUT and IN targets . Activity is plotted using a color map . In the default response , population response is capped at 70 sp/s . Differential signals are normalized to their maximum response in the Posture Task ( au ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13141 . 010 The neural basis of feedback processing for voluntary motor actions has not received much attention for the last 30 years . At that time , the prevailing view was that task-dependent changes in long-latency motor responses , that occur 50 to 100 ms after a mechanical disturbance , were provided by a transcortical feedback pathway from primary somatosensory to primary motor cortex ( Desmedt , 1977; Brooks , 1986 ) . The timing of mechanical responses in M1 are sufficiently fast to generate these long-latency responses . Classic work by Evarts and Tanji illustrated that perturbation responses in M1 could be modulated when the disturbance was the cue for the monkey to push or pull a lever ( Evarts and Tanji , 1974 ) . Further studies implicated the dentate nucleus in this task-dependent feedback processing ( Meyer-Lohmann et al . , 1975; Vilis et al . , 1976; Strick , 1983; Hore and Vilis , 1984 ) , likely through its connections to M1 ( Orioli and Strick , 1989; Dum and Strick , 2003 ) . Based on this evidence , it has been generally assumed that there was minimal overlap in cortical circuits involved in online feedback processing from the limb ( S1 and M1 ) , and the broad parietal and frontal cortical circuits involved in motor planning and initiation . The present study provides the first examination of the relative timing and magnitude of perturbation-related activity across sensory and motor cortices in non-human primates . Here we show that even when the animal is not engaged in a limb motor task , limb afferent feedback is still rapidly transmitted to many sensory and motor cortical regions , beginning in S1 and then spreading to adjacent cortical regions ( Figure 5 ) . The magnitude of the perturbation response is also greatest in S1 and diminishes spatially across the cortex . This gradient in the timing and magnitude of perturbation-related activity across the cortical surface may be generated by intra-cortical communication , although subcortical sensory feedback pathways may also be involved ( Cappe et al . , 2007; Padberg et al . , 2009 ) . While muscle spindle afferents are assumed to play a dominant role for eliciting the perturbation-related activity observed in the present study ( Lucier et al . , 1975 ) , cutaneous afferents likely also contribute , particularly in S1 , signaling skin stretch caused by the perturbation . However , the very fastest responses are likely due to muscle afferents even in area 3b ( Heath et al . , 1976 ) . Neurons in S1 with cutaneous receptive fields are broadly tuned to the direction of movement during reaching ( Cohen et al . , 1994; Prud’homme et al . , 1994 ) , much like neurons in M1 with receptive fields from shoulder and elbow muscles ( Scott and Kalaska , 1997 ) . Correspondingly , we expect both muscle and cutaneous afferents will be broadly tuned to loads applied to the shoulder and elbow , and contribute to feedback responses observed in the present study . Yet the differential physiological contributions of each of these feedback sources to control are an interesting question , warranting further investigation . M1 was the first cortical region to display changes in perturbation-related activity when the applied load was used to initiate movements to the OUT versus the IN target , and occurred prior to changes in proximal limb muscle activity . This suggests M1 is the primary cortical source for implementing this aspect of control ( along with dentate ) . While changes in perturbation-related activity was only observed in PMd at the same time or later than that observed in limb muscles , this cortical area did show changes in baseline activity for the IN and OUT targets before perturbation onset . One can also see a small ( non-significant ) dip in the differential signal in Figure 4C at about 40 ms post-perturbation . Thus , PMd may play some role in this process of using sensory feedback for action selection . In contrast , all parietal regions examined in the present study displayed no change in perturbation-related activity until ~150 ms when differences in limb motion begin to emerge ( Pruszynski et al . , 2014 ) , suggesting none of these regions were involved in this aspect of control . However , a more thorough examination of area 3a , that has direct projections onto motoneurons of limb muscles , is warranted ( Rathelot and Strick , 2006 ) . Of particular note is that engagement of the limb in a motor task increases the perturbation response immediately in A5 ( ~25 ms ) . In contrast , the initial response was unaltered in other cortical regions , but they then increased 15 ms later or more , potentially driven by A5 . Parietal area 5 has been implicated in somatomotor and visuomotor processing for limb motor actions ( Mountcastle et al . , 1975; Chapman et al . , 1984; Kalaska , 1996; Debowy et al . , 2001; Reichenbach et al . , 2014 ) . Preparatory activity in this cortical region reflects the selective use of the limb for a subsequent motor action ( Cui and Andersen , 2011 ) . Our data suggests a potential role of A5 in online control , consistent with recent studies demonstrating TMS over medial intraparietal sulcus in humans ( approximate analog to A5 in NHPs ) prolongs corrective responses to mechanical disturbances during reaching ( Reichenbach et al . , 2014 ) . Thus , posterior parietal cortex is not only important for online control of visual feedback ( Desmurget and Grafton , 2000 ) , but also somatosensory feedback . The fact that perturbation-related activity was altered between the movie and postural tasks initially suggests a role in the control policy rather than state estimation . However , changes in perturbation-related activity in A5 were related to whether the animal was engaged or not in a behavioral task . There was no change in its response when using sensory feedback to select a new goal , as the animal remained engaged in a task before and after this selection . Thus , it may instead play a role in linking state estimation to the control policy for online control . The present study explored how task engagement and target selection alters the spatiotemporal pattern of perturbation-related activity across sensory and motor cortices . Our results suggest that multiple areas are activated in response to sensory feedback , and activity in each area reflects processing of different aspects of the task . This concurrent processing of information across different areas could cumulatively shape the observed output generated in each task ( Ledberg et al . , 2007; Cisek and Kalaska , 2010; Siegel et al . , 2015 ) . Task-dependent changes in perturbation-related activity in cortex preceded corresponding changes in muscle responses , and thus , these cortical circuits are the likely source for these task-dependent changes in motor output . However , long-latency motor responses are extremely complex ( see Shemmell et al . , 2010; Pruszynski and Scott , 2012 for reviews ) and consider many factors such as limb mechanics ( Kurtzer et al . , 2008; 2009 ) , goal-directed corrections associated with the shape of spatial goal ( Nashed et al . , 2012 ) , presence of obstacles in the environment , and selection of alternate goals ( Nashed et al . , 2014 ) . We predict that distributed frontoparietal circuits ( and cerebellum ) also provide the neural substrate to generate these other complex corrective responses , a hallmark of highly skilled motor behaviors ( Scott , 2012 ) . Three male non-human primates ( Macaca mulatta , 10–17 kg ) were trained to perform whole limb visuomotor tasks while attached to an exoskeleton robot ( KINARM , BKIN Technologies , Kingston , Ontario , Canada ) . The robot permitted combined flexion and extension movements of the shoulder and elbow in the horizontal plane and applied loads to the shoulder and/or the elbow independently . Two monkeys ( Monkey X & A ) used a right-arm robot and one monkey ( Monkey P ) used a left-arm robot . Targets and hand visual feedback were presented to the monkeys , in the horizontal plane , using an overhead monitor and a semitransparent mirror . Hand position was represented by a white circle ( 5 mm diameter ) positioned at the tip of the index finger . The Queen’s University Animal Care Committee approved all experimental procedures ( Protocol 1348 ) . Throughout the experiment different combinations of shoulder and/or elbow torques were applied to the monkey’s arm . Three tasks were performed and varied in how the monkeys were required to respond to the perturbations . In the Posture Task ( Herter et al . , 2007 ) , the monkey was instructed to maintain its hand at a central target ( visual: 12 mm diameter , acceptable window: 16 mm diameter ) . At a random time ( 1000–1500 ms ) , the limb was perturbed with one of nine combinations of loads applied to the shoulder and/or elbow ( flexor , extensor or null ) , and the monkey had to return its hand to the target within 750 ms of the perturbation time ( Figure 1A ) . Each perturbation lasted 300ms and the size of the load varied with the size of the monkey ( Monkey P & X , 0 . 24 Nm and monkey A , 0 . 32 Nm ) . Load magnitudes were adjusted in the bi-articular load directions to compensate for larger hand motions induced in these directions ( Herter et al . , 2007 ) . Each load combination was presented randomly in a block of trials and the monkey was required to complete 10 blocks in a set . In the Movie Task ( Omrani et al . , 2014 ) , monkeys were not required to do anything in response to the perturbation . Task-related visual feedback ( i . e . target position and hand position ) was replaced by a movie and the monkeys were trained to quietly watch the movie . The robot moved the hand to the central target at the beginning of each trial . The hand was then perturbed using the same 9 load combinations as in the Posture Task ( Figure 1B ) . The monkey was rewarded irrespective of its response to the perturbation . In the IN/OUT task ( Pruszynski et al . , 2014 ) , the monkey started each trial by maintaining its hand at a central target ( 12 mm diameter ) , and moved its hand to a second target ( 2 . 5 cm radius ) following a perturbation ( the load remained on for 1500 ms ) . The perturbation was the load combination from the Posture Task that elicited the largest response in the neuron/muscle presently being recorded . The location of the second target was strategically chosen such that the load either pushed the hand in ( IN ) , or away ( OUT ) from it ( Figure 1C; location of the second target could also remain aligned with the initial central target , but data not analyzed in this study ) . The monkey had to move to the second target within 750 ms and remain there for an additional 1 s . IN and OUT target trials were randomly interleaved and 20 repeat trials were recorded for each target . Normally an experimental session was composed of a fixed order of tasks: first the Posture Task , then the Movie Task , followed by another repeat of the Posture Task and finally the IN/OUT task . A reduced version of the experiment was performed near the end of the recording session; in which one set of the Posture and the Movie Tasks were randomly presented followed by one set of the IN/OUT task . In recording sessions from S1 , receptive field properties of the neurons were investigated following the last experimental block . Neural data was recorded from shoulder/elbow regions of the primary somatosensory areas 1&3 ( S1 ) , primary somatosensory area 2 ( A2 ) , parietal area 5 ( A5 ) , primary motor cortex ( M1 ) and dorsal premotor cortex ( PMd ) , using standard extracellular recording techniques ( Herter et al . , 2007; Omrani et al . , 2014; Pruszynski et al . , 2014 ) . The neural data was initially sorted online for single units ( Plexon Inc . , Dallas ) , then confirmed and examined further offline using the Plexon offline sorter . Recording chamber and penetration sites were chosen using monkey atlas coordinates ( Paxinos et al . , 2008 ) and also MRI imaging ( for Monkeys X & A ) . Single tungsten microelectrodes ( FHC , Bowdoin ) were advanced in the cortex until neural activity was recorded . We verified the location of shoulder and elbow areas of M1 by eliciting muscle twitches using microstimulation ( Stoney et al . , 1968 ) and the response of neurons to passive movement of the joints . The shoulder/elbow regions in other cortical areas were generally at the same laterality as that in M1 . We tested sensory receptive fields of neurons particularly in the first few sessions of recording in a new cortical area to make sure we were in the shoulder/elbow representation . While recording in the sensory cortices , we also tried to differentiate the best modality of stimulus for each neuron ( e . g . cutaneous , soft touch , deep touch or joint movement ) . However , the robotic device attached to the arm made it difficult to dissociate whether sensory responses during manual examination were related to muscle or cutaneous afferents . The majority ( ~85% ) of our recordings from A5 were performed over the convexity of the cortex ( area 5d ) . Neural recordings in S1 were equally on the surface ( putative Area 1 ) , bank of post-central sulcus ( putative area 3b ) and deep in post-central sulcus ( putative area 3a ) . In one monkey ( P ) , we verified our recording areas post-mortem . We used Paraformaldehyde to perfuse the monkey and its brain . Right before removing the chamber from the skull , we inserted several pins to known coordinates within the chamber . We then photographed the brain and sketched the location of the sulci ( Figure 1—figure supplement 1 ) . Post mortem penetration locations have yet to be performed in the other monkeys . We also recorded electromyographic ( EMG ) activity of proximal arm muscles during the tasks . The EMG recordings were scored from 1 to 5 ( based on recording quality , gain of the signal , signal-to-noise ratio , and whether the muscle looked active in the task ) . Muscles that scored 3 and higher were included in our analysis . EMG signals were band-pass filtered ( 10–150 Hz , two-pass , third-order Butterworth ) and full-wave rectified . Each trial was aligned based on the perturbation onset . The EMG data related to posture and movie tasks were presented previously ( Omrani et al . , 2014 ) . For the IN/OUT task , we recorded EMG from 3 to 6 proximal limb muscles in 9 sessions in one monkey ( Monkey P ) . Nineteen samples ( representing all the major muscles involved in flexion and extension of the shoulder and elbow joints ) were identified as good quality ( score 3 or higher on subjective rating scale out of 5 ) and had significant perturbation responses ( p<0 . 05 ) . Spike times were extracted from the Plexon files into Matlab ( Mathworks , Natick ) . Spike-density functions were generated by convolving spike time-stamps with asymmetric double-exponential kernels ( 1 ms rise- and 20 ms fall-time , [Thompson et al . , 1996] ) . We consider cell activity 50–100 ms post-perturbation corresponding roughly to the long-latency epoch for muscle activity . The load combination with the largest response ( 50 to 100 ms post-perturbation ) was then selected as the neuron’s preferred-torque direction ( PTD ) . The neuron’s activity in its PTD was then compared to its activity in the null load condition ( catch trial ) using a two-sample t-test . If the comparison was significant , the activity of the cell in its PTD was used for further analyses ( SPSS , IBM , New York ) . Single cell activities in each area were averaged to calculate the perturbation population response for each area . We determined the first point in time that the activity of a cell/muscle/population passed a defined threshold ( baseline + 3 SD of baseline activity ) and remained above this threshold for at least 20ms ( to avoid capturing random transient responses ) . The baseline period consisted of cell/population activity 100 ms prior to the perturbation ( average population baseline for each area is represented as insets in Figure 1E ) . A similar approach was used to identify when population signals associated with different conditions ( Posture versus Movie Tasks , or IN and OUT targets ) were significantly different ( differential signals ) . We compared the onset times of perturbation responses and differential signals across cortical areas using a bootstrap technique , resampling ( with replacement ) cells in each population 10000 times , and then calculating the response onset for each iteration . We also rank ordered the onset across different areas in each iteration and calculated the percentage of times activity in one area preceded that of others . In calculating the percentages , we also included iterations where the population or difference signal did not pass 3SD .	Humans and other animals can change a movement they are making in a split second , such as when a basketball player has to move around an unexpected opponent to shoot a ball through the hoop . These on-the-fly corrections rely on information about the movement that comes in from the senses . However , it was unclear how the brain and spinal cord process this sensory information to guide movement . Omrani et al . have now recorded electrical activity from the brains of monkeys while the animals tried to keep their hand at a target . Each monkey wore a robotic exoskeleton that would occasionally move the monkey’s arm . Even if the monkey was not engaged in a motor task , a small nudge of the limb by the robot caused neural activity to spread rapidly throughout the sensory and motor regions of the cerebral cortex ( the outer layer of the brain ) . In some trials , when the monkey was actively trying to keep its hand at a target , the robot would again nudge the monkey’s arm slightly . Omrani et al . observed that within 25 milliseconds of this nudge , the activity in an area of the cortex called parietal area 5 responded even more , suggesting that this area was using information from the senses to actively deal with the change in arm position . This change in activity then spread to other parts of the brain . In another set of trials , the monkey was trained to move to a second target if the robot nudged its arm . In this case , the activity in an area called the primary motor cortex increased even more , likely supporting the monkey’s ability to rapidly move to this second target . Overall , the study by Omrani et al . highlights the complex way that sensory feedback is processed in the cerebral cortex , supporting our ability to perform highly skilled motor actions .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology" ]	2016	Distributed task-specific processing of somatosensory feedback for voluntary motor control
Recent findings indicate a high level of specialization at the level of microcircuits and cell populations within brain structures with regards to the control of fear and anxiety . The hippocampus , however , has been treated as a unitary structure in anxiety and fear research despite mounting evidence that different hippocampal subregions have specialized roles in other cognitive domains . Using novel cell-type- and region-specific conditional knockouts of the GABAA receptor α2 subunit , we demonstrate that inhibition of the principal neurons of the dentate gyrus and CA3 via α2-containing GABAA receptors ( α2GABAARs ) is required to suppress anxiety , while the inhibition of CA1 pyramidal neurons is required to suppress fear responses . We further show that the diazepam-modulation of hippocampal theta activity shows certain parallels with our behavioral findings , suggesting a possible mechanism for the observed behavioral effects . Thus , our findings demonstrate a double dissociation in the regulation of anxiety versus fear by hippocampal microcircuitry . Fear and anxiety are distinct emotional states induced by different environmental triggers ( acute , objectively harmful stimuli versus the possibility of unidentifiable , obscure threats , respectively ) and resulting in distinguishable defensive behaviors ( freezing , fight or immediate active avoidance versus alertness and risk-assessment [Tovote et al . , 2015; Davis et al . , 2010] ) . Recent circuit-focused studies demonstrate that the neurocircuitry and neuronal cell populations mediating fear and anxiety show some overlap ( Botta et al . , 2015; Jennings et al . , 2013 ) , but also significant divergence ( Kheirbek et al . , 2013; Yamaguchi et al . , 2013 ) . Here , we investigated whether fear and anxiety converge or diverge at the level of microcircuits within the hippocampus ( HPC ) . The intra-HPC circuitry includes three subregions ( CA1 and CA3 areas , and dentate gyrus ( DG ) ) , with predominantly unidirectional excitatory projections from DG to CA3 to CA1 . Additionally , the principal neurons in each subregion are tightly regulated via activity of GABAA receptors ( GABAARs ) . Out of the five GABAAR subtypes expressed in the HPC , the α2-containing GABAARs ( α2GABAAR ) have been strongly and consistently implicated in anxiety and fear ( Vollenweider et al . , 2011; Löw et al . , 2000; Smith et al . , 2012 ) . α2GABAARs are expressed on principal neurons in all three HPC subregions and mediate fast phasic inhibition ( Fritschy and Mohler , 1995; Hörtnagl et al . , 2013 ) . To address the question of how HPC microcircuits contribute to anxiety versus fear , we generated three gene-targeted mouse lines in which the α2GABAARs are deleted selectively in the pyramidal neurons of the CA1 or of the CA3 , or in the granule cells of the DG . Reducing the expression of synaptic GABAARs later than approximately 2–3 weeks postnatally typically causes no baseline anxiety phenotype ( Earnheart et al . , 2007; Shen et al . , 2012 ) , in line with the concept of a developmental origin of anxiety ( Gross et al . , 2002; Gross and Hen , 2004 ) . The cre driver lines used in the current study express cre recombinase later in development , and our conditional knockout mice display no spontaneous anxiety or fear phenotype . To assess which part of intrahippocampal circuitry is essential for modulation of anxiety and fear , we thus used a mixed genetic-pharmacological approach: We systemically administered a nonselective positive allosteric modulator of GABAA receptors , known to induce anxiolysis and fear reduction , in cell type- and region-specific α2 knockout mice . Our findings demonstrate a double-dissociation within the HPC with regards to fear and anxiety , where inhibition of CA3 and DG is required for anxiolysis , while the inhibition of CA1 is required for the reduction of fear . A floxed Gabra2 allele was generated by placing two loxP sites 1 kb apart flanking exon 5 ( 221 bp ) ( see Witschi et al . , 2011 for details; Figure 1A ) . The α2 conditional knockout mice were generated by crossing mice homozygous for the floxed Gabra2 allele ( α2F/F mice ) with α2F/F mice carrying one of the following three cre recombinase transgenes: CaMKIIα cre ( T29-1 mice; Tsien et al . , 1996 ) to generate a CA1 pyramidal neuron selective knockout ( α2CA1KO ) , Grik4 cre ( G32-4 mice; Nakazawa et al . , 2002 ) to generate a CA3 pyramidal neuron selective knockout ( α2CA3KO ) and POMC cre ( McHugh et al . , 2007 ) to generate a DG granule cell selective knockout ( α2DGKO ) . 10 . 7554/eLife . 14120 . 003Figure 1 . Targeted reduction of α2 expression in CA1 , CA3 or DG . ( A ) Generation of the α2F/F control and α2CA1KO , α2CA3KO , α2DGKO mice . ( B ) Top: False color images showing the α2 staining intensity in immunohistochemically stained sections . Cooler colors = less staining . Bottom: Semi-quantitative comparisons of α2 staining . ( C ) α2 mRNA expression ( see sample ROI’s on the right ) , expressed as% of α2F/F control . p<0 . 05 , p<0 . 01 , *p<0 . 001 compared to corresponding α2F/F group . DOI: http://dx . doi . org/10 . 7554/eLife . 14120 . 00310 . 7554/eLife . 14120 . 004Figure 1—figure supplement 1 . Immunohistochemical localization of GABAAR α2 subunits in conditional knockout mice bred on a 129X1/SvJ background . ( A ) False color images showing α2 staining intensity in immunohistochemically stained sections in α2F/F mice . Cooler colors = less staining . ( B ) Same as A for α2CA1KO , ( C ) Same as A for α2CA3KO , ( D ) Same as A for α2DGKO . DOI: http://dx . doi . org/10 . 7554/eLife . 14120 . 00410 . 7554/eLife . 14120 . 005Figure 1—figure supplement 2 . Expression of α1 and α5 subunits in conditional knockout mice . ( A ) Top: Representative gray-scale images showing the α1 staining intensity in immunohistochemically stained sections from the hippocampus , cortex and the amygdala of α2F/F , α2CA1KO , α2CA3KO and α2DGKO mice . Bottom: Semi-quantitative comparisons of α1 staining intensity between α2F/F , α2CA1KO , α2CA3KO and α2DGKO mice in CA1 , CA3 , DG , cortex and the amygdala . ( B ) Same as A for α5 . DOI: http://dx . doi . org/10 . 7554/eLife . 14120 . 00510 . 7554/eLife . 14120 . 006Figure 1—figure supplement 3 . Expression of α3 and α4 mRNA in conditional knockout mice . ( A ) Gabra3 expression in CA1 , CA3 , DG , cortex and the amygdala of α2F/F , α2CA1KO , α2CA3KO and α2DGKO mice expressed as percentage of α2F/F control . ( B ) Same as A for Gabra4 . p<0 . 05 in comparison to the corresponding α2F/F control group . DOI: http://dx . doi . org/10 . 7554/eLife . 14120 . 00610 . 7554/eLife . 14120 . 007Figure 1—figure supplement 4 . Miniature inhibitory postsynaptic currents . Mean ( ± S . E . M . ) decay time , frequency , amplitude of mIPSCs recorded from ( A ) the CA1 pyramidal neurons in α2CA1KO and α2F/F sections , ( B ) the CA3 pyramidal neurons in α2CA3KO and α2F/F sections , ( C ) the DG granule cells in α2DGKO and α2F/F sections . p<0 . 01 , p<0 . 001 . CTRL=vehicle-treated , DZP=diazepam-treated . DOI: http://dx . doi . org/10 . 7554/eLife . 14120 . 00710 . 7554/eLife . 14120 . 008Figure 1—figure supplement 5 . Tests of hippocampal function . ( A ) Mean ( ± S . E . M . ) percentage of time spent freezing during the tone 24 hr after auditory fear-conditioning using a delay ( black ) or trace ( white ) protocol . p<0 . 05 , *p<0 . 001 , different from the corresponding delay group . No significant differences were found in planned pair-wise comparisons of the genotypes within the trace-conditioned groups or within the delay-conditioned groups . ( B ) Mean ( ± S . E . M . ) percentage of time spent freezing during re-exposure to the context 24 hr after contextual fear-conditioning . p<0 . 001 , different from α2F/F . ( C ) Mean ( ± S . E . M . ) distance to platform during probe trials . p<0 . 05 , **p<0 . 001 , different from α2F/F; gray: α2CA3KO , red: α2DGKO . DOI: http://dx . doi . org/10 . 7554/eLife . 14120 . 008 Within the HPC , the reduction in α2 expression was limited to the targeted regions at both the protein ( Figure 1B; Figure 1—figure supplement 1 ) and mRNA ( Figure 1C ) level for each genotype ( Table 1 , Sect . 1 , 2 ) . The knockdowns extended through the septo-temporal axis of the hippocampus , although they were more pronounced in dorsal regions ( Figure 1B; Figure 1—figure supplement 1 ) . α2 mRNA expression was also reduced in the cortex of α2CA1KO mice , but this reduction was not observed at the protein level at the time point examined ( 10–11 weeks ) . CaMKIIα cre-mediated recombination is progressive over time in cortex ( Fukaya et al . , 2003 ) , and the mRNA expression might be already reduced at this time point and this reduction may start to be reflected at the functional protein level later in development . The expression of other GABAAR subunits in the regions of interest was largely unaffected ( Figure 1 , Figure 1—figure supplement 2 , Figure 1—figure supplement 3; Table 1 ) . 10 . 7554/eLife . 14120 . 009Table 1 . Results of omnibus statistical tests of experiments for the general characterization of α2CA1KO , α2CA3KO and α2DGKO mice . DOI: http://dx . doi . org/10 . 7554/eLife . 14120 . 0091 . ImmunohistochemistryOne-Way ANOVA; Factor: Genotypeα2 Subunitα1 Subunitα5 SubunitCA1F ( 3 , 10 ) =9 . 44P=0 . 003F ( 3 , 9 ) =0 . 08p=0 . 97F ( 3 , 9 ) =0 . 10p=0 . 96CA3F ( 3 , 10 ) =5 . 05P=0 . 02F ( 3 , 9 ) =0 . 01p=0 . 99F ( 3 , 9 ) =0 . 13p=0 . 94DGF ( 3 , 10 ) =9 . 08P=0 . 003F ( 3 , 9 ) =0 . 00p=1 . 00F ( 3 , 9 ) =0 . 06p=0 . 98CortexF ( 3 , 10 ) =1 . 90p=0 . 19F ( 3 , 9 ) =0 . 01p=0 . 83F ( 3 , 9 ) =0 . 01p=0 . 99AmygdalaF ( 3 , 10 ) =1 . 28p=0 . 34F ( 3 , 9 ) =0 . 21p=0 . 89F ( 3 , 9 ) =0 . 36p=0 . 782 . Quantitative PCROne-Way ANOVA; Factor: Genotypeα2 Subunitα3 Subunitα4 SubunitCA1F ( 3 , 16 ) =10 . 66p<0 . 001F ( 3 , 16 ) =0 . 10p=0 . 96F ( 3 , 16 ) =2 . 48p=0 . 1CA3F ( 3 , 16 ) =10 . 53p<0 . 001F ( 3 , 16 ) =0 . 67p=0 . 58F ( 3 , 16 ) =4 . 74p=0 . 02DGF ( 3 , 16 ) =7 . 32p=0 . 003F ( 3 , 12 ) =1 . 96p=0 . 17F ( 3 , 12 ) =2 . 90p=0 . 08CortexF ( 3 , 16 ) =15 . 69p<0 . 001AmygdalaF ( 3 , 16 ) =3 . 01p=0 . 063 . Slice ElectrophysiologyTwo-Way Mixed Factorial ANOVA; Factors: Genotype ( between-subjects ) , Drug ( within-subjects ) CA1AmplitudeFrequencyDecay TimeGenotypeF ( 1 , 38 ) =1 . 83p=0 . 18F ( 1 , 38 ) =2 . 80p=0 . 10F ( 1 , 38 ) =1 . 94p=0 . 17DrugF ( 1 , 38 ) =1 . 49p=0 . 23F ( 1 , 38 ) =9 . 38p=0 . 004F ( 1 , 38 ) =107 . 96p<0 . 001Genotype x Drug InteractionF ( 1 , 38 ) =2 . 91p=0 . 09F ( 1 , 38 ) =3 . 11p=0 . 08F ( 1 , 38 ) =0 . 82p=0 . 37CA3AmplitudeFrequencyDecay TimeGenotypeF ( 1 , 21 ) =2 . 36p=0 . 14F ( 1 , 21 ) =1 . 85p=0 . 19F ( 1 , 21 ) =0 . 95p=0 . 34DrugF ( 1 , 21 ) =0 . 66p=0 . 42F ( 1 , 21 ) =0 . 09p=0 . 77F ( 1 , 21 ) =30 . 54p<0 . 001Genotype x Drug InteractionF ( 1 , 21 ) =0 . 21p=0 . 65F ( 1 , 21 ) =1 . 90p=0 . 18F ( 1 , 21 ) =0 . 25p=0 . 62DGAmplitudeFrequencyDecay TimeGenotypeF ( 1 , 21 ) =1 . 50p=0 . 23F ( 1 , 21 ) =1 . 58p=0 . 22F ( 1 , 21 ) =0 . 01p=0 . 91DrugF ( 1 , 21 ) =2 . 58p=0 . 12F ( 1 , 21 ) =1 . 58p=0 . 22F ( 1 , 21 ) =47 . 42p<0 . 001Genotype x Drug InteractionF ( 1 , 21 ) =0 . 54p=0 . 47F ( 1 , 21 ) =1 . 33p=0 . 26F ( 1 , 21 ) =0 . 30p=0 . 594 . Tests of Hippocampal FunctionDelay – Trace Fear ConditioningTwo-Way Factorial ANOVA; Factors: Genotype ( between-subjects ) , Condition ( between-subjects ) % FreezingGenotypeF ( 3 , 49 ) =0 . 38p=0 . 77ConditionF ( 1 , 49 ) =41 . 57p<0 . 001Genotype x Cond . InteractionF ( 3 , 49 ) =0 . 71p=0 . 55Contextual Fear ConditioningOne-Way ANOVA; Factor: Genotype ( between-subjects ) % FreezingGenotypeF ( 3 , 35 ) =5 . 47p=0 . 003Morris Water MazeTwo-Way Factorial ANOVA; Factors: Genotype ( between-subjects ) , Day ( within-subjects ) Time to platformGenotypeF ( 3 , 80 ) =5 . 17p=0 . 01DayF ( 5 , 80 ) =244 . 12p<0 . 001Genotype x Day InteractionF ( 5 , 80 ) =1 . 36p=0 . 19 The conditional knockouts of the Gabra2 gene did not cause major changes in the frequency , amplitude or decay kinetics of the miniature inhibitory postsynaptic currents ( mIPSCs ) , and did not impair the response to diazepam in any of the genotypes ( Figure 1—figure supplement 4; Table 1 , Sect . 3 ) . Cognitive tests of hippocampal function revealed no gross anomalies in any of the knockouts ( Figure 1—figure supplement 5; Table 1 , Sect . 4 ) . Thus , the knockout of the Gabra2 gene in CA1 , CA3 or DG was limited to the targeted regions and the targeted receptor , and did not cause major impairments in baseline inhibitory synaptic activity or the diazepam-induced changes of inhibitory synaptic responses in principal neurons of the targeted regions in the hippocampus . Anxiety-related behavior was measured using two validated tests of anxiety ( Treit et al . , 2010 ) : Elevated plus maze ( EPM ) and light/dark box ( LDB ) . In EPM , enhancing GABAAR-mediated responses with diazepam increased open arm activity in α2F/F and α2CA1KO , but not in α2CA3KO or α2DGKO mice ( Figure 2A , Figure 2—figure supplement 1A; Table 2 , Sect . 1 ) . While the diazepam effects were clear in α2F/F and α2CA1KO mice , the lack of these effects especially in α2CA3KO mice was partially due to an increase in percent open arm time in the vehicle condition , as well as a decrease in the same measure in the diazepam condition . To clarify these findings , we repeated this test with mice this time bred on a 129X1/SvJ background under slightly different testing conditions ( see Methods; the conditional knockouts bred on the 129X1/SvJ background show similar distribution of GABAAR α2 subunits as the C57BL/6J; Figure 1—figure supplement 1 ) . Differences between vehicle-treated groups were smaller in this strain , and yet the pattern of diazepam effects were identical to those reported with the C57BL/6J strain , with significant anxiolytic-like effects in α2F/F and α2CA1KO , but not in α2CA3KO or α2DGKO mice ( Figure 2—figure supplement 1B; Table 2 , Sect . 1’ ) . 10 . 7554/eLife . 14120 . 010Figure 2 . Behavioral tests of anxiety and locomotor activity . ( A ) Left: Activity heat maps on the EPM of representative α2F/F , α2CA1KO , α2CA3KO and α2DGKO mice treated with vehicle or diazepam . Right: Percentage ( Mean ± S . E . M . ) of time spent in the open arms of the EPM . ( B ) Left: Activity heat maps on the LDB of representative α2F/F , α2CA1KO , α2CA3KO and α2DGKO mice treated with vehicle or diazepam . The dark ( red ) and lit ( yellow ) compartments of the box are outlined for ease of visualization . Right: Percentage ( Mean ± S . E . M . ) of time spent in the lit compartment of the LDB . ( C ) Mean ( ± S . E . M . ) distance travelled in the open field . DOI: http://dx . doi . org/10 . 7554/eLife . 14120 . 01010 . 7554/eLife . 14120 . 011Figure 2—source data 1 . Raw data for elevated plus maze and light/dark box figures . DOI: http://dx . doi . org/10 . 7554/eLife . 14120 . 01110 . 7554/eLife . 14120 . 012Figure 2—figure supplement 1 . Additional measures in tests of anxiety-like behavior . ( A ) Left: Percentage of entries into the open arms in elevated plus maze in F/F control , α2CA1KO , α2CA3KO and α2DGKO mice bred on the C57BL/6J background . Right: Total distance travelled in the elevated plus maze . ( B ) Left: Percentage of time spent in the open arms , Right: Percentage of entries into the open arms in F/F control , α2CA1KO , α2CA3KO and α2DGKO mice bred on the 129x1/SvJ background . ( C ) Number of entries into the light compartment in light/dark box in F/F control , α2CA1KO , α2CA3KO and α2DGKO mice bred on the 129×1/SvJ background . +p<0 . 06 , p<0 . 05 , *p<0 . 01 in comparison to the corresponding vehicle group . DOI: http://dx . doi . org/10 . 7554/eLife . 14120 . 01210 . 7554/eLife . 14120 . 013Table 2 . Results of omnibus statistical tests of measured parameters in behavioral tests of anxiety and general locomotion . DOI: http://dx . doi . org/10 . 7554/eLife . 14120 . 0131 . Elevated Plus Maze ( C57BL/6J ) Two-Way Factorial ANOVA; Factors: Genotype ( between-subjects ) , Drug ( between-subjects ) % Open Arm Time% Open Arm EntriesDistance TravelledGenotypeF ( 3 , 75 ) =1 . 41p=0 . 25F ( 3 , 75 ) =0 . 13p=0 . 94F ( 3 , 75 ) =0 . 85 F ( 3 , 69 ) = 1 . 11p=0 . 47DrugF ( 1 , 75 ) =16 . 48p<0 . 001F ( 3 , 75 ) =2 . 33p=0 . 13F ( 3 , 75 ) =0 . 61p=0 . 44Genotype x Drug InteractionF ( 3 , 63 ) =1 . 54p=0 . 21F ( 3 , 75 ) =1 . 47p=0 . 23F ( 3 , 75 ) =1 . 13p=0 . 341’ . Elevated Plus Maze ( 129X1/SvJ ) Two-Way Factorial ANOVA; Factors: Genotype ( between-subjects ) , Drug ( between-subjects ) % Open Arm Time% Open Arm EntriesDistance TravelledGenotypeF ( 3 , 55 ) =0 . 17p=0 . 92F ( 3 , 55 ) =1 . 59p=0 . 20F ( 3 , 55 ) =1 . 17p=0 . 33DrugF ( 1 , 55 ) =11 . 49P=0 . 001F ( 1 , 55 ) =3 . 4649p=0 . 07F ( 1 , 55 ) =1 . 28p=0 . 29Genotype x Drug InteractionF ( 3 , 55 ) =2 . 28p=0 . 09F ( 3 , 55 ) =0 . 50p=0 . 69F ( 3 , 55 ) =0 . 82p=0 . 492 . Light / Dark BoxTwo-Way Factorial ANOVA; Factors: Genotype ( between-subjects ) , Drug ( between-subjects ) % Time in LightEntries to LightGenotypeF ( 3 , 73 ) =5 . 03p=0 . 003F ( 3 , 73 ) =0 . 84p=0 . 48Drug ( F ( 1 , 73 ) =26 . 00p<0 . 001F ( 1 , 73 ) =1 . 45p=0 . 23Genotype x Drug InteractionF ( 3 , 73 ) =5 . 53p=0 . 002F ( 3 , 73 ) =2 . 17p=0 . 093 . Open FieldTwo-Way Factorial ANOVA; Factors: Genotype ( between-subjects ) , Drug ( between-subjects ) Distance TravelledGenotypeF ( 3 , 56 ) =1 . 78p=0 . 16DrugF ( 1 , 56 ) =0 . 04p=0 . 84Genotype x Drug InteractionF ( 3 , 56 ) =0 . 43p=0 . 73 LDB experiments were conducted only on mice bred on the 129X1/SvJ background , as diazepam did not lead to consistent anxiolytic-like effects in control mice of C57BL/6J background . Similar to EPM , diazepam increased the time animals spent in the larger lit compartment of the LDB in α2F/F and α2CA1KO mice , but not in α2CA3KO or α2DGKO mice ( Figure 2B; Figure 2—figure supplement 1C; Table 2 , Sect . 2 ) . The changes ( or the lack thereof ) in open-arm and lit-compartment activity could not be attributed to nonspecific effects on general locomotor activity , as overall locomotion was not affected by diazepam ( Figure 2C; Figure 2—figure supplement 1A; Table 2 , Sect . 1 , 3 ) . These findings suggest that anxiety-like behaviors are under control of α2GABAAR-mediated inhibition of principal neurons in DG and CA3 , whereas a similar level of inhibition of CA1 pyramidal neurons does not affect anxiety-like behavior . Anxiety and fear are regulated by overlapping but somewhat distinct circuits in the brain ( e . g . , Jennings et al . , 2013 , Kheirbek et al . , 2013 , Botta et al . , 2015 ) . Next we tested whether HPC regulation of fear is mediated by an overlapping HPC microcircuit . We used two tests , both of which involve a distinct harmful stimulus ( a mild electric shock ) , to test fear-related behavior . In the fear-potentiated startle ( FPS; Figure 3A ) test , all experimental groups showed stable baseline startle responses that increased in magnitude with louder white noise bursts during the habituation trials ( Figure 3—figure supplement 1A–D ) , showing that a startle magnitude at 85dB ( i . e . , the magnitude of the testing phase stimulus ) is not bound by floor or ceiling effects . On test day , all vehicle-treated mice had increased startle amplitudes when the startle stimulus was preceded by the previously fear-conditioned tone ( i . e . , FPS; Figure 3B–E; Table 3 , Sect . 1 ) . When GABAAR activity was increased , the magnitude of FPS ( % ) was reduced significantly in α2F/F control mice . Interestingly , this reduction in% FPS was also observed in α2CA3KO and α2DGKO mice , but not in α2CA1KO mice ( Figure 3F; Table 3 , Sect . 1 ) . In a separate test , we evaluated the baseline shock sensitivity of each genotype to the 0 . 4 mA shock used in FPS , and found no difference between the genotypes ( Figure 3—figure supplement 1E ) . 10 . 7554/eLife . 14120 . 014Figure 3 . Behavioral tests of fear . ( A ) Representative recordings of “Startle Stimulus Only” and “Tone + Startle Stimulus” trials in a α2F/F control mouse treated with vehicle . The increased startle amplitude in “Tone + Startle” trials represents fear-potentiation of the startle response . ( B-E ) Mean ( ± S . E . M . ) startle amplitude in “No Tone” and “Tone” startle trials in vehicle and diazepam-treated ( B ) α2F/F , ( C ) αCA1KO , ( D ) α2CA3KO , ( E ) α2DGKO mice . ( F ) Mean ( ± S . E . M . ) percent FPS in trials preceded by the tone . Asterisks represent significant difference from the corresponding vehicle group . ( G ) Mean ( ± S . E . M . ) number of licks recorded in the pretest session of the VCT where drinking is not punished ( This session does not involve drug administration ) . ( H ) Mean ( ± S . E . M . ) number of licks recorded in the test session where every 20th lick is punished in vehicle- or diazepam-treated mice . Asterisks represent significant difference from the corresponding vehicle group . p<0 . 05 , p<0 . 01 , p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 14120 . 01410 . 7554/eLife . 14120 . 015Figure 3—source data 1 . Raw data for fear-potentiated startle and Vogel conflict test figures . DOI: http://dx . doi . org/10 . 7554/eLife . 14120 . 01510 . 7554/eLife . 14120 . 016Figure 3—figure supplement 1 . Additional measures in tests of fear-related behavior . ( A-D ) Mean ( ± S . E . M . ) startle amplitude in response to different intensities of white noise bursts measured in arbitrary units based on the displacement of load cells during the first three days ( Habituation sessions ) of the FPS experiment in ( A ) α2F/F , ( B ) α2CA1KO , ( C ) α2CA3KO , ( D ) α2DGKO mice . ( E ) Mean ( ± S . E . M . ) shock sensitivity measured by the average motion registered during 0 . 4 mA shocks . ( F ) Mean ( ± S . E . M . ) number of unpunished licks on test day in vehicle- and diazepam-treated α2F/F control mice . DOI: http://dx . doi . org/10 . 7554/eLife . 14120 . 01610 . 7554/eLife . 14120 . 017Table 3 . Results of omnibus statistical tests of measured parameters in behavioral tests of fear . DOI: http://dx . doi . org/10 . 7554/eLife . 14120 . 0171 . Fear-Potentiated StartleWithin-Genotype ComparisonsTwo-Way Factorial ANOVA; Factors: Tone/No Tone ( within-subjects ) , Drug ( between-subjects ) α2F/Fα2CA1KOToneF ( 1 , 28 ) =33 . 75p<0 . 001F ( 1 , 20 ) =49 . 17p<0 . 001DrugF ( 1 , 28 ) =0 . 30p=0 . 59F ( 1 , 20 ) =0 . 51p=0 . 48Tone x Drug InteractionF ( 1 , 28 ) =9 . 95p=0 . 004F ( 1 , 20 ) =0 . 02P=0 . 89α2CA3KOα2DGKOToneF ( 1 , 18 ) =16 . 60p<0 . 001F ( 1 , 20 ) =54 . 46p<0 . 001DrugF ( 1 , 18 ) =0 . 09p=0 . 77F ( 1 , 20 ) =0 . 73p=0 . 40Tone x Drug InteractionF ( 1 , 18 ) =5 . 16p=0 . 04F ( 1 , 20 ) =10 . 24p=0 . 01Between-Genotype ComparisonsTwo-Way Factorial ANOVA; Factors: Genotype ( between-subjects ) , Drug ( between-subjects ) % FPSGenotypeF ( 3 , 86 ) =1 . 17p=0 . 91DrugF ( 1 , 86 ) =16 . 69p<0 . 001Genotype x Drug InteractionF ( 3 , 86 ) =2 . 44p=0 . 072 . Vogel Conflict TestPretest ( Unpunished ) DrinkingOne-Way ANOVA; Factor: Genotype ( between-subjects ) Number of licksGenotypeF ( 3 , 63 ) =0 . 63p=0 . 60Test ( Punished ) DrinkingTwo-Way Factorial ANOVA; Factors: Genotype ( between-subjects ) , Drug ( between-subjects ) Number of licksGenotypeF ( 3 , 59 ) =1 . 00p=0 . 40DrugF ( 1 , 59 ) =14 . 57p<0 . 001Genotype x Drug InteractionF ( 3 , 59 ) =1 . 21p=0 . 31 In our FPS test , the fear-reducing effect of diazepam was due not only to a reduction in startle responses in the “tone + startle” condition , but also to an increase in startle magnitude in the “startle stimulus only” condition . While it complicates the interpretation of the data , this effect of diazepam in C57BL/6J mice was documented previously ( Smith et al . , 2011 ) . It should be noted that the startle magnitudes in “tone+startle” condition in diazepam-treated groups was not bound by a ceiling effect , as magnitudes up to 10a . u . ’s were observed; thus , the reduction in% FPS in diazepam-treated α2F/F , α2CA3KO and α2DGKO mice is not simply a result of the increased startle in the “startle stimulus only” condition . Still , we conducted a second test of fear , Vogel Conflict Test ( VCT ) , to further clarify these findings . The genotype-linked contrast between the FPS and the findings from tests of anxiety could be due to different circuitry underlying fear and anxiety , or simply due to different circuitries being involved in unconditioned ( EPM and LDB ) anxiety versus long-term fear memory . Thus , we chose the VCT , which does not rely on long-term memory ( although effects of within-session working memory cannot be eliminated ) , as our second test of fear . In VCT , water-deprived animals are allowed to drink from a spout which delivers electric shocks to the tongue , creating a conflict between the desire to drink and the desire to avoid a painful stimulus . A reduction in fear of the shock is reflected in increased drinking in the presence of shock . When drinking was unpunished , we found no difference between genotypes ( Figure 3G; Table 3 , Sect . 2 ) . In contrast , on test day ( punished drinking; Figure 3H; Table 3 , Sect . 2 ) elevation of GABAAR activity by diazepam significantly increased punished drinking in α2F/F , α2CA3KO and α2DGKO mice . Similar to FPS , the fear-reducing effect of diazepam was abolished in α2CA1KO mice . We also tested the possible effects of diazepam on unpunished drinking and found no nonspecific effects on water consumption ( Figure 3—figure supplement 1F ) . Our findings suggest that the double dissociation observed between FPS/VCT and EPM/LDB is likely due to a divergence in the HPC circuitry mediating fear versus anxiety rather than being a consequence of the involvement of memory processes . Previous studies linked theta range oscillations in the HPC to behavioral manifestations of anxiety . For instance , theta activity is enhanced in the ventral HPC ( vHPC ) with anxiety ( Adhikari et al . , 2010 ) and pharmacological manipulations that reduce anxiety also reduce the frequency of HPC theta elicited by brain stem stimulation ( McNaughton et al . , 1986 , McNaughton and Coop , 1991 , Engin et al . , 2009 , Yeung et al . , 2012 see also Wells et al . , 2013 ) . GABAergic anxiolytic drugs additionally reduce the slope of the function that relates brain stem stimulation current to theta frequency ( McNaughton et al . , 2007; Engin et al . , 2008 ) . As our tests of anxiety indicated that increasing GABAAR function can reduce anxiety in control and α2CA1KO , but not in α2CA3KO and α2DGKO mice , we next tested whether the effect of these manipulations on HPC theta range activity is consistent with the observed behavioral effects . Whereas our genetic manipulations span the whole HPC , previous studies implicated specifically the vHPC in anxiety-related behaviors ( Bannerman et al . , 2002; Bannerman et al . , 2004; Fanselow and Dong , 2010 ) . Thus , we recorded separately from the dorsal ( dHPC ) and vHPC ( Figure 4A ) , hypothesizing that the recordings from vHPC might be more closely linked to anxiety . 10 . 7554/eLife . 14120 . 018Figure 4 . Evoked theta oscillations in the vHPC . ( A ) Stimulation and recording sites and representative traces showing vHPC theta activity before and after diazepam injection in a α2F/F mouse . ( B-E ) Mean ( ± S . E . M . ) peak frequency in the theta range at different stimulation intensities before ( Baseline ) , and 30 min ( 30min post-DZP ) and 60 min ( 60min post-DZP ) following diazepam injection in ( B ) α2F/F , ( C ) α2CA1KO , ( D ) α2CA3KO , ( E ) α2DGKO mice . Asterisks represent significant difference from the baseline at the given stimulating current , with top ones for 30 min and the lower ones for 60 min post-injection . ( F ) Change from baseline slope in the stimulation intensity – peak frequency function 30 min ( left ) or 60 min following diazepam injection . #p<0 . 09 , p<0 . 05 , p<0 . 01 , p<0 . 001 compared to corresponding α2F/F group . DOI: http://dx . doi . org/10 . 7554/eLife . 14120 . 01810 . 7554/eLife . 14120 . 019Figure 4—source data 1 . Raw data for peak frequency and stimulation intensity – peak frequency slope figures . DOI: http://dx . doi . org/10 . 7554/eLife . 14120 . 01910 . 7554/eLife . 14120 . 020Figure 4—figure supplement 1 . Frequency and power of theta range oscillations . A-D Mean ( ± S . E . M . ) peak frequency in the theta range in dHPC recorded before , 30min after and 60min after diazepam administration in ( A ) α2F/F , ( B ) α2CA1KO , ( C ) α2CA3KO , ( D ) α2DGKO mice . p<0 . 01 , *p<0 . 001 compared to pre-diazepam baseline at the given stimulating current . ( E ) Mean ( ± S . E . M . ) change from baseline slope in the stimulation intensity – peak frequency function for dHPC recordings 30 min ( left ) or 60 min following diazepam injection , p<0 . 05 compared to corresponding α2F/F group . ( F ) Mean ( ± S . E . M . ) theta power in the vHPC normalized to baseline power 60 min following diazepam injection . ( G ) Same as F for dHPC recordings . *p<0 . 05 compared to corresponding α2F/F group . DOI: http://dx . doi . org/10 . 7554/eLife . 14120 . 020 vHPC recordings indicated that the increase in GABAAR activity 30 min and 60 min following injection of diazepam reduced the frequency of theta oscillations in the 0 . 04–0 . 10 mA stimulation range in both α2F/F and α2CA1KO mice ( Figure 4B , C; Table 4 , Sect . 1 ) , in line with the anxiolytic-like behavioral effects of the same drug manipulation in these groups . In α2CA3KO and α2DGKO mice , the peak frequency in theta band was reduced following diazepam , but the reductions were smaller in size and were limited to higher stimulation intensities ( Figure 4D , E; Table 4 , Sect . 1 ) . The slope of the linear function that relates stimulation intensity to theta frequency was also reduced in all genotypes following diazepam injection ( Figure 4F ) , but 30min following injection ( the time point at which we conducted our behavioral tests ) , the magnitude of the reduction in slope was significantly smaller in α2CA3KO and α2DGKO mice compared to controls and α2CA1KOs ( Figure 4F; Table 4 , Sect . 2 ) . Thus , the well-validated effects of diazepam on elicited theta activity , which show parallels with anxiolytic behavioral effects , were dampened in α2CA3KO and α2DGKO mice . 10 . 7554/eLife . 14120 . 021Table 4 . Omnibus statistical tests of measured parameters in in vivo LFP recordings collected from ventral and dorsal hippocampus . DOI: http://dx . doi . org/10 . 7554/eLife . 14120 . 021VENTRAL HIPPOCAMPUS1 . Peak Theta FrequencyTwo-Way ANOVA; Factors: Stimulation intensity ( within-subjects ) , Time before/after drug ( within-subjects ) α2F/Fα2CA1KOStimulation intensityF ( 4 , 32 ) =9 . 13p<0 . 001F ( 4 , 24 ) =18 . 12p<0 . 001Time after drugF ( 2 , 32 ) =22 . 98p<0 . 001F ( 2 , 24 ) =37 . 01p<0 . 001Stimulation x TimeF ( 8 , 32 ) =4 . 27p<0 . 001F ( 8 , 24 ) =9 . 07p<0 . 001α2CA3KOα2DGKOStimulation intensityF ( 4 , 32 ) =8 . 14p<0 . 001F ( 4 , 32 ) =3 . 84P=0 . 02Time after drugF ( 2 , 32 ) =6 . 14P=0 . 02F ( 2 , 32 ) =2 . 47P=0 . 15Stimulation x TimeF ( 8 , 32 ) =3 . 14P=0 . 01F ( 8 , 32 ) =6 . 63p<0 . 0012 . Stimulation Intensity – Theta Frequency SlopeOne-Way ANOVA; Factor: Genotype ( between-subjects ) 30min post-diazepam60min post-diazepamGenotypeF ( 3 , 15 ) =3 . 94p=0 . 03F ( 3 , 73 ) =0 . 84p=0 . 483 . Normalized PowerOne-Way ANOVA; Factor: Genotype ( between-subjects ) 60min post-diazepamGenotypeF ( 3 , 15 ) =4 . 65p=0 . 02DORSAL HIPPOCAMPUS4 . Peak Theta FrequencyTwo-Way ANOVA; Factors: Stimulation intensity ( within-subjects ) , Time before/after drug ( within-subjects ) α2F/Fα2CA1KOStimulation intensityF ( 4 , 32 ) =8 . 19p<0 . 001F ( 4 , 32 ) =5 . 29p=0 . 01Time after drugF ( 2 , 32 ) =31 . 65p<0 . 001F ( 2 , 32 ) =3 . 03p<0 . 11Stimulation x TimeF ( 8 , 32 ) =8 . 87p=0 . 003F ( 8 , 32 ) =1 . 89p=0 . 10α2CA3KOα2DGKOStimulation intensityF ( 4 , 40 ) =19 . 19p<0 . 001F ( 4 , 32 ) =8 . 86p<0 . 001Time after drugF ( 2 , 40 ) =27 . 08p<0 . 001F ( 2 , 32 ) =9 . 25p=0 . 01Stimulation x TimeF ( 8 , 40 ) =4 . 30p<0 . 001F ( 8 , 32 ) =2 . 39p=0 . 045 . Stimulation Intensity – Theta Frequency SlopeOne-Way ANOVA; Factor: Genotype ( between-subjects ) 30min post-diazepam60min post-diazepamGenotypeF ( 3 , 17 ) =2 . 97p=0 . 14F ( 3 , 17 ) =4 . 32p=0 . 026 . Normalized PowerOne-Way ANOVA; Factor: Genotype ( between-subjects ) 60min post-diazepamGenotypeF ( 3 , 15 ) =1 . 02p=0 . 41 Interestingly , simultaneous recordings in the dHPC showed the opposite pattern . Diazepam reduced the frequency of theta in control , α2CA3KO and α2DGKO mice , but there was no main effect of diazepam injection in α2CA1KO mice ( Figure 4—figure supplement 1A–D; Table 4 , Sect . 4 ) . The reduction in slope following diazepam injection was also significantly smaller in α2CA1KO mice compared to controls ( Figure 4—figure supplement 1E ) . We found no parallels between behavioral tests of anxiety and fear , and power of evoked theta in dHPC or vHPC ( Figure 4—figure supplement 1F–G; Table 4 , Sect . 3 , 6 ) , consistent with previous reports ( McNaughton et al . , 2007 ) . Our findings indicate a double dissociation in the control of anxiety versus fear by HPC microcircuits , with increased α2GABAAR-mediated inhibition of principal neurons in the DG and CA3 nodes of the trisynaptic pathway required for suppression of anxiety , and increased inhibition of CA1 pyramidal neurons required for suppression of fear . Strikingly , as increased GABAAR activity was achieved through a systemic pharmacological manipulation , our findings suggest that even if GABAAR activity is simultaneously increased in all other parts of the circuitry underlying anxiety and/or fear responses , such as the amygdala , bed nucleus of the stria terminalis or the medial prefrontal cortex ( mPFC ) ( Tovote et al . , 2015 ) ( and in the hippocampus via other GABAA receptor subtypes ) , it is not sufficient to suppress anxiety and/or fear responses unless the corresponding HPC subregions are simultaneously inhibited via α2GABAA receptors . The findings cannot be explained by changes in expression of the α2GABAARs outside of the hippocampus . As seen in Figure 1 ( and Figure 1—figure supplement 1 ) , the conditional knockouts were highly specific to the target regions in α2CA3KO and α2DGKO mice . In α2CA1KO mice , there are no significant changes in the expression of the GABAAR α2 subunit protein in the cortex or the amygdala , although there is an apparent change in expression in especially the medial regions of the basolateral amygdala ( BLA ) in some sections ( see Figure 1 ) . While this raises the possibility that the effects ( or the lack thereof ) observed in α2CA1KO mice may be partially due to changes in α2 protein expression in the BLA , our findings from ongoing studies indicate that knocking down α2GABAARs in the BLA leads to a completely different pattern of results than that observed in α2CA1KO mice , making this possibility highly unlikely . Our findings also cannot be explained by confounding factors such as gross HPC dysfunction or nonspecific behavioral changes in the gene-targeted mice , as all genotypes showed normal baseline HPC function and response to diazepam , as measured by ex vivo electrophysiology ( Figure 1—figure supplement 3 ) and HPC-dependent behavioral tasks ( trace and context fear conditioning , Morris water maze; Figure 4—figure supplement 1 ) . We further report that the underlying mechanism for the suppression of anxiety and fear following the systemic elevation of GABAAR activity may be the effect of this manipulation on vHPC and dHPC theta range activity , respectively . All classes of clinically effective anxiolytic compounds reduce the frequency of evoked HPC theta activity , with no known false positives or false negatives , making this a strong physiological marker of anxiolytic action ( McNaughton et al . , 2007 ) . While most systemic pharmacological manipulations that reduce anxiety also reduce fear , previous studies did not investigate whether the effects of anxiolytic and fear-reducing manipulations on theta range activity may be distinct in this specific model . There is some evidence , however , suggesting that theta range activity recorded from dHPC may be more relevant for fear , while vHPC theta may be more relevant for anxiety . For instance , theta range activity recorded from dHPC in CA1 and the amygdala is synchronized during expression of conditioned fear responses , but not during anxiety-related responses ( Seidenbecher et al . , 2003 ) . Conversely , in recordings from vHPC , increases in theta power and synchrony with mPFC in the theta band were observed during the exploration of anxiogenic arenas , while this was not the case for recordings from the dHPC ( Adhikari et al . , 2010 ) . In a similar vein , we report that diazepam effects on vHPC theta were reduced in α2CA3KO and α2DGKO mice , while effects on dHPC theta were reduced in α2CA1KO mice , in parallel with the effects observed in behavioral tests of anxiety and fear , respectively . While our knockouts mostly span the septo-temporal axis of the hippocampus , these findings also point to the possibility that CA1 in the dHPC may be important for fear , while CA3 and DG in the vHPC may be important for anxiety , integrating our findings with previous reports ( Bannerman et al . , 2004; Fanselow and Dong , 2010 ) . The role of HPC in anxiety and fear was recognized early on ( e . g . , Gray , 1982 ) and has been demonstrated using different approaches ( Engin and Treit , 2007; Bannerman et al . , 2004; Maren , 2001 ) . However , despite concentrated efforts in the last 15 years to understand the roles of different HPC subregions in cognitive processes ( e . g . , ( Nakazawa et al . , 2002; McHugh et al . , 2007; Tsien et al . , 1996; Kesner , 2013a; Kesner , 2013b ) , and a shift in the focus of anxiety and fear research towards within-structure microcircuits and specific neuronal populations ( e . g . , Kim et al . , 2013 , Botta et al . , 2015 , Tye et al . , 2011 ) , a systematic analysis of the regulation of fear and anxiety by HPC microcircuitry had so far not been conducted . Our experiments extend previous research , which to a large extent treated the HPC as a unitary structure except anatomical distinctions along its septo-temporal axis ( Fanselow and Dong , 2010 ) , and ignored the possible specialization within the microcircuitry of the HPC . Furthermore , the distinction between anxiety and fear , and whether they are processes mediated by distinct neurocircuitry , has been a controversial question ( Perusini and Fanselow , 2015 ) . Here we show that anxiety and fear , distinguished via experimental conditions that cause each state ( we operationally define “tests of fear” as those involving a distinct harmful stimulus; e . g . , an electric shock ) , are mediated by distinct subregions within the HPC , lending further credence to the distinction between the two states and the overlapping but distinct neurobiological underpinnings of fear and anxiety . In conclusion , in addition to demonstrating a surprisingly essential role of the hippocampus in the pharmacological modulation of anxiety and fear , our study indicates distinct molecular mechanisms underlying regulation of two distinct negative valence systems ( i . e . , anxiety and fear ) , and provides defined cellular entry points into neuronal circuitry underlying complex behavioral states . For the generation of the floxed Gabra2 allele , see ( Witschi et al . , 2011 ) for details . Briefly , a 6 . 3 kb genomic fragment ( PstI-NcoI ) containing exons 5 ( 221 bp ) and 6 ( 83 bp ) of the Gabra2 gene was isolated . A 1 kb SphI-SphI fragment containing Exon 5 was then removed from the 6 . 3 kb fragment and was replaced by an oligo hybrid containing a loxP site in addition to the 1 kb SphI-SphI fragment with exon 5 . A neomycine resistance cassette ( NEO; FRT-Pol2-neo-bpA-FRT-loxP ) was subcloned into the SalI site . The vector was electroporated into embryonic stem cells ( C57BL6/N , Eurogentec ) , clones with correctly targeted alleles were injected into blastocysts ( Polygene , Rumlang , Switzerland ) , and the NEO was bred out ( Gabra2tm2 . 1Uru ) . Figure 1A shows the 6 . 3 kb PstI-NcoI fragment containing exon 5 ( 221 bp ) , flanked by two loxP sites , the single FRT site remaining following the excision of the NEO , and exon 6 ( 83 bp ) . The α2 conditional knockout mice were generated by crossing mice homozygous for the floxed Gabra2 allele ( Figure 1A top; α2F/F mice ) with mice that are homozygous for the floxed Gabra2 allele and carry one of the following three cre recombinase transgenes: CamKIIα cre ( T29-1 mice; Tsien et al . , 1996 ) to generate a CA1-selective knockout ( α2CA1KO ) , Grik4 cre ( Nakazawa et al . , 2002 ) to generate a CA3-selective knockout ( α2CA3KO ) and POMC cre ( McHugh et al . , 2007 ) to generate a DG-selective knockout ( α2DGKO ) . Mice were bred at McLean Hospital animal facility . For immunohistochemistry , ex vivo recordings and FPS experiments , the mice were kept on a 12-hr light/dark cycle with lights on at 07:00 am . For EPM , LDB , open field and VCT , the mice were maintained on a reverse 12-hr light/dark cycle with lights on at 07:00pm , and the tests were conducted during the dark phase . For in vivo electrophysiology experiments , a mix of male and female experimental mice were shipped from McLean Hospital to Yale University at 4 weeks of age , and were allowed to acclimatize to the new environment for at least 6 weeks before the experiments . Only male mice were used for all other experiments . α2F/F mice carrying the corresponding cre transgene were used as experimental animals , while the α2F/F , Cre- littermates were combined into a single α2F/F control group with approximately equal numbers from each breeding . All mice were maintained on either a C57BL/6J background or a 129X1/SvJ ( only for the mice used in light/dark box , the secondary elevated plus maze experiments and cognitive measures ) background . Mice were deeply anesthetized with sodium pentobarbital ( 200 mg/kg ) and were perfused transcardially with ice-cold phosphate-buffered saline ( PBS ) , followed by 150mM sodium phosphate buffer containing 4% paraformaldehyde and 15% picric acid . The brains were post-fixed in the same fixation solution for 4 hr , and were then processed for antigen retrieval . Briefly , whole brains were incubated in sodium citrate buffer ( pH 4 . 5 ) overnight at room temperature . Next day , blocks of tissue including the regions of interest were cut , placed in fresh sodium citrate buffer and were irradiated in a microwave for 90 sec . The brains were washed in PBS following microwave irradiation and were placed in 30% sucrose for cryoprotection . The brains were sectioned coronally into 40 µm-thick sections using a sliding microtome , and the sections were stored at -20C in an antifreeze solution until use . Immunoperoxidase staining was performed using diaminobenzidine as a chromophore . The sections were incubated in a 0 . 3% H2O2 solution for 30 min , followed by 2 hr blocking in 3% normal goat serum ( NGS ) , 0 . 25% Triton X-100 solution for 2 hr , and were incubated overnight at 4C in primary antibodies ( Guinea pig anti-α1 ( 1:20 , 000 ) , guinea pig anti-α2 ( 1:1 , 000 ) and guinea pig anti-α5 ( 1:1 , 000; Fritschy and Mohler , 1995 ) . diluted in Tris buffer containing 2% NGS and 0 . 2% Triton X-100 . The next day , the sections were washed and incubated in a biotinylated secondary antibody ( goat anti-guinea pig ( 1:300 ) , Jackson ImmunoResearch ) , and then in ABC complex solution ( Vectastain Elite kit; Vector Laboratories , Burlingame , CA ) at room temperature . The sections were incubated in 0 . 05% diaminobenzidine tetrahydrochloride ( Sigma-Aldrich , St . Lois , MO ) dissolved in Tris-Triton ( pH 7 . 7 ) containing 0 . 02% H2O2 for approximately 10 min at room temperature , washed in ice-cold PBS , mounted on gelatinized slides . The slides were air-dried , dehydrated and coverslipped with Eukitt ( Fluka , Sigma-Aldrich , St . Lois , MO ) . For semi-quantitative analysis of the DAB-stained tissue , sections spanning the length of the HPC were photographed at 4x and 10x magnification , and optical density was calculated in regions of interest using the Image J software . For each area , optical density measured on α2F/F control sections was set to 1 and all other measurements were expressed as a proportion of this . All procedures were carried out in an RNase free environment with surgical tools and bench space decontaminated with RNase AWAY ( Molecular Bioproducts , Carlsbad , CA ) . Hippocampal regions were dissected by laser capture microdissection ( LCM ) . Procedures for LCM were previously described by ( Chen et al . , 2014 ) . Briefly , fresh frozen brains were sectioned at 10 µm and mounted on uncoated glass slides . The slides were then treated with a series of dehydration steps ( acetone , ethanol , and xylene ) and then air-dried . Hippocampal regions of interest ( ROIs ) , as illustrated in Figure 1C , were captured with Arcturus XT ( Applied Biosystems , Carlsbad , CA ) onto CapSure LCM caps ( Applied Biosystems , Carlsbad , CA ) . For each brain , hippocampal ROIs were collected from 4–5 coronal sections , bilaterally . Total RNA was then purified with PicoPure RNA isolation kit ( Applied Biosystems , Carlsbad , CA ) from the caps , inspected by Synergy HT ( BioTek , Winooski , VT ) and normalized to 2 ng/µL concentration . Amygdala and cortical specimens were obtained by manual dissection . Briefly , after collecting brain sections for LCM , a 300 µm section was cut from the brain and submerged in ice cold PBS . The ROIs , as illustrated in Figure 1C , were dissected manually under a magnifying scope . Total RNA was then purified with RNeasy mini kit ( Qiagen , Valencia , CA ) , inspected by Synergy HT ( BioTek , Winooski , VT ) and normalized to 20ng/µL concentration . The following primer-sets were used for q-PCR: Gabra2: forward 5’-GCTGCTTCGAATCCAGGATGA-3’ , reverse 5’-AAATCCTCCAGGTGCATGGG-3’; Gabra3: forward 5’-CTTGGGAAGGCAAGAAGGTA-3’ , reverse GGAGCTGCTGGTGTTTTCTT-3’; Gabra4: forward 5’-AAAGCCTCCCCCAGAAGTT-3’ , reverse 5’-CATGTTCAAATTGGCATGTGT-3’ . The selectivity of the assays was tested by end-point gel electrophoresis and the primer efficiencies were between 85% and 105% . Reference gene ( β-tubulin , forward 5’-GCGCATCAGCGTATACTACAA-3’ , reverse 5’-TTCCAAGTCCACCAGAATGG-3’ ) was selected from a pool of 5 candidate reference genes ( β-Actin , Cyclin D , HGPRT , S19 , and β-tubulin ) based on an initial assessment experiment showing its stable expression across strains as well as good selectivity and efficiency . For each RNA specimen , first-strand cDNA was made from 20 ng total RNA by Transcriptor reverse transcriptase kit ( Roche Applied Science , Indianapolis , IN ) . For Gabra2 assay , 1ng cDNA was used in one LightCycler 480 SYBR Green I q-PCR reaction on the LightCycler® 480 Real-Time PCR System ( Roche Applied Science , Indianapolis , IN ) . For Gabra3 and Gabra4 assays , cDNA pre-amplification was performed with equal amount of cDNA input using TaqMan PreAmp Kit ( Life Technologies , Grand Island , NY ) and q-PCR was subsequently carried out using LightCycler 480 Probes Master ( Roche Applied Science , Indianapolis , IN ) . The CT values were assessed by the LC480 Software SW1 . 5 ( Roche Applied Science , Indianapolis , IN ) and relative expression values were calculated by the ∆∆CT-Method . Statistical analysis was performed by One-Way ANOVA with strain as independent variable , followed by Holm-Sidak t-test against α2F/F ( control ) group . One outlier and 3 failed PCR reactions were excluded from data analysis for Gabra3 and Gabra4 mRNA expression . Vibratome slices of the hippocampus ( 250–300 mm ) were prepared from male α2CA1KO , α2CA3KO or α2DGKO mice or corresponding littermate controls for each group ( 4 – 7 mice per group ) . Slices were continuously superfused in solution containing ( in mM ) : 119 NaCl , 2 . 5 KCl , 2 . 5 CaCl2 , 1 . 0 MgSO4 , 1 . 25 NaH2PO4 , 26 . 0 NaHCO3 , 10 glucose and equilibrated with 95% O2 and 5% CO2 ( pH 7 . 3–7 . 4 ) at 22o – 23o C . mIPSC were recorded in CA1 , CA3 or DG neurons ( i . e . , the site of the conditional knockout ) in the presence of 10 µM NBQX and 1 μM TTX . Diazepam 1–5 µM was added to the bath solution . Whole-cell recordings of mIPSCs were obtained from pyramidal neurons or granule cells under visual guidance ( DIC/infrared optics ) with an EPC-9 amplifier and Pulse v8 . 67 software ( HEKA Elektronik ) . Cells were classified as principal neurons based on spike frequency adaptation in response to prolonged depolarizing current injections . The recording patch electrodes ( 3–5 MW resistance ) contained ( in mM ) : 101 . 5 K-gluconate , 43 . 5 KCl , 1 MgCl2 , 0 . 2 EGTA , 10 HEPES , 2 MgATP , and 0 . 2 NaGTP ( adjusted to pH 7 . 2 with KOH ) . Currents were filtered at 1 kHz and digitized at 5 kHz . mIPSCs ( recorded in the presence of 1 mM TTX ) were analyzed with the Mini Analysis Program v6 . 0 . 7 ( Synaptosoft Inc . ) . The sample sizes for behavioral tests were calculated with power analyses based on previous findings ( Smith et al . , 2012 ) . The EPM test in the Smith et al . ( Smith et al . , 2012 ) study uses the same background and test parameters as the EPM test on C57Bl/6J mice in the current study . Thus , the mean and standard deviation values in the control group of the Smith et al . ( Smith et al . , 2012 ) study were used in a power analysis to calculate sample sizes for all tests of anxiety used in the current study ( Mean 1: 0 . 40 , Mean 2: 0 . 80 , Standard deviation values between 0 . 14–0 . 18 , alpha: 0 . 05 , intended power: 0 . 80 , which yielded target sample sizes of 8–13 ) . Similarly , the FPS test values for control mice in the same study were used to calculate sample sizes for all tests of fear in the current study ( Mean 1: 0 . 45 , Mean 2: 0 . 80 , Standard deviation values between 0 . 24–0 . 31 , alpha: 0 . 05 , intended power: 0 . 80 , which yielded target sample sizes of 7–13 ) . A sample size per group of 4–8 mice was chosen as the group size for in vivo electrophysiology experiments based on earlier studies measuring hippocampal theta oscillations in mice using similar parameters , in which these sample sizes yielded a statistical power of 0 . 80 or above ( Scott et al . , 2012 ) . Following the exclusion of mice based on criteria explained above ( see Materials and methods – In vivo electrophysiology ) , eventual sample sizes of 3–5 mice were used in statistical analyses . Data were expressed as means and standard errors of the mean ( S . E . M . ) and analyzed using the SAS statistical software version 9 . 1 ( SAS Institute , Inc . , Cary , NC ) and SigmaPlot software version 11 . 0 ( Systat Software , Inc . , Chicago , IL ) . With the exception of in vitro electrophysiology , data were analyzed with Two-Way Analyses of Variance ( ANOVAs ) using genotype and drug dose as the factors , followed , where the initial test is statistically significant , by post hoc Holm-Sidak tests for multiple comparisons ( unless noted otherwise ) . In vitro electrophysiology data were analyzed separately for each conditional knockout , with F/F Cre- littermates serving as controls . A two-way ANOVA was used for genotype and drug effects . The significance level for all tests was set at p<0 . 05 .	Fear and anxiety can be thought of as different but related emotional states . Fear is triggered by specific harmful situations , such as the immediate presence of a predator . Anxiety instead results from the possibility of an obscure threat , such as being in an exposed environment , which increases the chance of being detected by a predator . Evidence suggests that slightly different areas of the brain control fear and anxiety , but much remains unknown about the specific brain regions that help to regulate these two emotional states . One brain region that has been implicated in both anxiety and fear – as well as in learning and memory – is the hippocampus . Named after the Greek word for seahorse because of its shape , the hippocampus is made up of three subregions: CA1 , CA3 and the dentate gyrus . Each of these subregions has a distinct role in learning and memory . However , their individual contributions to the control of fear and anxiety were not known . An inhibitory receptor protein found in the surface of some hippocampal neurons had previously been shown to be involved in controlling fear and anxiety . Now , Engin et al . have studied three different groups of genetically modified mice , each of which lacks the receptor protein in a different subregion of the hippocampus . The mice completed tests that stimulated anxiety or fear , some while under the influence of the anxiety and fear-reducing drug diazepam . Notably , diazepam failed to reduce fear in animals that lacked the inhibitory receptor protein in the CA1 subregion of the hippocampus , suggesting that this subregion participates in the fear response . However , mice that lacked the receptor in the dentate gyrus or CA3 responded normally to the drug ( they showed reduced fear when given diazepam ) . In tests of anxiety , the picture was exactly the opposite . Diazepam failed to reduce anxiety in animals lacking the inhibitory receptor in the dentate gyrus or CA3 , indicating that these subregions are involved in the regulation of anxiety . However , the drug still reduced anxiety in mice that lacked the receptor protein in the CA1 subregion . Further studies are now needed to clarify how manipulating specific subregions of the hippocampus alters how it communicates with other brain structures to generate changes in anxiety or fear-related behaviors .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Modulation of anxiety and fear via distinct intrahippocampal circuits
Human flap endonuclease 1 ( FEN1 ) and related structure-specific 5’nucleases precisely identify and incise aberrant DNA structures during replication , repair and recombination to avoid genomic instability . Yet , it is unclear how the 5’nuclease mechanisms of DNA distortion and protein ordering robustly mediate efficient and accurate substrate recognition and catalytic selectivity . Here , single-molecule sub-millisecond and millisecond analyses of FEN1 reveal a protein-DNA induced-fit mechanism that efficiently verifies substrate and suppresses off-target cleavage . FEN1 sculpts DNA with diffusion-limited kinetics to test DNA substrate . This DNA distortion mutually ‘locks’ protein and DNA conformation and enables substrate verification with extreme precision . Strikingly , FEN1 never misses cleavage of its cognate substrate while blocking probable formation of catalytically competent interactions with noncognate substrates and fostering their pre-incision dissociation . These findings establish FEN1 has practically perfect precision and that separate control of induced-fit substrate recognition sets up the catalytic selectivity of the nuclease active site for genome stability . Biologically-critical , structure-specific 5’nucleases are highly conserved endo- or exo-nucleases that hydrolyze phosphodiester bonds that are one nucleotide into the 5’end of single-stranded ( ss ) /double-stranded ( ds ) -DNA junctions ( Figure 1A ) , including nicks , gaps , flaps , bubbles and four-way junctions ( Balakrishnan and Bambara , 2013; Finger et al . , 2012; Tsutakawa et al . , 2014; Tsutakawa and Tainer , 2012 ) . This conserved cleavage site despite diverse structures operates by uniformly binding to a bent junction to place the scissile phosphate near the active site ( Liu et al . , 2015; Orans et al . , 2011; Tsutakawa et al . , 2011 ) ( Figure 1B ) . Yet , the mechanism underlying this specificity remains unclear including the question of whether 5’nucleases actively distort the DNA or selectively bind to a DNA that bends spontaneously ( Craggs et al . , 2014; Sobhy et al . , 2013 ) . Such mechanistic knowledge not only pertains to biological understanding but also to strategies for the design of specific inhibitors as potential cancer drugs ( Exell et al . , 2016 ) . Catalysis is proposed to require changes in the protein conformation that assembles the active site ( Devos et al . , 2007; Lee et al . , 2015; Liu et al . , 2015; Orans et al . , 2011; Sakurai et al . , 2005; Tsutakawa et al . , 2011 ) and movement of the scissile phosphate closer to the catalytic metals ( Liu et al . , 2015; Orans et al . , 2011; Tsutakawa et al . , 2011 ) . Although possible steps in the substrate selection and cleaving process have been described ( Devos et al . , 2007; Lee et al . , 2015; Liu et al . , 2015; Orans et al . , 2011; Sakurai et al . , 2005; Tsutakawa et al . , 2011 ) , much of the control mechanisms by DNA and protein conformational changes that lead to exquisite catalytic selectivity and efficiency remain controversial and largely undetermined . 10 . 7554/eLife . 21884 . 003Figure 1 . Active junction bending by structure-specific 5’nucleases . ( A ) FEN1 cleavage reaction . Schematic showing the equilibration of a flap substrate junction from a single- to a double-flap and its subsequent cleavage by FEN1 to generate a nick that can be sealed by DNA ligase 1 . ( B ) Ordering of FEN1 upon DNA binding . FEN1 alone ( 1ULI . pdb ) ( Sakurai et al . , 2005 ) and in complex with bent DNA ( 3Q8L . pdb ) ( Tsutakawa et al . , 2011 ) , highlighting the various structural features of FEN1 and the regions that undergo through disorder-to-order transitioning upon DNA binding . ( C ) Active DNA versus DNA conformational capturing models for forming the FEN1 complex with the bent DNA conformer . Monitoring DNA bending of FEN1 and non-equilibrated DF-6 , 1 using the flap-labeling scheme ( NonEQ DF-6 , 1Flap ) ( D ) and internal labeling-scheme ( NonEQ DF-6 , 1dsDNA ) ( E ) . For each labeling , a schematic of the donor and acceptor positions ( upper panel ) and smFRET time traces of the substrate alone ( middle panel ) and in presence of FEN1 ( lower panel ) are shown; change in FRET upon DNA bending in each labeling scheme is highlighted . ( F ) Analysis of the structure of NonEQ DF-6 , 1 by MD simulations . The effective free energy profile ( PMF ) from adaptive biasing force calculations is shown . ( G ) Bending of equilibrated DF-6 , 1 ( EQ DF-6 , 1dsDNA ) by FEN1 . smFRET time traces of EQ DF-6 , 1dsDNA alone ( upper panel ) and in the presence of FEN1 ( middle panel ) and analysis of its DNA bending association rate constant ( kon-bending ) and dissociation rate constant ( koff-unbending ) ( lower panel ) are shown . kbending and kunbending were calculated by fitting an exponential function to the histogram from the population of dwell times of bent ( τbending ) and unbent ( τunbending ) conformers , respectively; error bars correspond to the standard deviation of the fit . kon-bending and koff-unbending are calculated from the slope of kbending from a linear regression fit and the mean of kunbending , respectively; the error bars correspond to the standard deviation of the fit . Kd-bending = koff-unbending/kon-bending . ( H ) Bending of nicked substrate using the internal labeling scheme ( NickdsDNA ) by EXO1 . A schematic of the donor and acceptor positions ( upper panel ) , smFRET time traces of NickdsDNA alone and in the presence of EXO1 ( middle panels ) and analysis of its kon-bending , koff-unbending and Kd-bending ( lower panel ) is presented . Donor and acceptor are at identical positions to those in DF-6 , 1dsDNA in Figure 1E . kon-bending , koff-unbending and Kd-bending were calculated as in 1G . All TIRF-smFRET experiments were acquired at 100 ms . DOI: http://dx . doi . org/10 . 7554/eLife . 21884 . 00310 . 7554/eLife . 21884 . 004Figure 1—figure supplement 1 . Bending of equilibrated and non-equilibrated DF-6 , 1 by FEN1 . ( A ) TIRF-based single-molecule FRET ( smFRET ) of FEN1 and NonEQ DF6 , 1Flap . Histograms of NonEQ DF6 , 1Flap alone ( high FRET ) and in the presence of FEN1 ( low FRET ) ( upper panel ) . Kd-bending ( lower panel ) is calculated by fitting the percentage of the bent conformer from the FRET histograms at various concentrations of FEN1 with a non-linear least squares regression; the percentages of unbent and bent DNA are calculated by fitting two Gaussians . The uncertainty in calculating the percentage of bent DNA and Kd-bending correspond to the standard deviation of triplicate measurements and the non-linear least squares regression fit , respectively . ( B ) Kd-bending of FEN1 on NonEQ DF-6 , 1dsDNA . Kd-bending is calculated as described in Figure 1—figure supplement 1A . ( C ) Kd-bending of FEN1 on EQ DF-6 , 1dsDNA . Kd-bending is calculated as described in Figure 1—figure supplement 1A . DOI: http://dx . doi . org/10 . 7554/eLife . 21884 . 00410 . 7554/eLife . 21884 . 005Figure 1—figure supplement 2 . Flap substrates exist as a stable extended conformer . ( A ) Effect of various Mg2+ concentration on DF-6 , 1 . TIRF-smFRET histograms of NonEQ DF6 , 1dsDNA ( left panel ) and NonEQ DF6 , 1Flap ( right panel ) with increasing concentration of MgCl2 ( 0 mM , 1 mM , 10 mM and 50 mM ) . The insensitivity of both labeling schemes to increasing divalent metal ion concentrations demonstrates that both labeling schemes report directly on the geometry of the duplex DNA . ( B ) Effect of various Mg2+ concentration on DF-12 , 1 . TIRF-smFRET histograms of NonEQ DF12 , 1dsDNA ( left panel ) and NonEQ DF12 , 1Flap ( right panel ) with increasing concentration of MgCl2 ( 0 mM , 1 mM , 10 mM and 50 mM ) . The sensitivity of the flap-labeling scheme to varying divalent metal ion concentrations but not the internal-labeling scheme demonstrates that the geometry of the duplex DNA is not influenced by the length of the 5’flap and that the flap-labeling scheme is inappropriate to describe the geometry of the duplex DNA only when the 5’flap length exceeds 6 nt . All TIRF-smFRET measurements were acquired at 160 ms temporal resolution . FWHM represents the full width at half maximum of the Gaussian peak . ( C ) Burst confocal-smFRET histograms from freely diffusing DNA in solution of NonEQ DF6 , 1dsDNA ( 0 . 5 nM ) ( upper panel ) and NonEQ DF6 , 1Flap ( 0 . 5 nM ) ( lower panel ) acquired at sub-ms temporal resolution . No enrichment of other FRET conformers was observed upon increasing the temporal resolution from 160 ms ( shown in Figure 1—figure supplement 1A , B ) to sub-ms . ( D ) Confocal-smFRET time traces of surface-immobilized NonEQ DF6 , 1Flap ( upper panel ) and NonEQ DF6 , 1dsDNA ( lower panel ) with 5 ms temporal resolution . DOI: http://dx . doi . org/10 . 7554/eLife . 21884 . 00510 . 7554/eLife . 21884 . 006Figure 1—figure supplement 3 . MD simulations of the conformational states and DNA bending energy of nick and various flap structures . ( A ) Definition of the DNA bending angle . The bending angle is calculated between two vectors . One vector was defined by using the center of the masses of two blocks of nucleotides: block I ( green ) and block II ( green ) and one vector was defined with the center of mass of block III ( blue ) and block VI ( blue ) . ( B , C , D and E ) are the averaged structure ( upper panel ) and histogram of the DNA bending angle ( lower panel ) for NonEQ DF-6 , 1 , Nick , EQ DF-6 , 1 and SF-6 , 0 , respectively , taken from the MD simulations . These MD simulations demonstrated that the extended DNA structure was the most energetically favorable conformer in all tested substrates and that these substrates could sample bent conformers up to 140° , which are equivalent to those observed for dsDNA ( Sharma et al . , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21884 . 00610 . 7554/eLife . 21884 . 007Figure 1—figure supplement 4 . Bulk cleavage , SPR binding and time-resolved bulk FRET of selected substrates . ( A ) Cleavage of NonEQ DF-6 , 1dsDNA ( 0 . 5 nM ) by bulk assays . A plot of initial rates ( v0 , nM . min−1 ) in relation to the FEN1 concentration fitted with a generalized non-linear least-squares regression using a Michaelis-Menten model . The values for v0 were estimated using bulk cleavage assays with different time intervals as described in the Materials and methods . This plot was used to determine the steady-state Km . Uncertainty in Km corresponds to the standard error of the fit . ( B ) SPR binding studies of FEN1 and NonEQ DF-6 , 1dsDNA ( left panel ) , SF-6 , 0dsDNA ( middle panel ) and DF-30 , 1blocked-dsDNA ( right panel ) . The sensorgram of binding of increasing concentrations of FEN1 is shown ( left panel ) . The maximum response units ( RU ) reached at each FEN1 concentration were fitted using the steady-state affinity model to obtain the equilibrium dissociation binding constant ( Kd-binding ) ( lower panel ) . The uncertainty corresponds to the standard deviation of N = 2 runs . 170 RU , 70 RU and 90 RU of NonEQ DF-6 , 1dsDNA , SF-6 , 0dsDNA and DF-30 , 1blocked-dsDNA were immobilized on the surface , respectively . ( C ) Bulk-FRET measurements of NonEQ DF-6 , 1 . The FRET efficiency of the donor-acceptor pair in NonEQ DF-6 , 1Flap at different FEN1 concentrations ( left panel ) and relative fluorescence lifetimes ( taudonor ( enzyme ) /taudonor ) at different FEN1 concentrations without acceptors ( middle panel ) are shown . The results show that FEN1 binding influenced the fluorescence intensity of the donor in the flap-labeling scheme . Kd-bending after correcting for the effect of the donor was similar to that obtained from uncorrected apparent FRET ( Figure 1—figure supplement 1C ) ; Kd-bending was calculated using a standard quadratic equation for the simple bimolecular association as described in Materials and methods and the uncertainty corresponds to the standard error of the fit . The uncertainty corresponds to the standard error of the fit . Bulk-FRET measurements of NonEQ DF-6 , 1dsDNA ( right panel ) . The relative fluorescence lifetime ( taudonor ( enzyme ) /taudonor ) OF DF-6 , 1dsDNA at different FEN1 concentrations without acceptors ( right panel ) , showing no effect on the donor fluorescence intensity upon FEN1 binding in the internal-labeling scheme . ( D ) Burst confocal-smFRET histograms from freely diffusing substrate missing the 3’flap but containing the 5’flap ( SF6 , 0dsDNA ) ( 0 . 5 nM ) in solution acquired at sub-ms temporal resolution ( upper panel ) . Kd-bending ( lower panel ) calculated as described in Figure 1—figure supplement 1A . DOI: http://dx . doi . org/10 . 7554/eLife . 21884 . 00710 . 7554/eLife . 21884 . 008Figure 1—figure supplement 5 . Active bending of nicked DNA by EXO1 . TIRF-smFRET histograms of NickdsDNA alone and in the presence of EXO1 ( left panel ) and Kd-bending ( right panel ) . Kd-bending is calculated as described in Figure 1—figure supplement 1A . The donor and acceptor in NickdsDNA are at identical positions to those in NonEQ DF6 , 1dsDNA . DOI: http://dx . doi . org/10 . 7554/eLife . 21884 . 008 Flap endonuclease 1 ( FEN1 ) and its substrate and product complexes provide a prototypic system for unveiling the extreme catalytic selectivity of structure-specific 5’nucleases . Whereas sequence-based specificity partially explains the secret of replication fidelity , key information is missing about the basis for the structure-based excision required at more than 10 million Okazaki fragment sites during human DNA replication . Strikingly , FEN1 maintains exquisite specificity with extreme efficiency that enhances the hydrolysis rate of target phosphodiester bonds by ~1017 , and in vitro reaction rates resemble those of enzyme-substrate encounters ( Finger et al . , 2009 ) . FEN1 recognizes dsDNA bearing a double-flap ( DF ) nick junction consisting of short ssDNA or ssRNA 5’flaps and strictly one nucleotide ( nt ) ssDNA 3’flap ( Figure 1A ) ( Finger et al . , 2009; Kao et al . , 2002 ) . DF intermediates are produced during Okazaki fragment synthesis on the lagging strand and during long-patch base excision repair ( Garg et al . , 2004; Liu et al . , 2005 ) . Mutations that reduce FEN1 expression or alter its activity are linked to cancers and genetic diseases ( Balakrishnan and Bambara , 2013; Henneke et al . , 2003; Kucherlapati et al . , 2002; Zheng et al . , 2007 , 2011 ) . In cells , the 5’flap is complementary to the template strand , enabling the junction to equilibrate and form a single nt 3’flap ( Figure 1A ) . Upon 5’flap cleavage , the 3’flap complements the newly unpaired template base to create a DNA ligase 1 sealable nick ( Figure 1A ) . FEN1 contacts duplex DNA from both sides of the flap junction through a 100° bend stabilized by the interaction of the superfamily-conserved hydrophobic wedge with the junction ( Figure 1B ) ( Tsutakawa et al . , 2011 ) . A superfamily-conserved helical gateway covered by a unique FEN1 helical cap forms a narrow cavity at the DNA junction . This gateway is suitable to select for threading ss 5’flaps with a free end ( Figure 1B ) ( Gloor et al . , 2010; Patel et al . , 2012; Sobhy et al . , 2013; Tsutakawa et al . , 2011 ) . Alternatively , clamping the 5’flap away from the active site is a proposed selection mechanism ( Orans et al . , 2011 ) . A small cavity makes contacts with the 3’flap and may impose specificity for the single nt 3’flap ( Figure 1B ) . Part of the cap-helical gateway , which contains catalytically indispensable residues , and the 3’flap-binding pocket appear disordered without DNA ( Sakurai et al . , 2005 ) , but it is ordered when bound to the 3’flap ( Figure 1B ) ( Chapados et al . , 2004; Tsutakawa et al . , 2011 ) , suggesting DNA-induced ordering of the cap-helical gateway . Comparison of substrate and product complexes shows that the scissile phosphate nucleotide is fully paired at the nick junction and away from the active site’s metal ions , suggesting that unpairing flanking nucleotides may move ssDNA into the active site ( Tsutakawa et al . , 2011 ) . By building upon these strong static structural and ensemble biochemical results , we reasoned that single-molecule experiments could resolve mechanistic unknowns by deconvoluting DNA bending , protein disorder-to-order transitioning , active-site assembly and incision . Like other 5’nucleases , FEN1 displays maximum catalytic efficiency for its cognate substrate but it is only residually active on substrates that vary only slightly ( Finger et al . , 2009; Kao et al . , 2002 ) . To define the mechanism for this catalytic efficiency and selectivity , we used single-molecule ( sm ) FRET at a millisecond to sub-millisecond temporal resolution to simultaneously measure in real time DNA conformational changes and catalysis when FEN1 encounters cognate or noncognate flap substrates as well as when the disorder-to-order transition or the active site is perturbed . A major question in DNA damage recognition is whether the DNA distortion observed in protein−DNA complexes occurs spontaneously and is captured by the protein ( termed conformational selection ) or if the protein actively sculpts the DNA into the distorted conformation ( Figure 1C ) . To determine which is the case for FEN1 , we started by establishing the conformational state of DF substrates alone using an ideal non-equilibrated ( NonEQ ) DF substrate containing 6 nt ssDNA 5’flap and 1 nt ssDNA 3’flap primers with no complementarity with the template strand ( NonEQ DF-6 , 1 ) . DNA bending was monitored by placing an Alexa Fluor-647 acceptor 12 nt into the upstream dsDNA and a Cy3 donor at the 5’flap end ( NonEQ DF-6 , 1Flap ) ( Figure 1D ) or 15 nt into the downstream dsDNA ( NonEQ DF-6 , 1dsDNA ) ( Figure 1E ) . The substrates contained biotin to allow for immobilization on a polyethylene glycol-coated coverslip via biotin-NeutrAvidin linkage ( Figure 1D , E ) . The experiments were performed using custom-built setups operating in either the total internal reflection fluorescence ( TIRF ) mode at a standard temporal resolution of 100 ms ( Sobhy et al . , 2011 ) as the primary method or the confocal mode for higher temporal resolution . The single-molecule time traces of the substrate alone showed a single FRET state with no transition from this state ( Figure 1D , E ) ; FRET efficiency histograms generated from multiple single molecules fit to a single Gaussian ( Figure 1—figure supplement 1A , B ) . The conformer of the duplex arms of the substrate was insensitive to variation in the concentration of divalent metal ions ( Mg2+ ) ( Figure 1—figure supplement 2A , B ) , type of metal ion ( Mg2+ versus Ca2+ ) ( Figure 1—figure supplement 1A , B and Figure 1—figure supplement 2A , B ) or 5’flap length ( Figure 1—figure supplement 2A , B ) . The flap-labeling scheme was sensitive to variation in the Mg2+ ion concentration when the 5’flap length exceeded 6 nt ( Figure 1—figure supplement 2B ) . This could explain why previous work suggested that a double-flap substrate is a dynamic structure ( Craggs et al . , 2014 ) . To test for short-lived alternative conformers , we used confocal-based smFRET to increase the temporal resolution to 5 ms on surface-immobilized DNA and to sub-ms on freely diffusing DNA in solution . Importantly , we found that substrate remained as a single conformer ( Figure 1—figure supplement 2C , D ) . Potential mean force molecular dynamics ( MD ) simulations showed that extended DNA ( ~165° ) was the most energetically favorable conformer in DF-6 , 1 ( Figure 1F and Figure 1—figure supplement 3A , B ) . Base stacking at the nick junction and between the 3’flap base and the first base on the 5’flap stabilized this extended conformer ( Figure 1—figure supplement 3B ) . The energetic cost required to bend the DNA up to ~140° was low ( Figure 1F ) and similar to that in dsDNA ( Sharma et al . , 2013 ) . This was followed by a rapid increase in the energy required to surpass a significant barrier of ~14 kcal/mole to break the base stacking and bend the DNA ( Figure 1F ) . These data suggest that DF substrate remains in an extended form that must be actively bent . In fact , adding FEN1 to DNA NonEQ DF-6 , 1 in both labeling schemes showed transitions to the bent states in a single step to form a stable FEN1−DNAbent complex that rarely dissociated during our 60 s standard acquisition time ( Figure 1D , E ) . We calculated the DNA bending dissociation constant ( Kd-bending ) from the FRET efficiency histogram-binding isotherm to be 3 . 9 ± 0 . 4 nM and 4 . 6 ± 0 . 6 nM for NonEQ DF-6 , 1Flap and NonEQ DF-6 , 1dsDNA , respectively ( Figure 1—figure supplement 1A , B ) . This dissociation constant agreed with the nM range of Km from bulk cleavage assays ( Figure 1—figure supplement 4A ) ( Finger et al . , 2009 ) and the DNA binding dissociation constant ( Kd-binding ) of FEN1 and NonEQ DF-6 , 1 as determined by surface plasmon resonance ( SPR ) ( Figure 1—figure supplement 4B ) . The change in FRET in both NonEQ DF-6 , 1Flap and NonEQ DF-6 , 1dsDNA was confirmed by time-resolved bulk FRET measurements ( Figure 1—figure supplement 4C ) . To mimic the in vivo junction , we used equilibrated ( EQ ) DF-6 , 1 ( Figure 1A ) . FEN1 actively bent EQ DF-6 , 1dsDNA to a similar extent and similar Kd-bending as DF-NonEQ 6 , 1dsDNA ( Figure 1—figure supplement 1B , C ) . Nonetheless , time traces showed multiple transitions between bent and unbent states ( Figure 1G ) . The reduced stability of the bent conformer in the equilibrated substrate suggests that a bound 3’flap could dissociate from the 3’flap-binding pocket . The dissociated 3’flap in the equilibrated substrate would pair with the template strand before FEN1 could rebind it while in the non-equilibrated substrate it would remain available for rebinding FEN1 . Dwell time analysis of the bent ( τbending ) and unbent ( τunbending ) states at increasing FEN1 concentrations indicated that the apparent first-order rate constant for DNA bending ( kbending = 1/τbending ) increased linearly while that for DNA unbending ( kunbending = 1/τunbending ) remained constant ( Figure 1G ) . This is the trend expected for a 1:1 binding equilibrium where kbending and kunbending correspond to the association and dissociation of FEN1 , respectively . Notably , the second-order association rate constant ( kon-bending ) calculated from the slope of the linear fit of the concentration dependence of kbending was diffusion-limited ( 1 . 4 ± 0 . 03×108 M−1 s−1 ) , and the average value of kunbending ( koff-unbending ) was 0 . 45 ± 0 . 05 s−1 ( Figure 1G ) . It is unclear what caused the much higher Kd-bending reported in our earlier work ( Sobhy et al . , 2013 ) , but we suggest that both slower association and faster dissociation rates were influenced . Nonetheless , the FRET states of NonEQ DF-6 , 1Flap alone and when bent by FEN1 and the relative comparison of bending the cognate with the noncognate substrates are similar under low and high Kd-bending conditions as shown below . To see if active bending of the ss/ds-DNA junction may be a conserved feature in 5’nucleases , we tested human mismatch repair exonuclease 1 ( EXO1 ) , which recognizes an ideal junction of either a nick or a 3’ overhang ( Orans et al . , 2011 ) . EXO1 actively bent a DNA nick with diffusion-limited kon-bending ( Figure 1H ) ; the donor and acceptor had identical positions to those in NonEQ DF-6 , 1dsDNA . Free MD simulations showed that the nick behaved similarly to flap substrates for bending angles in the 140°−180° range ( Figure 1—figure supplement 3C ) . The bent conformer of EXO1 had similar FRET to that of FEN1 ( Figure 1—figure supplement 1B and Figure 1—figure supplement 5 ) , consistent with the structures of their DNA complexes ( Orans et al . , 2011; Tsutakawa et al . , 2011 ) . To examine active-site assembly with respect to DNA bending , we replaced Ca2+ with Mg2+ to simultaneously monitor DNA bending and 5’flap cleavage using the flap-labeling scheme ( Figure 2A ) . Time traces indicated that FEN1 always bent NonEQ DF-6 , 1Flap before cleaving the Cy3-containing 5’flap; remarkably every DNA bending event led to a successful cleavage reaction ( Figure 2A , Figure 2—figure supplement 1A ) . We confirmed DNA bending before cleavage by the clear anti-correlated change in the donor and acceptor intensities ( Figure 2A ) . Direct comparison of donor fluorescence in the presence of FEN1 and either Mg2+ or Ca2+ ions indicated that there is a strong correlation between the loss of donor particles and the presence of Mg2+ that coincided with the introduction of FEN1 into the flow cell ( Figure 2—figure supplement 2A , B ) . This confirms that the loss of donor particles is due to 5’flap cleavage and not due to donor photobleaching . Analysis of FRET values before cleavage from individual time traces showed that FEN1 cleaved NonEQ DF-6 , 1Flap from a fully bent state ( Figure 2B ) and remained bent for 160 ± 7 ms prior to cleavage ( Figure 2C ) . FEN1 cleavage generates two products: 5’flap ssDNA and nicked dsDNA ( Figure 1A ) . Previous studies demonstrated that excess nicked dsDNA but not 5’flap ssDNA influences FEN1 activity , which suggests that only nicked dsDNA is a competitive inhibitor of FEN1 release ( Finger et al . , 2009; Tarantino et al . , 2015 ) . Consistent with these findings , we also observed that the lag time before cleavage is not influenced by the presence of excess 5’flap ssDNA ( Figure 2—figure supplement 2C ) . Furthermore , SPR showed only residual transient binding of FEN1 to ssDNA at concentrations that were orders of magnitude above Kd-bending of DF-6 , 1 ( Figure 2—figure supplement 2D ) . Our single-molecule cleavage measurement is not inhibited by lack of 5’flap ssDNA product release . Therefore single-turnover kcat ( kSTO ) could be determined directly from the lag time prior to cleavage ( kSTO = 1/τbefore cleavage ) . However , since 5’flap release would still contribute to the dwell time before cleavage , our single turnover should be treated as an apparent value . Notably , our kSTO ( 6 . 3 ± 0 . 2 s−1 ) was comparable to that determined by bulk cleavage assays ( kSTO 12 . 3 s−1−>5 s−1 ) ( Algasaier et al . , 2016; Stodola and Burgers , 2016 ) , with the slight difference explained by the lower reaction temperature in the single molecule assays . The diffusion-limited rates of DNA bending and cleavage before protein dissociation provide direct evidence that the reaction of FEN1 on a cognate substrate is limited by encounters between the enzyme and the substrate . 10 . 7554/eLife . 21884 . 009Figure 2 . Cleavage of cognate substrate by FEN1 . ( A ) Cleavage of NonEQ DF-6 , 1Flap . Schematic showing the simultaneous monitoring of DNA bending and 5’flap cleavage at the single-molecule level ( upper panel ) . A representative smFRET time trace with a zoomed-in view showing the cleavage of NonEQ DF-6 , 1Flap in which FEN1 never misses the opportunity to bend the DNA and cleave it ( lower panel ) . ( B ) FRET of the bent state before cleavage of NonEQ DF-6 , 1Flap fitted with a Gaussian distribution from multiple cleavage events . ( C ) Dwell times of the bent state prior to cleavage of NonEQ DF-6 , 1Flap fitted with a gamma distribution to calculate average dwell time ( τavg ) from the number of independent experiment N = 3; the uncertainty corresponds to the standard error of the mean . Single turnover kcat ( kSTO ) is = 1/τavg . Cleavage was performed at 50 ms temporal resolution . ( D ) Effect of low molecular weight viscogen ( glycerol ) on disorder-to-order transitioning in FEN1 . Graph showing relative kSTO of NonEQ DF6 , 1Flap cleavage upon addition of glycerol at increasing relative viscosity from N = 2–3 fitted with a linear regression to calculate the slope of the curve; the error corresponds to the standard error of the mean . kSTO was determined as in Figure 2C . ( E ) A representative smFRET time trace showing the cleavage of EQ DF-6 , 1Flap in which FEN1 never misses the opportunity to bend the DNA and cleave it . ( F ) A histogram showing the distribution of dwell times of the bent state prior to cleavage of EQ DF6 , 1Flap . τavg from N = 3 is calculated as in Figure 2C at a temporal resolution of 50 ms . DOI: http://dx . doi . org/10 . 7554/eLife . 21884 . 00910 . 7554/eLife . 21884 . 010Figure 2—figure supplement 1 . Single-molecule cleavage of cognate substrates and probing conformational changes to FEN1 by viscogens . ( A ) Representative TIRF-smFRET time traces showing the cleavage of NonEQ DF6 , 1Flap in which FEN1 always bends and cleaves the DNA . ( B ) Histograms showing the distribution of dwell times of bent state prior to the cleavage of NonEQ DF6 , 1Flap by FEN1 in the presence of varying concentrations of glycerol . For each concentration , the histogram was fitted and τavg was calculated from N = 4 as described in Figure 2C . ( C and D ) Histograms showing the distribution of dwell times of the bent state prior to the cleavage of NonEQ DF6 , 1Flap by FEN1 in the presence of 40% sucrose or 10% PEG-8000 , respectively . The histogram was fitted and τavg was calculated from N = 4 as described in Figure 2C . ( E ) Representative TIRF-smFRET time traces showing the cleavage of NonEQ DF6 , 1Flap in which FEN1 always bends and cleaves the DNA . DOI: http://dx . doi . org/10 . 7554/eLife . 21884 . 01010 . 7554/eLife . 21884 . 011Figure 2—figure supplement 2 . Controls for assigning donor loss to the cleavage and immediate departure of the 5’flap . ( A ) ( Left panel ) bar chart comparing the Cy3 donor signal lost in cognate substrate due to photobleaching ( +Ca2+ ) or to both photobleaching and incision ( +Mg2+ ) in presence of FEN1 . The donor loss increases significantly in the presence of Mg2+ as compared to what would be expected for its loss in the presence of Ca2+ . ( Right panel ) The detailed classification of all single-molecule time traces in the field of view for cognate substrate in the presence of Mg2+ as explained in Materials and methods ( right panel ) . This shows that most of the traces exhibited donor loss instantaneously only upon FEN1 bending , demonstrating that donor loss in the presence of Mg2+ is due 5’flap cleavage by FEN1 . The uncertainty corresponds to the standard deviation between multiple movies in the presence of either CaCl2 or MgCl2 . ( B ) Time-dependent loss of Cy3 molecules under binding ( Ca2+ ) or cleavage ( Mg2+ ) conditions , showing that sudden loss of Cy3 coincides with FEN1 introduction into the focal volume in the case of Mg2+ , while Cy3 loss in the case of Ca2+ is minimal and not abrupt . ( C ) Histograms showing the distribution of dwell times of the bent state prior to the cleavage of NonEQ DF6 , 1Flap by FEN1 in the presence of excess 10 nt ssDNA as a competitor , showing no effect on the cleavage dwell time when compared to cleavage of NonEQ DF6 , 1Flap by FEN1 ( Figure 2C ) . The histogram was fitted and τavg was calculated from N = 2 as described in Figure 2C . ( D ) SPR binding of FEN1 to 22 nt ssDNA , showing only residual binding of FEN1 at extreme concentrations of FEN1 . ( E ) Similar analysis to that described in Figure 2—figure supplement 2A ( left panel ) , showing that most of the loss of the donor molecules in the noncognate substrate SF-6 , 0 resulted from cleavage by FEN1 in the presence of Mg2 . DOI: http://dx . doi . org/10 . 7554/eLife . 21884 . 011 The lag time distribution prior to cleavage shows a rise and decay ( Figure 2C ) , suggesting that the underlying catalytic mechanism after the diffusion-limited DNA bending step involves two or more steps , as a single-step process will have a single exponential decay . We reasoned that these steps likely include 3’flap-induced disorder-to-order transitioning and cleavage chemistry . To test this idea , we employed glycerol and sucrose as low-molecular-weight viscogen to slow any local protein conformational change that mediates catalysis and/or product release . Increasing glycerol concentration decreased kSTO linearly with a slope of 1 . 5 ± 0 . 2 ( Figure 2D and Figure 2—figure supplement 1B ) ; a similar effect was observed with sucrose ( Figure 2—figure supplement 1C ) . Yet , kSTO was unaffected by polyethylene glycol-8000 , a high-molecular-weight viscogen that is too large to interfere with local protein conformational changes ( Figure 2—figure supplement 1D ) . kSTO is not influenced by 5’flap ssDNA product release , suggesting that 3’flap-induced protein ordering is a terminal step to verify the substrate before incision . The shape of the histograms in the presence of viscogen however remains a rise and decay ( Figure 2—figure supplement 1B ) , in contrast to the prediction of collapsing into a single exponential decay should protein-ordering acts in a single rate-limiting step . This suggests that the 3’flap-induced protein ordering is likely involves multistep processes that are being slowed down by the presence of viscogen and/or these multisteps control different rate-limiting steps during catalysis such as DNA unpairing and/or DNA shifting into the active site . Biologically relevant , the cleavage behavior from the first DNA bending and the kSTO were similar whether there was a deliberate mispaired 3’flap ( NonEQ DF-6 , 1; Figure 2A , C and Figure 2—figure supplement 1A ) or an equilibrating 3’flap ( EQ DF-6 , 1; Figure 2E , F and Figure 2—figure supplement 1E ) , consistent with bulk cleavage reactions ( Tarantino et al . , 2015 ) . From bulk measurements , it remains unclear how 3’flap-induced protein ordering operates in the case of the in vivo equilibrated DF substrate . The equilibrated junction may exist as a single 5’flap that requires active molding by FEN1 into a double 5’- and 3’-flap or as a DF with a readily available 3’flap for FEN1 capturing . To address this , we started by investigating the requirement of having a preformed 3’flap for inducing DNA bending . Removal of the 3’flap from NonEQ DF while maintaining its 5’flap ( a substrate termed single flap ( SF ) ) decreased FEN1 cleavage activity by 34 fold ( Finger et al . , 2009 ) . Time traces on surface-immobilized SF-6 , 0Flap accessed at 5 ms using confocal-based smFRET showed that FEN1 actively bent SF ( Figure 3A ) . The τbending was markedly reduced to ~43 ms ( koff-unbending = 23 . 3 ± 3 . 8 s−1 ) in contrast to that of the stable bent conformer in NonEQ DF-6 , 1 ( Figure 3A ) . However , the kon-bending remained limited by diffusion and similar to that of EQ DF-6 , 1dsDNA ( Figure 3A ) . Kd-bending was 50-fold higher than for NonEQ DF-6 , 1 ( Figure 3A ) , consistent with that observed from confocal-based smFRET with burst analysis of freely diffusing SF-6 , 0dsDNA ( Figure 1—figure supplement 4D ) and the markedly increased Kd-binding by SPR ( Figure 1—figure supplement 4B ) . These results show that a 3’flap is not required for DNA bending but it is critical for DNA binding stability . 10 . 7554/eLife . 21884 . 012Figure 3 . Active sculpting of the 3’end of a nick junction creates a 3’flap and drive protein ordering . ( A ) Confocal-based smFRET time traces of surface-immobilized SF6 , 0Flap alone ( left panel ) and in the presence of FEN1 ( middle panel ) acquired at 5 ms , showing rapid transitions from high to low FRET upon DNA bending . kon-bending and koff-unbending of SF6 , 0Flap by FEN1 ( right panel ) calculated as in Figure 1G . ( B ) FEN1 actively creates a 3’flap at the nick junction of cognate and noncognate substrates . Determining the status of the 3’flap in equilibrated DF and SF junctions by comparing the FRET states of various nick-junction positions in NonEQ DF6 , 1dsDNA , EQ DF6 , 1dsDNA and SF6 , 0dsDNA in the absence or presence of FEN1 . 0 . 5 nM DNA and saturating concentrations of FEN1 were used ( 1000 nM for SF6 , 0dsDNA and its one base pair shift construct and 200 nM for the remaining constructs ) . FRET values were determined by fitting the burst confocal-smFRET histograms from freely diffusing DNA at sub-ms temporal resolution with a Gaussian . FRET is reported as a percentage , and the uncertainty corresponds to the standard deviation of N = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 21884 . 01210 . 7554/eLife . 21884 . 013Figure 3—figure supplement 1 . All histograms corresponding to the data shown in Figure 3B . Histograms were fit with Gaussian distribution to get the mean of the distribution . Uncertainties correspond to the error of the fit . DOI: http://dx . doi . org/10 . 7554/eLife . 21884 . 013 We next established whether the equilibrated junction existed as a SF or DF by comparing the FRET states of substrates containing only 5’flap , nonequilibrated 5’- and 3’-flap , or equilibrated 5’- and 3’-flap . Interestingly , we found that the FRET values of the substrate alone in EQ DF-6 , 1dsDNA and in SF-6 , 0dsDNA were similar ( E ~ 0 . 27 ) , but less than that in NonEQ DF-6 , 1dsDNA ( E ~ 0 . 34 ) ( Figure 3B; Figure 3—figure supplement 1 ) . The geometry of the equilibrated DF and SF was slightly less extended than that of dsDNA ( E ~ 0 . 23 ) ( Figure 3B; Figure 3—figure supplement 1 ) . This suggests that the equilibrated junction existed as a SF , which was shown by free MD simulations not to equilibrate to a DF ( Figure 1—figure supplement 3D , E ) . The finding that FEN1 cleaves equilibrated and non-equilibrated DF substrates with equal activity ( Tarantino et al . , 2015 ) prompted us to propose that FEN1 actively converts the equilibrated junction from a SF to a DF by pulling out its 3’end to create a 3’flap . The nick position relative to the donor and acceptor would differ by one base pair if EQ DF-6 , 1dsDNA existed in a SF form compared to that in NonEQ DF-6 , 1dsDNA ( Figure 3B; Figure 3—figure supplement 1 ) . Creating a 3’flap would thus move the junction back to the identical position of that in NonEQ DF-6 , 1dsDNA ( Figure 3B; Figure 3—figure supplement 1 ) . In a control experiment , we showed that we were able to detect junction movement by one base pair because a deliberate difference of one base pair in the nick position of EQ DF-6 , 1dsDNA would result in a detectable difference in the FRET value of their bent conformers ( E ~ 0 . 46 versus E ~ 0 . 54 ) ( Figure 3B; Figure 3—figure supplement 1 ) . The bent conformer in EQ DF-6 , 1dsDNA had identical FRET to that in NonEQ DF-6 , 1 ( E ~ 0 . 54 ) ( Figure 3B; Figure 3—figure supplement 1 ) , demonstrating that the equilibrated nick junction must have moved by one base pair . These data therefore suggest that FEN1 actively sculpts the 3’end of its in vivo equilibrated nick junction to create a 3’flap and to drive its ordering . The active DNA bending and its subsequent use to induce protein ordering through active formation of a 3’flap suggest that there is an induced-fit mechanism between FEN1 and DNA that functions in a mutual way . The ability of FEN1 to actively create a 3’flap at the nick junction of its equilibrated DF substrate raises the possibility that it could also create a 3’flap at nick junctions of noncognate substrates . This mechanism would explain why the cleavage site is shifted by 1 nt in SF versus DF substrates ( Finger et al . , 2009 ) . We found that the FRET value of the bent conformer in SF-6 , 0dsDNA was also similar to that of NonEQ DF-6 , 1dsDNA ( Figure 3B; Figure 3—figure supplement 1 ) , demonstrating that the nick junction must have moved by one base pair . In a control experiment we showed that a shift of one base pair in the nick position in SF-6 , 0dsDNA resulted in a detectable difference in the FRET of the bent conformer as observed in the case of EQ DF-6 , 1dsDNA ( Figure 3B; Figure 3—figure supplement 1 ) . This indicates that FEN1 creates 3’flaps at noncognate nick junctions , suggesting that there is another requirement during substrate validation . In an in vivo-equilibrating junction , the nick structure would be maintained , while in noncognate substrates , a one-nucleotide mismatch would be added at the junction ( Figure 3B; Figure 3—figure supplement 1 ) . FEN1 discriminates against such a structure with 33-fold reduced activity ( Beddows et al . , 2012 ) . Here , the Kd-bending of DF-6 , 1 containing a one-nucleotide mismatch at the junction ( termed DF-7 , 1mismatch ( 1nt ) -Flap ) increased by seven fold ( Figure 4A ) , with the time traces showing a less-stable bent conformer ( Figure 4B and Figure 4—figure supplement 1A ) . Since FEN1 forms 3’flaps for both cognate and noncognate substrates , only the junctions that are fully paired are therefore stably bent . 10 . 7554/eLife . 21884 . 014Figure 4 . Verification of the bent DNA conformer by the 3’flap-induced protein ordering . ( A ) Bar chart comparing Kd-bending for FEN1-WT or FEN1-R47A on various non-equilibrating flap substrates using the internal labeling scheme . Used noncognate substrates include SF-6 , 0 , DF containing 1 nt mismatch at the nick junction ( DF-7 , 1mismatch ( 1nt ) ) , DF containing biotin-NeutrAvidin on the 5’flap to block 5’flap threading ( DF-30 , 1blocked ) and its SF version ( SF-30 , 0blocked ) , and DF containing 2 nt 3’flap ( DF-6 , 2 ) . ( B ) Bar chart comparing koff-unbending for FEN1-WT or FEN1-R47A on various non-equilibrating flap substrates using the internal labeling scheme . The lower estimate of koff-unbending for FEN1-WT on DF-6 , 1 corresponds to the 60 s acquisition time where transitions were rarely detected . Kd-bending and koff-unbending are calculated as in Figure 1—figure supplement 1A and Figure 1G , respectively . koff-unbending was determined from multiple FEN1 concentrations except for FEN1-R47A on SF-6 , 0 and FEN1 on DF-7 , 1mismatch ( 1nt ) , which were determined from two and one concentration , respectively . The smFRET technique and temporal resolutions used in Figure 4A , B are described in Figure 4—figure supplement 1 . ( C ) R47 acts as a sensor that couples structuring of the 3’flap-binding pocket and the cap-helical gateway . R47 in the hydrophobic wedge mediates multiple interactions , where it stacks against the first base pair on the 3’flap side of the junction while its side chain C-caps the α2 in the gateway ( highlighted in green ) and stacks with K128 on α5 in the cap ( highlighted in purple ) ( 3Q8L . pdb ) ( Tsutakawa et al . , 2011 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21884 . 01410 . 7554/eLife . 21884 . 015Figure 4—figure supplement 1 . Bending kinetics of various noncognate substrates . ( A ) TIRF-smFRET time traces of DF-7 , 1mismatch ( 1nt ) -Flap alone and in the presence of FEN1 . ( B ) Surface-immobilized confocal-smFRET time traces of NonEQ DF-6 , 1Flap in the presence of FEN1-R47A at 5 ms temporal resolution . ( C ) The effect of blocking 5’flap threading on DNA bending by FEN1 . Schematic showing the strategy used to block 5’flap threading into the cap-helical gateway by introducing NeutrAvidin/biotin linkage at the 5’end of the 5’flap of DF-30 , 1 ( termed of DF-30 , 1blocked-dsDNA ) prior to the addition of FEN1 ( upper panel ) . Surface-immobilized confocal-smFRET time traces of DF-30 , 1blocked-dsDNA alone and in the presence of FEN1 at 5 ms temporal resolution ( lower panel ) . The substrate was immobilized by surface-coated-NeutrAvidin via the biotin group on the 5’flap . ( D ) A bar chart comparing final bent FRET states of DF-30 , 1trapped-dsNDA , DF-30 , 1blocked-dsDNA , SF-30 , 0trapped-dsDNA and SF-30 , 0blocked-dsDNA using burst confocal-smFRET histograms from freely diffusing substrates acquired at sub-ms temporal resolution . FEN1 concentrations were 5000 nM for SF-30 , 0internal-blocked , 1000 nM for DF-30 , 1internal-blocked , 1000 nM for SF-30 , 0internal-trapped and 200 nM for DF-30 , 1internal-trapped . To trap a threaded 5’flap , FEN1 was first pre-incubated with the substrate before NeutrAvidin was added to bind the biotin on the 5’flap ( as shown in the schematic in the upper panel ) . The FRET value in each condition represents the average of N = 3 and the uncertainty corresponds to the standard error of their fits . ( E ) TIRF-smFRET time traces of DF6 , 2Flap alone and in the presence of FEN1 . DOI: http://dx . doi . org/10 . 7554/eLife . 21884 . 015 The combined requirement for 3’flap and base pairing at the junction suggests that signaling occurs via ordering from the 3’flap-binding pocket to the distant gateway where the 5’flap is recognized and cleaved . The superfamily semi-conserved R47 in the hydrophobic wedge is poised to mediate this coordination: it stacks against the first base pair on the 3’flap side of the junction while its side chain C-caps the α2 in the gateway and stacks with K128 on α5 in the cap ( Tsutakawa et al . , 2011 ) ( Figure 4C ) . Mutating R47 to A ( FEN1-R47A ) disabled FEN1’s cleavage on DF substrate to a similar extent as wild-type ( wt ) -FEN1’s cleavage on SF-6 , 0 ( Tsutakawa et al . , 2011 ) . To test this allosteric signaling idea , we maintained the 3’flap binding using NonEQ DF-6 , 1dsDNA while altering R47 using FEN1-R47A . The defects in Kd-bending and koff-unbending resemble those in wt-FEN1 on SF-6 , 0 ( Figure 4A , B and Figure 4—figure supplement 1B ) . We determined that FEN1-R47A engaged the 3’flap because Kd-bending and koff-unbending increased when SF-6 , 0dsDNA was used rather than NonEQ DF-6 , 1dsDNA ( Figure 4A , B ) . We next investigated the communication between the 3’flap-induced protein ordering and the distant gateway with respect to 5’flap recognition . It has been postulated that the 5’flap may thread through the cap-helical gateway and that this threading is needed for catalysis ( Gloor et al . , 2010; Patel et al . , 2012; Sobhy et al . , 2013 ) or that the 5’flap may be clamped away from the active site for catalysis ( Orans et al . , 2011 ) . To test possible threading and its coordination with the 3’flap-induced protein ordering , we used a modification that prevented 5’flap threading . Blocking the threading by immobilizing a DF substrate through a biotin attached at the end of a 30 nt ssDNA 5’flap ( termed DF-30 , 1blocked-dsDNA ) impaired DNA bending to comparable Kd-bending and koff-unbending of SF-6 , 0 ( Figure 4A , B and Figure 4—figure supplement 1C ) , consistent with the markedly reduced Kd-binding captured by SPR ( Figure 1—figure supplement 4B ) . Notably , the bent DNA in the unthreaded substrate was distorted , but it did not reach the same final FRET state as when the 5’flap was not blocked ( Figure 4—figure supplement 1D ) . This significant DNA distortion was masked in our previous experiment that relied on a flap labeling scheme to infer to the geometry of the blocked-threaded complex ( Sobhy et al . , 2013 ) . Importantly , our new results indicate that initial DNA bending by FEN1 did not require threading , but that full bending required 5’flaps , if present , to be able to thread . The increased Kd-bending of the unthreaded substrate upon removal of its 3’flap ( SF-30 , 0blocked-dsDNA ) ( Figure 4A ) indicates that 3’flap binding did not require 5’flap threading . However , the ability of the 5’flap to thread is required for the 3’flap-induced protein ordering to form the stably and correctly bent DNA conformer . Collectively , these results demonstrate that FEN1 bends both cognate and noncognate substrates and that Kd-bending is higher for noncognate substrates . This is consistent with our previous findings under high Kd-bending conditions ( Sobhy et al . , 2013 ) . They further showed that FEN1 stabilizes the cognate substrate through remarkable selectivity for its key features of a fully paired nick junction , a 3’flap and a 5’flap while promoting the dissociation of noncognate substrates . Our observation of FEN1’s ability to significantly bend the DNA in the blocked-threaded complex challenges our previous conclusion that 5’flap threading is strictly required to induce DNA bending ( Sobhy et al . , 2013 ) . Our new results are consistent with a model in which FEN1 actively bends DNA to interact with the ss/ds-DNA junctions and subsequently verifies these interactions by the 3’flap-induced protein ordering . To test the mechanism for assembly of catalytically competent active sites for cognate substrate incision , we compared the lifetime of the bent conformer to the lag time for cleaving correct versus incorrect substrates . In SF-6 , 0 , the lifetime of the bent conformer was ~3 . 5 fold shorter than the required lag time prior to cleaving the cognate substrate ( Figure 3A and Figure 2C ) . Traces of single-molecule cleavage showed that SF-6 , 0 underwent multiple cycles of DNA bending and unbending before a successful DNA bending event led to 5’flap cleavage ( Figure 5A; Figure 5—figure supplement 1A; Figure 2—figure supplement 2E ) . These abortive DNA bending events are masked in bulk cleavage assays , which leads to underestimation of both kSTO and the accuracy of FEN1 cleavage . Following the FEN1 cleavage reaction at the single-molecule level clearly leads to additional information . Similar results were observed in the cleavage of DF-6 , 1 by FEN1-R47A ( Figure 5B and Figure 5—figure supplement 1B ) . These results show that destabilizing the bent DNA intermediate to rates that are limiting for catalysis reduces the probability of assembling catalytically competent active sites . 10 . 7554/eLife . 21884 . 016Figure 5 . Cleavage of noncognate substrates by FEN1 . ( A , B , C and D ) Representative smFRET time traces showing the cleavage of FEN1 on SF6 , 0Flap , FEN1-R47A on NonEQ DF6 , 1Flap , FEN1 on DF-7 , 1mismatch ( 1nt ) -Flap and FEN1 on DF-6 , 2Flap , respectively and the distribution of dwell times of the bent state prior to cleavage ( τavg ) . Cleavage occurs after a random number of missed cleavage opportunities from a bent conformer as illustrated in blue arrows in Figure 5A . τavg is calculated from N = 4 as in Figure 2C at a temporal resolution of 50 ms . DOI: http://dx . doi . org/10 . 7554/eLife . 21884 . 01610 . 7554/eLife . 21884 . 017Figure 5—figure supplement 1 . FEN1 cleavage of various noncognate substrates . ( A ) Representative TIRF-smFRET time traces showing the cleavage of SF6 , 0Flap in which FEN1 cleaves the DNA after multiple unsuccessful bending events . ( B ) Representative TIRF-smFRET time traces showing the cleavage of NonEQ DF6 , 1Flap by FEN1-R47A in which FEN1-R47A cleaves the DNA after multiple unsuccessful bending events . ( C ) Representative TIRF-smFRET time traces showing the cleavage of DF6 , 2Flap in which FEN1 cleaves the DNA after multiple unsuccessful bending events . ( D ) Representative TIRF-smFRET time traces showing the cleavage of DF7 , 1mismatch ( 1nt ) -Flap in which FEN1 cleaves the DNA after multiple unsuccessful bending events . ( E ) Representative TIRF-smFRET time traces showing the cleavage of NonEQ DF6 , 1Flap by FEN1-Y40A in which FEN1-Y40A cleaves the DNA after multiple unsuccessful bending events . ( F ) Bar chart comparing koff-unbending of the DNA bending events that did not lead to 5’flap cleavage in presence of Mg2+ in the cases of FEN1-WT on noncognate substrates and FEN1-Y40A and FEN1-R47A on NonEQ DF-6 , 1 . The koff-unbending values were similar to those reported under no cleavage condition in the presence of Ca2+ that are presented in Figure 4B . The slight decrease in rates in the cases of FEN1-WT on SF-6 , 0 and FEN1-R47A on NonEQ DF-6 , 1 in the presence of Mg2+ relative to Ca2+ is a result of averaging due to the lower temporal resolution in TIRF-based smFRET ( Mg2+ case ) versus immobilized confocal-based smFRET ( Ca2+ case ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21884 . 017 Interestingly , we also observed similar bending behavior without cleavage of FEN1 on noncognate substrates under conditions in which the lifetime of the bent conformer does not limit catalysis . DF-7 , 1mismatch ( 1nt ) -Flap and DF containing 2 nt 3’flap ( DF-6 , 2Flap ) exhibited koff-unbending that was ~13–15 fold slower than that of SF-6 , 0 and ~3–4 fold longer than kSTO of the cognate substrate ( Figure 4B and Figure 2C ) , yet FEN1 still bent these substrates multiple times without cleaving them ( Figure 5C , D and Figure 5—figure supplement 1C , D ) . FEN1 therefore likely has intrinsic mechanisms that block the probable formation of catalytically competent active sites in noncognate substrates to inhibit off-target incision . We reasoned that there are two possible mechanisms for controlling incision that can be tested experimentally . The 3’flap-induced protein ordering could act once per DNA bending event , locking the DNA into either a catalytically competent or incompetent conformation . In this mechanism , the kSTO after DNA bending should be similar between cognate and noncognate substrates regardless of whether or not the lifetime of the bent conformer is limiting . Alternatively , the protein could lock the DNA into a bent conformer and go through multiple cycles of disorder-to-order transitioning to search for a catalytically competent conformation of protein and DNA . In this mechanism , the kSTO would be slower for noncognate substrates , particularly under conditions when the lifetime of the bent conformer exceeded that required for cleavage . We found that the kSTO of FEN1 in all tested noncognate substrates was similar and comparable to that in the cognate substrate ( Figure 5A–D and Figure 2C ) . This indicates that the 3’flap-induced protein ordering locked the DNA into either a catalytically competent conformation to be immediately incised or into an incompetent conformation that led to immediate DNA release from the bent conformation . Directly observing FEN1 conformation will lead to further understanding on how it prevents cleavage in noncognate substrates . Given that FEN1 actively bends the DNA , it is unclear how FEN1 positions the junction before it locks down the DNA conformation . We therefore tested the role of key active-site residues . Mutating individual gateway residues Y40 , K93 , R100 or one of the metal-coordinating aspartic acidic residues ( D181 ) to alanine markedly reduced the bent conformer’s stability ( Figure 6A ) . These results revealed a direct role for active-site residues in stabilizing the bent DNA conformer . Interestingly , kon-bending was reduced up to 11 fold in gateway mutants , with R100 and to a lesser extent Y40 being critical residues ( Figure 6B ) . This result is surprising , given that these gateway residues appear to be disordered prior to DNA binding . Thus , active-site residues evidently contribute to active positioning of the junction while the 5’flap is being threaded through the unstructured cap-helical gateway . Moreover , D181A had only a minor effect on kon-bending , implying that the metal ions do not interact with the phosphates during DNA bending and 5’flap threading . Importantly , this could provide a mechanism that protects the 5’flap from nonspecific cleavage . 10 . 7554/eLife . 21884 . 018Figure 6 . Role of active-site residues in active positioning of the 5’flap and the junction . ( A ) Bar chart comparing koff-unbending of WT-FEN1 and the FEN1 active-site mutants Y40A , K93A , R100A and D181A on NonEQ DF6 , 1Flap . koff-unbending is calculated as described in 1G . ( B ) Bar chart comparing kon-bending of wild-type FEN1 on EQ DF6 , 1Flap and the FEN1 mutants Y40A , K93A , R100A and D181A on NonEQ DF6 , 1Flap . kon-bending is calculated as described in Figure 1G . ( C ) A representative smFRET time trace showing the cleavage of NonEQ DF6 , 1Flap by FEN1-Y40A after multiple trials of DNA bending ( left panel ) and a histogram showing the distribution of dwell times of the bent state prior to cleavage ( right panel ) . τavg from N = 4 is calculated as in Figure 2C at a temporal resolution of 50 ms . ( D ) Model for control of catalytic selectivity by the DNA mutual-induced fit mechanism in FEN1 . DNA sculpting: FEN1 actively bend a variety of structures to verify the key features of its cognate DF substrates of fully paired ss/dsDNA nick junction , threaded 5’flap into the cap-helical gateway and 3’flap . Protein ordering: FEN1 actively pulls the 3’end of the nick junction to create a 3’flap and drive protein ordering , which in turn orders the active site and locks the DNA conformation . Decision: the active site and locked DNA conformer are always in catalytically competent form in cognate substrate , while they are primarily in catalytically incompetent form in noncognate substrates ( no 5’flap threading , no 3’flap , mispair junctions ) and FEN1 mutants ( R47A , K93A and R100A ) . DNA release or catalysis: the DNA will shift or unpair to move the scissile phosphate into the active site for cleavage as probed by the flap/junction positioning-residue Y40 , while in noncognate substrates FEN1 promotes DNA dissociation . DOI: http://dx . doi . org/10 . 7554/eLife . 21884 . 01810 . 7554/eLife . 21884 . 019Figure 6—figure supplement 1 . Sequences of DNA constructs used in this study . Schematic of the substrates with internal-labeling and flap-labeling schemes used in this study including their sequences . A legend of the symbols used is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 21884 . 019 FEN1-Y40A is the only active-site mutation that retains activity , albeit with a 100-fold reduced kSTO ( Algasaier et al . , 2016 ) . Y40 is proposed to play various roles in substrate positioning , including placing FEN1 at the junction , at the 5’flap in the cap-helical gateway and at the scissile phosphate in the active site ( Algasaier et al . , 2016; Tsutakawa et al . , 2011 ) . We observed that FEN1-Y40A cleaves DF-6 , 1 after multiple cycles of DNA bending and unbending ( Figure 6C , Figure 5—figure supplement 1E ) . This highlights the extreme selectivity of FEN1 to local variation in the positioning of the junction and 5’flap for proper DNA lockdown . Unlike noncognate substrates or FEN1-R47A , the kSTO in FEN1-Y40A increased by about three fold ( Figure 6C ) ; nonetheless , koff-unbending was still about two-fold slower than its kSTO ( Figure 6A , C ) . This indicates the presence of another step after proper DNA lockdown that involves active positioning of the scissile phosphate for incision . Critical cellular processes such as DNA replication and repair are regulated by the molecular properties encoded in interacting macromolecules whereby distinct dynamic conformations correspond to different functional outcomes . The mechanisms for these dynamic changes that occur during macromolecular interactions are the subject of intense interest with two major proposed mechanisms , 'induced fit' and 'conformational selection’ . Findings so far have suggested that many processes are regulated by conformational change before substrate binding by the substrate’s selective binding to the active form of the enzyme , indicating that functional or conformational selection is in play ( Boehr et al . , 2009; Vogt et al . , 2014 ) . However the high precision required for DNA replication and repair has consistently raised issues of whether or not these might involve unusual mechanisms of chemistry or physics . In this context , how DNA-repair enzymes specifically recognize and remove damage in DNA is a decades-long debate . Does the damage destabilize the DNA duplex leading to disruption of the DNA structure ( extrahelical base flipping or DNA bending ) before its subsequent capture by the repair enzymes or do these enzymes actively sculpt the DNA as part of their recognition of the damage ? With the FEN1 single-molecule results , a picture emerges of induced conformational changes to both substrate and protein playing key roles in stabilizing a transition state that has been thoroughly vetted using multiple checks and is poised for catalysis with remarkable specificity ( Figure 6D ) . In this process , FEN1 can differentiate between substrates whose incision is good for the cell ( cognate ) or toxic ( noncognate ) even if these substrates have small differences in their binding affinities . Active DNA bending does not create a significant energy barrier as evident by diffusion-limited on-rates in FEN1 and EXO1 . We propose that different members of the 5’nuclease family share similar DNA-bending-induced disorder-to-order transitioning but differ in the mechanisms that couple this transitioning with active-site assembly . In FEN1 , the coupling of protein transitions of the 3’flap-binding pocket and the 5’flap-binding helical gateway with DNA sculpting uncovers how dynamic protein segments are critical contributors to substrate binding and catalytic selection . As part of the active DNA sculpting process , we observed single-molecule measurements consistent with a mutual induced-fit mechanism , with the protein bending the DNA and the bent DNA inducing a protein-conformational change ( Figure 6D ) . Substrate distortion by ~90° and DNA-induced conformational changes in proteins are features that extend beyond 5’nucleases ( Gwon et al . , 2014a , 2014b; Yan et al . , 2015 ) . More generally , the FEN1-type induced-fit mechanism may be central to detecting chemically subtle but biologically critical differences between correct and incorrect substrates for multiple DNA and RNA processes . Here , nuclease precision in replication comes from induced fit that regulates the compatibility of the distorted DNA conformer with active-site assembly and its off rate to allow cleavage of cognate but not noncognate substrates ( Figure 6D ) . The diffusion-limited bending of the cognate substrate by FEN1 and its cleavage from the first encounter represents a practically perfect precision reaction whose rate is limited by diffusion . The stochastic cleavage behavior of noncognate substrates after multiple cycles of DNA bending and dissociation of FEN1 has a fundamental bearing on how enzyme specificity is understood . In cases in which an enzyme encounters noncognate substrates and cleaves them after multiple trials , results on substrate specificity from classical biochemical techniques that monitor product formation become misleading . Furthermore , multiple attempts to cleave noncognate substrates are likely to be insignificant inside the cell . These findings advance our insight into a previously unidentified mechanism in structure-specific nucleases for extreme specificity towards their cognate substrates inside the cell . The 3’flap binding pocket is distant from the active site , raising the question why FEN1 , along with some other structure-specific nucleases , utilizes long-range DNA-induced conformational coupling , in contrast with local coupling as observed in EXO1 ( Gwon et al . , 2014a , 2014b; Orans et al . , 2011; Sakurai et al . , 2005; Tsutakawa et al . , 2011; Yan et al . , 2015 ) . FEN1 cleaves 5’flaps containing RNA , DNA or mismatches of various lengths . We anticipate that long-range DNA-induced conformational coupling could provide a mechanism that enhances flexibility in nuclease substrates . The action of 3’flap-induced protein ordering as a key step that locks the FEN1 interaction with the junction could provide the advantage of limiting the sampling time between the disordered protein form and noncognate substrates that could otherwise lead to nonspecific cleavage . This active site control via a long-range induced-fit mechanism suggests why mutations distant from the FEN1 active site have a dramatic effect on genomic stability and disease states ( Sun et al . , 2017; Zheng et al . , 2007 ) . Thus , evolutionary selection against toxic and mutagenic DNA instability may have developed the unusual DNA-induced conformational coupling seen in FEN1 as a previously unrecognized part of repair and replication nuclease fidelity . Human FEN1 ( amino acids: 2–380 ) was cloned into a pE-SumoPro expression vector ( Lifesensors ) , which encodes an N-terminal His6-Tag followed by SUMO protein . This clone was used for recombinant expression FEN1 in BL21 ( DE3 ) E . coli strain . FEN1 was purified by multiple chromatographic steps involving two sequential Ni-NTA columns interspersed by sumo protease cleavage . Purity was further increased by running FEN1 on heparin column and Hioload superdex-75 gel filtration column . FEN1 was dialyzed against buffer ( 50 mM Tris-HCl pH 7 . 5 , 50% glycerol , 300 mM NaCl , 10 mM BME ) , flash frozen and stored at −80°C . Human EXO1 catalytic domain ( amino acids: 2–352; referred to as EXO1 in this study ) was cloned into a pE-SumoPro expression vector ( Lifesensors ) . EXO1 was expressed and purified using a similar protocol to that described for FEN1 . The purified EXO1 was dialyzed against buffer ( 50 mM Tris-HCl pH 7 . 5 , 50% glycerol , 300 mM NaCl , 10 mM BME ) , flash frozen and stored at −80°C . DNA oligos , modified and unmodified , were purchased from Integrated DNA Technologies ( IDT ) . All sequences used to make the substrate structures are shown in Figure 6—figure supplement 1 . To prepare the DNA substrates , we mixed equimolar concentrations of oligos in TE-100 and annealed by heating to 95°C followed by slow cooling to room temperature . Properly annealed substrates were purified on non-denaturing PAGE gel and extracted using the crush and soak method followed by ethanol precipitation . Experiments were performed in a microfluidic flow chamber made by sandwiching a polyethylene spacer ( 100 μm thick polyethylene double-sided spacer SA-S-1L from Grace Biolabs ) between a quartz slide and a glass coverslip with inlet and outlet tubing channels . The glass coverslip was functionalized and passivated by a combination of 1:100 molar ratio of biotinylated polyethylene glycol ( Biotin-PEG-SVA MW 5 , 000 ) and polyethylene glycol ( mPEG-SVA MW 5000 ) ( Laysan Bio Inc . ) . The flow chamber was incubated with NeutrAvidin ( 0 . 2 mg/ml ) just prior to the experiment for 2 min and then washed excessively with reaction buffer ( 50 mM HEPES pH 7 . 5 , 1 mM DTT , 5% glycerol and 0 . 1 mg/mL BSA , 100 mM KCl and 10 mM CaCl2 ) ; in the cleavage assays , CaCl2 was replaced with 10 mM MgCl2 . This was followed by incubation with biotin-labeled DNA substrate ( 100 pM ) until a sufficient surface coverage of fluorescent-labeled substrate was achieved . This was followed by washing with imaging buffer ( described below ) containing the appropriate divalent metal ion , CaCl2 for bending or MgCl2 for bending and cleavage . To minimize the effect of photobleaching and photo-blinking , we used an oxygen scavenging solution as described earlier ( Aitken et al . , 2008 ) leading to the enzymatic removal of oxygen by a 6 mM proto-catechuic acid ( PCA ) ( Sigma-Aldrich , Japan , P5630 ) and 60 nM protocatechuate-3 , 4-dioxygenase ( PCD ) system . Trolox ( Sigma-Aldrich ) was added at 2 mM concentration to reduce the photo-blinking by quenching the triplet state . The imaging buffer contained the reaction buffer and the aforementioned oxygen scavenging solution . The experiments were performed on a custom-built TIRF-FRET setup as described earlier ( Sobhy et al . , 2013 ) . For data analysis , the spatial mapping of the donor and acceptor emission channels was first calibrated using fluorescent beads that were imaged in TIRF mode . This generated a transformation matrix file , which was then used in the subsequent analysis of fluorescent molecules to map the donor and acceptor positions . The fluorescent molecules were registered as Gaussian point spread functions ( PSF ) around the brightest pixel in both channels and aligned with each other using the transformation matrix file . Donor and acceptor intensities were extracted by using software as described previously ( Holden et al . , 2010 ) and the apparent FRET efficiency was subsequently calculated . Any molecules with aberrant emission in brightness were excluded from further analysis . Histograms of FRET efficiencies were obtained from alternating the excitation of donor and acceptor 2c-ALEX as described previously ( Kapanidis et al . , 2005 ) . The vbFRET package implemented in Matlab ( Bronson et al . , 2009 ) was used for dwell time analysis . The association ( τbending ) and disassociation ( τunbending ) dwell times were generated by idealizing and fitting the single molecule traces with two FRET states modeled by vbFRET ( bent and unbent states ) . Histograms that were generated from dwell times in each state were fit with a single exponential decay function to generate kbending ( 1/τunbending ) and kunbending ( 1/τbending ) , respectively . The experiments were performed on a custom-built confocal epifluorescence microscope setup ( Cotlet et al . , 2005 ) . Fluorophores were excited with a 532 nm line of a pulsed laser diode operating at 20 MHz ( 100 ps pulse width , LDH-P-FA-530L , PicoQuant ) or a 50 mW 532 nm Cobolt Samba laser through a microscope objective . A water immersion objective ( UPLSAPO60XW NA 1 . 2 , Olympus ) and an oil immersion objective ( UPLSAPO100XO NA 1 . 4 , Olympus ) were used for solution-phase smFRET and smFRET on surface-immobilized molecules , respectively . A circularly polarized beam was obtained by inserting a Berek compensator ( Mo . no . 5540 , Newport ) in the excitation beam path . The laser beam was made Gaussian and expanded to fill the back aperture of the objective lens before introducing it into the microscope using a spatial filter with a 30 µm pinhole . The laser beam was reflected off the surface of a longpass dichroic Di02 R532−25 × 36 ( Semrock Inc . ) into the objective . The excitation power at the sample plane was set to 400Wcm−2 or 255 Wcm−2 for solution-phase smFRET or smFRET on surface-immobilized molecules respectively . In the detection path , emitted fluorescence passed through the dichroic and was focused onto a 100 µm pinhole by the tube lens of the microscope ( IX71 , Olympus ) and recollimated using a lens . A longpass BLP01-532R-25 filter ( Semrock Inc . ) was used to remove scattered laser light , and then the beam was split into donor and acceptor channels using a dichroic FF635-Di01−25 × 36 ( Semrock Inc . ) . The donor and acceptor paths were equipped with a band-pass FF01-580/60-25-D ( Semrock Inc . ) and a longpass LP02-664RU-25 filter ( Semrock Inc . ) , respectively , before being focused onto single-photon avalanche diodes ( τSPAD , PicoQuant ) . Fluorescence intensity trajectories were recorded by a time-correlated single-photon counting ( TCSPC ) module ( HydraHarp 400 , PicoQuant ) in the time-tagged time-resolved ( TTTR ) mode , which allowed for recording the arrival time of each photon emitted by the fluorophores . The SymPhoTime software ( PicoQuant ) was used for the data acquisition as well as for controlling the excitation lasers and TCSPC module . The solution-phase smFRET experiments were performed in a home-made flow-through chamber by sandwiching a paraffin film spacer ( 0 . 13 mm thick , Bemis Inc . ) between two glass coverslips . The glass coverslips were functionalized and passivated with polyethylene glycol ( mPEG-SCA , MW5000 ) ( Laysan Bio Inc . ) prior to the construction of the flow-through chamber . Samples were prepared by mixing imaging buffer with appropriate dilutions of the stock enzyme and stock DNA solutions . The solution was allowed to flow onto the flow cell by pipetting the solution into one side of the chamber while applying suction to the opposite end . Then , the flow cell was placed on the microscope . In all solution-phase smFRET experiments , the reaction buffer was used with the addition of 10 mM CaCl2 and 2 mM Trolox . In the solution-phase smFRET experiments , the excitation laser was focused approximately 40 µm above the surface of the bottom cover slip . The fluorescence intensity trajectories on the donor and acceptor channels were recorded for 15 min to obtain between 3500–8000 bursts from individual DNA molecules . SymPhoTime script was used to analyze the bursts and generate burst histograms . The intensity trajectories were first binned to 0 . 5 ms , and bursts above 35 total counts were considered for the analysis . The FRET efficiency was calculated by the integrated intensity of each burst in the donor and acceptor channels . OriginPro was used to fit the histograms of the FRET efficiency to Gaussian peaks . The smFRET experiments on surface-immobilized molecules were performed using either the microfluidic flow chamber used in the TIRF-based FRET experiments or pre-made sticky-Slide VI0 . 4 microfluidic chambers ( ibidi GmbH ) , with cover slips identical to those used in the TIRF-based FRET experiments . The DNA substrates were immobilized on the glass cover slips according to the procedures described above . The smFRET experiments were performed in the presence of the oxygen scavenger and the triplet quencher used in the TIRF-based FRET experiments . The excitation laser was focused on the surface of the cover slip using back reflection . Fluorescence intensity trajectories of individual molecules were acquired by first scanning a 10 × 10 µm section of the coverslip using a scanning piezo stage . Then , individual molecules were manually chosen from the image and the trajectories were sequentially acquired , with the laser focus dwelling on each point for 10 s . The SymPhoTime software was used for the image acquisition and stage positioning . Once fluorescence intensity trajectories were acquired , SymPhoTime was used to generate traces by binning the data to either 2 , 5 or 10 ms and then exporting the donor and acceptor counts . A custom-written MATLAB script was used to generate traces from data exported from SymPhoTime and subsequently to select regions before photobleaching ( Harris , 2017 ) a copy is archived at https://github . com/elifesciences-publications/ConfocalFret ) . Then , the FRET efficiency trace was calculated using the intensity trajectories of the donor and acceptor , and the histograms of the FRET efficiency were generated from the selected regions of the traces . Aberrant traces were excluded for further analysis . The selected regions were exported by the MATLAB script into files readable by HaMMy , a software used for analysis of single-molecule FRET trajectories using hidden Markov modeling ( McKinney et al . , 2006 ) . The FRET trajectories were analyzed by a two-state model using HaMMy . Another custom-written MATLAB script was then used to collate the results from HaMMy and generate lists of dwell times for low FRET ( that is , bent ) and high FRET ( that is , unbent ) states . These lists were imported into OriginPro , histogrammed and fitted to a single exponential decay . Cleavage experiments were performed by TIRF-based smFRET at a temporal resolution of either 50 or 100 ms . The surface-immobilized substrate was pre-incubated in the flow chamber with imaging buffer containing 10 mM MgCl2 . In the case of cleavage of NonEQ DF6 , 1Flap and EQ DF6 , 1Flap by wild-type FEN1 , image acquisition started before FEN1 reached the microfluidic chamber . In all other cleavage experiments , imaging started after the protein had reached the microfluidic chamber but before it reached the focal volume . This delay in acquisition was to reduce particle loss due to acceptor photobleaching since the waiting time before cleavage markedly increased under these suboptimal conditions . Selective loss of the donor signal was confirmed by direct excitation of the acceptor at the end of the cleavage experiment . Due to short lag time before cleavage , manual counting of frames in the bent state was used to calculate the cleavage dwell time for each trace . All particles in the field of view were grouped into the following five different categories as shown in Figure 2—figure supplement 2A . ( a ) Molecules with aberrant intensity that suffered from strong noise , photoblinking , step bleaching or deviation from the average intensity . These particles were excluded from further analysis . ( b ) Molecules that had acceptor photobleaching before the loss of the donor signal . These molecules do not influence dwell time analysis , which depends only on the donor signal . They were therefore excluded . ( c ) Molecules that went to the low FRET bent state followed by a single-step loss in the donor signal . This formed the bulk of the traces of cognate substrates . ( d ) Molecules that lost their donor signal without going into the low FRET bent state . This excluded minority population could result from donor photobleaching events and/or 5’flap cleavage that occurs at a faster rate than the acquisition time . ( e ) Molecules that stayed in the unbent high-FRET state within our imaging time . These molecules were also excluded from the cleavage analysis as they exhibit no FEN1 binding . The following criteria were used to perform dwell time analysis for the selected particles . ( a ) A minimum FRET change of 0 . 2 between unbent and bent frames was applied as a filtering criterion before selecting traces for dwell time analysis . ( b ) Each selected trace was checked for anti-correlated behavior between the donor and acceptor upon change in the FRET efficiency . ( c ) With noncognate substrates , the dwell time was calculated by counting the number of frames spent in the lower FRET state before the donor signal was lost in the last bent step . MATLAB was used to calculate the mean of the cleavage dwell time by fitting with gamma distribution function and the error in the mean by bootstrap . In the viscosity measurement , a falling balls viscometer ( Gilmont ) was used to calculate the absolute viscosity . The density of the solution was measured from the mass of 1 ml of the same solution as used to calculate the viscosity . The measurements were performed on a QuantaMaster 800 spectrofluorometer ( Photon Technology International Inc . ) coupled with a supercontinuum fiber laser source . The fluorophore lifetime was determined by the time-correlated single-photon counting ( TCSPC ) method . The excitation was carried out at 535 nm and emission was collected at 568 nm using 5 nm bandwidths for the excitation and emission . A longpass filter with 550 nm cut-on was placed on the emission side to prevent scattered light . The instrument response function ( IRF ) was determined using a colloidal silica suspension . The decay time traces were acquired at 10 , 000 counts . The measurements were performed at room temperature using the same 5’ flap constructs and buffer composition as in the smFRET experiments . The determination of the lifetimes was done using IRF reconvolution and a multi-exponential decay function incorporated in the FluoFit software package ( PicoQuant ) . The donor lifetime curve was fitted to two-exponential decay . The best fit was selected based on reduced chi-square and randomness of the residuals . The FRET efficiency , EFRET , refers to the conformational change resulting from the action of the enzyme on the substrate . EFRET is calculated from the measured lifetime of the donor in the donor-only and donor-acceptor substrates at the respective enzyme concentration using the following Equation:EFRET=1− ( tauDA−EnzymetauD−Enzyme ) where tauD−Enzyme andtauDA−Enzyme are the amplitude-weighted average lifetimes of the donor excited-state in the donor-only and donor-acceptor substrates in the presence of the enzyme , respectively . The dissociation bending constant ( Kd-bending ) was calculated by fitting the data to a standard quadratic equation for simple bimolecular association ( S+F ⇌ SF ) under equilibrium conditions:E=E0+ ( Ef−E0 ) ( ( S+F+Kd−bending ) 2− ( S+F+Kd−bending ) 2−4SF2S ) where E is the FRET value at any protein concentration , E0 and Ef are the initial and final FRET values , and S and F represent the substrate and FEN1 concentrations , respectively . SPR binding was performed on a Biacore T100 ( GE Healthcare Inc . ) . Biotinylated DNA substrates were immobilized on S-series streptavidin sensor chips in HBS-EP buffer according to the manufacturer’s recommendations . The response unit ( RU ) of the immobilized substrate is stated in the figure legends . FEN1 was dialyzed overnight at 4°C against the smFRET reaction buffer containing CaCl2 ( no oxygen-scavenging solution was added ) . Serial dilutions of FEN1 were made using the same reaction buffer . For each concentration , the run started with a surface-regeneration injection of reaction buffer +1 M NaCl at a flow rate of 100 µL/min for 120 s , followed by protein sample injection at a flow rate of 20 µL/s for 90 s for DF-6 , 1 or 20 µL/s for 120 s for the other substrates . The sensorgrams were corrected for bulk refractive index and residual nonspecific binding to the surface using a blank flow cell . The sensorgrams were processed using Biacore T100 Evaluation Software ( GE Healthcare Inc . ) . The maximum RUs reached at each FEN1 concentration were fitted using the steady-state affinity mode to obtain the equilibrium dissociation constant ( Kd-binding ) for each DNA substrate . Reaction mixtures containing 0 . 5 nM Cy5-labeled NonEQ DF-6 , 1dsDNA ( Cy5 was placed 15-nt away from the nick junction on the downstream dsDNA on the 5’flap primer ) in 1X reaction buffer ( 50 mM HEPES-KOH pH 7 . 5 , 100 mM KCl , 0 . 1 mg/ml BSA , 5% ( v/v ) glycerol , 10 mM MgCl2 and 1 mM DTT ) were pre-incubated at 37°C before the initiation of the cleavage reaction with the addition of varying concentrations of FEN1 . Each reaction mixture was incubated further at 37°C and equal aliquots were removed and quenched by equal volumes of 2X denaturing buffer ( 90% deionized formamide , 100 mM EDTA ) at the following time intervals ( 0 , 0 . 17 , 0 . 5 , 1 , 1 . 5 , 2 , 5 , 10 mins ) . These samples were run on 20% denaturing PAGE gels , which were imaged using a Typhoon TRIO Variable Mode Imager ( GE Healthcare , Life Sciences ) . The product formation was quantified using the ImageJ gel analysis tool . For each FEN1 concentration , the concentration of the product formed was plotted against time to estimate the initial rate ( v0 , nM . min−1 ) by taking the slope of the linear part . These v0 values were plotted against the FEN1 concentration and Km was determined by nonlinear least-squares fitting using a Michaelis-Menten model . All simulations were performed with the AMBER 15 . 0 molecular dynamics package using the Parm14 force field with parmbsc0 nucleic acid modifications ( Case et al . , 2015; Maier et al . , 2015 ) . In total , four substrate DNA models ( denoted NonEQ DF-6 , 1 , SF-6 , 0 , EQ DF-6 , 1 and nicked DNA ) were generated . A dsDNA 47-mer with sequence d ( 5’-TGACCGTTGTTTGACGGTCGTGAGGAGGAAAGTTCCTCCTACGGCAG-3’ ) •d’ ( 5’-CTGCCGTAGGAGGAACTTTCCTCCT ( 25 ) C ( 26 ) A ( 27 ) CGACCGTCAAACAACGGTCA-3’ ) , identical to the one used in the smFRET experiments , was first constructed in a canonical B-DNA conformation with the UCSF CHIMERA program ( Pettersen et al . , 2004 ) . The model NonEQ DF-6 , 1 was built by adding 5’ ssDNA flap ( 5’-TTTTTA-3’ ) and a single-nucleotide 3’ flap ( G-3’ ) at the junction of the two DNA duplexes ( between bases C26 and T25 ) . The SF-6 , 0 and EQ DF-6 , 1 models were constructed by adding 5’ flap ssDNA ( 5’-TTTTA-3’ ) at base C26 and 5’ flap ssDNA ( 5’-TTTTAC-3’ ) at base A27 , respectively . Each system was solvated with TIP3P water ( Jorgensen and Jenson , 1998 ) with a minimum distance of 15 . 0 Å from the DNA to the edge of the periodic simulation box . The systems were then neutralized by the addition of Na+ counterions . Additionally , 100 mM NaCl concentration was introduced to mimic physiological conditions . First , the water and ions were subjected to 3000 steps of steepest descent and 1500 steps of conjugate gradient minimization while restraining all DNA atoms with a force constant of 2 kcal/mol Å2 . All restraints were then released . The particle mesh Ewald ( PME ) method ( Darden et al . , 1993 ) was used to treat the long-range electrostatic interactions . The cutoff for non-bonded interactions was set to 10 Å . All bonds involving hydrogen atoms were constrained using the SHAKE algorithm . We imposed a 1-fs simulation time step during equilibration . The temperature of the simulated systems was then gradually increased to 300 K over 50 ps in the NVT ensemble . Subsequently , equilibration dynamics was carried out in the NPT ensemble ( p=1 atm and T = 300 K ) for an additional 50 ns . Then , 800-ns production runs were carried out for each simulation system in the isothermal isobaric ensemble ( p=1 atm and T = 300 K ) . We utilized hydrogen mass repartitioning ( HMR ) as a method to increase the simulation time step to 4 fs during the production runs ( Hopkins et al . , 2015 ) . The substrate-bending angle for each system was defined and computed as described in Figure 1—figure supplement 3A . Data were analyzed with the CPPTRAJ code in AMBER15 ( Case et al . , 2015 ) and TCL scripts in VMD ( Humphrey et al . , 1996 ) . The effective free-energy profile ( potential of mean force; PMF ) for bending the NonEQ DF-6 , 1 DNA substrate was estimated using the adaptive biasing force ( ABF ) method ( Comer et al . , 2015; Hénin et al . , 2010 ) with the COLVARS module of NAMD 2 . 11 ( Phillips et al . , 2005 ) . ABF is a widely used enhanced sampling approach , which computes average forces along a predefined reaction coordinate ( RC ) and then applies an adaptive biasing potential to flatten the underlying free energy landscape . As a result , all points along the RC can be sampled efficiently . In our case , the RC was defined as the bending angle between the two-dsDNA fragments of the substrate . The exact definitions of the two vectors and the fragments are shown in Figure 1—figure supplement 3A . First , we carried out a 20-ns targeted molecular dynamics ( TMD ) simulation ( Schlitter et al . , 1994 ) , with a force constant of 100 kcal/mol Å2 applied to all nucleic acid heavy atoms . This simulation transformed the DNA from a straight conformation ( bending angle of ~180° ) to the bent conformation observed in the FEN1/DNA complex ( bending angle of ~90° ) . The target configuration of the bent DNA was directly taken from the FEN1/DNA X-ray structure 3Q8L ( Tsutakawa et al . , 2011 ) . In the ABF simulations , the reaction coordinate was segmented into nine discrete windows with a confining wall potential ( k = 50 kcal/mol ) placed at the boundaries . Snapshots collected from the TMD trajectory were used to seed the ABF windows . Each window was further subdivided into small 0 . 20 bins . Force averages were then accumulated into the bins and continuously updated in the course of the ABF simulation . Cancellation of the averaged forces through the gradual introduction of an adaptive bias led to enhanced sampling and overcoming of energy barriers along the RC . Since the instantaneous forces may fluctuate considerably , the application the adaptive bias was delayed until an adequate number of force samples was collected ( 2000 samples ) . PMF reconstruction was then accomplished by integration of the averaged forces from the bins .	When a cell divides it must copy its genetic information , which is found in the form of strands of DNA . Damage to the DNA may lead to cancer or a number of genetic diseases . However , every time a cell divides more than 10 million toxic “flaps” of excess DNA are generated . A protein called flap endonuclease 1 ( FEN1 ) keeps the DNA in good repair by cutting off the flaps in a highly specific and selective manner . Many proteins that interact with DNA are attracted to specific genetic sequences within the DNA strands . However , this is not the case for FEN1 and several other “structure-specific” proteins that help to repair and replicate DNA strands . So how do these proteins select the correct regions of DNA to interact with ? Rashid et al . used single-molecule fluorescence measurements to examine how purified FEN1 proteins interact with DNA flaps . The results show that FEN1 can perfectly recognize and correctly remove flaps through a process called “mutual-induced fit” , where the DNA and FEN1 are shaped by each other to produce a highly specific structure . Further work is now needed to examine whether other proteins that are related to FEN1 use a similar process to ensure that they always cut DNA in the same way . More detailed and direct examination of the structure of FEN1 through other experimental methods may also help to reveal how the mutual-induced fit process enables FEN1 to achieve such high levels of precision . This could increase our understanding of how problems with FEN1 and similar proteins lead to different genetic diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2017	Single-molecule FRET unveils induced-fit mechanism for substrate selectivity in flap endonuclease 1
Trial by trial covariations between neural activity and perceptual decisions ( quantified by choice Probability , CP ) have been used to probe the contribution of sensory neurons to perceptual decisions . CPs are thought to be determined by both selective decoding of neural activity and by the structure of correlated noise among neurons , but the respective roles of these factors in creating CPs have been controversial . We used biologically-constrained simulations to explore this issue , taking advantage of a peculiar pattern of CPs exhibited by multisensory neurons in area MSTd that represent self-motion . Although models that relied on correlated noise or selective decoding could both account for the peculiar pattern of CPs , predictions of the selective decoding model were substantially more consistent with various features of the neural and behavioral data . While correlated noise is essential to observe CPs , our findings suggest that selective decoding of neuronal signals also plays important roles . In most sensory systems , neurons encode sensory stimuli by responding selectively to a particular range of stimulus parameters , as typically characterized by tuning curves ( Dayan and Abbott , 2001 ) . In turn , the pattern of activation across a population of such neurons provides information about the most likely stimulus that may have occurred ( Dayan and Abbott , 2001 ) . Whether or not a sensory neuron contributes to perceptual decisions generally depends on whether that neuron is selective to the stimulus dimensions relevant to the task at hand , and how much weight is given to the activity of that neuron in population decoding . One method for assessing the potential contribution of a sensory neuron to perception involves measuring the trial-by-trial covariation between neural activity and perceptual decisions , as typically quantified by computing the choice probability ( CP ) ( Britten et al . , 1996; Dodd et al . , 2001; Uka and DeAngelis , 2004; Purushothaman and Bradley , 2005; Gu et al . , 2007 , 2008; Nienborg and Cumming , 2009 , 2010; Liu et al . , 2013 ) . When a sensory neuron shows a significant CP , there is a stereotypical relationship between response , tuning , and choice: neurons tend to respond more strongly when the subject reports perceiving the stimulus as having a value that is more preferred by the neuron . Tested properly , such effects are typically found to be independent of the stimulus value itself ( Britten et al . , 1996; Uka and DeAngelis , 2004 ) . While the phenomenology is rather consistent across many studies , the interpretation of CPs has remained controversial ( Nienborg and Cumming , 2010; Cohen and Kohn , 2011 ) . Some studies have suggested that the pattern of CPs across a population of neurons can provide insight into how responses of neurons with different tuning properties are selectively weighted in the decision process , that is selective decoding ( Britten et al . , 1996; Uka and DeAngelis , 2004; Purushothaman and Bradley , 2005; Gu et al . , 2007 , 2008 ) . Other studies have pointed out that correlated noise among neurons is necessary to observe significant CPs in large populations , suggesting that CPs are dominated by correlated noise and may not carry any useful information about decoding strategy ( Nienborg and Cumming , 2010; Cohen and Kohn , 2011; Nienborg et al . , 2012 ) . As an extreme example , neurons that are not involved in the decision process can exhibit significant CPs solely through correlations with other neurons that do contribute ( Cohen and Newsome , 2009 ) . Thus , a critical issue is whether CPs can reflect selective decoding of sensory neurons . A recent theoretical study potentially unifies these divergent perspectives ( Haefner et al . , 2013 ) , demonstrating mathematically that CPs could reflect both the structure of correlated noise and selective decoding of neurons . However , experimental evidence that can dissociate these causes has been lacking . Here , we take advantage of a peculiar pattern of CPs exhibited by multisensory neurons that represent translational self-motion ( i . e . , heading ) . Some neurons in areas MSTd ( Gu et al . , 2006 , 2008 ) and VIP ( Chen et al . , 2013 ) have matched heading preferences in response to visual and vestibular stimuli ( ‘congruent’ cells ) , whereas others prefer widely disparate headings ( ‘opposite’ cells ) . Opposite cells could be decoded such that they provide evidence in favor of either their visual or their vestibular heading preference . We showed previously that congruent and opposite cells have CPs with opposite polarities in a visual heading discrimination task ( Gu et al . , 2008; Chen et al . , 2013 ) , and we suggested that this may result from selectively decoding both congruent and opposite cells according to their vestibular heading preferences ( Gu et al . , 2008 ) . This system provides a valuable test bed for exploring the roles of noise correlations and selective decoding in producing CPs . Using simulations , we explore whether the peculiar pattern of CPs exhibited by multisensory neurons can be explained solely by correlated noise or whether selective decoding is also involved . Our results suggest that selective decoding can play important roles in shaping the pattern of CPs across a population of sensory neurons . To study how correlated noise and selective decoding affect the CPs of sensory neurons , we first considered a simple model in which only one sensory cue ( e . g . , visual ) was involved . We generated a population of 1000 hypothetical neurons with cosine tuning for heading . All neurons in this population had tuning curves with the same amplitude and width , but differed in their heading preferences . For simplicity , half of the neurons preferred leftward heading ( −90° ) while the other half preferred rightward heading ( +90° ) so that tuning curves for any pair of neurons had either identical slopes or opposite slopes around a straight-forward heading reference . Other distributions of heading preferences ( e . g . , uniform or bimodal ) did not substantially alter our conclusions . We used a maximum likelihood decoder ( Sanger , 1996; Dayan and Abbott , 2001; Jazayeri and Movshon , 2006; Gu et al . , 2010; Fetsch et al . , 2011 ) to estimate heading from simulated population activity , and we required the population activity to discriminate between headings that were slightly leftward or rightward relative to a straight-forward reference heading . Specifically , a likelihood function over heading was computed from the population activity on each trial . The decoder then made a ‘leftward’ choice if the area under the likelihood function for leftward headings exceeded the area under the curve for rightward headings , and vice versa for a ‘rightward’ choice . We then computed each model neuron's CP for the ambiguous stimulus condition ( i . e . , straight forward motion , 0° , ‘Materials and methods’ ) , and we explored how correlated noise and selective decoding affected CPs . As a prelude to considering the multisensory situation , we consider two extreme cases in which CPs of a group of neurons are driven mainly by correlated noise or by selective decoding . In both schemes , structured noise correlations are necessary to observe significant CPs , but the models differ in terms of which pools of neurons are correlated and how they are decoded . For the ‘pure-correlation’ model ( Figure 1A , B ) , only correlated noise is needed to produce CPs that are significantly different from the chance level of 0 . 5 . In this model , we divided the population of neurons into two groups , each of which contained an equal number of neurons preferring leftward and rightward headings . The first group of neurons ( pool 1 in Figure 1A ) contributed to the decoder's heading report ( decoding weight = 1 ) , while responses from the other group ( pool 2 ) were ignored by the decoder ( decoding weight = 0 ) . We then examined the CPs of pool 2 neurons as a function of their correlations with pool 1 . Although the signals from pool 2 neurons did not contribute to the decoder output , they still exhibited significant CPs as long as their noise was correlated with that of pool 1 neurons ( Cohen and Newsome , 2009 ) . 10 . 7554/eLife . 02670 . 003Figure 1 . Comparison of models in which choice probabilities ( CPs ) arise through either correlated noise or selective decoding . Each model consists of two pools of neurons ( 500 neurons each ) with equal numbers of neurons that prefer leftward and rightward headings . In the 'pure-correlation' model ( A and B ) , neurons in pool 2 make no contribution to the decision and activity within or across pools is correlated according to the relationship illustrated in panel B . In the 'selective decoding' model ( C and D ) , neurons shared correlated noise within each pool but not across pools . Neurons in pool 1 were always given a decoding weight of 1 , while neurons in pool 2 were given weights ranging from 0 to 1 . Solid curves in D: responses of pool 2 were decoded according to each neuron's preferred stimulus; dashed curves: pool 2 responses were decoded relative to each neuron's anti-preferred stimulus . Dashed black horizontal line: CP = 0 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 02670 . 003 We introduced correlated noise having a structure that is based on experimental observations from heading-selective neurons in both cortical ( Gu et al . , 2011; Chen et al . , 2013 ) and subcortical ( Liu et al . , 2013 ) areas . Specifically we assumed that noise correlations among pairs of model neurons are a linear function of their tuning similarity , or signal correlation ( rsignal , Figure 1B , insets ) . In this correlation structure , neurons with similar heading tuning ( positive rsignal ) generally have positive noise correlations , whereas neurons with dissimilar tuning ( negative rsignal ) tend to have negative noise correlations . This correlation structure was applied to all pairs of neurons , regardless of whether they were from pool 1 or pool 2 . For the simulations , the average noise correlation across all neurons was close to zero , consistent with some previous findings ( Gu et al . , 2011 ) , but the results are not sensitive to this mean value ( Liu et al . , 2013 ) . Rather , the critical factor is the slope of the relationship between rnoise and rsignal ( Liu et al . , 2013 ) : larger slopes lead to greater CPs among pool 2 neurons ( Figure 1B ) . Hence , in this pure-correlation model , CPs of pool 2 neurons are driven exclusively through correlations with neurons in pool 1 that contribute to the decision process ( Cohen and Newsome , 2009; Nienborg and Cumming , 2010 ) . For the ‘selective decoding’ model , noise among neurons within each pool was correlated in the same manner as described above , but there were no correlations between neurons in different pools ( Figure 1C ) . In this case , significant CPs for pool 2 neurons require that these neurons make a contribution to the decision ( Figure 1D ) . We manipulated two aspects of the contribution of each pool 2 neuron to the decoder output: magnitude and polarity . The magnitude reflects how strongly each neuron's activity influenced the decoder and was implemented mathematically by multiplying each neuron's contribution by a value in the range [0–1] . If the weight value is 1 , then pool 2 neurons contribute equally as pool 1 neurons . As the weight is reduced toward zero , the contribution of pool 2 neurons diminishes and eventually is eliminated . The polarity ( or sign ) of the weight determines whether each neuron provides the decoder with evidence for ( positive polarity ) or against ( negative polarity ) its heading preference . It may appear counterintuitive to consider that neurons might provide evidence against their stimulus preference , but the need to consider this case will arise later in the multisensory version of the model due to the presence of opposite cells . These neurons have different heading preferences for the two sensory modalities , so they can provide evidence in favor of their stimulus preference for one modality or the other . In this selective decoding model , the magnitude of CP increases with the weight applied to pool 2 neurons . In addition , whether the CP value is greater or less than 0 . 5 depends on the polarity of the contribution of pool 2 neurons ( Figure 1D ) . Interpreting responses as evidence in favor of the preferred heading produces CP >0 . 5 ( solid curves ) while decoding responses as evidence against the preferred heading leads to CP <0 . 5 ( dashed curves ) . Hence , the two models generate CPs for pool 2 neurons through different mechanisms . In the pure-correlation model , CPs of pool 2 neurons are produced through correlations with pool 1 neurons that are involved in the decision process ( Figure 1B ) . In the selective decoding model , pool 2 neurons have CPs that depend on how strongly they contribute to the decision , as well as the polarity of the contribution of each neuron to the decision ( Figure 1D ) . We next consider a more complicated case in which two different sensory cues are involved in a perceptual decision . For example , both visual ( optic flow ) and vestibular signals provide information about the direction of self-motion , or heading ( Angelaki and Cullen , 2008; Britten , 2008 ) . Previous studies have reported that neurons in multiple cortical areas ( e . g . , MSTd , VIP , VPS ) are tuned for heading , and tend to prefer either the same or opposite headings defined by optic flow and vestibular cues ( Page and Duffy , 2003; Gu et al . , 2006 , 2008; Chen et al . , 2011a , 2011b ) . We refer to these as congruent cells and opposite cells , respectively ( Figure 2A ) . For opposite cells , the preferred heading is different for the two sensory modalities , thus raising the fundamental question of how these cells may be decoded . In a multimodal heading discrimination task ( Gu et al . , 2008 ) , we showed previously that CPs of MSTd neurons have a peculiar dependence on the congruency of visual/vestibular heading tuning ( Figure 2B ) . For congruent cells ( cyan symbols in Figure 2B ) , CPs were consistently >0 . 5 when heading judgments were based on either vestibular or visual cues . In contrast , CPs for opposite cells tended to be >0 . 5 in the vestibular task condition but <0 . 5 in the visual condition ( magenta symbols in Figure 2B ) . We suggested previously ( Gu et al . , 2008 ) that this peculiar pattern of CPs might result from decoding the responses of MSTd neurons according to their vestibular heading preferences , a form of selective decoding . In what follows , we evaluate whether this pattern of CPs is compatible with a multisensory version of either the pure correlation or selective decoding models described above . 10 . 7554/eLife . 02670 . 004Figure 2 . Responses of multisensory neurons and multisensory versions of the pure-correlation and selective-decoding models . ( A ) Heading tuning curves from two example MSTd neurons measured during a fine heading discrimination task ( Gu et al . , 2008 ) : one congruent cell ( left ) and one opposite cell ( right ) . Red and blue data show responses measured for the visual and vestibular conditions , respectively . ( B ) Choice probability as a function of congruency for MSTd neurons tested in the vestibular ( left ) and visual ( right ) conditions ( adapted , with permission , from Supplement Figure 8A and Figure 6C of Gu et al . ( 2008 ) ; respectively ) . Cyan and magenta symbols denote data for congruent and opposite cells , respectively ( unfilled symbols: intermediate cells ) . ( C and D ) Multisensory version of the pure-correlation model ( 500 model neurons in each pool ) . Pool 1 consists of all congruent cells ( same slope tuning curves for the two cues ) , whereas pool 2 contains all opposite neurons . Correlated noise within or across pools depends only on the similarity of tuning for cue 1 . ( E and F ) In the selective decoding model , neurons were correlated according to the similarity of tuning for both cues ( ‘Materials and methods’ ) . This rule generated correlated noise within each pool but not between pools . Neurons in pool 1 were always given a full weight of 1 in the decoding , whereas the decoding weights of neurons in pool 2 ranged from 0 to 1 ( different colors in F ) . Dashed black horizontal line: CP = 0 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 02670 . 00410 . 7554/eLife . 02670 . 005Figure 2—figure supplement 1 . Predictions from a variant of the pure-correlation model in which correlated noise depends only on signal correlations from the visual tuning curves . This modification reverses the patterns of CPs across stimulus conditions , as compared to Figure 2C , D . This pattern of results is not consistent with experimental data . DOI: http://dx . doi . org/10 . 7554/eLife . 02670 . 00510 . 7554/eLife . 02670 . 006Figure 2—figure supplement 2 . Predictions from a variant of the selective decoding model in which responses are decoded according to the visual heading tuning of each neuron , instead of the vestibular tuning . This modification reverses the CP patterns across stimulus conditions , as compared to Figure 2E , F , and is inconsistent with experimental data . DOI: http://dx . doi . org/10 . 7554/eLife . 02670 . 00610 . 7554/eLife . 02670 . 007Figure 2—figure supplement 3 . Predictions from a “hybrid” model ( see text for details ) in which correlated noise was assigned according to vestibular signal correlations , and heading was decoded relative to the vestibular heading tuning of each neuron . CP patterns were roughly similar to those seen in the pure-correlation model ( Figure 2C , D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02670 . 007 We again divided 1000 model neurons into two pools of equal size , with one pool consisting of congruent cells and the other consisting of opposite cells ( Figure 2C , E ) . For simplicity , these neurons again preferred either leftward or rightward headings . In the pure-correlation model , only pool 1 neurons ( congruent cells ) provided inputs to the decoder’s decision process , whereas pool 2 neurons ( opposite cells ) were given no weight . However , opposite cells can still exhibit CPs as long as they are correlated with congruent cells ( Figure 2D ) . For a pair of neurons that includes an opposite cell , correlated noise could depend on the similarity of vestibular tuning between the members of the pair or on the similarity of visual tuning ( or on both , as discussed further below ) . If correlated noise is dependent on the similarity of vestibular tuning ( i . e . , vestibular signal correlation ) , opposite cells show CPs that are >0 . 5 in the vestibular condition and <0 . 5 in the visual condition ( Figure 2D ) , roughly similar to experimental data from MSTd neurons ( Figure 2A , B ) . The intuition for this result is straightforward . When a congruent cell fires more spikes than average , an opposite cell that shares the same vestibular tuning will also fire more spikes than average and will have a CP >0 . 5 in the vestibular condition . In the visual condition , this correlation structure will lead to an opposite cell responding more when the animal chooses its non-preferred visual heading , thus producing a CP <0 . 5 . Note that this prediction of the pure correlation model only mimics real data if opposite cells are correlated with congruent cells having matched vestibular heading preferences . In contrast , if correlated noise depends on the similarity of visual heading preferences , then the pattern of results will be reversed for the vestibular and visual conditions ( Figure 2—figure supplement 1 ) , which is clearly inconsistent with the experimental data ( Figure 2B ) . Finally , correlations between the two pools of the pure-correlation model could depend on the similarity of heading preferences for both the visual and vestibular modalities ( ‘Materials and methods’; Equation 2 ) . If correlated noise depends equally on both visual and vestibular signal correlations ( i . e . , Equation 2 with avestibular = avisual ) , then correlations between opposite cells and congruent cells become effectively zero because the two terms of Equation 2 cancel . In this case , the pure-correlation model becomes equivalent to the selective decoding model with zero weight placed on opposite cells , as considered below . In the selective decoding model , we assume that correlated noise among a pair of neurons depends on the similarity of both vestibular and visual tuning ( ‘Materials and methods’ ) , as demonstrated previously for pairs of MSTd neurons ( Gu et al . , 2011 ) . Because we assume avestibular = avisual in Equation 2 , the resulting effective noise correlation between mixed pairs of congruent and opposite cells will be zero ( denoted by ‘ × ’ between the two pools in Figure 2E ) . Thus , use of the selective-decoding model with this particular correlation structure effectively eliminates correlated noise between the two pools while allowing the decoding weights of pool 2 to vary . Under these conditions , non-zero decoding weights must be applied to pool 2 in order to observe CPs for opposite cells ( Figure 2F ) . Note that , for the selective decoding model , we do not consider situations in which correlated noise between the two pools depends only on vestibular or visual heading preferences . In such cases , there would be correlated noise between pairs of neurons drawn from the two pools , and the CPs of opposite cells would depend on both readout weights and noise correlations . Thus , the two models would no longer be conceptually distinct . We shall evaluate these assumptions in the following sections . To examine the predictions of the selective decoding model , we again manipulated two aspects of the readout . With regard to magnitude of the decoding weights , a decoding weight different from zero was essential to produce CPs for opposite cells that were different from the chance level ( green symbols/lines vs other colors in Figure 2F ) . With regard to polarity ( or sign ) of the decoding weights , decoding responses of opposite cells with respect to their vestibular heading preference led to CP >0 . 5 in the vestibular condition and <0 . 5 in the visual condition ( Figure 2F ) , which was similar to that seen in the real data . On the other hand , if responses of model neurons were decoded with respect to their visual heading preferences , the pattern of CPs across stimulus conditions would reverse and would be incompatible with that observed for MSTd neurons ( Figure 2—figure supplement 2 ) . Hence , both models can produce CP patterns for congruent and opposite cells that are similar to those seen in the real data , but through critically different mechanisms . In the pure-correlation model , CPs of opposite cells arise solely through correlations with congruent cells having matched vestibular tuning , even though opposite cells do not contribute directly to the decision . In the selective decoding model , there is no effective correlation between mixed pairs of congruent/opposite cells ( i . e . , no correlated noise between pools ) . In this case , to qualitatively match the pattern of CP results from real neurons , the activity of opposite cells must be given weight in the decoding and these cells must be decoded selectively according to their vestibular preferences . Another way to summarize the distinction between models is that the pure correlation model mimics experimental CP results by virtue of modality specificity in the structure of correlated noise , whereas the selective decoding model achieves this by modality specificity in the decoding weights . For completeness , we also considered a hybrid model that combines features from both of the above models . Noise correlations were linearly dependent only on the similarity of vestibular tuning , as in the pure-correlation model . This produced correlations between the two pools ( congruent and opposite cells ) , unlike in the selective decoding model . In addition , a readout weight was assigned to the opposite cells , as in the selective decoding model . Under these conditions , we found that the predicted patterns of CPs largely resembled those from the pure-correlation model , as if the decoding weights on pool 2 played little role ( Figure 2—figure supplement 3 ) . In the following analyses , we only considered comparisons between the pure-correlation model and the selective decoding model , as these are conceptually distinct . Although this distinction is useful for exploring the relative roles of correlated noise and selective decoding in producing CPs , we recognize that both factors may contribute . Which model best matches experimental data on correlated noise ? Our previous study ( Gu et al . , 2011 ) showed that rnoise measured for pairs of MSTd neurons depended approximately equally on rsignal computed from both visual and vestibular tuning curves ( Figure 3A ) . This dependence on rsignal for both modalities is not due to strong covariance between the two signal correlations because rsignal values for visual and vestibular tuning are only weakly correlated ( Figure 3—figure supplement 1 ) . Note also that noise correlations are generally negative for neurons with opposite tuning ( negative signal correlations ) in our heading discrimination data sets ( Gu et al . , 2011; Chen et al . , 2013; Liu et al . , 2013 ) , such that subtracting responses of oppositely tuned neurons is not expected to have the benefit seen in other systems ( Romo et al . , 2003 ) . 10 . 7554/eLife . 02670 . 008Figure 3 . Comparison of the structure of correlated noise between models and data . ( A ) Data from pairs of neurons recorded from area MSTd ( Gu et al . , 2011 ) . Noise correlation is plotted against signal correlations obtained from the vestibular ( left column ) or visual ( right column ) tuning curves . Black and gray symbols denote pairs with matched ( black ) or mismatched ( gray ) congruency . Open symbols represent undefined pairs . ( B ) Predicted noise correlations as a function of signal correlation based on fits of the selective decoding model . Format as in panel A . ( C ) Predicted noise correlations as a function of signal correlation by for the pure-correlation model fit , for which noise correlations depend only on vestibular signal correlation . DOI: http://dx . doi . org/10 . 7554/eLife . 02670 . 00810 . 7554/eLife . 02670 . 009Figure 3—figure supplement 1 . Comparison of vestibular and visual signal correlations for 127 pairs of neurons simultaneously recorded from area MSTd by Gu et al . ( 2011 ) . There is a fairly weak correlation between the two variables ( R = 0 . 29 , p=0 . 001 , Spearman correlation ) . Histograms along the top and right side show marginal distributions of the two signal correlations . DOI: http://dx . doi . org/10 . 7554/eLife . 02670 . 00910 . 7554/eLife . 02670 . 010Figure 3—figure supplement 2 . Noise correlation structure of the pure correlation model computed from the signal correlations of all distinct pairings of 129 neurons that were recorded previously by Gu et al . ( 2011 ) . For the pure-correlation model , correlated noise is determined by the signal correlation in the vestibular condition , such that pairs of cells with matched ( black ) and mismatched ( gray ) congruency have opposite relationships between rnoise and rsignal in the visual condition . DOI: http://dx . doi . org/10 . 7554/eLife . 02670 . 01010 . 7554/eLife . 02670 . 011Figure 3—figure supplement 3 . Noise correlation structure of the selective decoding model computed from the signal correlations of all distinct pairings of 129 neurons that were recorded previously by Gu et al . ( 2011 ) . For the selective decoding model , correlated noise depends on rsignal in both stimulus conditions . As a result , the relationship between rnoise and rsignal is strong for pairs with matched congruency in both stimulus conditions ( black ) , and this relationship is weak for pairs with mismatched congruency ( gray ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02670 . 011 Here , we have sorted these previous data ( Gu et al . , 2011 ) into two groups according to whether the congruency of visual/vestibular tuning is matched or mismatched for the two members of each pair of neurons: ‘matched congruency’ pairs consist of either two congruent cells or two opposite cells , whereas ‘mismatched congruency’ pairs consist of one congruent cell and one opposite cell ( see ‘Materials and methods’ for classification procedure ) . Pairs that could not be classified into either group were labeled as ‘undefined’ ( open symbols , Figure 3A ) . We found that the dependence of rnoise on both visual and vestibular signal correlations was weak for the mismatched congruency pairs ( vestibular rsignal: slope = 0 . 033 , CI = [−0 . 5–0 . 27] , R = 0 . 08 , p=0 . 8; visual rsignal: slope = −0 . 057 , CI = [−0 . 424–0 . 258] , R = −0 . 12 , p=0 . 7 , type II linear regression ) , whereas this dependence was quite robust for the matched congruency pairs ( vestibular rsignal: slope = 0 . 192 , CI = [0 . 082–0 . 288] , R = 0 . 61 , p=0 . 001; visual rsignal: slope = 0 . 186 , CI = [0 . 069–0 . 28] , R = 0 . 62 , p<0 . 001 , type II linear regression ) . This pattern of results is consistent with the correlation structure assumed in the selective decoding model . To evaluate whether the experimental data significantly favor the selective decoding model over the pure correlation model , we fit the data from these 127 pairs of neurons with two correlation structures: ( 1 ) rnoise depended only on vestibular rsignal , rnoise = avestibular * rsignal_vestibular , as in the pure correlation model; ( 2 ) rnoise depended on both vestibular and visual signal correlations , rnoise = avestibular * rsignal_vestibular + avisual *rsignal_visual , as assumed in the selective decoding model . We then used linear regression to fit the data with both correlation structures . Importantly , we found that the model with coefficients for both vestibular and visual signal correlations provided a significantly better fit to the data , after accounting for the difference in the number of parameters ( p=0 . 0003 , sequential F-test ) . The coefficients of the best-fitting model were avestibular = 0 . 12 and bvisual = 0 . 09 , respectively . Thus , the empirical correlation data are significantly better fit with a correlation structure in which rnoise depends on rsignal for both vestibular and visual tuning curves . To help visualize why the second structure above better fits the data , we used the measured signal correlations for each pair of neurons , along with the fitted coefficients , to predict rnoise for each pair of neurons . When rnoise is predicted based on both visual and vestibular signal correlations ( Figure 3B ) , the dependence of rnoise on rsignal was much weaker for mismatched congruency pairs ( vestibular rsignal: slope = 0 . 074 , 95% CI = [0 . 044–0 . 109] , R = 0 . 82 , p=0 . 004; visual rsignal: slope = −0 . 01 , 95% CI = [−0 . 08–0 . 066] , R = −0 . 09 , p=0 . 8 , type II linear regression ) than for matched congruency pairs ( vestibular rsignal: slope = 0 . 19 , 95% CI = [0 . 177–0 . 203] , R = 0 . 99 , p<<0 . 001; visual rsignal: slope = 0 . 18 , 95% CI = [0 . 157–0 . 195] , R = 0 . 96 , p<<0 . 001 , type II linear regression ) , similar to the MSTd data ( Figure 3A ) . In contrast , when rnoise is only dependent on vestibular rsignal , the predicted correlation structure is quite different ( Figure 3C ) . While rnoise is perfectly correlated with vestibular rsignal for all neurons ( by assumption ) , the mismatched congruency pairs reveal roughly equal but opposite dependencies on vestibular and visual signal correlations ( vestibular rsignal: slope = 0 . 16 , 95% CI = [0 . 161–0 . 161] , R = 1 , p<<0 . 001; visual rsignal: slope = −0 . 12 , 95% CI = [−0 . 220–0 . 018] , R = −0 . 65 , p=0 . 04 , type II linear regression ) . Thus , the available data on noise and signal correlations compare more favorably with the assumptions of the selective decoding model than the pure correlation model . We now evaluate whether predictions of the pure correlation and selective decoding models are compatible with choice probability data obtained from area MSTd neurons during a fine heading discrimination task ( Gu et al . , 2008 ) . The Gu et al . dataset consisted of 129 single neurons that were not recorded simultaneously . Thus , to generate model population responses for decoding analyses , we generated responses for a model population of 1000 neurons , each of which had visual and vestibular tuning curves that were obtained by drawing data ( with replacement ) from the sample of 129 real MSTd neurons . Responses of the 1000 model neurons on each simulated trial were generated from a covariance matrix that was based on the two different correlation structures described in the previous section , with parameters that were obtained by fits to the MSTd data . This yielded noise and signal correlations similar to those described in Figure 3 ( Figure 3—figure supplements 2 and 3 ) . With these constraints , we can decode the simulated population responses ( ‘Materials and methods’ ) and make predictions of CPs and neuronal thresholds . With our biologically-constrained versions of the pure correlation and selective decoding models , we now consider the patterns of choice probabilities predicted by each model and how they compare to data from MSTd neurons . In the pure-correlation model ( Figure 4A ) , the average CP for congruent cells is significantly greater than 0 . 5 for both the vestibular ( 0 . 65 ± 0 . 06 SD ) and visual ( 0 . 65 ± 0 . 04 SD ) conditions ( p<0 . 001 , t test ) . For opposite cells , the average CP is significantly >0 . 5 in the vestibular condition ( 0 . 623 ± 0 . 039 SD , p<0 . 001 , t test ) and significantly <0 . 5 in the visual condition ( 0 . 372 ± 0 . 047 SD , p<0 . 001 ) . This pattern of CPs is qualitatively similar to that observed for MSTd neurons ( Figure 2B ) . 10 . 7554/eLife . 02670 . 012Figure 4 . Predictions of choice probabilities from the two models . ( A ) The pattern of CPs predicted by the pure-correlation model . Format as in Figure 2B . ( B ) A family of weighting profiles used to consider various degrees of contribution of opposite cells to the selective decoding model . Each curve shows the decoding weight as a function of congruency between visual and vestibular heading tuning . Each curve corresponds to a specific value of the Readout Index ( RI ) . ( C ) Predicted average CPs from the selective decoding model for a subset of the RI values illustrated in ( B ) . ( D ) The pattern of CPs across neurons in the selective decoding model for an RI value of 0 . 5 . Cyan symbols: congruent cells; Magenta symbols: opposite cells; Unfilled symbols: intermediate cells . Solid squares: mean CP . Dashed horizontal line: CP = 0 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 02670 . 012 In the selective decoding model , we need to consider how the weight applied to opposite cells influences the pattern of CPs . However , unlike the simple hypothetical model of Figure 2 , real neurons exhibit a range of congruencies between visual and vestibular tuning . To explore a range of possible relative weightings of opposite cells , we used a sigmoidal function ( having a single parameter called the Readout Index , RI ) to selectively weight the contributions of congruent and opposite cells ( Figure 4B , ‘Materials and methods’ ) . For RI values near 1 . 0 , all neurons contribute equally to the decoder output , regardless of congruency . As RI declines , opposite cells are given gradually less weight in the decoding ( Figure 4B ) . For small values of RI , opposite neurons in the selective decoding model show CPs that approach 0 . 5 , as expected since these neurons are given little weight in the decision process . However , as RI increases , the average CP for opposite cells increases in the vestibular condition and decreases in the visual condition ( Figure 4C ) . For RI values near 0 . 5 ( Figure 4D ) , the pattern of CPs across the population of simulated neurons resembles that exhibited by MSTd neurons ( Figure 2B ) . Together , these results ( Figure 4 ) demonstrate that both the pure correlation and selective decoding models are capable of mimicking the patterns of CPs shown by MSTd neurons in the visual and vestibular conditions ( although the pure correlation model assumes a correlation structure different from that seen experimentally ) . We now consider whether the models make distinct predictions regarding CPs for the combined condition , in which visual and vestibular cues are presented together . As described previously ( Gu et al . , 2008 ) , when animals judge heading based on congruent combinations of visual and vestibular cues , congruent cells tend to have CPs >0 . 5 , whereas opposite cells have CPs that cluster around the chance level of 0 . 5 ( Figure 5A ) . Interestingly , although the average CP of opposite cells was approximately 0 . 5 for both models in the combined condition ( pure-correlation model: 0 . 486 ± 0 . 13 SD; selective decoding model: 0 . 491 ± 0 . 06 SD ) , the SD of CP across the population of opposite cells was much greater for the pure-correlation model than for the selective decoding model ( Figure 5B , C ) . Indeed , a non-parametric test showed that the dispersion of the CPs was significantly greater for the pure correlation model ( p<<0 . 001 , Ansari–Bradley test ) . The reason for this difference is apparent: CPs of opposite cells in the pure correlation model clearly have a bimodal distribution ( puni<<0 . 001 , pbi>0 . 05 , modality test , middle column of Figure 5D ) . In contrast , CPs of opposite cells in the selective readout model have a unimodal distribution centered around 0 . 5 ( puni>0 . 05 , modality test , right column in Figure 5D ) , which is similar to that seen for MSTd neurons ( puni>0 . 05 , modality test , left column in Figure 5D ) . Thus , the pattern of CPs in the combined condition favors the selective decoding model . 10 . 7554/eLife . 02670 . 013Figure 5 . Analysis of choice probabilities for the combined condition in which both visual and vestibular heading cues are present . ( A ) CP values plotted as a function of congruency index for neurons from area MSTd tested in the Combined condition ( adapted , with permission , from Figure 6A of Gu et al . , 2008 ) . ( B and C ) CP as a function of congruency for model neurons from the pure-correlation model , and model cells from the selective decoding model , respectively . ( D ) Distributions of CPs for opposite cells from area MSTd , the pure-correlation model and the selective decoding model . ( E–G ) CP values plotted as a function of neuronal discrimination thresholds for real MSTd neurons ( adapted , with permission , from Figure 6B of Gu et al . , 2008 ) , units from the pure-correlation model and units from the selective decoding model , respectively . ( H ) Correlation coefficient of the best linear fit to the relationship between CP and neuronal threshold for MSTd data ( filled black circles ) , pure-correlation model ( open black circles ) and selective decoding model ( open red circles ) . Error bars represent 95% confidence intervals . DOI: http://dx . doi . org/10 . 7554/eLife . 02670 . 01310 . 7554/eLife . 02670 . 014Figure 5—figure supplement 1 . Bimodality of CP for opposite cells in the cue combined condition . Top panels: four examples of simulations of the pure-correlation model having different values of avestibular ranging from 0 . 01 to 0 . 12 . Both the magnitude and bimodality of the CP distribution increases with avestibular . To quantify bimodality , we chose cells with Congruency Index in the range [−1–0 . 5] , and applied K-means clustering to generate two clusters . The vertical distance , d , between the centroids of the two clusters was taken as an index of the separation of the two clusters . Bottom row: ( left ) Summary of d values as a function of avestibular . Asterisks represent significant bimodality , as assessed by a modality test ( ‘Materials and methods’ ) . ( right ) Relationship between the CP of congruent cells ( congruency index from 0 . 5 to 1 ) and avestibular . DOI: http://dx . doi . org/10 . 7554/eLife . 02670 . 01410 . 7554/eLife . 02670 . 015Figure 5—figure supplement 2 . Same format as in Figure 5—figure supplement 1 , but results are shown for the selective decoding model . Top row: simulations for four different values of the Readout Index ( RI ) . Bottom row: A bimodal distribution of CP is only seen for RI values >= 0 . 8 , and CP of congruent cells is independent of RI . DOI: http://dx . doi . org/10 . 7554/eLife . 02670 . 015 Additional analyses suggest that this difference in the shape of the CP distribution between models is fairly robust to variations in the key model parameters . For the pure correlation model , a bimodal distribution of CPs in the combined condition is a robust result over a range of values of avestibular , including values much smaller than needed to fit our noise correlation data ( Figure 5—figure supplement 1 ) . For the selective decoding model , a unimodal distribution of CPs is predicted for a wide range of RI values , and significant bimodality only occurs for RI values ≥0 . 8 , which are not consistent with behavioral thresholds as described below ( Figure 5—figure supplement 2 ) . A possible explanation for the near-chance CPs of opposite cells in the combined condition is that the sensitivity these cells is low relative to that in the single cue conditions and thus opposite cells contribute less to the decision process ( see insets in Figure 5A , left panel ) . Indeed , for real MSTd neurons , congruent cells tend to have low thresholds and high CPs whereas opposite cells tend to have high thresholds and low CPs ( Figure 5E; R = −0 . 65 , 95% CI: [−0 . 47 to −0 . 78] , p<<0 . 001 , linear regression ) . Although both models also show a similar dependency of CP on neuronal sensitivity in the combined condition ( p<<0 . 001 , linear regression , Figure 5F , G ) , the strength of the correlation is significantly weaker for the pure correlation model ( R = −0 . 41 , 95% CI: [−0 . 32 to −0 . 48] ) than for the selective readout model ( R = −0 . 67 , 95% CI: [−0 . 61 to −0 . 72] , Figure 5H ) . This also appears to be largely due to the bimodal distribution of CPs for opposite cells in the pure correlation model , which is not observed for the real MSTd data . Thus , while both models produce similar patterns of CPs in the vestibular and visual conditions ( Figure 4 ) , the pattern of CPs from the selective decoding model was more analogous to the measured data in the combined condition . Finally , we compute the predicted psychophysical performance by decoding data from real MSTd neurons using each readout model , and we compare the results to animals’ behavioral performance . In a fine heading discrimination task , we have previously shown that heading sensitivity in the combined condition increases when the two cues have comparable sensitivity , and that the effects are close to that predicted from optimal cue integration theory ( Gu et al . , 2008; Fetsch et al . , 2009 , 2011 ) . In Figure 6A , we have replotted the animals' behavioral data ( solid circles and dashed curve ) . The average psychophysical thresholds were 2 . 03°±0 . 09° ( mean ± SEM ) and 2 . 12° ± 0 . 1° for the vestibular and visual conditions , respectively . In the combined condition , the average threshold was reduced to 1 . 44° ± 0 . 06° , a 29% improvement compared to the best single cue , and was very close to the prediction ( 1 . 43° ± 0 . 06° ) from optimal cue integration theory . 10 . 7554/eLife . 02670 . 016Figure 6 . Comparisons between model population threshold and psychophysical performance of the animals . ( A ) Average thresholds are shown for the vestibular , visual , and combined conditions , along with the prediction from optimal cue integration theory . Data are shown for the average of two monkeys ( filled symbols , dashed curve ) , for predictions of the pure-correlation model ( black open symbols and solid curve ) , and for predictions of the selective decoding model with RI = 0 . 5 ( red symbols and curve ) . ( B ) Predicted thresholds from the selective decoding model are plotted as a function of Readout Index for vestibular , visual , and combined conditions , as well as the optimal prediction from the single-cue thresholds . ( C ) Thresholds from the selective-decoding model ( with RI = 0 . 5 ) as a function of population size . Solid curves: model predictions; dashed horizontal lines: average performance of two animals . DOI: http://dx . doi . org/10 . 7554/eLife . 02670 . 01610 . 7554/eLife . 02670 . 017Figure 6—figure supplement 1 . Comparison of neuronal sensitivity between the visual ( ordinate ) and vestibular ( abscissa ) stimulus conditions for 30 congruent cells . Arrows and numbers indicate geometric mean values . Overall , visual responses were somewhat more sensitive than vestibular responses , although the animals’ behavioral performance was similar between conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 02670 . 017 Unlike the animals' behavior , decoder performance based on the pure-correlation model predicted mismatched thresholds for the visual and vestibular conditions ( Figure 6A , open symbols and solid black curve ) . The average threshold predicted for the visual condition ( 1 . 24° ± 0 . 02° , mean ± SEM ) was 42% lower than that for the vestibular condition ( 2 . 16° ± 0 . 14° , mean ± SEM ) . This may be due to the fact that the average neuronal threshold of congruent MSTd neurons tends to be lower for the visual condition ( 5 . 5° ) than the vestibular condition ( 7 . 1° ) , although this difference did not reach significance ( p=0 . 106 , t test , N = 30 , Figure 6—figure supplement 1 ) . Despite the performance mismatch between the two single-cue conditions , sensitivity was still improved during the combined condition ( 1 . 10° ± 0 . 05° ) , as expected by optimal cue integration predictions ( 1 . 07° ± 0 . 02° ) . For the selective decoding model , we examined how performance changes as a function of the weighting applied to opposite cells ( Readout Index , Figure 6B , ‘Materials and methods’ ) . In the vestibular condition , the decoder's discrimination threshold decreased modestly as more weight was applied to opposite cells because more cells contribute to the likelihood function ( Figure 6B , blue curve ) . In the visual condition , because heading was decoded with respect to the vestibular preference of each neuron , a greater contribution of opposite cells tends to drive the decoder to make choices that are opposite to that being signaled by congruent cells . Consequently , predicted thresholds in the visual condition rise precipitously as opposite cells are given more weight ( Figure 6B , red curve ) . From these data , we can see that a Readout Index value of ∼0 . 5 produces roughly matched visual and vestibular heading thresholds , similar to the animals’ behavior ( arrow in Figure 6B ) . With this weighting of opposite cells , the selective decoding model predicts similar heading thresholds for vestibular ( 2 . 33° ± 0 . 04° , mean ± SEM ) and visual ( 2 . 40° ± 0 . 1° ) conditions , as well as thresholds in the combined condition ( 1 . 58° ± 0 . 01° ) that are very close to the optimal prediction ( 1 . 57° ± 0 . 02° , Figure 6B , green and brown curves ) . Critically , we note that the value of the readout index ( 0 . 5 ) that allows the selective-decoding model to mimic behavioral performance is the same value that we separately found to produce a pattern of CPs that approximates the experimentally observed data ( Figure 5C , G ) . Thus , converging lines of evidence suggest a readout in which opposite cells contribute , but substantially less than congruent cells . The above simulation results were based on an arbitrarily sized population of neurons ( n = 1000 ) . Not surprisingly , as the population size is varied in the simulations , predicted psychophysical thresholds decline as a function of population size in all stimulus conditions ( Figure 6C ) . In our simulations , performance reaches a plateau at a population size of a few hundred neurons , and is roughly comparable to the animals' behavioral performance over a broad range of pool sizes . It must be noted that the extent to which performance asymptotes with population size is likely to depend on the exact structure of correlated noise , the extent to which the true decoder has full and accurate knowledge of the correlation structure , the extent to which correlated noise mimics stimulus variations and thus can be removed by decoding , and the degree of heterogeneity of the tuning curves in the population ( Ecker et al . , 2011; Beck et al . , 2012 ) . Importantly , however , the key experimental features that we have sought to understand here are related to the sign of CP for opposite cells and not simply the magnitude of CPs; thus , the basic qualitative nature of our findings is not likely to be altered by the considerations above . Choice probability measures the trial-by-trial correlation between the activity of a single neuron and perceptual decisions . While the measurement itself is straightforward , the interpretation of CPs has been varied and somewhat controversial ( Nienborg and Cumming , 2010; Nienborg et al . , 2012 ) . One possible interpretation of a significant CP is that variability in the response of a sensory neuron drives variability in perceptual decisions across trials—this is the so-called ‘bottom-up’ interpretation ( Parker and Newsome , 1998 ) . If this were true , then CPs would at least partially reflect the contribution of each neuron to the decision , and would be shaped by selective decoding of sensory signals . Along these lines , some studies have suggested that the pattern of CPs observed reflects the strategy by which sensory signals are decoded to perform specific tasks ( Uka and DeAngelis , 2004; Purushothaman and Bradley , 2005; Gu et al . , 2007 , 2008 ) . Somewhat analogous conclusions were drawn in a previous study which showed that ‘detect probabilities’ depend on tuning preferences in a change-detection task , which may also be compatible with the notion of selective decoding ( Bosking and Maunsell , 2011 ) . An alternative ( but not mutually exclusive ) possibility is that CPs are mainly driven by top-down feedback signals from parts of the brain involved in making decisions ( Nienborg and Cumming , 2009 , 2010 ) . If this is the case , then the pattern of CPs need not be directly related to the way that sensory signals are decoded to perform a task . Regardless of the relative roles of bottom-up and top-down signals in generating CPs , it is broadly recognized that CPs should not be observable in large neural populations unless noise is correlated among neurons ( Shadlen et al . , 1996; Cohen and Newsome , 2009; Nienborg and Cumming , 2010; Cohen and Kohn , 2011; Nienborg et al . , 2012 ) . Indeed , a recent study ( Liu et al . , 2013 ) provided the first experimental evidence that the difference in magnitude of CPs between two brain areas coincides with a difference in the structure of correlated noise between areas . The controversy regarding whether CPs can reflect decoding strategy or just correlated noise was recently resolved by an important theoretical study ( Haefner et al . , 2013 ) , which shows that CPs are determined by both factors . However , whether the pattern of CPs in a population reflects decoding strategy or not will depend on the specific details of the decoding weights , correlation structure , population size , etc . This theory shows that the decoding weights could be inferred from CPs if the full structure of correlated noise is known with sufficient accuracy and precision . Using multisensory heading perception as a model system , we show that the pattern of CPs exhibited by neurons in area MSTd is more compatible with a model in which both selective decoding and correlated noise contribute to the generation of CPs than a model in which CPs are determined solely from correlations . Multisensory representations may have advantages for studying CPs because of the presence of neurons that show opposite tuning for the two cues . If such neurons are decoded as providing evidence in favor of either their visual or vestibular heading preference , then the sign of their CP ( whether it is > or <0 . 5 ) may reverse depending on the decoding strategy . We suspect that this feature of our model system has provided us with additional leverage to dissociate models that emphasize correlations vs selective decoding . In our selective decoding model , we assume that responses of all MSTd neurons , both congruent and opposite , are decoded relative to their vestibular heading preference . This accounts for opposite cells having CPs <0 . 5 in the visual condition , as seen in the real data . Why might responses be decoded according to the vestibular heading preference ? One possibility is that this allows the system to estimate heading in a manner that is robust to the presence of moving objects in a scene . Indeed , we have recently shown that a strategy of decoding both congruent and opposite cells according to their vestibular preferences can provide a near-optimal solution to the problem of marginalizing over object motion in order to extract heading in a robust manner ( Kim et al . , 2014 ) . Moreover , adjusting the relative weighting of opposite to congruent cells can allow the population code to tradeoff robustness to object motion against increased sensitivity during cue integration ( Kim et al . , 2014 ) . Thus , the selective decoding strategy that we employ here may provide a flexible way to decode self-motion signals efficiently under conditions in which moving objects may or may not distort optic flow . Correlation matrices that describe interneuronal noise correlations for a simulated population of neurons were constructed by assigning correlated noise ( rnoise ) to each pair of neurons ( either hypothetical or real ) according to the measured relationship between rnoise and the signal correlation between the pair of tuning curves , that is rsignal ( Gu et al . , 2011 ) . Signal correlation , rsignal , was computed as the Pearson correlation coefficient between the tuning curves ( mean firing rates ) for a pair of neurons . We previously showed that the relationship between rnoise and rsignal for MSTd neurons could be well described by Gu et al . ( 2011 ) : ( 2 ) rnoise , i , j=avestibular×rsignal , vestibular , i , j+avisual×rsignal , visual , i , j , where avestibular was 0 . 12 and avisual was 0 . 09 ( Gu et al . , 2011 ) . For population decoding simulations based on real neurons ( Figures 4–6 ) , correlated noise between pairs of neurons was assigned according to Equation 2 with the parameters above . For simulations based on hypothetical neurons , both coefficients in Equation 2 were set to 0 . 1 for simplicity . Single-trial responses of model neurons to each heading stimulus were generated according to the assumption of proportional Gaussian noise , with response variance set to be 1 . 5 times the mean firing rate to approximate the general behavior of cortical neurons ( Shadlen et al . , 1996; Gu et al . , 2008; Cohen and Newsome , 2009 ) . To generate simulated population responses that incorporated correlated noise among neurons , we incorporated the estimated correlation matrix and generated population activity for each stimulus modality ( k ) , trial ( tr ) , and heading ( θ ) according to the following equation ( Shadlen et al . , 1996; Cohen and Newsome , 2009 ) : ( 3 ) responsek , tr ( θ ) =<responsek , tr ( θ ) >+Q×rrand×1 . 5×<responsek , tr ( θ ) > , where Q is the square root of the correlation matrix , rrand is a random vector of standard normal deviates with the same length as the number of neurons ( Matlab function ‘normrnd’ , zero mean , unit variance ) , and ‘<>’ represents the mean value . We typically produced 200 trials of responses for each heading ( θ ) . As shown previously , neurons in area MSTd have vestibular heading tuning that can be either congruent with or opposite to their visual heading tuning ( Gu et al . , 2006 , 2008 ) . Over the narrow range of headings used in the discrimination task , congruent cells generally have monotonic visual and vestibular tuning curves with matched slopes , whereas opposite cells generally have oppositely signed slopes ( Figure 2A ) . To quantify tuning congruency , we compute a Pearson correlation coefficient between firing rate and heading for each tuning curve . From these , we compute a Congruency Index for each neuron , which is the product of the correlation coefficients for the visual and vestibular tuning curves . Thus , congruent cells will have a positive Congruency Index , whereas opposite cells with have a negative Congruency Index . A pair of MSTd neurons can therefore have matched congruency ( congruent–congruent pairs or opposite–opposite pairs ) , or mismatched congruency ( congruent–opposite pairs ) . We categorized each pair of neurons as matched or mismatched by computing the product of their two congruency indices . Specifically , matched congruency pairs were classified as those having a product >0 . 2 ( black symbols , Figure 3A ) , and mismatched pairs are those having a product <−0 . 2 ( gray symbols , Figure 3A ) . The remaining cell pairs with products of Congruency Indices that fall in the range from −0 . 2 to +0 . 2 were classified as ‘undefined’ ( open symbols , Figure 3A ) . These criteria are rather stringent , but we found that they reliably classify cells pairs . To test whether a distribution of CPs contains a single mode ( unimodal ) or two modes ( bimodal ) , we used a multimodality test based on the kernel density estimate method ( Gu et al . , 2006; Takahashi et al . , 2007 ) . Watson's U2 statistic , corrected for grouping , was computed as a goodness-of-fit test statistic to obtain a p value through a bootstrapping procedure . This test generates two p values , with the first one ( puni ) for the test of unimodality and the second one ( pbi ) for the test of bimodality . If puni >0 . 05 , the distribution is defined as unimodal . If puni <0 . 05 , the hypothesis of unimodality is rejected . If pbi >0 . 05 as well , the distribution is considered bimodal . To transform responses from a population of neurons into quantitative predictions of behavioral sensitivity and choice probabilities , we decoded simulated population activity by computing a likelihood function as in previous studies ( Sanger , 1996; Dayan and Abbott , 2001; Jazayeri and Movshon , 2006; Gu et al . , 2010; Fetsch et al . , 2011 ) . For each stimulus modality ( k ) , the likelihood over heading ( θ ) given the observed population activity on a particular trial ( tr ) was given by: ( 4 ) Log ( Lk , tr ( θ ) ) =∑i=1nresponsek , tr , i ( θ ) ×Log ( <responsek , tr , i ( θ ) > ) −∑i=1n<responsek , tr , i ( θ ) > . The first term describes the summation of each cell’s contribution to the log likelihood function , which corresponds to the response on each trial weighted by the logarithm of each cell’s tuning curve . The second term is the sum of all tuning curves , to counter biases associated with a non-uniform distribution of stimulus preferences . This formulation embodies two assumptions . First , it assumes Poisson firing statistics . We have also tried alternative decoders that are based on Gaussian noise , such as the Fisher linear discriminant ( Dayan and Abbott , 2001 ) , and the results are almost identical . Thus , the details of the spiking statistics have little effect on our conclusions . Second , this formulation does not assume that the decoder has knowledge of the structure of correlated noise among neurons in the population , that is , a factorized decoder ( Averbeck et al . , 2006 ) . While this assumption very likely affects the absolute sensitivity of the decoder , it is unlikely to alter the basic pattern of predicted results regarding CPs ( Gu et al . , 2010 ) . However , further investigation is needed to systematically examine the detailed differences between decoders with and without knowledge of correlations , as well as the effects of different forms of correlated activity . Given these assumptions , our decoder is optimal in the maximum-likelihood sense , and does not otherwise assume a specific ( and perhaps substantially suboptimal ) ‘pooling model’ , as was done in some previous simulations of CPs ( Shadlen et al . , 1996 ) . The decoder determined whether a tested heading was leftward or rightward relative to straight ahead by comparing the area under the computed likelihood function for leftward headings and rightward headings . If the summed likelihood for rightward headings was greater than that for leftward headings , the decoder would report ‘right’ , and vice versa . Choice probability for each neuron in the simulated population was consequently computed for the ambiguous straight-forward heading ( 0° ) ( Britten et al . , 1996; Shadlen et al . , 1996; Gu et al . , 2007 , 2008 ) . The precision ( threshold ) of the decoder was also computed from each simulated psychometric function , which was analyzed using methods identical to those applied to the human and monkey behavior ( Gu et al . , 2010 ) . Heading information was decoded in two ways . According to our hypothesis that both congruent and opposite cells are decoded according to their vestibular heading preference ( Gu et al . , 2008 ) , our main method of decoding involved using the vestibular heading tuning curve for each neuron in the formulation of Equation 4 . In addition , for some analyses , heading was also decoded relative to the heading preference of each neuron in each stimulus condition . In this case , responses from the visual condition were decoded based on visual tuning curves , and so on . For the selective decoding model , we implemented a function that controlled the contribution of each model neuron to the decoder output based on each neuron’s congruency value , as given below: ( 5 ) weighti=1−e− ( congruencyi/2 ) ×RI1−e−1 . Here , weighti denotes the decoding weight of the ith neuron , congruencyi denotes the Congruency Index ( described above ) for the ith neuron , and RI represents a Readout Index that ranges from 0 to 1 in steps of 0 . 1 ( Figure 4B ) . In the selective decoding model , the computed decoding weight of each neuron was then multiplied by that neuron's contribution to the likelihood function ( Equation 4 ) before the likelihood contributions are summed across neurons . Thus , the computation of the likelihood under the selective decoding model was given by: ( 6 ) Log ( Lk , tr ( θ ) ) =∑i=1nweighti×responsek , tr , i ( θ ) ×Log ( <responsek , tr , i ( θ ) > ) −∑i=1n<responsek , tr , i ( θ ) > .	Even the simplest tasks require the brain to process vast amounts of information . To take a step forward , for example , the brain must process information about the orientation of the animal's body and what the animal is seeing , hearing and feeling in order to determine whether any obstacles stand in the way . The brain must integrate all this information to make decisions about how to proceed . And once a decision is made , the brain must send signals via the nervous system to the muscles to physically move the foot forward . Specialized brain cells called sensory neurons help to process this sensory information . For example , visual neurons process information about what the animal sees , while auditory neurons process information about what it hears . Other sensory neurons—called multisensory neurons—can process information coming from more than one of an animal's senses . For more than two decades , researchers have known that the firing of an individual sensory neuron can be linked to the decision that an animal makes about the meaning of the sensory information it has received . The ability to predict whether an animal will make a given decision based on the firing of individual sensory neurons is often referred to as a ‘choice probability’ . Measurements of single neurons have often been used to try to work out how the brain decodes the sensory information that is needed to carry out a specific task . However , it remains unclear whether choice probabilities really reflect how sensory information is decoded in the brain , or whether these measurements are just reflecting coordinated patterns of background ‘noise’ among the neurons as the decisions are being made . Gu et al . set out to help resolve this debate by examining choice probabilities in the multisensory neurons in one area of the brain . A series of experiments was conducted to see how these neurons process information , both from the eyes and the part of the inner ear that helps control balance , to work out the direction in which an animal was moving . By performing computer simulations of the activity of groups of neurons , Gu et al . found that choice probability measurements are better explained by the models whereby these measurements did reflect the strategy that is used to decode the sensory information . Models based solely on patterns of correlated noise did not explain the data as well , though Gu et al . suggest that this noise is likely to also contribute to the observed effects . Following on from the work of Gu et al . , a major challenge will be to see if it is possible to infer how the brain extracts the relevant information from the different sensory neurons . This may require recordings from large groups of neurons , but it might help us to decipher how patterns of activity in the brain lead to decisions about the world around us .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2014	Contribution of correlated noise and selective decoding to choice probability measurements in extrastriate visual cortex
The kinase Bub1 functions in the spindle assembly checkpoint ( SAC ) and in chromosome congression , but the role of its catalytic activity remains controversial . Here , we use two novel Bub1 inhibitors , BAY-320 and BAY-524 , to demonstrate potent Bub1 kinase inhibition both in vitro and in intact cells . Then , we compared the cellular phenotypes of Bub1 kinase inhibition in HeLa and RPE1 cells with those of protein depletion , indicative of catalytic or scaffolding functions , respectively . Bub1 inhibition affected chromosome association of Shugoshin and the chromosomal passenger complex ( CPC ) , without abolishing global Aurora B function . Consequently , inhibition of Bub1 kinase impaired chromosome arm resolution but exerted only minor effects on mitotic progression or SAC function . Importantly , BAY-320 and BAY-524 treatment sensitized cells to low doses of Paclitaxel , impairing both chromosome segregation and cell proliferation . These findings are relevant to our understanding of Bub1 kinase function and the prospects of targeting Bub1 for therapeutic applications . During eukaryotic cell division , the spindle assembly checkpoint ( SAC ) contributes to ensure the accuracy of chromosome segregation . This evolutionarily conserved surveillance mechanism monitors the status of kinetochore ( KT ) -microtubule ( MT ) interactions and delays anaphase onset until all chromosomes have undergone bipolar attachment to the spindle . Both KT-MT interactions and SAC activity are regulated by several KT-associated protein kinases , including Aurora B , Monopolar spindle 1 ( Mps1 ) and Budding uninhibited by benzimidazoles 1 ( Bub1 ) ( Maciejowski et al . , 2010; Meraldi and Sorger , 2005; Santaguida et al . , 2011 ) . SAC activity depends on a diffusible inhibitor of the ubiquitin ligase anaphase-promoting complex/cyclosome ( APC/C ) , termed mitotic checkpoint complex ( MCC ) ( Foley and Kapoor , 2013; Lara-Gonzalez et al . , 2012; Musacchio , 2011; Sacristan and Kops , 2015 ) . Once the SAC is silenced in response to chromosome biorientation , APC/C activation then triggers the onset of chromatid separation and mitotic exit , respectively ( Funabiki and Wynne , 2013 ) . The serine/threonine kinase Bub1 is one of the first proteins to accumulate at unattached kinetochores ( Jablonski et al . , 1998 ) . Its recruitment is governed by Mps1-dependent phosphorylation of MELT motifs on the KMN complex member KNL-1 ( London et al . , 2012; Overlack et al . , 2015; Shepperd et al . , 2012; Vleugel et al . , 2013; Yamagishi et al . , 2012 ) . Bub1 has been implicated in the regulation of chromosome cohesion , KT-MT interactions and SAC function . In particular , Bub1 was shown to be important for the centromere/KT recruitment of Shugoshin proteins ( Sgo1 and Sgo2 ) , the chromosomal passenger complex ( CPC ) comprising Aurora B kinase , CENP-E , CENP-F , BubR1 , Mad1 and Mad2 ( Boyarchuk et al . , 2007; Kitajima et al . , 2005; Klebig et al . , 2009; Liu et al . , 2013; Perera et al . , 2007; Taylor and McKeon , 1997 ) . So far , only few substrates of Bub1 have been identified . Best characterized is the phosphorylation of histone H2A on threonine 120 ( T120 ) ( Kawashima et al . , 2010 ) . Phosphorylation of this site by Bub1 can be demonstrated not only in vitro but also in living cells ( Kawashima et al . , 2010; Lin et al . , 2014; Sharp-Baker and Chen , 2001 ) . Histone H2A phosphorylation on T120 triggers the centromere localization of Sgo1 , which in turn recruits the CPC subunit Borealin ( Kawashima et al . , 2010; Liu et al . , 2013; Tsukahara et al . , 2010; Yamagishi et al . , 2010 ) . Centromere recruitment of the CPC is further enhanced by the kinase Haspin , which phosphorylates histone H3 at T3 and triggers the centromere binding of the CPC component Survivin ( Du et al . , 2012; Kelly et al . , 2010; Wang et al . , 2010 ) . Another intriguing potential substrate of Bub1 is the APC/C co-activator Cdc20 ( Lin et al . , 2014; Tang et al . , 2004a ) . Whether Bub1 phosphorylates Cdc20 in living cells remains to be determined , but recent studies strongly suggest that Bub1 binding to Cdc20 is important for SAC function ( Di Fiore et al . , 2015; Vleugel et al . , 2015 ) . Genetic , biochemical or siRNA-mediated depletion of Bub1 protein clearly interferes with chromosome alignment and SAC activity , but the importance of Bub1 catalytic activity has long been subject to debate ( Bolanos-Garcia and Blundell , 2011; Elowe , 2011; Funabiki and Wynne , 2013 ) . For example , while a Bub1 mutant completely lacking the kinase domain is checkpoint proficient in Saccharomyces cerevisiae ( Fernius and Hardwick , 2007 ) , conflicting data have been reported on the importance of Bub1 kinase activity in fission yeast Schizosaccharomyces pombe ( Rischitor et al . , 2007; Vanoosthuyse et al . , 2004; Yamaguchi et al . , 2003 ) . Similarly , in Xenopus egg extracts , catalytically inactive Bub1 can sustain the SAC ( Sharp-Baker and Chen , 2001 ) , although kinase-proficient Bub1 may be more efficient ( Boyarchuk et al . , 2007; Chen , 2004 ) . In mammalian cells , several studies point to the conclusion that Bub1 mutants devoid of catalytic activity are able to restore many , albeit not all , aspects of chromosome congression and SAC function ( Klebig et al . , 2009; McGuinness et al . , 2009; Perera and Taylor , 2010a; Ricke et al . , 2012 ) . To address the role of Bub1 kinase activity in mammalian mitosis , we have made use of two novel small molecule inhibitors , BAY-320 and BAY-524 . Using biochemical and cellular assays , we show that these ATP-competitive inhibitors potently and specifically block human Bub1 both in vitro and in living cells . By comparing phenotypes provoked by Bub1 kinase inhibition and Bub1 protein depletion , we are able to differentiate between catalytic and non-catalytic functions of Bub1 . Our data indicate that Bub1 catalytic activity is largely dispensable for chromosome alignment and SAC function , arguing that Bub1 largely operates as a scaffolding protein . However , even though Bub1 inhibition per se exerts only minor effects on mitotic fidelity , BAY-320 and BAY-524 treatment sensitizes cells to clinically relevant low doses of Paclitaxel , resulting in remarkable impairment of chromosome segregation and cell proliferation . The chemical synthesis of small molecule inhibitors against Bub1 has recently been described ( Hitchcock et al . , 2013 ) . In this study , we used the two substituted benzylpyrazole compounds , 2-[5-cyclopropyl-1- ( 4-ethoxy-2 , 6-difluorobenzyl ) -4-methyl-1H-pyrazol-3-yl]-5-methoxy-N- ( pyridin-4-yl ) pyrimidin-4-amine and 2-[1- ( 4-ethoxy-2 , 6-difluorobenzyl ) -5-methoxy-4-methyl-1H-pyrazol-3-yl]-5-methoxy-N- ( pyridin-4-yl ) pyrimidin-4-amine , abbreviated as BAY-320 and BAY-524 , respectively ( Figure 1A ) . In vitro inhibition of Bub1 by BAY-320 and BAY-524 was demonstrated by monitoring both Bub1 autophosphorylation and phosphorylation of histone H2A on T120 ( Kawashima et al . , 2010 ) ( Figure 1B ) . In presence of 2 mM ATP , both compounds inhibited the recombinant catalytic domain of human Bub1 ( amino acids 704–1085 ) with an IC50 of 680 ± 280 nM and 450 ± 60 nM , respectively ( Supplementary file 1 ) . When tested against a panel of 222 protein kinases , BAY-320 showed only modest cross reactivity with other kinases , even when used at a concentration of 10 μM ( Supplementary file 2 ) . Furthermore , quantitative measurements of BAY-320 interactions with 403 human kinases , using an active site-directed competition-binding assay , showed exquisite binding selectivity for Bub1 ( Supplementary file 3 ) . 10 . 7554/eLife . 12187 . 003Figure 1 . BAY-320 and BAY-524 inhibit Bub1 kinase . ( A ) Chemical structure of ATP-competitive inhibitors BAY-320 and BAY-524 . ( B ) In vitro kinase assays showing dose-dependent inhibition of Bub1 kinase activity towards histone H2A . The assays were performed by mixing human wild-type ( WT ) or kinase-dead ( KD ) LAP-Bub1 , ectopically expressed in and purified from mitotic HEK 293T cells , with recombinantly expressed histone H2A as a substrate , γ-32P-ATP and increasing doses of the Bub1 inhibitors BAY-320 and BAY-524 . After 30 min at 30°C , reactions were stopped and analyzed by gel electrophoresis . Bub1 autophosphorylation and H2A phosphorylation were visualized by autoradiography ( 32P ) and protein levels monitored by Coomassie brilliant blue staining ( CBB ) . Histone H2A-T120 phosphorylation ( pT120-H2A ) was detected by phospho-antibody probing of Western blots ( WB ) and Bub1 was monitored as control . ( C , D ) Inhibition of Bub1 reduces histone H2A-T120 phosphorylation . Asynchronous cultures of HeLa S3 ( left panels ) and RPE1 cells ( right panels ) were treated with the proteasomal inhibitor MG132 for 2 hr , followed by the addition of 3 . 3 μM nocodazole and increasing doses of BAY-320 ( C ) or BAY-524 ( D ) for 1 hr . The cells were fixed and analyzed by immunofluorescence microscopy ( IFM ) . Scatter plots show centromeric levels of pT120-H2A ( n = 19–28 cells per condition ) . Bars represent mean values . ( E ) HeLa S3 cells were synchronized by thymidine block , released for 10 hr in the presence of solvent ( control ) , 3 μM BAY-320 or 7 μM BAY-524 and analyzed by quantitative IF ( top panels ) . Cells transfected with mock ( Gl2 ) or Bub1 siRNA-oligonucleotides for 48 hr were synchronized and analyzed in parallel ( bottom panels ) . The cells were stained with antibodies raised against Bub1 and pT120-H2A . Human CREST serum was used to identify centromeres and DNA was stained with DAPI; scale bars represent 10 µm . ( F ) Histograms showing the average signal intensities of centromeric pT120-H2A observed in the experiments described in ( E ) ; n = 73–107 cells per condition . Error bars represent standard error of the mean ( SEM ) . ( G ) To monitor the efficacy of Bub1 kinase inhibition within cells , HeLa S3 cells were synchronized by thymidine block and released for 14 hr in the presence of 3 . 3 μM nocodazole as well as Bub1 inhibitors or solvent as indicated . Prometaphase-arrested cells were harvested by shake-off and mitotic cell extracts were treated with or without phosphatase inhibitor for 30 min at 30°C . Histone isolation was followed by Western blot analysis of pT120-H2A . Equal loading was monitored by Ponceau S staining . DOI: http://dx . doi . org/10 . 7554/eLife . 12187 . 00310 . 7554/eLife . 12187 . 004Figure 1—figure supplement 1 . BAY-320 and BAY-524 inhibit Bub1 kinase . ( A , B ) BAY-320 and BAY-524 treatment coordinately reduces histone H2A-T120 phosphorylation as well as Aurora B centromere/KT binding , until maximal Bub1 inhibition is reached at 10 μM . Asynchronous cultures of HeLa S3 ( left panels ) and RPE1 cells ( right panels ) were treated with the proteasomal inhibitor MG132 for 2 hr , followed by the addition of 3 . 3 μM nocodazole and increasing doses of BAY-320 ( A ) or BAY-524 ( B ) for 1 hr . The cells were fixed and analyzed by immunofluorescence microscopy ( IFM ) . Scatter plots show centromere/KT levels of pT120-H2A and Aurora B ( n = 19–28 cells per condition ) . Bars represent mean values . ( C ) Untreated HeLa cells ( red ) or HeLa cells treated with nocodazole for 16 hr , followed by various concentrations of BAY-320 ( green ) or solvent ( black ) for 1 hr , were fixed and analyzed by quantitative in-cell western . Plot shows total pT120-H2A signal intensity . Grey area highlights the concentration range between 3 and 10 μM . The IC50 ( reflecting the inhibition of Bub1 kinase activity compared to control and normalized to cell number ) was determined to be 379 +/- 156 nM . DOI: http://dx . doi . org/10 . 7554/eLife . 12187 . 004 To test whether BAY-320 and BAY-524 also inhibit Bub1 in intact cells , increasing doses of inhibitors were applied to mitotically synchronized hTERT-RPE1 ( RPE1 ) and HeLa cells , and phospho-histone H2A-T120 staining at kinetochores was monitored by immunofluorescence ( Figure 1C–F and Figure 1—figure supplement 1A , B ) and in-cell western assays ( Figure 1—figure supplement 1C ) . These studies revealed that near-maximal inhibition of Bub1 could be achieved by using BAY-320 at 3–10 μM and BAY-524 at 7–10 μM and these concentrations were therefore used in all future experiments on intact cells . To corroborate the above immunofluorescence data , histones were purified from control and inhibitor-treated cells . Examination of histone H2A phosphorylation by Western blotting revealed that treatment of cells with either BAY-320 or BAY-524 drastically reduced T120 phosphorylation ( Figure 1G ) . Thus , BAY-320 and BAY-524 act as potent and selective inhibitors of Bub1 kinase in both biochemical and cellular assays and thus constitute attractive tools to study Bub1 catalytic function during mitosis . Next , we set out to directly compare the impact of Bub1 kinase inhibition with the previously reported consequences of Bub1 depletion ( Boyarchuk et al . , 2007; Johnson et al . , 2004; Kitajima et al . , 2005; Klebig et al . , 2009; Logarinho et al . , 2008; Meraldi and Sorger , 2005; Tang et al . , 2004b ) or genetic Bub1 knock-out ( Jeganathan et al . , 2007; Perera and Taylor , 2010a; Perera et al . , 2007; Ricke et al . , 2012 ) . In a first series of experiments , we used time-lapse imaging to compare progression through mitosis in asynchronously growing HeLa and RPE1 cells in response to either Bub1 inhibition or siRNA-mediated Bub1 depletion . In line with previous results ( Kitajima et al . , 2005; Tang et al . , 2004b ) , depletion of Bub1 from HeLa cells significantly prolonged duration of mitosis , due to delayed chromosome alignment and delays in prometa- and metaphase ( Figure 2A and C , Figure 2—figure supplement 1A ) . In stark contrast , treatment with either BAY-320 or BAY-524 provoked at most minor effects on mitotic progression , marked by a short delay of anaphase onset ( Figure 2B and C , Figure 2—figure supplement 1B and C ) . Furthermore , in contrast to aneuploid HeLa cells , diploid RPE1 cells were not significantly affected by either Bub1 inhibition or depletion ( Figure 2D and Figure 2—figure supplement 1D ) . Efficiency of siRNA-mediated depletion was monitored by Western blotting ( Figure 2—figure supplement 1E ) . Flow-cytometric analyses confirmed that Bub1 depletion from HeLa cells causes an increase in the G2/M population of HeLa but not RPE1 cells and that Bub1 inhibition by BAY-320 or BAY-524 did not detectably affect cell cycle profiles in either cell line ( Figure 2—figure supplement 1F and G ) . We conclude that the inhibition of Bub1 kinase activity in either HeLa or RPE1 cells produces at most subtle effects on mitotic progression , whereas Bub1 depletion exerts more profound effects , at least in HeLa cells . These results are consistent with the demonstration that Bub1 kinase activity is not required for the development and viability of mice ( Perera and Taylor , 2010b; Ricke et al . , 2012 ) . 10 . 7554/eLife . 12187 . 005Figure 2 . Inhibition of Bub1 kinase activity barely affects mitotic progression . ( A , B ) Representative stills from time-lapse recordings of asynchronously growing cultures of HeLa S3 cells stably expressing GFP-tagged histone H2B . The cells were either treated with Bub1 inhibitors ( 3 μM BAY-320 and 7 μM BAY-524 ) or transfected with control ( Gl2 ) or Bub1 siRNA for 48 hr prior to time-lapse microscopy . Scale bars represent 10 µm . ( C , D ) Graphs show the cumulative frequency of mitotic duration determined by cell rounding/flattening . Indicated averages represent the time spent in mitosis ( n = 100 cells per condition ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12187 . 00510 . 7554/eLife . 12187 . 006Figure 2—figure supplement 1 . Inhibition of Bub1 kinase activity barely affects mitotic progression . ( A , B ) Representative stills from time-lapse recordings described in Figure 2 . Micrographs show fluorescence or DIC images of HeLa S3 cells stably expressing GFP-tagged histone H2B . Time is indicated relative to mitotic entry . ( C , D ) Hela S3 cells stably expressing GFP-H2B were treated with solvent ( control ) or Bub1 inhibitors at indicated doses and monitored by fluorescence time-lapse imaging . Dot plots show the time from mitotic entry to anaphase onset; bars represent mean values ( n = 80 cells per condition ) . ( E ) Representative Western blots show Bub1 depletion efficiencies . Asynchronous cultures of HeLa S3 or RPE1 cells were transfected with control ( Gl2 ) or Bub1 siRNA-oligonucleotides for 48 hr , harvested and analyzed by Western blotting; α-tubulin served as loading control . ( F , G ) Cell cycle distribution of exponentially growing HeLa S3 ( F ) and RPE1 ( G ) cells upon treatment with solvent ( control ) , 3 μM BAY-B320 , 7 μM BAY-B524 or after Bub1 protein depletion ( siGl2 served as control ) . After 48 hr of treatment or siRNA oligonucleotide transfection cells were permeabilized and DNA was stained with propidium iodide . Cellular DNA content was determined using flow cytometry and frequencies of G1 , S and G2/M phases were determined . Considering that Bub1 inhibition did not influence cell cycle profiles in HeLa and RPE1 cells and that the increase in the G2/M population after depletion of Bub1 from HeLa cells correlated with an increase in mitotic duration ( Figure 2 ) , we anticipate that the increase in the corresponding G2/M population mostly reflects mitotic cells . DOI: http://dx . doi . org/10 . 7554/eLife . 12187 . 006 One of the most interesting effects of Bub1 depletion described so far relates to sister chromatid cohesion ( Boyarchuk et al . , 2007; Fernius and Hardwick , 2007; Kitajima et al . , 2005; Tang et al . , 2004b ) . In particular , depletion of Bub1 was shown to cause persistent arm cohesion and a redistribution of Sgo proteins from centromeres to chromosome arms ( Kitajima et al . , 2005 ) . To directly demonstrate a role for Bub1 kinase activity in sister chromatid cohesion , we analyzed chromosome spreads prepared from mitotic HeLa cells or RPE1 cells after treatment with Bub1 inhibitors ( Figure 3A , Figure 3—figure supplement 1A ) or Bub1-specific siRNA for comparison ( Figure 3B and Figure 3—figure supplement 1B ) . While mitotic chromosome spreads from nocodazole-treated control cells showed the expected X-shape structure , indicative of centromere cohesion , most cells treated with either Bub1 inhibitors or Bub1 siRNA showed sister chromatids whose arms remained paired ( Figure 3A–C and Figure 3—figure supplement 1 ) . Moreover , centromeric levels of Sgo1 and Sgo2 were reduced to ~20% of control values in BAY-320 or BAY-524 treated cells ( Figure 3D and E ) and , concomitantly , a significant redistribution of Sgo2 to chromosome arms could be observed ( Figure 3F and G ) . We thus conclude that Bub1 catalytic activity contributes to the regulation of sister chromatid cohesion and the localization of Sgo proteins . 10 . 7554/eLife . 12187 . 007Figure 3 . Inhibition of Bub1 affects Sgo1 and Sgo2 localization and chromatid cohesion . ( A , B ) HeLa S3 cells were synchronized by thymidine block and released for 12 hr in the presence of 3 . 3 μM nocodazole as well as solvent ( control ) , 3 μM BAY-320 or 7 μM BAY-524 . Cells transfected with mock ( Gl2 ) or Bub1 siRNA-oligonucleotides for 48 hr were synchronized and analyzed in parallel . Micrographs show representative chromosome spreads prepared from mitotic cells . Insets show magnifications of chromosomes; they illustrate representative chromatid cohesion states . ( C ) Quantification of results of the experiments described in ( A ) and ( B ) as well as Figure 3—figure supplement 1; n = 200 cells per condition . ( D ) HeLa S3 cells were released from a thymidine arrest into solvent , 3 μM BAY-320 or 7 μM BAY-524 . The cells were fixed and stained for Sgo1 , Sgo2 , CREST and DNA ( DAPI ) and analyzed by IFM . Scale bars represent 10 µm . ( E ) Histogram showing average centromeric Sgo levels observed in the experiments described in ( A ) ; n = 43–120 cells per condition . Error bars represent SEM . ( F ) Asynchronous cultures of RPE1 cells were treated with indicated doses of Bub1 inhibitors for 3 hr , fixed and analyzed by IFM . Scale bar represents 5 µm . ( G ) Dot plot showing the quantitative results of the experiment shown in ( F ) . Sgo2 levels at centromeres and chromosome arms were determined in metaphase cells ( n = 150 centromere/arm regions from 15 different cells ) . Bars represent mean values . DOI: http://dx . doi . org/10 . 7554/eLife . 12187 . 00710 . 7554/eLife . 12187 . 008Figure 3—figure supplement 1 . Inhibition of Bub1 affects chromatid cohesion . ( A , B ) RPE1 cells were synchronized by thymidine block ( 4 mM ) and released for 12 hr in the presence of 3 . 3 μM nocodazole as well as solvent ( control ) , 3 μM BAY-320 or 7 μM BAY-524 . The cells transfected with mock ( Gl2 ) or Bub1 siRNA-oligonucleotides for 48 hr were synchronized and analyzed in parallel . Micrographs show representative chromosome spreads prepared from mitotic cells . Insets show magnifications of chromosomes; this illustrates representative chromatid cohesion states . Data relate to Figure 3C . Scale bars represent 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12187 . 008 In addition to preserving sister chromatid cohesion , Sgo1 and Sgo2 play important roles in the recruitment of the CPC , comprising Aurora B kinase ( Kawashima et al . , 2007; Tsukahara et al . , 2010 ) . This prompted us to investigate the impact of Bub1 inhibition on Aurora B localization and activity . Consistent with the marked effects on centromere localization of Sgo1/2 , we also observed significant effects of Bub1 inhibition on Aurora B localization . After treatment of HeLa cells with BAY-320 or BAY-524 , all CPC subunits examined were partially displaced from centromeres ( Figure 4A and B , Figure 4—figure supplement 1A ) . While Bub1 inhibition reduced centromeric levels of Aurora B , Borealin and INCENP by ~50% ( Figure 4A and B , left panels ) , depletion of Bub1 lowered centromere levels of these CPC components by ~70% ( Figure 4A and B , right panels ) . We emphasize that , due to a lack of resolution , these experiments do not discriminate between centromere- and KT-associated pools of the CPC . 10 . 7554/eLife . 12187 . 009Figure 4 . Bub1 inhibition affects localization and activity of the CPC . ( A , C ) Untreated or siRNA transfected ( siBub1 , siGl2 for control ) HeLa S3 cells were synchronized by thymidine block and released for 10 hr , as indicated ( BAY-320 was used at 3 μM , BAY-524 at 7 μM ) . Cells were fixed and stained for Aurora B , Borealin , INCENP , pS7-CENP-A , pS10-histone H3 , MCAK , CREST and DNA ( DAPI ) and analyzed by IFM . Scale bars represent 10 µm . ( B , D ) Histograms show quantitative results of the experiments described in ( A , C ) . Measurements represent centromeric levels except for pS10-histone H3 signals , which was monitored along chromosome arms ( n = 40–113 cells per condition ) . Scale bars represent 10 µm , error bars represent SEM . ( E ) FRET experiments were performed on HeLa Kyoto cells stably expressing chromatin ( H2B ) - or centromere ( CENP-B ) -fused FRET reporters for Aurora B activity . Cells were synchronized in mitosis by 6 hr treatment with 3 . 3 μM nocodazole , before the indicated inhibitors and 20 μM MG132 were added prior to live fluorescence microscopy . Heat-map represents the phosphorylation status of the reporter . Scale bar represents 10 µm . ( F ) Left panel: scatter plot depicts CFP/FRET emission ratios of reporter targeted to chromatin ( H2B; n = 23–52 cells per condition ) . Right panel: scatter plot depicts TFP/FRET emission ratios of reporter targeted to centromeres ( CENP-B , n = 16–34 cells per condition ) . Bars represent mean values; *p<0 . 001 ( from unpaired two-tailed Student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12187 . 00910 . 7554/eLife . 12187 . 010Figure 4—figure supplement 1 . Bub1 inhibition affects localization and activity of the CPC . ( A , B ) Panels relate to the quantitative data shown in Figure 4B and D . Scale bars represent 10 µm . ( C ) HeLa Kyoto cells stably expressing the chromatin-targeted Aurora B FRET reporter were transfected with control ( Gl2 ) or Bub1 siRNA oligonucleotides for 48 hr and synchronized by 6 hr treatment with 3 . 3 μM nocodazole , before 20 μM MG132 were added prior to live fluorescence microscopy . Top panel: heat-map represents the phosphorylation status of the reporter . Scale bar represents 10 μm . Bottom panel: dot plot shows measured CFP/FRET emission ratios ( n = 14–19 cells per condition ) . Bars represent mean values; p < 0 . 001 ( from unpaired two-tailed Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 12187 . 010 To examine the impact of Bub1 inhibition on the catalytic activity of Aurora B at both centromeres and chromosome arms , we next monitored phosphorylation of CENP-A Ser7 ( Zeitlin et al . , 2001 ) and histone H3 Ser10 ( Hirota et al . , 2005; Hsu et al . , 2000 ) , respectively . Compared to control cells , both Bub1 inhibition and depletion reduced CENP-A and histone H3 phosphorylation by ~50% and ~10–20% , respectively ( Figure 4C and D , Figure 4—figure supplement 1B ) , suggesting that interference with Bub1 primarily affects Aurora B activity at centromeres . This conclusion was corroborated by showing that both inhibition and depletion of Bub1 reduced the centromere association of the Aurora B effector protein MCAK ( Andrews et al . , 2004 ) by ~50% ( Figure 4C and D ) . Furthermore , use of biosensors for Aurora B activity ( Fuller et al . , 2008 ) revealed a reduction in fluorescence resonance energy transfer ( FRET ) ratios for sensors tethered to either chromosome arms ( through fusion to H2B ) or centromeres ( through fusion to histone CENP-B ) ( Figure 4E and F , Figure 4—figure supplement 1C ) . Collectively , these observations demonstrate that Bub1-dependent phosphorylation plays a major role in the regulation of Aurora B localization and activity . However , neither Bub1 inhibition nor Bub1 depletion resulted in complete removal of Aurora B from centromeres , prompting us to examine the relative contributions of Bub1 and Haspin to the process of CPC recruitment . While inhibition of Bub1 by BAY-320 or BAY-524 or inhibition of Haspin by 5-Iodotubercidin ( De Antoni et al . , 2012 ) similarly reduced centromere levels of the CPC components Aurora B , Borealin and INCENP to ~40% , combined inhibition of both kinases resulted in a ~80% reduction in CPC levels at centromeres ( Figure 5A and B , Figure 5—figure supplement 1A ) . As an important control , treatment of cells with only BAY-320 or BAY-524 did not detectably affect the phosphorylation of the Haspin substrate histone H3 ( T3 ) , attesting to the specificity of the two Bub1 inhibitors ( Figure 5A and B ) . 10 . 7554/eLife . 12187 . 011Figure 5 . Bub1 and Haspin inhibition exert additive effect on centromere association of CPC . ( A ) HeLa S3 cells were released from a thymidine block into 3 . 3 µM nocodazole , before they were additionally treated for 2 hr with the proteasomal inhibitor MG132 and indicated kinase inhibitors . The Haspin inhibitor 5-iodotubercidin ( 5-ITu [De Antoni et al . , 2012] ) was used at a concentration of 2 . 5 µM , BAY-320 at 3 μM and BAY-524 at 7 μM . Cells were fixed , stained for pT3-H3 , Aurora B , Borealin , INCENP , CREST and DNA ( DAPI ) and analyzed by IFM . Anti-pT3-H3 antibody was used to monitor Haspin and Bub1 inhibition , respectively . Scale bar represents 10 µm . ( B ) Histograms show average centromeric ( AurB , Borealin , INCENP ) or chromosome arm ( pT3-H3 ) signal intensities observed in the experiments shown in ( A ) ; n = 20–100 cells per condition . Error bars represent SEM , **p < 0 . 001 ( from unpaired two-tailed Student’s t-test ) . ( C ) RPE1 cells expressing endogenously EGFP-tagged Aurora B were incubated with the indicated drugs for several hours before EGFP signals were recorded by live fluorescence imaging . Scale bar represents 5 µm . ( D ) Scatter plots depict Aurora B-EGFP signal intensities at centromeres or arms after treatment with indicated drugs ( n = 84–185 centromeres/arm regions from 5–6 cells per condition ) . Bars represent mean values . For comparison , dashed lines mark the mean values of Aurora B-EGFP signal intensities at arms and centromeres in control cells . Measurements relate to the experiment shown in ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12187 . 01110 . 7554/eLife . 12187 . 012Figure 5—figure supplement 1 . Bub1 and Haspin inhibition exert additive effect on centromere association of CPC . ( A ) Panels relate to the quantitative data shown in Figure 5B . Scale bars represent 5 µm . ( B ) Asynchronous cultures of RPE1 cells expressing endogenously EGFP-tagged Aurora B were treated with indicated doses of Bub1 inhibitors for 3 hr , fixed and analyzed by IFM . Scale bar represents 5 µm . ( C ) Dot plots show the quantitative results of the experiment shown in ( C ) . EGFP-Aurora B and Sgo2 levels at centromeres and chromosome arms were determined in metaphase cells ( n = 100 centromeres or arm regions from 10 different cells ) . Bars represent mean values . DOI: http://dx . doi . org/10 . 7554/eLife . 12187 . 012 To quantify CPC localization over chromosome arms , analysis of fixed cells proved inadequate . We therefore used an RPE1 cell line expressing one endogenous allele of Aurora B tagged with EGFP ( von Schubert et al . , 2015 ) to monitor the subcellular localization of this kinase in living cells . Following Bub1 inhibition , Aurora B-EGFP levels at chromosome arms increased approximately twofold , concomitant with the described reduction of Aurora B at centromeres ( Figure 5C and D ) ( Boyarchuk et al . , 2007; Ricke et al . , 2012 ) . Interestingly , this change in localization showed a strong correlation with the redistribution of Sgo2 ( Figure 5—figure supplement 1B and C ) . In contrast , treatment of cells with the Haspin inhibitor 5-Iodotubercidin did not induce any significant redistribution of Aurora B from centromeres to chromosome arms; instead , inhibition of Haspin caused an overall reduction of EGFP signals at both centromeres and chromosome arms ( Figure 5C and D ) . Combined inhibition of Bub1 and Haspin displaced Aurora B from both centromeres and chromosome arms ( Figure 5C and D ) , in line with the analysis of fixed cells described above . Taken together , these data corroborate the notion that Bub1 and Haspin cooperate in the recruitment of CPC to centromeres through phosphorylation of histone H2A-T120 and histone H3-T3 , respectively . In addition , they reveal a role for Bub1 kinase activity , but not Haspin , in restricting CPC localization to the centromere . Considering the role of Aurora B kinase in the regulation of KT-MT interactions and SAC signaling , the above results raised the question of what contributions Bub1 activity might possibly make to chromosome congression and/or the SAC . Although our initial analyses had not revealed a major impact of BAY-320 or BAY-524 on the overall timing of mitotic progression ( Figure 2 ) , we considered the possibility that inhibition of Bub1 might provoke compensatory effects on mitotic timing , notably a delay in congression and a concomitant acceleration of mitotic exit . According to such a scenario , effects on timing might conceivably cancel each other . In support of this possibility , we emphasize that the inhibition of mitotic kinases with pleiotropic functions have previously been shown to provoke opposing phenotypes ( Santaguida et al . , 2011; von Schubert et al . , 2015 ) . To explore the possibility of compensatory effects of Bub1 inhibition , we thus carried out more detailed analyses of mitotic progression , notably KT-recruitment of SAC components , SAC signaling and chromosome congression . Depletion of Bub1 is known to weaken SAC signaling in human cells ( Klebig et al . , 2009; Meraldi and Sorger , 2005; Perera et al . , 2007 ) . To test the impact of Bub1 catalytic activity on SAC function , we first analyzed KT levels of Mad1 , Mad2 and BubR1 in BAY-320 or BAY-524 treated cells . With the possible exception of a very minor effect on BubR1 , the localization of none of these SAC proteins was significantly affected by Bub1 inhibition ( Figure 6A and B , Figure 6—figure supplement 1A ) . In sharp contrast , and in agreement with previous reports ( Boyarchuk et al . , 2007; Johnson et al . , 2004; Overlack et al . , 2015; Sharp-Baker and Chen , 2001 ) , Bub1 depletion decreased KT recruitment of all three proteins by 80–90% ( Figure 6A and B , Figure 6—figure supplement 1B ) . Thus , the recruitment of several SAC components to KTs strongly depends on Bub1 protein , but not Bub1 kinase activity . 10 . 7554/eLife . 12187 . 013Figure 6 . Bub1 inhibition marginally affects SAC signaling . ( A ) Inhibition of Bub1 kinase does not significantly affect recruitment of SAC effectors to unattached KTs . HeLa S3 cells were synchronized by thymidine block and released for 10 hr in the presence of solvent ( control ) , 3 μM BAY-320 or 7 μM BAY-524 . Cells transfected with mock ( Gl2 ) or Bub1 siRNA-oligonucleotides for 48 hr were synchronized and analyzed in parallel . The cells were fixed and stained for Bub1 , Mad1 , closed Mad2 ( C-Mad2 ) , CREST and DNA ( DAPI ) and analyzed by IFM . ( B ) Histogram shows average KT levels of indicated proteins ( n = 20–50 cells per condition ) observed in the experiment shown in ( A ) . Error bars represent SEM . ( C ) RPE1 cells expressing endogenously tagged Bub1-EGFP were synchronized in mitosis by overnight treatment with the Eg5 inhibitor STLC ( 10 μM ) and subsequently treated with 3 . 3 µM nocodazole and 20 μM MG132 as well as solvent ( control ) , 10 μM BAY-320 or 10 μM BAY-524 . Bub1-EGFP KT levels were recorded by 1 sec time-lapse microscopy . After 5 sec , a single KT pair was bleached and fluorescence recovery was monitored . Traces illustrate average fluorescence recovery at KT pairs ( n = 10–16 KT pairs per condition ) ; shaded areas represent standard deviation ( SD ) . Half-times and plateaus were determined by non-linear curve fitting based on a one-phase association . ( D , E ) Asynchronously growing cultures of HeLa S3 ( D ) or RPE1 ( E ) cells stably expressing GFP-tagged histone H2B were either directly treated with 3 . 3 µM nocodazole and the kinase inhibitors BAY-320 ( 3 μM ) and BAY-524 ( 7 μM ) or transfected with control ( Gl2 ) or Bub1 siRNA for 48 hr prior to addition of nocodazole . Cell fates ( continued arrest , apoptosis or slippage ) and duration of mitotic arrest were determined by fluorescence time-lapse imaging ( n = 150 cells per condition , accumulated from 3 independent experiments ) . Frequencies of observed cell fates as well as average times of arrest are indicated . ( F ) Asynchronously growing HeLa S3 cells or HeLa cells stably expressing GFP-tagged histone H2B were treated with 3 . 3 µM nocodazole and 0 . 5 µM of the Mps1 inhibitor Reversine as well as solvent ( control ) , 3 and 10 μM BAY-320 , 7 and 10 μM BAY-524 or 2 . 5 µM of the Aurora B inhibitor ZM-447439 ( ZM ) ( left panel ) . Alternatively , cells were transfected with control ( Gl2 ) or Bub1 siRNA oligonucleotides for 48 hr prior to addition of 3 . 3 μM nocodazole and 0 . 5 µM Reversine ( right panel ) . The cells were monitored by fluorescence time-lapse microscopy and the time elapsed from nuclear envelope breakdown to SAC override and mitotic slippage was determined . Traces illustrate the cumulative frequency of mitotic duration before slippage ( n = 50 cells per condition ) . ( G ) Asynchronously growing RPE1 cells stably expressing GFP-tagged histone H2B were treated and analyzed as described in ( F ) . Scale bars represent 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12187 . 01310 . 7554/eLife . 12187 . 014Figure 6—figure supplement 1 . Panels relate to the quantitative data shown in Figure 6D . Scale bars represent 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12187 . 01410 . 7554/eLife . 12187 . 015Figure 6—figure supplement 2 . Tagging strategy and IFM analysis of Bub1 KT levels . ( A ) Schematic illustrating the targeting strategy used to introduce the EGFP open-reading frame into the indicated locus of RPE1 cells . The EGFP open reading frame was fused to the 3’ exon of one allele of the BUB1L gene . ( B ) HeLa cells were synchronized by thymidine block and released for 10 hr in the presence of solvent ( control ) , 3 μM BAY-320 or 7 μM BAY-524 . Cells were fixed and stained for Bub1 , CREST and DNA ( DAPI ) and analyzed by IFM . Scale bar represents 10 µm . ( C ) Histogram shows average Bub1 KT levels as measured in the experiment shown in ( B ) ; n = 20–22 cells per condition . DOI: http://dx . doi . org/10 . 7554/eLife . 12187 . 015 The association of Bub1 with unattached KTs is dynamic ( Howell et al . , 2004 ) , raising the question of how Bub1 turnover at KTs is regulated . In the case of the SAC kinase Mps1 , autophosphorylation constitutes a major mechanism for controlling Mps1 levels at KTs ( Hewitt et al . , 2010; Jelluma et al . , 2010; von Schubert et al . , 2015 ) , and a recent study suggests that Bub1 turnover at KTs is also regulated by autophosphorylation ( Asghar et al . , 2015 ) . To determine whether Bub1 dynamics at KTs is affected by inhibition of Bub1 activity , we made use of an RPE1 cell line harboring one allele of Bub1 tagged by EGFP at the endogenous locus ( Figure 6C , Figure 6—figure supplement 2A ) . After the treatment of cells with nocodazole to assure complete MT depolymerization and full SAC activation ( Santaguida et al . , 2011; Yang et al . , 2009 ) , Bub1 levels and turnover at KTs were measured by immunofluorescence microscopy and fluorescence recovery after photobleaching ( FRAP ) , respectively . In comparison to control cells , neither BAY-320 nor BAY-524 detectably affected steady-state Bub1 levels at KTs ( Figure 6—figure supplement 2B and C ) , in line with a recent report ( Liu et al . , 2015 ) . More importantly , FRAP experiments revealed only minor effects of Bub1 inhibition on Bub1 dynamics at KTs ( Figure 6C ) . The extent of fluorescence recovery after FRAP was not significantly different in control cells and inhibitor treated cells , revealing an immobile fraction of ~42% , in excellent agreement with previous data ( Asghar et al . , 2015; Howell et al . , 2004 ) . The half-time of Bub1 recovery at KTs after FRAP was ~18 sec in controls , again in good agreement with previous data ( Asghar et al . , 2015; Howell et al . , 2004 ) . However , whereas Asghar and colleagues observed a ~50% reduction in the half-time of recovery of an exogenously expressed , catalytically inactive EGFP-Bub1 mutant , we found that recovery of endogenously tagged wild-type EGFP-Bub1 was only marginally accelerated by Bub1 inhibition ( half-time reduced from 18 s to 12–15 s ) ( Figure 6C ) . When considering this discrepancy , it is important to bear in mind that our data reflect turnover of chemically inhibited wild-type Bub1 expressed at endogenous levels , whereas Ashghar and colleagues monitored mutant versions of overexpressed Bub1 , raising the possibility that their results may have been influenced by expression levels and/or mutation-induced structural alterations . We conclude that the effects of Bub1 activity on Bub1 turnover at KTs are at most minor , particularly when compared to the striking effects of Mps1 activity on Mps1 dynamics at KTs ( Hewitt et al . , 2010; Jelluma et al . , 2010; von Schubert et al . , 2015 ) . As a further read-out for the effects of Bub1 inhibition on SAC activity , we used live cell imaging to monitor the responses of nocodazole-arrested HeLa and RPE1 cells to BAY-320 or BAY-524 and compared these to the responses seen in Bub1-depleted cells ( Figure 6D and E ) . Over a 24 hr observation period , the percentage of HeLa cells maintaining a SAC arrest dropped from 17% in controls to 4% and 2% in response to Bub1 inhibition by BAY-320 and BAY-524 , respectively ( Figure 6D , left panel ) . These shifts in cell fates were largely compensated by increases in the percentages of cells undergoing apoptosis , from 74% in controls to 94% in Bub1-inhibited cells . In contrast , although the duration of mitosis was slightly reduced upon Bub1 inhibition , the extent of mitotic slippage remained at less than 10% under all conditions . In RPE1 cells , maintenance of SAC arrest over 24 hr was more pronounced , but again the percentage of arrested cells dropped from 61% in controls to 51%/44% in response to Bub1-inhibition , with increasing proportions of cells undergoing apoptosis or mitotic slippage ( Figure 6E , left panel ) . For comparison , depletion of Bub1 from either HeLa or RPE1 cells resulted in a 2–3 fold increase in mitotic slippage at the expense of apoptosis , while the proportion of cells sustaining an arrest remained roughly constant ( Figure 6D and E , right panels ) . Collectively , these results indicate that Bub1 activity contributes to the maintenance of maximal SAC activity , but that Bub1 protein levels are more important , most likely reflecting the observed role of Bub1 in the KT recruitment of SAC components ( Figure 6A ) . Importantly , we also compared the requirements for Bub1 activity and Bub1 protein in a cellular background in which SAC activity was partially compromised by the treatment of HeLa or RPE1 cells with a low dose of Reversine , a widely used inhibitor of the SAC kinase Mps1 ( Santaguida et al . , 2010 ) . In agreement with the results described above , Bub1 inhibition marginally reduced the time that Reversine-treated cells remained arrested before overriding nocodazole-induced arrest ( Figure 6F and G , left panels ) . Addition of Aurora B or Plk1 inhibitors , used as positive controls , led to the expected shortening of the duration of mitotic arrest ( Figure 6F and G , left panels ) ( Saurin et al . , 2011; von Schubert et al . , 2015 ) . Similarly , Bub1 depletion also caused a drastic shortening of arrest ( Figure 6F and G , right panels ) . Taken together with previous studies ( Klebig et al . , 2009; Perera and Taylor , 2010a; Perera et al . , 2007; Ricke et al . , 2012 ) , these observations demonstrate that the scaffolding function of Bub1 is required for the SAC , but its catalytic activity is largely dispensable . To analyze the impact of Bub1 inhibition on chromosome alignment , we treated cells with the Eg5 inhibitor Monastrol ( Kapoor et al . , 2000 ) and then monitored the restoration of KT-MT attachments during spindle bipolarization in response to drug washout ( Figure 7A and B ) . While nearly 28% of Bub1-depleted cells failed to completely align all chromosomes , more than 90% of Bub1-inhibited cells showed complete alignment that was indistinguishable from control cells . Inhibition of Aurora B , analyzed for control , resulted in the expected impairment of alignment ( Figure 7A and B ) . To complement these assays , we also used immunofluorescence microscopy to quantify the frequency of micronucleation , a read-out for chromosome segregation errors , in HeLa and RPE1 cells . While partial inhibition of Aurora B kinase provoked an increase in micronucleation in both cell lines , as expected ( Gohard et al . , 2014; Tao et al . , 2009 ) , Bub1 inhibition only marginally increased the frequency of micronucleation ( Figure 7C ) . This result supports the view that Bub1 inhibition causes surprisingly mild defects in chromosome congression or segregation ( Figure 2 and 7A ) . Further corroborating this conclusion , we found that BAY-320 or BAY-524 treatment exerted no significant effects on the kinetochore recruitment of the motor protein CENP-E ( Figure 7D and E , left panels ) . In contrast , Bub1 depletion reduced CENP-E levels at KTs by ~40% ( Figure 7D and E , right panels ) , in agreement with previous reports ( Johnson et al . , 2004; Sharp-Baker and Chen , 2001 ) . Taken together , these results show that Bub1 kinase activity is largely dispensable for chromosome congression and segregation . It follows that even though Bub1 inhibition results in a marked reduction of Aurora B levels at centromeres ( Figure 4 ) , these levels are still sufficient to ensure largely faithful chromosome segregation . Conversely , Bub1 protein is clearly important for efficient chromosome congression , presumably reflecting the role of Bub1 in CENP-E recruitment to KTs . 10 . 7554/eLife . 12187 . 016Figure 7 . Bub1 inhibition does not significantly affect chromosome congression . ( A ) HeLa S3 cells were transfected with control ( Gl2 ) or Bub1 siRNA-oligonucleotides for 48 hr , synchronized by thymidine block and released for 12 hr in the presence of the Eg5 inhibitor monastrol to induce the formation of monopolar spindles . The capacity of spindle bipolarization and metaphase plate formation was tested by monastrol wash-out and addition of MG132 and indicated drugs for 2 hr ( n = 170–200 cells ) . Percentages indicate the frequencies of depicted spindle morphologies . ( B ) Histograms show the frequencies of full , partial ( ≤5 unaligned chromosomes ) or failed metaphase chromosome alignments that were observed in the experiment shown in ( A ) . ( C ) HeLa S3 and RPE1 cells were treated for 16 hr with the indicated drugs , fixed and analyzed by IFM . Histograms show the frequency of micronucleation among interphase cells ( n = 300 cells per condition ) . ( D ) Depletion but not inhibition of Bub1 kinase affects recruitment of CENP-E to unattached kinetochores . Untreated HeLa S3 cells or cells transfected with control ( Gl2 ) or Bub1 siRNA-oligonucleotides ( for 48 hr ) were synchronized by thymidine block and released for 10 hr in the presence or absence of 3 μM BAY-320 or 7 μM BAY-524 . The cells were fixed and stained for CENP-E , CREST , DNA ( DAPI ) and analyzed by IFM . ( E ) Histograms show average CENP-E KT levels observed in prometaphase cells . Data relate to micrographs shown in ( D ) . Error bars represent SEM . Scale bars represent 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 12187 . 016 Interference with the SAC proteins Mps1 or BubR1 was previously shown to exert synergistic effects with Paclitaxel treatment of tumor cells , significantly elevating the frequency of chromosome missegregation and lethality ( Janssen et al . , 2009; Lee et al . , 2004 ) . Thus , we asked how inhibition of Bub1 kinase activity by BAY-320 or BAY-524 would impact on cells in which MT dynamics was compromised by low doses of Paclitaxel . Importantly , when used at clinically relevant doses of 1–4 nM , Paclitaxel induces spindle defects and aneuploidy without delaying mitotic progression ( Brito and Rieder , 2009; Chen and Horwitz , 2002; Ikui et al . , 2005; Janssen et al . , 2009 ) . While single treatment with 1–4 nM Paclitaxel produced modest impairment of cell proliferation , the concomitant application of the Bub1 inhibitors , BAY-320 at 3 μM or BAY-524 at 7 or 10 μM , clearly exacerbated inhibition of proliferation . Effects were particularly drastic in aneuploid HeLa cells ( Figure 8A and B , top panels ) , while diploid RPE1 cells were less affected ( Figure 8A and B , bottom panels ) . For comparison , we also examined the effects of combining low dose Paclitaxel treatment with partial inhibition of Mps1 by Reversine ( Janssen et al . , 2009 ) . This analysis shows that the combination of Paclitaxel with either Mps1 or Bub1 inhibition produced similar synergistic effects , albeit with cell-type specific differences ( Figure 8A and B ) . Using extensive dose-response analyses , synergy between BAY-320 and Paclitaxel treatment was further confirmed for both HeLa ( Figure 8C ) and non-small cell lung cancer cells ( Figure 8D ) . In future , it will be interesting to determine to what extent combined treatments differentially affect aneuploid versus diploid cells ( Janssen et al . , 2009; Kops et al . , 2004; Maia et al . , 2015 ) . 10 . 7554/eLife . 12187 . 017Figure 8 . BAY-320 and BAY-524 treatment sensitizes cells to low doses of Paclitaxel . ( A ) Micrographs show colony formation of HeLa ( top panel ) and RPE1 cells ( bottom panel ) treated for 7 days with solvent ( control ) or the indicated kinase inhibitors in the presence or absence of 4 nM Paclitaxel . ( B ) Histograms quantify colony formation in HeLa ( top panels ) and RPE1 cells ( bottom panels ) treated with the indicated kinase inhibitors in the presence or absence of 1–4 nM Paclitaxel for 7 days . ( C , D ) IC50-Isobolograms confirm the synergistic effect of BAY-320 and Paclitaxel on cell survival . HeLa cells ( C ) or NCI-H1299 non-small cell lung cancer cells ( D ) were grown in the presence various concentrations of BAY-320 ( 0 . 1–10 μM ) and paclitaxel ( 1–100 nM ) in mono ( Z1 , Z11 ) and in nine different fixed-ratio combinations ( Z2-Z10 ) . IC50 values were determined and the respective BAY-320 and Paclitaxel concentrations plotted in IC50 Isobolograms ( left panel ) . The grey dashed lines indicate the results expected for additivity . Combination indices ( CIs ) were calculated according to the median-effect model of Chou-Talalay ( Chou , 2006 ) and plotted over fixed-ratio combinations Z2-Z10 ( right panel ) . The red dashed line indicates a CI of 0 . 8 ( defined as upper limit for a synergistic interaction ) . ( E ) Time-lapse stills of HeLa cells expressing H2B-GFP illustrate chromosome segregation defects that were used to classify cell fates in the experiments described in ( F ) ; arrowheads point to chromosome bridges and lagging chromosomes . ( F ) HeLa ( top panels ) and RPE1 cells ( bottom panels ) stably expressing H2B-GFP were treated with solvent ( control ) or the indicated kinase inhibitors in the presence or absence of 1–4 nM Paclitaxel and monitored by fluorescence time-lapse imaging . Histograms show the frequencies of chromosome segregation defects , following the classification illustrated in ( E ) ( n = 100 cells per condition ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12187 . 017 To assess whether the observed impairment of proliferation results from errors in chromosome segregation ( Kops et al . , 2004 ) , we scored HeLa and RPE1 cells expressing GFP-H2B for mild and severe chromosomal defects , as illustrated in Figure 8E . Following application of the above Paclitaxel and Bub1/Mps1 inhibitor treatments , the frequencies of chromosomal defects were monitored by fluorescence time-lapse imaging and quantified ( Figure 8F ) . Consistent with the micronucleation data described above ( Figure 7 ) , Bub1 inhibition alone did not significantly elevate the frequency of chromosome missegregation in either HeLa or RPE1 cells ( Figure 8F ) . For comparison , interference with the SAC by inhibition of Mps1 led to a marked increase in segregation defects in both cell lines , as expected ( Figure 8F ) . Most importantly , HeLa cells displayed an even higher frequency of severe chromosome segregation defects when Bub1 inhibition was combined with 1–4 nM Paclitaxel , comparable to the consequences of combined Mps1 inhibition and Paclitaxel treatment ( Figure 8F ) . In contrast , Bub1 inhibition only marginally elevated the rate of Paclitaxel-induced chromosome missegregation in RPE1 cells , while combinatorial treatment with Reversine still resulted in a high rate of mild segregation defects . Considering the strong correlation in the data shown in Figures 7 and 8 , it is tempting to conclude that chromosome segregation errors constitute the most likely cause for the observed impairment of cell proliferation when Bub1 inhibition is combined with low dose Paclitaxel treatment . Thus , although inhibition of Bub1 kinase activity per se exerts only minor effects on SAC functionality , chromosome segregation and mitotic progression , treatment with BAY-320 or BAY-524 sensitizes cells to low , clinically relevant doses of Paclitaxel . These findings are clearly relevant for the potential therapeutic use of Bub1 inhibitors . Clarification of the role of Bub1 activity in mitosis has previously been hampered by the absence of specific inhibitors . While genetic or siRNA-mediated depletion experiments generally suffer from poor temporal resolution , small molecule inhibitors offer a unique opportunity for acute kinase inactivation . The only previously described inhibitor of Bub1 is the bulky ATP analog 2OH-BNPP1 ( Kang et al . , 2008; Krenn et al . , 2012; Liu et al . , 2015; Nyati et al . , 2015 ) , but neither the specificity nor the efficacy of this compound in intact cells have been thoroughly characterized . Here , we document that both BAY-320 and BAY-524 effectively inhibit Bub1 kinase activity , in intact cells as well as in vitro . In addition to inhibiting phosphorylation of histone H2A-T120 , both compounds cause a persistence of chromosome arm cohesion , validating their efficacy ( Kawashima et al . , 2010; Liu et al . , 2015 ) . Moreover , in vitro assays performed on a large panel of kinases showed that inhibition of off-target substrates required at least 20x higher concentrations of BAY-320 than inhibition of Bub1 , and for one potentially relevant off-target , the kinase Haspin , we show that intracellular phosphorylation of a major substrate of this kinase is not decreased by inhibitor concentrations that effectively inhibit Bub1 . Furthermore , a binding assay performed on a panel of 403 human kinases documents exquisite selectivity of BAY-320 for Bub1 . Thus , we are confident that these new Bub1 inhibitors constitute highly effective and specific tools to explore the role of Bub1 kinase activity . The major conclusion emerging from the present study is that the overall impact of Bub1 inhibition on mitotic progression is surprisingly mild , clearly less severe than the impact of Bub1 depletion . This reinforces the notion that the requirement for Bub1 during chromosome congression and segregation primarily reflects a scaffolding function ( Fernius and Hardwick , 2007; Klebig et al . , 2009; Perera and Taylor , 2010b; Rischitor et al . , 2007; Sharp-Baker and Chen , 2001 ) . It is difficult to exclude that a small fraction of total Bub1 kinase activity might be refractory to inhibition and suffice for functionality , but we emphasize that genetic elimination of Bub1 kinase activity is compatible with mouse development , arguing similarly against an essential role of Bub1 kinase activity for mitotic progression ( Ricke et al . , 2012 ) . Thus , inhibitor studies and genetic data concur to indicate that lack of Bub1 kinase activity produces only mild disturbances of mitotic progression . We show that Bub1 inhibition by BAY-320 or BAY-524 results in loss of Shugoshin and CPC subunits proteins from centromeres/KTs , and that CPC levels at these locations are further reduced upon simultaneous inhibition of Bub1 and Haspin , in line with the established roles of these kinases in the phosphorylation of histones H2A-T120 and H3-T3 , respectively ( Wang et al . , 2011; Yamagishi et al . , 2010 ) . Furthermore , both Sgo2 and Aurora B relocalize to chromosome arms when Bub1 is inhibited . Phosphorylation of Sgo2 by Aurora B promotes Sgo2 interaction with phosphatase 2A ( PP2A ) ( Tanno et al . , 2010 ) , and PP2A in turn protects cohesin proteins against phosphorylation ( Kitajima et al . , 2006; Riedel et al . , 2006 ) . Thus , colocalization of Sgo2 and Aurora B in Bub1-inhibited cells provides a straightforward explanation for the observed persistence of arm cohesion ( Kitajima et al . , 2005; Perera et al . , 2007 ) . However , alternative mechanisms should not be excluded and it might be rewarding to explore a possible connection between Bub1 kinase and the Sororin-Wapl pathway ( Peters and Nishiyama , 2012 ) . It may appear surprising that the persistent chromosome arm cohesion observed in Bub1 inhibited cells did not markedly prolong mitotic timing ( Figure 2 ) . However , we note that depletion of Wapl also causes persistent cohesion without significantly affecting mitotic progression ( Lara-Gonzalez and Taylor , 2012 ) . In contrast to Bub1 inhibition , depletion of Bub1 markedly extended mitotic timing ( Figure 2 ) . One straightforward explanation for this observation is that Bub1 depletion , but not Bub1 inhibition , caused the displacement of CENP-E from KTs , a motor protein required for efficient chromosome congression ( Bancroft et al . , 2015; Putkey et al . , 2002; Tanudji et al . , 2004 ) . A mechanism centered on CENP-E may also explain the observation that Bub1 depletion exerted a more extensive mitotic delay in the hypertriploid HeLa cells than in diploid RPE1 cells . Considering the important role of Aurora B in the regulation of KT-MT interactions ( Carmena et al . , 2012; Funabiki and Wynne , 2013 ) , it is remarkable that the observed reduction of centromere/KT-associated CPC caused by Bub1 inhibition did not exert a more profound effect on the fidelity of chromosome segregation . This suggests that approximately half the normal centromere/KT levels of CPC are sufficient to confer functionality . In line with this conclusion , we note that partial impairment of CPC recruitment to centromeres/KTs did not abolish viability or trigger extensive defects in chromosome segregation in budding yeast or chicken DT40 cells ( Campbell and Desai , 2013; Yue et al . , 2008 ) . Inhibition of Bub1 kinase activity did not significantly reduce the KT recruitment of Mad1 , Mad2 and BubR1 and barely affected the ability of nocodazole-treated cells to maintain a SAC arrest . Even when SAC activity was compromised by partial inhibition of the SAC kinase Mps1 , Bub1 inhibition triggered only minor weakening of SAC signaling . In striking contrast , Bub1 depletion produced a drastic weakening of the SAC in this sensitized background . For comparison , combined inhibition of Mps1 and either Plk1 or Aurora B resulted in a complete SAC shutdown and immediate mitotic exit , in line with previous results ( Saurin et al . , 2011; von Schubert et al . , 2015 ) . Collectively these findings confirm that mitotic functions of Bub1 depend primarily on Bub1 protein rather than kinase activity . In future , it will be interesting to explore whether Bub1 activity contributes to purported non-mitotic functions of Bub1 ( Nyati et al . , 2015; Yang et al . , 2012 ) . Inhibition of SAC kinases has emerged as a potentially attractive strategy to kill tumor cells ( Janssen et al . , 2009; Salmela and Kallio , 2013 ) . Several inhibitors of the SAC kinase Mps1 were shown to exert anti-tumor effects in mouse models ( Colombo et al . , 2010; Kusakabe et al . , 2015; Tannous et al . , 2013; Tardif et al . , 2011 ) , but toxicity associated with single agent therapy remains a concern ( Martinez et al . , 2015 ) . Instead , combination of anti-SAC compounds with MT-targeting agents may constitute a more rewarding strategy ( Jemaà et al . , 2013; Maia et al . , 2015 ) . Our present data suggest that it may be attractive to use inhibitors of Bub1 in combinatorial therapy . While BAY-320 and BAY-524 had comparatively little effect on mitotic progression when used as single agents , they showed extensive anti-proliferative activity , accompanied by strong increases in chromosome segregation errors , when combined with therapeutic doses of Paclitaxel . A plausible explanation for this synergy is that Paclitaxel increases KT-MT attachment errors to levels that can no longer be corrected when Aurora B/CPC is partially displaced upon Bub1 inhibition . Interestingly , these synergistic effects were substantially more pronounced in aneuploid HeLa cells than in near-diploid RPE1 cells , suggesting a potential therapeutic window . These findings clearly encourage further exploration of the potential use of Bub1 inhibitors for therapeutic applications . BAY-320 and BAY-524 were synthesized as described previously ( Hitchcock et al . , 2013 ) . For biochemical and cellular experiments BAY-320 and BAY-524 were used from stock solutions in dimethyl sulfoxide ( DMSO ) . Working concentration of Bub1 inhibitors are indicated in Figures and Figure legends , respectively . Inhibitory activities BAY-320 and BAY-524 towards Bub1 in presence of 2 mM ATP were quantified as previously published ( Hitchcock et al . , 2013 ) . A time-resolved fluorescence energy transfer ( TR-FRET ) kinase assay was used to measure phosphorylation of the synthetic peptide Biotin-Ahx-VLLPKKSFAEPG ( C-terminus in amide form , Biosyntan , Berlin , Germany ) by the recombinant catalytic domain of human Bub1 ( amino acids 704–1085 ) . Recombinant human Bub1 ( 704–1085 ) was expressed in Hi5 insect cells with an N-terminal His6-tag and purified by affinity- ( Ni-NTA ) and size exclusion chromatography . BAY-320 was screenedin vitro , at 10 µM and 10 µM ATP , against a panel of 222 kinases using the Eurofins kinase profiler screen ( Millipore ) . In addition , BAY-320 was screened , at 300 and 1000 nM , in an active site-directed competition-binding assay measuring 403 human kinases ( Lead Hunter , DiscoverX Kinome Scan ) . HEK 293T cells were transfected with plasmids coding for LAP-tagged Bub1 wild-type ( WT ) or the K821R kinase-dead ( KD ) mutant ( kindly provided by G . Kops , Utrecht , Netherlands ) ( Suijkerbuijk et al . , 2012 ) . After induction of mitotic arrest ( 18 hr incubation with 1 μg/ml of nocodazole ) , the cells were harvested and lysed in kinase lysis buffer ( 50 mM HEPES pH7 . 5 , 150 mM NaCl , 5 mM EDTA , 0 . 5% NP-40 , 1 mM Na3VO4 , 1 mM β-glycerophosphate , 1 mM NaF and complete protease inhibitor ( Roche ) ) . Lysates were cleared by centrifugation for 15 min at 21 , 000 g , 4°C , and LAP-Bub1 proteins isolated by a 2 hr incubation with S-protein-agarose ( Novagen , EMD Chemical , CA , USA ) . Beads were washed six times in lysis buffer containing increasing concentrations of NaCl ( 150 mM , 200 mM , 300 mM , 400 mM , 500 mM and 600 mM ) and three times in kinase buffer ( 20 mM HEPES pH7 . 5 , 100 mM KCl , 10 mM MgCl , 1 mM Na3VO4 , 1 mM -glycerophosphate , 1 mM NaF , 1 mM DTT ) . The bead-bound LAP-Bub1 was then aliquoted and used for kinase assays in 30-μl reaction volumes . Kinase reactions were carried out at 30°C in kinase buffer in the presence of 100 μM ATP , 5 μCi γ-32P-ATP , 1 μg recombinant histone H2A ( NEB , Frankfurt am Main , Germany ) as substrate , and serial dilutions of Bub1 inhibitors . The reactions were stopped after 30 min by the addition of sample buffer and heating to 95°C . The samples were then resolved by SDS-PAGE and visualized by autoradiography and Western blotting . HeLa S3 cells , HeLa S3 cells expressing histone H2B-GFP ( Silljé et al . , 2006 ) , HeLa Kyoto cells expressing a FRET reporter for Aurora B fused to histone H2B ( van der Waal et al . , 2012 ) and HEK293T cells were grown under standard conditions in DMEM-Glutamax medium ( Invitrogen , CA , USA ) , supplemented with 10% heat-inactivated fetal calf serum ( FCS ) ( PAN Biotech , Aidenbach , Germany ) and penicillin-streptomycin ( Pen-Strep; 100 IU/ml and 100 mg/ml respectively , Gibco Life Technologies , Zug , Switzerland ) . hTERT-RPE1 cells and hTERT-RPE1 cells expressing histone H2B-GFP ( kind gift of Stephen Taylor , University of Manchester , UK ) were cultured in F12 DMEM nutrient mixture F-12 HAM ( Sigma Aldrich , MO , USA ) supplemented with 10% heat-inactivated FCS , L-glutamine ( 2 mM; PAN Biotech , Aidenbach , Germany ) , sodium bicarbonate ( 0 . 35%; Sigma-Aldrich , MO , USA ) and Pen-Strep . NCI-H1299 cells were grown under standard conditions in RPMI-1640 medium supplemented with L-glutamine ( Biochrome , Berlin , Germany ) and 10% fetal calf serum ( Biochrome , Berlin , Germany ) . All cell lines were routinely tested for mycoplasma , using PCR ( by the lab in Basel ) or the MycoAlert Mycoplasma Detection Assay ( by the lab in Berlin ) . HeLa cells ( ACC-57 ) were obtained from the German Collection of Microorganisms and Cell Cultures , Braunschweig , and authentication was done at provider prior to shipment; NCI-H1299 ( CRL-5803 ) were obtained from ATCC and authentication was done by STR profiling ( authentication service at German Collection of Microorganisms and Cell Cultures , Braunschweig ) . Thymidine arrest was performed for 24 hr and cells were either released into fresh medium for 10 hr or into medium supplemented with Nocodazole for 12–14 hr . Thymidine ( Sigma-Aldrich ) was used at 2 mM if not stated otherwise , Nocodazole ( Sigma-Aldrich ) at 3 . 3 μM if not stated otherwise , RO-3306 at 10 μM ( Calbiochem , Darmstadt , Germany ) , Paclitaxel ( Calbiochem ) at 1–4 nM , Reversine ( Enzo Life Sciences , Lausen , Switzerland ) at 0 . 25 and 0 . 5 μM , ZM-447439 ( Tocris Bioscience , [Ditchfield et al . , 2003] ) at 1 . 25 , 2 . 5 and 5 . 0 μM , 5-Iodotubercidin ( 5’Itu , Santa Cruz Biotechnology , TX , US ) at 2 . 5 μM , Monastrol ( Enzo Life Sciences ) at 150 μM and MG132 ( Calbiochem ) at 10 and 20 μM . Transient transfections of HEK293T cells with plasmids and small interfering RNA ( siRNA ) duplexes were performed using TransIT-LT1 transfection reagent ( Mirus Bio , Madison , WI ) and Oligofectamine ( Invitrogen ) , respectively , according to manufacturers protocols . The following siRNA duplex oligonucleotides were used: siGl2 CGTACGCGGAATACTTCGA ( Elbashir et al . , 2001 ) , siBub1 CCAGGCTGAACCCAGAGAGTT ( Tang 2004 ) . All siRNA duplex oligonucleotides were ordered from Qiagen ( Hilden , Germany ) . HeLa S3 or RPE1 cells were incubated with kinase inhibitors or depleted of the indicated proteins for 48 hr . Cell suspensions were then fixed with 70% ice-cold ethanol and incubated with 0 . 2 mg/ml RNase ( Sigma-Aldrich ) and 5 μg/ml propidium iodide ( Sigma-Aldrich ) . Cellular DNA content was determined by flow cytometry using FACSCanto II ( BD Biosciences Clontech , San Jose , CA , USA ) and FlowJo ( Treestar , Ashland , OR , USA ) instruments . Cells extracts were prepared on ice for 30 min in Tris lysis buffer ( 20 mM Tris , pH 7 . 4 , 150 mM NaCl , 0 . 5% IGEPAL CA-630 , 30 μg/ml RNAse , 30 μg/ml DNAse , 1 mM DTT , protease inhibitors cocktail ( Roche , Basel , Switzerland ) and phosphatase inhibitor cocktails ( cocktails 2 and 3 , Sigma-Aldrich ) . Lysates were cleared by centrifugation for 15 min at 21 , 000 g , 4°C , and proteins were resolved by SDS-PAGE and analyzed by Western blotting . HeLa S3 cells were as described above and mitotic cells were collected by shake-off . The cells were the washed with cold PBS and lysed at 4°C for 30 min using histone lysis buffer ( 50 mM Tris pH 7 . 8 , 300 mM NaCl , 1% IGEPAL CA-630 ) . Nuclei were collected by centrifugation ( 110 g , 4°C , 10 min ) and washed three times with histone lysis buffer . After an additional wash with Tris-EDTA ( 100 mM Tris , 1 mM EDTA ) , the nuclear pellet was incubated for 2 hr in 0 . 4 M HCl at 4°C . After high-speed centrifugation of the sample , 6 volumes of acetone were added to the supernatant , followed by overnight incubation at -20°C . Histones were collected by centrifugation , washed with acetone , air-dried and resolved by SDS-PAGE . Antibodies used for Western blotting: anti-Bub1 ( [Hanisch et al . , 2006] or ab9000 , Abcam , Cambridge , UK ) , anti-pT120-H2A ( Active Motif , Carlsbad , CA , USA ) and anti-α-tubulin ( DM1A , Sigma-Aldrich ) . Antibodies used for immunofluorescence microscopy: anti-Mad1 ( clone 117–468 [Fava et al . , 2011] ) , anti-cMad2 ( clone 107–276 [Fava et al . , 2011] ) anti-Borealin ( Klein et al . , 2006 ) , anti-INCENP ( clone 58–217 , ab23956 , Abcam ) , anti-Bub1 ( antibody against Bub1 hybridoma ( clone 62–406 ) was produced after mice were injected with Bub1 recombinant protein spanning residues 1–318 , anti-Bub1 ( ab9000 , Abcam ) , CREST anti-human auto-immune serum ( Immunovision , Springdale , AR , USA ) , anti-Aurora B ( AIM-1 , BD Biosciences , San Jose , CA , USA ) , anti-Bub1 ( ab9000 , Abcam ) , anti-CENP-E ( 1H12 , Abcam ) , anti-Mad2 ( A300-301A , Bethyl Laboratories , Montgomery , TX , USA ) , anti-Sgo1 ( Abnova , Taipei , Taiwan ) , anti-Sgo2 ( Bethyl Laboratories ) , anti-pT120-H2A ( Active Motif , Carlsbad , CA , USA; ABIN482721 ) , anti-pS7CENP-A ( clone NL41 , Merck Millipore , Billerica , MA , USA ) , anti-pT3-H3 ( clone 9714 , Cell Signaling Technology , Danvers , MA , USA ) and anti-pS10-H3 ( Millipore , Billerica , MA , USA ) . The polyclonal MCAK ( R120 ) antibody was raised in rabbits by immunization with bacterially expressed His-MCAKaa588-725 . For immunofluorescence experiments , all primary antibodies were detected with AlexaRed-594- , AlexaRed-564- , and AlexaGreen-488-labeled secondary anti-mouse and anti-rabbit antibodies ( Invitrogen , Carlsbad , CA , USA ) or Cy5-conjugated donkey antibodies ( Dianova , Hamburg , Germany ) . For Western blotting , signals were detected using HRP-conjugated anti-mouse or anti-rabbit antibodies ( Pierce , Rockford , IL , USA ) . HeLa cells were seeded into 96-well plates for 5 hr at 37°C ( ca . 25’000 cells/well ) . Then , cells were treated with Nocodazole for 16 hr and varying concentrations of test compounds for 1 hr . The cells were fixed , washed and blocked with buffer before incubating with the primary antibody ( Phospho-Histone H2A; ABIN482721; 1:200 ) overnight at 2–8°C . After washing , secondary IRDye-labeled antibody mix with cell stains was added for 1 hr and washed again . Plates were scanned with a LiCor Odyssey Infrared Imager at 800 nm for P-H2A and at 700 nm for Draq5/Sapphire , a cell stain . The signal ratio ( 800/700 nm ) for cells treated only with Nocodazole was set to 100% and the corresponding ratio for untreated cells to 0% . The IC50 value was then determined by curve fitting ( using a four parameter fit ) . For fluorescence microscopy cells were grown on coverslips and fixed in PTEMF buffer ( 20 mM PIPES , pH 6 . 8 , 0 . 2% Triton X-100 , 10 mM EGTA , 1 mM MgCl2 , 4% formaldehyde ) or methanol at -20°C ( for CENP-A pS7 ) , respectively . Images of randomly selected cell were acquired as z-stacks using a DeltaVision microscope ( GE Healthcare ) on an Olympus IX71 base ( Applied Precision , WA , USA ) , equipped with a Plan Apochromat N 60x/NA1 . 42 oil immersion objective ( Olympus ) and a CoolSNAP HQ2 camera ( Photometrics ) . Serial optical sections were deconvolved and projected using SoftWorx software ( GE Healthcare ) . Images were quantified as previously described ( von Schubert et al . , 2015 ) using automated pipelines run by Cell Profiler software ( Carpenter et al . , 2006 ) . Results from 2–3 independent experiments were pooled and statistical analysis was done with GraphPad Prism software . Error bars on histograms illustrate SEM . Scale bars represent 10 μm . For time-lapse imaging , cells were imaged using a Nikon ECLIPSE Ti microscope equipped with a CoolLED pE-1 illumination system and a 20x/NA0 . 75 air Plan Apochromat objective ( Nikon ) in a climate-controlled environment . Images were acquired at multiple positions at indicated time intervals . MetaMorph 7 . 7 software ( MDS Analytical Technologies , Sunnyvale , CA , USA ) was used for acquisition and processing of data . FRET , FRAP , and high sensitivity microscopy ( monitoring endogenously EGFP-tagged proteins ) experiments were carried out using a spinning disk confocal system ( Intelligent Imaging Innovations ) based on a Zeiss Axio Observer stand equipped with a Photometric Evolve 512 back-illuminated EMCCD camera , 63x/NA1 . 4 plan apochromat objective and diode lasers and run by SlideBook software . FRET analyses were carried out by excitation with a 440 nm diode laser and by recording of CFP ( CFP signal ) and YFP ( FRET signal ) fluorescence emission in z-stacks . Background-corrected FRET ratios ( CFP signal/FRET signal ) were calculated in ImageJ using the Ratio Plus plugin . FRAP analysis of EGFP-Bub1 was performed with a 488 diode laser on one KT pair per cell . Overall bleaching was corrected using the signal intensities at a cytoplasmic region not targeted for photobleaching ( average of the first 4 frames ) . Fluorescence recovery half-times and plateaus were determined by non-linear curve fitting based on a one-phase association in Prism software ( GraphPad ) . Asynchronous cell cultures ( 50 , 000/well ) were plated on 6-well plates ( Falcon ) . After 7 days of proliferation in the presence of the indicated drugs , the cells were fixed with ice-cold methanol at -20°C and stained with 0 . 1% Cresyl Violet according to standard procedures . Dried culture plates were scanned and intensities measured using ImageJ after black-and-white-conversion and inversion of the images . BAY-320 plus Paclitaxel combination studies were conducted with HeLa and NCI-H1299 cells . Cells were plated into 384-well plates at 600 ( HeLa ) or 700 ( NCI-H1299 ) cells per well . After 24 hr , cells were treated with BAY-320 ( concentration range , 1E-07 M to 1E-05 M ) and Paclitaxel ( concentration range , 1E-09 M to 1E-07 M ) for single compound treatments , and in nine different fixed-ratio combinations of BAY-320 ( D1 ) and Paclitaxel ( D2 ) ( 0 . 9xD1+0 . 1xD2 , 0 . 8xD1+0 . 2xD2 , 0 . 7xD1+0 . 3xD2 , 0 . 6xD1+0 . 4xD2 , 0 . 5xD1+0 . 5xD2 , 0 . 4xD1+0 . 6xD2 , 0 . 3xD1+0 . 7xD2 , 0 . 2xD1+0 . 8xD2 , 0 . 1xD1+0 . 9xD2 ) . Cell viability was assessed after 96 hr exposure , using the Cell Titre-Glo Luminescent Cell Viability Assay ( Promega ) . IC50 values were determined by means of a 4-parameter fit after normalization of measurement data to vehicle ( DMSO ) -treated cells ( =100% ) and readings taken immediately before compound addition ( =0% ) . IC50 isobolograms were plotted with the actual concentrations of the two compounds on the x- and y-axis , and the combination index ( CI ) was calculated according to the median-effect model of Chou-Talalay ( Chou , 2006 ) . A CI of ≤0 . 8 was defined as more than additive ( i . e . synergistic ) interaction , and a CI of ≥1 . 2 was defined as antagonistic interaction . For gene targeting , homology arms to human Bub1 ( BUB1 ) gene were amplified from RPE1 cell genomic DNA . Targeting constructs allowing the insertion of an EGFP tag C-terminal to Bub1 were assembled by 4-piece ligation in a NotI-digested pAAV vector . Recombinant adenovirus-associated virus ( rAAV ) particles were generated as previously described ( Berdougo et al . , 2009 ) . RPE1 cells were infected with 3 ml of viral supernatant for 48 hr and then expanded into fresh medium for an additional 48 hr . FACS sorting was used to select EGFP-positive cells , as previously described ( Collin et al . , 2013 ) . To facilitate detection of fluorescence at mitotic stages , cells were synchronized with RO-3306 ( 10 μM ) for 18 hr and released into nocodazole ( 50 nM ) for 2 hr , before they were trypsinized and subjected to sorting in the continued presence of nocodazole ( 10 nM ) . Infected or uninfected cells were filtered ( 30 μm , Partec ) and EGFP-positive cells ( 488 excitation , 514/30 emission filter ) were isolated on an Aria IIIu ( BD ) cell sorter by selecting the 514/30 channel against a 585/42 filter detecting cellular autofluorescence . Single cells were sorted into 96-well plates filled with conditioned medium and positive clones screened for by fluorescence microscopy .	The DNA in our cells is packaged into structures called chromosomes . When a cell divides , these chromosomes need to be copied and then correctly separated so that both daughter cells have a full set of genetic information . Errors in separating chromosomes can lead to the death of cells , birth defects or contribute to the development of cancer . Chromosomes are separated by an array of protein fibers called the mitotic spindle . A surveillance mechanism known as the spindle assembly checkpoint prevents the cell from dividing until all the chromosomes have properly attached to the spindle . A protein called Bub1 is a central element of the SAC . However , it was not clear whether Bub1 works primarily as an enzyme or as a scaffolding protein . Baron , von Schubert et al . characterized two new molecules that inhibit Bub1’s enzyme activity and used them to investigate what role the enzyme plays in the spindle assembly checkpoint in human cells . The experiments compared the effects of these inhibitors to the effects of other molecules that block the production of Bub1 . Baron , von Schubert et al . ’s findings suggest that Bub1 works primarily as a scaffolding protein , but that the enzyme activity is required for optimal performance . Further experiments show that when the molecules that inhibit the Bub1 enzyme are combined with paclitaxel – a widely used therapeutic drug – cancer cells have more difficulties in separating their chromosomes and divide less often . The new inhibitors used by Baron , von Schubert et al . will be useful for future studies of this protein in different situations . Furthermore , these molecules may have the potential to be used as anti-cancer therapies in combination with other drugs .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2016	Probing the catalytic functions of Bub1 kinase using the small molecule inhibitors BAY-320 and BAY-524
Millions of naïve T cells with different TCRs may interact with a peptide-MHC ligand , but very few will activate . Remarkably , this fine control is orchestrated using a limited set of intracellular machinery . It remains unclear whether changes in stimulation strength alter the programme of signalling events leading to T cell activation . Using mass cytometry to simultaneously measure multiple signalling pathways during activation of murine CD8+ T cells , we found a programme of distal signalling events that is shared , regardless of the strength of TCR stimulation . Moreover , the relationship between transcription of early response genes Nr4a1 and Irf8 and activation of the ribosomal protein S6 is also conserved across stimuli . Instead , we found that stimulation strength dictates the rate with which cells initiate signalling through this network . These data suggest that TCR-induced signalling results in a coordinated activation program , modulated in rate but not organization by stimulation strength . Effector differentiation of a naïve CD8+ T cell begins when its T cell receptor ( TCR ) recognizes a peptide-MHCI ligand complex . If the interaction is strong enough , a cascade of signalling events follows that allows the naïve T cell to differentiate and expand into a pool of effector cells . Signal transduction downstream of the TCR involves a highly diverse network of post-translational protein modifications that ultimately drive transcriptional , translational , metabolic and cytoskeletal changes in the cell . It is estimated that fewer than 0 . 01% of naïve CD8+ T cells can recognize a particular foreign peptide-MHCI complex ( Jenkins and Moon , 2012 ) . Despite the diversity of rearranged TCRs on these naïve cells and the extensive range of antigenic peptides that may be presented , the intracellular machinery within each naïve T cell is able to sense the strength of the receptor-ligand interaction and mount an appropriate response . Previous work has demonstrated that the strength of stimulation a T cell receives upon binding a peptide-MHC ligand complex determines its fate in the thymus and its probability of activating in the periphery . During thymic selection , T cells that weakly recognize self-peptides are retained , while those that strongly recognize self-peptides undergo negative selection and are removed ( Daniels et al . , 2006; Hogquist et al . , 1994; Juang et al . , 2010; Prasad et al . , 2009 ) . In the periphery , the population response to stimuli of different strengths can vary in speed , magnitude and phenotype ( Denton et al . , 2011; King et al . , 2012; Moreau et al . , 2012; Ozga et al . , 2016; Palmer et al . , 2016; Zehn et al . , 2009 ) . Work from our group and others indicates that these observations may be explained by the fact that stimulation strength controls the rate with which individual cells activate transcriptional and proliferative processes ( Hommel and Hodgkin , 2007; Richard et al . , 2018 ) . This then raises the question , how does stimulation strength impact signalling downstream of the TCR , and how does this relate to transcriptional activation ( Balyan et al . , 2018 ) ? Many previous studies have examined signalling mediators and their coordinated network during naïve T cell activation ( Kannan et al . , 2012; Krishnaswamy et al . , 2014; Voisinne et al . , 2019 ) . Signalling through the TCR ( Courtney et al . , 2018; Navarro and Cantrell , 2014 ) begins with LCK and Fyn-mediated tyrosine phosphorylation of ITAM motifs on the invariant CD3 subunits . This creates a high affinity binding site for ZAP70 , which , upon phosphorylation and activation , leads to the generation of the LAT-SLP76 signalosome . From here , signalling activates multiple cascades including MAPK ( MEK1/2-ERK1/2 ) , PDK1-PI3K , calcium , and NFκB ( including IκBα-p65 ) pathways , each amplified and propagated via a series of phosphorylation events or other post-translational modifications . Signal transduction pathways can be broadly categorized as digital , with distinct ‘on’ or ‘off’ outcomes , or analogue , giving rise to a graded response ( Conley et al . , 2016; Zikherman and Au-Yeung , 2015 ) . In digital signalling , once the threshold for activation is surpassed , an output signal of constant intensity is produced . In analogue signalling , higher intensity of the originating stimulus results in a proportionally higher intensity of the output signal . Previous work has demonstrated that ligand potency determines the extent of signalling at various proximal and distal nodes ( Palmer et al . , 2016; Rosette et al . , 2001 ) . Studies focused on digital signalling nodes showed that stimulation strength affects the percentage of cells that phosphorylate ERK ( Altan-Bonnet and Germain , 2005; Das et al . , 2009; Tian et al . , 2007 ) , PKD2 ( Navarro et al . , 2014 ) , IκBα and the p65 component of NFκB ( Kingeter et al . , 2010 ) . Most of these studies have looked at each signalling molecule separately . It therefore remains unclear how ligand potency affects the coordination of signalling downstream of the TCR in naïve T cells . TCR-induced responses are rapid and often transient , and responding cell populations can be heterogeneous . Single-cell approaches are therefore well-suited to examining this system . Mass cytometry provides single-cell resolution , uses antibody-mediated measurements that can detect post-translational signalling protein modifications , and can achieve high-dimensionality through simultaneous measurement of up to 40 epitopes ( Bandura et al . , 2009; Bendall et al . , 2011; Lou et al . , 2007; Ornatsky et al . , 2010 ) . Previous mass cytometry studies of T cell signalling have demonstrated that small differences in proximal signalling molecules are propagated and amplified in downstream targets ( Mingueneau et al . , 2014 ) and that the interplay of ‘activatory’ versus ‘inhibitory’ signalling components determines the response of effector T cells to different antigen doses ( Wolchinsky et al . , 2014 ) . In this study , we designed a mass cytometry panel probing surface receptors and elements of key signalling pathways ( Figure 1 ) to examine the effect of stimulation strength on naïve CD8+ T cell responses . We used a minimal antigen presentation system to ask how modulating only the strength of the TCR-pMHC interaction affects signalling pathways without the influence of variable costimulatory factors or feedback from other cell types . Our multi-dimensional approach allowed us to determine how ligand potency impacts the synchronization of multiple parallel pathways . Through simultaneous measurement of S6 phosphorylation and early mRNA transcripts , we also examined the concurrent activation of these markers of translational and transcriptional processes . Our data suggest that the coordination of the TCR-induced signalling pathways that we tested does not vary with stimulation strength . Instead , strength of stimulation determines the rate with which T cells initiate this programme . We used the OT-I TCR transgenic mouse on a recombination-activating gene ( RAG ) -deficient background as a model for evaluating the impact of stimulation strength on TCR signalling pathways . In this model , all peripheral CD8+ T cells recognize the ovalbumin peptide SIINFEKL . Variants of SIINFEKL with altered potency for the OT-I TCR allow manipulation of the strength of TCR stimulation ( Daniels et al . , 2006; Hogquist et al . , 1994; Hong et al . , 2018; Rosette et al . , 2001 ) . In this study , we used the high potency SIINFEKL ( N4 ) , intermediate potency SIITFEKL ( T4 ) , and low potency SIIGFEKL ( G4 ) peptides , as well as an unrelated control peptide , ASNENMDAM ( NP68 ) . We designed a custom mass cytometry antibody panel to detect five surface markers of T cell identity and activation , eight phosphorylated signalling proteins with corresponding total proteins , and IκBα , which is degraded in response to stimulation ( Figure 1; Materials and methods ) . The antibodies labelled key components of major signalling pathways , including proximal signalling ( pZAP70[Y319]/pSyk[Y352] , pSLP76[Y128] , pLCK[Y505] and pPLCγ1[Y783] ) , the MAPK pathway ( pERK1/2[T202/Y204] ) , the PDK1-PI3K and mTOR pathways ( pAKT[S473] , pS6[S235/S236] ) , the NFκB pathway ( IκBα ) and the IL2 pathway ( pSTAT5[Y694] ) . All of these phosphorylation sites indicate active signalling with the exception of the inhibitory Y505 phosphorylation of LCK ( D'Oro and Ashwell , 1999; Marth et al . , 1988 ) . Measurement of total proteins allowed us to determine whether changes in phospho-protein staining were due to differences in signalling or protein expression levels . We isolated naïve CD8+ T cells from OT-I Rag1-/- splenocytes before stimulating with ligands of various strengths . To monitor signalling while naïve CD8+ T cells transitioned to activated T cells , and to relate signalling to changes in mRNA and protein expression during this process , cells were stimulated for 1 , 2 , 4 and 6 hr . We used a minimal , controlled system of peptide addition , allowing T cells to present antigens to each other . We also added exogenous IL2 to mitigate effects of potency-dependent differences in paracrine IL2 ( Au-Yeung et al . , 2017; Denton et al . , 2011; Marchingo et al . , 2014; Tan et al . , 2017; Verdeil et al . , 2006; Voisinne et al . , 2015 ) and provide all cells with an effector-promoting environment ( Pipkin et al . , 2010; Verdeil et al . , 2006 ) . This system was chosen in order to examine the cell-intrinsic effects of TCR stimulation strength on signalling pathways . Peptides were added at 1 µM since peptide titration revealed minimal differences in the percentage of cells phosphorylating S6 and ERK between 100 nM and 1 µM stimulation conditions ( Figure 2—figure supplement 1 ) . Using this reductionist stimulation system , we previously found that stimulation strength determined the rate with which naïve T cells initiated transcriptional activation but that cells activated by all ovalbumin-derived ligands were proliferating and expressing the effector molecule CD44 by two days ( Richard et al . , 2018 ) . Stimulated cells were stained with metal-conjugated antibodies and markers for dead cells and DNA before profiling by mass cytometry . We gated events detected by the mass cytometer in a hierarchical manner to select single , living cells that were TCRβ+ and CD8α+ before examining individual signalling molecules ( Figure 2a ) . While gating for single cells based on DNA content ( Figure 2—figure supplement 2a ) , we noted that a substantial percentage of events contained more than one cell-equivalent of DNA , particularly among cells stimulated with the strongest peptide , N4 ( Figure 2—figure supplement 2b ) . Separating events by DNA content revealed two distinct populations , such that events containing two cell-equivalents of DNA had higher phosphorylated and total protein staining ( Figure 2b , Figure 2—figure supplement 2c–d ) . It is therefore likely that this population contained multiple cells ( i . e . doublets ) , potentially even cells engaged in TCR-peptide-MHC interactions with their neighbours . After normalization of each phosphorylated protein to the DNA signal detected in the same mass cytometry event , signal intensities for events with two cell-equivalents of DNA had similar ranges to those of events with one for most markers ( Figure 2—figure supplement 2e ) . This provided further evidence that events with two cell-equivalents of DNA were doublets . Supporting the hypothesis that these doublets represented actively conjugated cells , a greater proportion of doublets than singlets showed signalling behaviour . In addition , from 1 to 4 hr after stimulation , pSLP76 signal was higher in doublet events even after normalization , suggesting that SLP76 was preferentially phosphorylated in cells actively engaged in TCR-peptide-MHC interactions . Because it was not possible to discern which proteins were signalling in which cell in a multiplet event , for subsequent analyses we included only singlet events . We next examined the kinetics of individual signalling molecules within our mass cytometry data . Total levels of signalling proteins did not substantially change over a 6 hr stimulation , while expression of effector proteins CD44 and CD25 increased in a time- and potency-dependent manner ( Figure 3—figure supplement 1 ) . In the presence of exogenous IL2 , ligand potency did not strongly influence the rate with which individual cells phosphorylated STAT5 ( Figure 3—figure supplement 2a ) . In the absence of exogenous IL2 , STAT5 phosphorylation was associated with ligand potency , such that weak G4-stimulated cells showed no STAT5 phosphorylation ( Figure 3—figure supplement 2b ) , likely due to autocrine/paracrine IL2 rapidly secreted by strongly stimulated cells ( Tan et al . , 2017 ) . The percentages of cells degrading IκBα or phosphorylating S6 or ERK1/2 were not impacted by the presence of exogenous IL2 . The percentages of cells phosphorylating pAKT[S473] were subtly increased by IL2 particularly under stimulation with low potency ligands ( Figure 3—figure supplement 2b ) . This may reflect the mechanism proposed by Ross et al . whereby JAK signalling induced by IL2 ultimately stimulates mTORC2 phosphorylation of AKT[S473] ( Ross et al . , 2016 ) . Together , these data indicate the selectivity of IL2 effects on T cell signalling pathways . Phosphorylation of proximal , membrane-recruited mediators ZAP70 and PLCγ1 was only detectable in a small percentage of cells at any point during this time course , preventing further interpretation ( Figure 2b ) . For LCK , the percentage of cells with inhibitory phosphorylation of pY505 decreased at 6 hr in a potency-dependent manner , suggesting that stronger stimuli resulted in greater LCK activity only at this late time point ( Figure 3—figure supplement 2a ) . For SLP76 , the high potency ligand N4 induced a greater percentage of signalling cells and greater signalling intensity within these cells between 1 and 4 hr , whereas signalling was minimal in cells stimulated with intermediate ( T4 ) or low potency ( G4 ) ligands ( Figure 2b , Figure 3—figure supplement 2a ) . Examination of the kinetics of individual distal signalling molecules revealed two distinct patterns . We defined these as transient if the percentage of signalling cells increased and subsequently decreased during the time course of high potency stimulation , and sustained if maximal signalling was ongoing at 6 hr ( Figure 3a ) . ERK1/2 , AKT and IκBα displayed transient signalling . While ERK1/2 and AKT are phosphorylated in response to TCR stimulation , IκBα is degraded , releasing NFκB subunits and permitting their translocation to the nucleus ( Paul and Schaefer , 2013 ) . Therefore , a reduction in IκBα+ cells indicates active signalling by this node . For these three signalling mediators , the percentage of cells actively signalling was maximal at 2 hr when stimulated with the highest potency N4 peptide but was delayed until 4 or 6 hr when stimulated with the lower potency T4 or G4 peptides ( Figure 3a ) . After 2 hr , signalling via these proteins declined in strongly stimulated cells , resulting in a convergence with more weakly stimulated cells . In addition , the maximum percentage of cells signalling through these nodes was substantially higher in strongly stimulated cells . This may indicate repeated node activation after high potency stimulation , such that a greater proportion of cells were signalling at any given time of measurement . Thus , for these transiently signalling proteins , ligand potency was associated with the maximal proportion of signalling cells and the speed with which this proportion was reached on a populational level . In contrast to these transient signalling events , S6 phosphorylation induced by TCR stimulation was sustained within our time course ( Figure 3a ) . Under N4 stimulation , there was a rapid initial increase in the percentage of pS6+ cells before a plateau . This pattern may be indicative of the signalling protein approaching saturation . The appearance of pS6+ cells was slower after stimulation with lower potency ligands , but the proportions of pS6+ cells approached convergence between N4 , T4 and G4 stimulations at 6 hr . Thus , for S6 , the rate with which cells exhibited active signalling , but not the maximal proportion of signalling cells was associated with ligand potency . We observed a bimodal distribution of pERK1/2 measurements in our mass cytometry data , consistent with previous reports ( Altan-Bonnet and Germain , 2005; Das et al . , 2009; Tian et al . , 2007 ) . The extent of ERK1/2 phosphorylation in pERK1/2+ cells , as determined by the median marker intensity , was unaffected by ligand potency ( Figure 3—figure supplement 3a ) . This confirmed that ERK1/2 exhibits digital signalling behaviour , that is on a per cell basis there is either an ‘on’ or ‘off’ state . Similarly , pS6[S235/S236] signal was also bimodally distributed ( Figure 2a , Figure 3—figure supplement 3b ) . The intensity of pS6 signal in pS6+ cells slightly increased over time . Normalization to total S6 protein intensity mitigated this effect ( particularly under strong stimulation ) and suggested that ligand potency may subtly affect the rate of S6 phosphorylation within individual cells ( Figure 3—figure supplement 3c ) in addition to the percentage of pS6+ cells at early time points . Since S6 phosphorylation at S235/S236 is driven by both S6K1 downstream of mTORC1 , and RSK downstream of MEK1/2-ERK1/2 ( Pende et al . , 2004; Roux et al . , 2007; Salmond et al . , 2009 ) , we were interested in how each of these pathways contributed to its digital behaviour in strongly stimulated cells . To this end , we treated cells with inhibitors of MEK1/2 ( MEK162 , Lee et al . , 2010 ) and mTOR ( rapamycin , Pollizzi and Powell , 2015 ) before stimulation with N4 peptides ( Figure 3—figure supplement 3d ) . The per-cell phosphorylation of S6 decreased moderately in response to rapamycin ( Figure 3b–c ) , with little difference between doses of 20 nM and 2 μM ( Figure 3—figure supplement 3e ) , but the bimodal distribution of pS6 and the percentage of pS6+ cells were not disturbed . In contrast , while S6 phosphorylation also decreased within each cell in response to MEK162 , this response was dose-dependent between 0 . 5 and 5 μM , and at the highest dose ( 5 μM ) , the bimodality was disturbed . In addition , even low doses of MEK162 halved the percentage of cells with phosphorylated S6 ( Figure 3d ) . Combined MEK162 and rapamycin resulted in severe inhibition of S6 phosphorylation ( Figure 3b ) . Neither MEK162 , rapamycin , nor the combination of these two substantially impacted CD44 expression at the concentrations of inhibitors used ( Figure 3—figure supplement 3f ) , but a synergistic inhibition of cellular proliferation was observed after 2 days ( Figure 3—figure supplement 3g ) . Thus , simultaneous activity of the MEK and mTOR pathways is required for phosphorylating S6[S235/S236] and proliferative responses , and MEK signalling is essential for S6 digital phosphorylation . These data emphasize the coordinated nature of signalling downstream of the TCR . To take advantage of the simultaneous measurements in mass cytometry data , we next tested for differential abundance of multidimensional cellular phenotypes , taking into account all of the signalling markers measured ( Lun et al . , 2017; Figure 4a , Figure 4—figure supplement 1 ) . We first defined 1585 fine-grained phenotypic states in the high dimensional space . We then compared the abundance of cells within these phenotypic states between unstimulated and all stimulated conditions ( Materials and methods ) . Clustering of significantly differentially abundant states revealed two main signalling phenotypes . Phenotype A , defined as pS6+ pSTAT5+ pERK1/2+ , was most prevalent in the high potency ( N4 ) -stimulated cells . Phenotype B , defined as pS6+ pSTAT5+ pERK1/2- , appeared under all stimulation conditions with similar prevalence ( Figure 4a ) . Phenotypes A and B were paralleled by the pSTAT5- phenotypes A’ and B’ , respectively . Phenotypes A and A’ were transient , whereas phenotypes B and B’ were sustained . Subpopulation analysis confirmed that the high potency ligand N4 was capable of inducing a greater abundance of phenotypes A’ ( pS6+ pSTAT5- pERK1/2+ ) and A ( pS6+ pSTAT5+ pERK1/2+ ) up to 4 hr after stimulation ( Figure 4b ) . In contrast , abundances of phenotypes B ( pS6+ pSTAT5+ pERK1/2- ) and B’ ( pS6+ pSTAT5- pERK1/2- ) increased at a similar rate between 1 and 6 hr after stimulation with all ovalbumin-derived ligands and were not associated with ordered ligand potency . To complement these signalling phenotypes , we further investigated their relationship with surface expression of the effector protein CD44 , which was an important contributor to phenotypic cluster separation ( Figure 4a ) . We examined the 16 ( 24 ) possible states defined by combinations of the markers CD44 , pS6 , pSTAT5 and pERK1/2 . The strongest peptide , N4 , was capable of inducing a large proportion of CD44- pS6+ pSTAT5- pERK+ cells after 1 hr of stimulation ( Figure 4c , Figure 4—figure supplement 2 ) . This was accompanied by an increasing population of CD44- pS6+ pSTAT5+ pERK+ cells , which reached its maximum abundance 2 hr after stimulation . Cells stimulated by the weaker peptides T4 and G4 showed dramatically reduced abundances of these cellular phenotypes , and their maxima were delayed . For all stimuli , a high abundance of CD44- pS6+ pSTAT5+/- pERK1/2- cells was seen between 4 and 6 hr . Cells expressed CD44 by 4 hr after strong and intermediate stimulation ( N4 and T4 ) and by 6 hr after weak stimulation ( G4 ) . From these data , we inferred the coordination of distal TCR-induced signalling . We propose that within our stimulation system , activating cells initially phosphorylate S6 and ERK1/2 , followed by STAT5 , after which ERK1/2 becomes dephosphorylated , followed by STAT5 in some cells . In the early hours after stimulation , signalling cells express CD44 at the same level as unstimulated cells but begin upregulating CD44 expression by 4–6 hr . As time progresses and cells shift to sustained phenotypes , those activated with reduced potency ligands begin to phenotypically resemble those stimulated with high potency ligand ( Figure 5 ) . To formally test this order of activation events , we constructed activation trajectories of cells under each stimulation condition across all time points , based on their expression of pS6 ( Figure 5—figure supplement 1a ) . We then asked in what order along these trajectories pS6 , pERK1/2 , pSTAT5 , and CD44 activation events were initiated ( Figure 5—figure supplement 1b–c ) . We found that pS6 appeared first , followed by pERK1/2 , pSTAT5 , and finally CD44 . Of note , the start of ERK1/2 phosphorylation corresponded to the most dramatic increase in S6 phosphorylation , supporting evidence that ERK activation drives full S6[S235/S236] phosphorylation ( Figure 3 ) . The order of activation events was shared across stimulation conditions ( p=0 . 00174 compared to random orders of events , Materials and methods ) . The signalling molecules pAKT , pLCK and IκBα were less dynamic along the trajectory , precluding precise determination of their order of activation particularly in weakly stimulated cells , but visualizing their changes along the trajectory further suggested shared patterns between stimuli ( Figure 5—figure supplement 1d ) . We therefore conclude that ligand potency controls the rate with which cells achieve certain signalling states and that the order of these signalling events is preserved regardless of stimulation strength . Finally , we shifted our focus further downstream to examine the relationship between signalling at the ribosomal protein S6 and mRNA expression of early response transcription factors . These two activation events indicate initiation of translational and transcriptional processes , which are required for the biosynthetic programs of T cell activation ( Araki et al . , 2017; Howden et al . , 2019; Tan et al . , 2017 ) . S6 is a ribosomal protein whose phosphorylation reflects , though it does not regulate , TCR-induced translation ( Salmond et al . , 2015; Salmond et al . , 2009 ) . Nr4a1 ( Nur77 ) and Irf8 encode transcription factors that are rapidly expressed upon T cell activation ( Moran et al . , 2011; Nelson et al . , 1996 ) , and we previously found that their transcripts are upregulated at 1 and 3 hr , respectively , after strong N4 stimulation ( Richard et al . , 2018; Figure 6—figure supplement 1a ) . To examine these translational and transcriptional characteristics simultaneously , we activated naïve OT-I CD8+ T cells with ligands of various potencies before measurement of pS6 and mRNA molecules using combined phosphoflow and RNA flow cytometry ( Figure 6a , Figure 6—figure supplement 1b ) . Stimulation time courses with the different potency ligands suggested that Nr4a1 transcripts were upregulated before phosphorylation of S6 and downregulated after , while Irf8 transcripts were upregulated after S6 phosphorylation ( Figure 6b , Figure 6—figure supplement 1c ) . This order of events appeared consistent across stimuli . The percentage of pS6+Nr4a1+ cells was maximal between 1 and 2 hr after stimulation with the highest potency peptide N4 , after 4 hr stimulation with the intermediate potency peptide T4 , and after 6 hr stimulation with the lowest potency peptide G4 . Likewise , the percentage of pS6+Irf8+ cells was maximal after 2 hr stimulation with N4 , 4 hr stimulation with T4 and 6 hr stimulation with G4 peptides ( Figure 6b ) . Similar to the multi-dimensional signalling phenotypes we measured by mass cytometry , these altered kinetics of phosphorylation and transcript upregulation indicate that stimulation strength controls their rate of activation . These results suggest that the relationship between signalling events is conserved under different strengths of stimulation , even among the differing signal transduction pathways controlling transcription and translation . Upon TCR activation both the transcriptional and translational machinery are deployed in a coordinated manner , which may improve efficiency of protein production enabling the naïve CD8+ T cell to differentiate and proliferate . The interaction of adhesion molecules LFA-1 and ICAM-1 assists the formation of a stable immunological synapse , augments TCR-induced signalling , and continues to promote differentiation even after initial activation ( Gérard et al . , 2013; Verma and Kelleher , 2017 ) . LFA-1 is constitutively expressed by naïve T cells , and TCR stimulation drives both redistribution and conformational changes that enhance its binding to the ligand ICAM-1 ( Capece et al . , 2017; Dustin and Springer , 1989; Verma and Kelleher , 2017 ) . Palmer et al . previously demonstrated that LFA-1-ICAM-1 interactions improve conjugate formation during T cell stimulation with peptide-loaded splenocytes , particularly for low potency ligands ( Palmer et al . , 2016 ) . However , it remained unclear whether ICAM-1 was expressed in our stimulation system and whether this integrin interaction could thereby play a role . We therefore measured ICAM-1 on the surface of T cells 6 hr after addition of pure peptides of various potencies ( Figure 7—figure supplement 1a ) and found that all T cells expressed ICAM-1 , regardless of their stimulation status . These data suggest that integrin adhesion likely contributes to T cell activation along with TCR stimulation and exogenous IL2 in our system . In contrast to this reductionist system , many additional factors impact T cell activation in vivo . Most fundamentally , naïve T cells are activated in the lymph node by professional antigen-presenting cells ( APCs ) , such as dendritic cells , instead of other T cells . These APCs express costimulatory ligands in addition to peptide-MHC complexes , which can further tune naïve T cell responses ( Chen and Flies , 2013; Hubo et al . , 2013 ) . For example , in our stimulation system , the costimulatory ligand CD80 remained largely absent after 6 hr of stimulation ( Figure 7—figure supplement 1a ) . In contrast , mature bone marrow-derived dendritic cells ( BMDCs ) expressed high levels of CD80 , along with additional costimulatory molecules ( Figure 7—figure supplement 1b–c ) . To test how signalling responses to ligands of different strengths might be impacted by the additional signalling conferred by professional APCs , we stimulated naïve T cells with mature BMDCs loaded with peptides of various potencies . Exogenous IL2 was included to maintain comparability with the T:T stimulation system . Signalling molecules pZAP70 , pSLP76 , pERK1/2 , pS6 and pSTAT5 , as well as CD44 expression , were measured by flow cytometry . We found that activation was in general less strong in the presence of peptide-pulsed professional APCs than pure peptides ( Figure 7 ) , perhaps due to reduced ligand availability as only half of the cells in the culture carried ligand ( Materials and methods ) . The potency-dependent kinetics of pERK1/2 , pSLP76 and CD44 resembled those observed in the T:T stimulation system , while pZAP70 remained undetectable . pS6 was upregulated over time under stimulation with high , medium and low potency ligands . pSTAT5 was upregulated over time with all stimuli , including the null peptide NP68 , suggesting that simply mixing naïve T cells with BMDCs enhanced IL2 signalling . These results indicate that the rate-based mechanism we observed in the T:T stimulation system is further tuned at particular signalling nodes by more complex antigen presentation . In this study , we examined the coordination of signalling pathways downstream of TCR activation using a custom mass cytometry panel as well as protein and RNA flow cytometry . The use of multidimensional measurements allowed us to probe the simultaneous activation of multiple signalling and transcriptional processes . This enabled comparisons of the impact of ligand potency on not only individual activation events but also their coordination . We found that the strength of TCR stimulation controlled the rate of appearance of the multi-dimensional signalling and transcriptional phenotypes that we profiled . Stimulation strength altering the rate of T cell activation has been observed in previous studies from our group and others investigating transcriptomic , proliferation , and protein characteristics ( Hommel and Hodgkin , 2007; Richard et al . , 2018; Rosette et al . , 2001 ) . Taken together , these data suggest that by controlling the probability that a cell will initiate activation responses , signal strength can modulate the average speed and magnitude of a population response . Our signalling results indicate that if such an activation switch exists , it lies very proximal to the TCR . An important outstanding question is the mechanism by which the TCR translates ligand strength into the probability of downstream signalling . One model explaining the threshold for T cell response that could propagate to a rate-based mechanism is kinetic proofreading ( McKeithan , 1995 ) . This theory postulates that the ligand must remain bound to the receptor for sufficient time for signalling accumulation to surpass a critical event and propagate downstream . Indeed , multiple reports have suggested that naïve T cells require sustained interaction with presented antigen to achieve optimal proliferation , though the necessary duration differs by study and likely depends on the presence of additional factors such as IL2 ( Balyan et al . , 2017; Curtsinger et al . , 2003; Iezzi et al . , 1998; Kaech and Ahmed , 2001; Prlic et al . , 2006; van Stipdonk et al . , 2003; van Stipdonk et al . , 2001; Wong and Pamer , 2001 ) . Refinements to the kinetic proofreading model suggest that not a single interaction but rather the cumulative interaction lifetime of a series of early binding events controls signal accumulation ( Dushek et al . , 2009; Liu et al . , 2014 ) . Biophysical investigations of the impact of force on binding events between the TCR and pMHC ( as well as CD8 ) have described catch bonds formed with high potency ligands that extend interaction lifetimes ( Das et al . , 2015; Hong et al . , 2018; Liu et al . , 2014; Sibener et al . , 2018; Wu et al . , 2019 ) , though this observation has not been universal ( Limozin et al . , 2019 ) and merits further investigation . Extrapolating from this lifetime theory , altered ligand potency could change the probability of long or rapidly repeated binding events , thereby controlling the probability that an individual cell activates . If such a lifetime-based mechanism exists , T cells must then translate variation in binding lifetime to the presence or absence of downstream signalling . Palmer and colleagues found ligand potency-associated differences in CD3ζ chain and ZAP70 phosphorylation ( Daniels et al . , 2006; Palmer et al . , 2016 ) , which may allow potency-dependent accumulation of signal before propagation downstream . Supporting this hypothesis , John James demonstrated that the number of CD3 ITAM motifs in a synthetic receptor influenced the rate but not the magnitude of signalling within individual Jurkat cells ( James , 2018 ) . Likewise , Mukhopadhyay et al . found that the presence of multiple ζ chain ITAMs , as well as ZAP70 , increases the efficiency of phosphorylation in a HEK 293T cell reconstitution system , although these phosphorylation events do not account for the apparent switch-like ultrasensitive behaviour of T cell signalling ( Mukhopadhyay et al . , 2016 ) . One mechanism that could explain the switch-like behaviour is the zero-order ultrasensitivity model ( Ferrell and Ha , 2014; Goldbeter and Koshland , 1981 ) , wherein negative regulators act in combination with activators to enhance responsiveness when signalling molecules operate close to saturation . In this way , a relatively small change in the binding lifetime of a pMHC ligand could be amplified by altering a kinase/phosphatase ratio to switch between the presence and absence of downstream signalling . An alteration in the local relative abundances of phosphatase CD45 and kinase LCK described by the kinetic segregation model ( Davis and van der Merwe , 2006; Razvag et al . , 2018 ) represents an intriguing candidate for controlling a zero-order ultrasensitivity mechanism ( Hui and Vale , 2014 ) , although a subsequent study has refuted its requirement for T cell activation ( Al-Aghbar et al . , 2018 ) , suggesting other mechanisms . Control may also be mediated by the phosphatase PTPN22 , which can dephosphorylate CD3ζ chains , ZAP70 and LCK ( Wu et al . , 2006 ) , as absence of PTPN22 results in increased proportions of activated cells , particularly under weak stimulation ( Salmond et al . , 2014 ) . Alternatively , Lo et al . showed that the slow phosphorylation of a tyrosine residue in LAT is a possible candidate for this rate-limiting step , since substitution of a single residue that enhances this phosphorylation improves T cell response to low potency ligands ( Lo et al . , 2019 ) . Through our single-cell measurements , we confirmed ( Altan-Bonnet and Germain , 2005; Das et al . , 2009; Tian et al . , 2007 ) that ERK1/2 is phosphorylated with ‘on or off’ states , characteristic of digital signalling . In addition , we found that S6[S235/S236] is also phosphorylated in a similar manner . However , whilst the extent of ERK1/2 phosphorylation during the ‘on’ state was constant , the extent of S6 phosphorylation subtly increased both with time and strength of stimulus . The parallel subtle increase in total S6 expression over time implies induction of S6 protein production during T cell activation . These data suggest that dividing signalling proteins into digital or analogue can be complicated by changes in total protein levels that may attribute analogue properties to a digital signal . Previous work has shown that in addition to mTORC1 signalling , the MEK/ERK pathway contributes to S6[S235/S236] phosphorylation ( Pende et al . , 2004; Roux et al . , 2007 ) , particularly in naïve T cells ( Krishnaswamy et al . , 2014 ) . Therefore , combinatorial effects of mTORC1 and MEK signalling might be expected to influence both S6 phosphorylation and other downstream T cell activation phenotypes . We found that chemical inhibition of both of these pathways blocked S6 phosphorylation , implying that they contribute in a non-redundant manner . A dose titration with MEK162 indicated that MEK/ERK signalling is critical for phosphorylation of S6[S235/S236] . Even at low doses , MEK162 reduced the percentage of pS6+ cells , suggesting that it may modulate the rate of response . Furthermore , in our trajectory analysis of multiple signalling markers , a steep increase in S6 phosphorylation coincided with the appearance of pERK+ cells under all peptide stimuli ( Figure 5—figure supplement 1a–b ) . These data raise the possibility that digital phosphorylation of ERK propagates through RSK to S6[S235/S236] . We additionally explored the effects of rapamycin and MEK162 on naïve T cell proliferation . Although rapamycin had little effect on S6 phosphorylation , it had a profound effect on T cell proliferation , which may be due to several different mechanisms . First , rapamycin can also impact mTORC2 signalling after prolonged treatment ( Sarbassov et al . , 2006 ) , and naïve T cells may be particularly susceptible ( Delgoffe et al . , 2011 ) . Second , mTORC1 affects many additional pathways other than ribosomal activity ( Pollizzi and Powell , 2015; Salmond , 2018 ) . Finally , even when signalling through S6K1 , mTORC1 can influence proliferation through pS6-independent mechanisms ( Salmond et al . , 2015 ) . We also found that MEK inhibition with MEK162 synergized with rapamycin to further dampen T cell proliferation , highlighting the interconnected nature of the signalling pathways downstream of the TCR . Many signalling molecules exhibited transient behaviour at the population level ( ERK1/2 , IκBα , AKT ) , while pS6 accumulated over the course of our 6 hr experiments . Stimulation strength strongly influenced the proportion of cells exhibiting transient signalling behaviours between 1 and 4 hr after activation , but by 6 hr , cells activated with any of the ovalbumin-derived peptide ligands exhibited a similar signalling phenotype . This potency-dependent difference in the maximal proportions of cells signalling may be due to either repeat or sustained signalling with strong ligands , the latter of which has been observed for calcium fluxes induced by TCR stimulation ( Chen et al . , 2010; Le Borgne et al . , 2016; Wülfing et al . , 1997 ) . For example , although under weak G4 stimulation only a very small percentage ( 15 . 4% ) of cells were found to be pERK1/2+ at any given time , the majority ( 72 . 1% ) of G4-stimulated cells achieved digital activation of pS6[S235/236] by 6 hr ( Figure 3a ) . Given that we found full S6 phosphorylation after strong stimulation requires MEK signalling , we hypothesize that this pathway is active in all stimulation conditions but that ERK activation events occur with reduced frequency or duration with weak stimuli and thus many were missed in our snapshot measurements . Future investigations using ERK reporters and ERK inhibition in weakly stimulated cells would be needed to test this prediction . Consistent with this proposed mechanism , single-cell studies in epithelial and HEK293 cell lines have observed oscillating ERK phosphorylation with frequency and duration dependent on the concentration or frequency of EGF stimulation ( Albeck et al . , 2013; Ryu et al . , 2018 ) . Such an effect on digital ERK activation may be modulated by multi-step activation of the upstream mediator SOS dependent on its dwell-time after activation-induced recruitment to the plasma membrane ( Huang et al . , 2019 ) . Interestingly , using a light-inducible ERK activation system in epithelial cells , Aoki et al . demonstrated divergent transcriptional effects of sustained versus transient ERK activation ( Aoki et al . , 2013 ) . It therefore remains possible that different ERK targets in T cells , such as translational machinery , microtubule remodelling , and transcription factors ( e . g . ELK1 , SAP1 , SAP2 ) ( Navarro and Cantrell , 2014 ) are differentially affected by stimulation strength , warranting further investigation of additional downstream components . Examination of the coordinated activation of transcriptional and translational signalling pathways also revealed conservation of this order of events . Biosynthetic processes are critical for naïve T cells to differentiate into effector cells ( Araki et al . , 2017; Tan et al . , 2017 ) , and thus , carefully controlled simultaneous activation would ensure efficient , consistent effector differentiation of activated cells . Under stimulation with peptide-loaded BMDCs , ligand potency determined the percentages of T cells undergoing certain activation events ( pERK1/2 , pSLP76 and CD44 ) , similar to observations in our reductionist stimulation system . In contrast , phosphorylation of S6 was not associated with ligand potency after stimulation with peptide-loaded BMDCs . Unlike naïve and recently activated T cells , BMDCs express high levels of costimulatory molecules that can impact TCR-induced signalling . For example , ligation of the costimulatory receptor CD28 at the same time as the TCR results in amplification of signalling pathways including NFAT , NFκB and AP-1 , and can enhance both the sensitivity and ultimate division potential of naïve T cell activation ( Esensten et al . , 2016; Heinzel et al . , 2017; Marchingo et al . , 2014; Yang et al . , 2017 ) . Further exploration of how individual costimulatory ligands impact the coordination and initiation rate of the TCR-induced signalling programme will be important for dissecting these additional inputs . Despite the increased complexity of BMDC peptide presentation , this in vitro system is nevertheless still far-removed from in vivo T cell activation , where the microenvironment is increasingly complex . Additional variables such as cytokine and nutrient availability and cell-cell interactions can further tune the T cell response in vivo ( Curtsinger and Mescher , 2010; Kedia-Mehta and Finlay , 2019 ) . Moreover , strongly stimulated T cells undergo prolonged retention in the lymph node ( Ozga et al . , 2016; Zehn et al . , 2009 ) and may out-compete weakly stimulated T cells for cytokines and nutrients ( Wensveen et al . , 2012 ) , suggesting that stimulation strength and the microenvironment are not independent . Our controlled in vitro systems allowed us to identify effects of stimulation strength on TCR-induced pathways alone , as well as in the context of BMDC-mediated costimulation , without confounding by other in vivo factors and feedback . By delineating the impact of stimulation strength in low-complexity systems , these data can form the basis for interpretation of future studies where additional variables may be explored . In this study , we measured 22 markers of protein expression and active signalling . While other unmeasured signalling mediators may respond to altered stimulation strength in a different manner , our data demonstrate a strict choreography of the distal signalling processes that we examined . Stimulation strength was associated with the rate with which cells embarked on this regimented programme . This suggests that using a limited set of signalling machinery in a single coordinated programme , T cells can finely tune their responses to different ligands through modulation of the rate of signalling . Key resources are detailed in Supplementary file 2 . CD8+ T cells were isolated from OT-I Rag1-deficient mice ( OT-I Rag1tm1Bal on a C57BL/6 background ) , which underwent confirmation of genotype prior to study . BMDCs were generated from wild-type C57BL/6 mice . Experiments used both male and female mice 9–25 weeks old . Mice were bred and maintained within University of Cambridge animal facilities . For T cell isolation , single cell suspensions of splenocytes were produced via homogenization of the spleen through a 70 µM nylon strainer . CD8+ T cells were isolated using the Mouse CD8α+ T cell Isolation Kit ( MACS Miltenyi Biotec ) . Cells were cultured in RPMI 1640 ( Gibco ) , 10% FBS ( Biosera ) , penicillin-streptomycin ( Sigma ) , sodium pyruvate ( Gibco ) , L-glutamine ( Sigma ) , β-mercaptoethanol ( Gibco ) and 20 ng/ml recombinant mouse IL-2 ( Peprotech ) . For stimulation , the following peptides were used at the concentrations indicated: SIINFEKL ( N4 ) , SIITFEKL ( T4 ) , SIIGFEKL ( G4 ) , and ASNENMDAM ( NP68 ) ( Cambridge Bioscience ) . Bone marrow derived dendritic cells ( BMDCs ) were generated based on a published protocol by Abcam . Femurs and tibias were sterilized in 70% ethanol and flushed with cold BMDC culture media , consisting of RPMI 1640 ( Gibco ) , 10% FBS ( Biosera ) , penicillin-streptomycin ( Sigma ) , L-glutamine ( Sigma ) and β-mercaptoethanol ( Gibco ) . The suspension of bone marrow progenitor cells was passed through a 70 µM nylon strainer and plated in 10 cm petri dishes in BMDC culture media supplemented with 20 ng/ml GM-CSF ( Peprotech ) . Fresh BMDC culture media with 20 ng/ml GM-CSF was added on day three and replaced on day six and , if needed , day 8 . Immature dendritic cells were harvested from day 7 to day 9 . Maturation was induced by culturing immature dendritic cells for 1 day in BMDC culture media with 20 ng/ml GM-CSF , as well as 50 ng/ml LPS ( Thermofisher Scientific ) and 20 ng/ml IL4 ( Abcam ) . Differentiation into immature and mature BMDCs was verified by flow cytometry . To stimulate T cells , mature BMDCs were pulsed with 1 µM of peptide for 1 hr at 37°C , washed , mixed with naïve T cells at a ratio of 1:1 , and cultured for the times indicated . Purified naïve CD8+ T cells were analysed by mass cytometry . In experiment 1 , cells from four age-matched mice ( two males and two females ) were used , representing four biological replicates . In experiment 2 , more stimulation conditions were included . This necessitated more cells for each biological replicate than could be obtained from a single mouse . Therefore , each biological replicate ( one male , one female ) was composed of cells from a pair of age- and gender-matched mice . Staining for mass cytometry was performed using sequential MaxPar reagent kits ( Fluidigm ) in the following steps . Live cells were stained with 5 µM Cell-ID Cisplatin for 5 min at 37°C and rested for 15–30 min before stimulation with 1 µM N4 , T4 , G4 , or NP68 peptides , or left unstimulated . In Experiment 1 , cells were stimulated for 1 and 2 hr . In experiment 2 , cells were stimulated for 1 , 2 , 4 and 6 hr . See Supplementary file 1 for replicate structure . Cells were fixed with Maxpar Fix I Buffer for 10 min at room temperature . Cells stimulated under different conditions were barcoded using the Cell-ID 20-Plex Pd Barcoding Kit and pooled for staining to minimise confounding technical differences . In experiment 1 , all cells from each mouse were pooled into a batch . In experiment 2 , in addition to pooling within a biological replicate , four samples were shared across the two pools to enable batch normalization for differential abundance analysis as described below . Cells were blocked with FCR blocking reagent ( Biolegend , clone 93 ) and stained with metal-conjugated surface antibodies ( Supplementary file 3 ) . Surface-stained cells were permeabilized with methanol ( Fisher Scientific ) and stained with metal-conjugated antibodies against intracellular targets ( Supplementary file 3 ) , all diluted in Maxpar Cell Staining Buffer . Stained cells were then fixed with 1 . 6% formaldehyde ( Thermofisher ) and stained overnight with 125 nM Cell-ID Intercalator-Ir in Maxpar Fix and Perm Buffer . Cells were analysed on a Helios CyTOF system ( Fluidigm ) . Data within each cell pool were normalized and debarcoded using the Fluidigm CyTOF software . Metal-conjugated antibodies were custom-conjugated where not already commercially available ( Supplementary file 3 ) . All custom-conjugated antibody clones were tested using phosphoflow cytometry before and after metal-conjugation . When allocating metals to antibody targets , brighter metals were assigned to antibodies that exhibited weaker phosphoflow staining or to those without clear bimodal expression . Metal channels that receive significant cross-over from other channels were also allocated antibodies with stronger signals . For each protein target , antibodies against the total protein and its phosphorylated version were conjugated to metals differing by more than one mass unit to avoid spillover . Antibodies targeting phosphorylated proteins were validated using phosphoflow ( with and without metal conjugation ) under different stimulating conditions , including anti-CD3 coated plate ( 1 µg/ml , BD Biosciences , clone 145–2 C11 ) , PMA ( 50 nM , Sigma-Aldrich ) and ionomycin ( 1 µg/ml , Sigma-Aldrich ) , N4 peptide ( 1 µM ) , and pervanadate ( 1 mM , prepared using sodium orthovanadate , Sigma-Aldrich ) ( Supplementary file 4 ) . To confirm the specificity of the antibody clones targeting ZAP70 , LCK and SLP76 , we transfected HEK 293 T cells , which lack endogenous expression of these proteins , with vectors encoding the proteins and tested antibody binding via flow cytometry and immunofluorescence ( Supplementary file 4a ) . To confirm the specificity of the antibody clone targeting PLCγ1 ( 3H1C10 ) , we performed siRNA knockdown in T cells and tested antibody binding by flow cytometry ( Supplementary file 4a ) . ( Knockdown was validated by western blotting with a WB-specific antibody clone ( D9H10 ) . ) Further validation of phospho-specific antibodies was performed using signalling inhibitors as detailed in the antibody specificity tables ( Supplementary file 4 ) . All metal-conjugated antibodies were tested by mass cytometry prior to experimentation . By testing antibodies on fixed cells , fixed and barcoded cells , and live cells ( surface markers only ) , we confirmed there was no additional loss of antibody activity through the addition of the barcoding step . Two surface markers performed less well when fixed ( CD62L-160Gd clone MEL-14 and CD69-143Nd clone H1 . 2F3 , both Fluidigm ) . The CD62L-160Gd antibody was excluded from the mass cytometry panel . The CD69-143Nd antibody was excluded from the analyses in experiment 1 , and excluded from the staining panel in experiment 2 . All metal-conjugated antibodies were titrated for optimal performance and key signalling antibodies were tested in a time-course assay under different stimulatory conditions to determine the optimal times for running the full panel . For mass cytometry analysis in FlowJo ( v10 ) , debarcoded samples were gated in a hierarchical manner: EQ bead exclusion followed by selection of intact cells based on DNA content , single cells based on the event length and DNA content , living cells based on cisplatin staining , and finally CD8α+ TCRβ+ cells . For activation-induced markers , positive/negative status was defined based on comparison with unstimulated cells . Normalization of antibody-measured signals to DNA signal , as well as phospho-protein to total protein signals , was performed in R using the ncdfFlow ( v2 . 30 . 1 ) ( Gopalakrishnan , 2019 ) and flowCore ( v1 . 50 . 0 ) ( Hahne et al . , 2009 ) Bioconductor packages . The signal of each marker in each event was normalized to the signal from the 191Ir DNA channel or appropriate total protein channel within that event . Normalized ratios were then scaled to the median 191Ir DNA or appropriate total protein signal from one selected sample for visualization purposes . To test for differential abundances , mass cytometry data from experiment 2 was processed using the ncdfFlow ( v2 . 30 . 1 ) , flowCore ( v . 1 . 50 . 0 ) and cydar ( v1 . 8 . 0 ) ( Lun et al . , 2017 ) Bioconductor Packages in R . A logicle transformation ( default parameters except w = 0 . 1 ) was applied to raw intensity data . Data from the two batches were range-normalized based on the four samples that were included in both batches using the normalizeBatch function ( with parameters p=0 . 001 , fix . zero = TRUE ) . After normalization , one technical replicate from each of these four repeated samples was carried forward for analysis . All samples were then pooled before constructing the sequential gating strategy: removal of residual EQ beads , removal of events with high event length , retaining events with a single cell-equivalent of DNA , removal of dead cells , removal of cells with TCRβ signal more than 5 MAD below the median , and removal of cells with CD8 signal more than 5 MAD below the median . Cells from each sample were down-sampled to the number in the smallest sample ( 10 , 982 ) . Only signalling proteins and selected surface markers were included in differential abundance testing ( to avoid invariant and non-biological markers ) : pSTAT5 , pAKT , pSLP76 , pLCK , IκBα , pPLCγ1 , pERK1/2 , pZAP70 , pS6 , CD8α , CD44 , CD25 , TCRβ , CD45 . To test for differential abundance of cells with any combinatorial phenotype , agnostic to cellular density or clustering patterns , we employed cytometry differential abundance testing in R ( cydar , Lun et al . , 2017 ) . This method takes advantage of the consistent staining achieved with sample barcoding , along with the count-based nature of single cell data , to find regions of the high-dimensional marker space occupied significantly more or less frequently by cells from a particular condition . This is achieved by filling the marker space with hyperspheres , comparing cellular abundances within each hypersphere across conditions , and controlling the false discovery rate across the marker space . Cells were assigned to hyperspheres and counted using the prepareCellData , neighborDistances , and countCells functions from cydar ( default parameters except countCells tol = 0 . 4 and downsample = 200 ) . Hyperspheres were included in differential abundance analysis if they contained more than 50 cells on average . Differential abundance was assessed using the edgeR ( v3 . 26 . 8 ) Bioconductor package ( Lun et al . , 2016; Robinson et al . , 2010 ) with a robust quasi-likelihood GLM fit ( Lun et al . , 2017 ) including the biological replicate of origin as a blocking factor for each sample in an analysis of deviance test to identify hyperspheres that changed in abundance in any stimulation condition compared to the unstimulated control . The spatial FDR was controlled at 0 . 05 to define significantly differentially abundant hyperspheres . See Supplementary file 5 for full summary statistics from differential abundance testing . For trajectory analysis , each biological replicate was analysed separately . A logicle transformation ( default parameters except w = 0 . 1 ) was applied to raw intensity data . Cells within each replicate were then gated using the sequential strategy described above . The MAD threshold for TCRβ+ cells was relaxed in gating biological replicates from experiment 1 due to a wider distribution in this dataset . To construct trajectories , equal numbers of cells stimulated by each ovalbumin-derived ligand were pooled with unstimulated cells . This created one sample per ligand ( N4 , T4 and G4 ) per biological replicate from which to construct trajectories . For each trajectory , cells were ordered by intensity of pS6 as this marker was observed to increase with activation over real time . Colouring cells by real time point confirmed enrichment of cells sampled at early times at the beginning of the trajectory and later times at the end ( Figure 5—figure supplement 1a ) . To generate plots in Figure 5—figure supplement 1b–c , a loess curve was fitted to intensity measurements of the indicated markers across 2000 randomly sampled cells from each trajectory ( span = 0 . 2 ) . To determine the trajectory interval in which each activation event started , trajectories were downsampled to 5000 cells each , and a sliding window encompassing 5% of the trajectory was established to move across the trajectory from least to most activated in steps of 1% . The first window in which the mean intensity of cells was more than one standard deviation away from the mean intensity in the starting window was deemed the initiation of the activation event . Events that displayed a shift in mean intensity across the trajectory but fell short of the threshold ( CD44 under G4 stimulation ) , were considered to be last in the ordering . If more than one activation event failed to meet this threshold in a given trajectory , or if two events shared an initiation window , it was not possible to robustly declare the order . We then computed the probability that orders of signalling events would be shared between each pair of trajectories to the observed extent or more if orders were random . To do this , we compared the mean-squared-distance ( MSD ) between the orders in trajectory 1 and trajectory 2 to a distribution of MSDs between the orders in trajectory 1 and permuted orderings of trajectory 2 . Both biological replicates from experiment 2 that contained all time-points revealed identical orders of activation events across all stimulation conditions . The two biological replicates in experiment 1 that included 0 , 1 and 2 hr of stimulation also revealed the same order of activation of pS6 and pERK , while pSTAT5 and CD44 were not sufficiently activated by 2 hr to determine their ordering . It was not possible to order events in the remaining biological replicates from experiment one that included only one stimulated timepoint . To test BMDC maturation , cells were stained with live-dead marker ( Zombie-NIR or Zombie-Aqua Fixable Viability Kit , Biolegend ) in PBS before staining in incubation buffer ( 1% FBS in PBS ) with FCR ( FC receptor ) blocking antibody ( Biolegend , clone 93 ) and antibodies against CD11c , MHC II , CD80 , CD86 , CD40 and ICAM1 ( Supplementary file 2 ) . Cells were acquired on a BD LSRFortessa . Data were analysed in FlowJo ( v10 ) gating on single , live cells . Mature BMDCs were consistently >90% CD11c+ and MHC II+ ( Figure 7—figure supplement 1b ) . To measure CD80 and ICAM-1 expression on activated T cells , T cells were stained with live-dead marker ( Zombie-NIR Fixable Viability Kit , Biolegend ) and antibodies against CD80 and ICAM1 in the same way . To test the impact of inhibiting the MEK and mTOR pathways on cell proliferation ( Figure 3—figure supplement 3g ) , cells were stained with eBioscience Cell Proliferation Dye eFluor-450 ( ThermoFisher ) , pre-treated for 2 hr with MEK162 ( 1 µM and 5 µM ) , rapamycin ( 200 nM ) or combined MEK162 ( 1 µM or 5 µM ) and rapamycin ( 200 nM ) , and stimulated with 1 µM N4 or NP68 peptides for 2 days . Cells were then stained with live-dead marker ( Zombie-NIR Fixable Viability Kit , Biolegend ) and acquired on a BD LSRFortessa . Data were analysed in FlowJo ( v10 ) gating on live , single cells . For phosphoflow cytometry experiments in Figure 2—figure supplement 1 , Figure 3b–d , Figure 3—figure supplements 2b and 3d–g , and Figure 7 , after stimulation , cells were fixed in 4% paraformaldehyde ( Electron Microscopy Sciences ) at room temperature for 15 min and washed in PBS . Cells were permeabilized with 90% ice-cold methanol ( Fisher Scientific ) for 30 min on ice or overnight at −20°C . Cells were washed in PBS and resuspended in 100 µL incubation buffer containing FCR blocking antibody ( Biolegend , clone 93 ) , stained with the primary antibodies of interest ( Supplementary file 2 ) , and incubated for 1 hr at room temperature . In cases where the primary antibody was not conjugated to a fluorophore , the cells were then washed , resuspended in 100 µL incubation buffer containing FCR blocking antibody and secondary antibody ( Supplementary file 2 ) and incubated for 30 min at room temperature . Cells were washed in incubation buffer prior to data acquisition on a BD LSRFortessa . Data were analysed in FlowJo ( v10 ) gating on single , live cells . To test the impact of titrating peptides on the phosphorylation of ERK and S6 ( Figure 2—figure supplement 1 ) , cells stained with live-dead marker ( Zombie-NIR Fixable Viability Kit , Biolegend ) were stimulated with N4 , T4 , G4 and NP68 peptides at concentrations of 10 nM , 100 nM and 1 µM for 2 and 4 hr . To test the impact of adding or withholding exogenous IL2 on phosphorylation of STAT5 , S6 , ERK , and AKT , and degradation of IκBα ( Figure 3—figure supplement 2b ) , cells stained with a live-dead marker ( Zombie-NIR Fixable Viability Kit , Biolegend ) were stimulated with 1 µM of peptides for 4 hr . To test the impact of inhibiting the MEK/ERK and mTOR/S6 pathways ( Figure 3b–d , Figure 3—figure supplement 3d-f ) , cells stained with live-dead marker ( Zombie-NIR Fixable Viability Kit , Biolegend ) were pre-treated with the MEK inhibitor MEK162 ( binimetinib/ARRY-162/ARRY-438162 , Selleckchem ) , mTOR inhibitor rapamycin ( Sigma-Aldrich ) or combined MEK162 and rapamycin for 2 hr . MEK162 was added at 0 . 5 µM , 1 µM and 5 µM , rapamycin was added at 20 nM , 200 nM and 2 µM . For combined drug treatments , MEK162 was added at 0 . 5 µM , 1 µM and 5 µM with rapamycin at 200 nM . DMSO as a vehicle control was added at 1:1000 , corresponding to the amount in the 200 nM dose of rapamycin and the 1 µM dose of MEK162 . Cells were stimulated with 1 µM of N4 or NP68 peptides for 4 hr . To test naïve T cell stimulation with peptide-loaded BMDCs , T cells were stained with live-dead marker ( Zombie-NIR Fixable Viability Kit , Biolegend ) before co-culture with BMDCs . To combine phosphoflow with RNA flow cytometry ( Figure 6 and Figure 6—figure supplement 1b-c ) , purified naïve CD8α+ T cells were stained using a live-dead marker ( Zombie Aqua Fixable Viability Kit , Biolegend ) . To achieve a sufficient number of cells , isolated naïve CD8+ T cells from three age- and gender-matched mice were pooled for each biological replicate . Cells were stimulated for 0–6 hr with 1 µM N4 , T4 , G4 , or NP68 peptides . At the end of stimulation , cells were immediately moved on to ice and washed with cold PBS . Cells were fixed and permeabilized using the Primeflow RNA Assay Kit ( ThermoFisher Scientific ) , blocked with FCR blocking reagent ( Biolegend , clone 93 ) and stained with antibodies against pS6[S235/236] ( BD Biosciences clone N7-548 ) for 30 min . Cells were stained with the following PrimeFlow probe sets ( Thermofisher Scientific ) : Nr4a1 AF647 ( Type1 , VB1-12484-204 ) , Irf8 AF750 ( Type 6 , VB6-3197312-210 ) , and Rpl39 AF488 ( Type 4 , VB4-3120826-204 ) as a control . The use of Rpl39 as a control gene was previously described in naïve and recently activated CD8+ T cells ( Richard et al . , 2018 ) . Cells were acquired on a BD LSRForessa and analysed in FlowJo ( v10 ) . Cells were gated on single , live cells that expressed Rpl39 , to ensure cells were permeabilized and probes hybridized and amplified . Analysis code for mass cytometry data is available at https://github . com/MarioniLab/SignallingMassCytoStimStrength ( Ma , 2020; copy archived at https://github . com/elifesciences-publications/SignallingMassCytoStimStrength/ ) .	Amongst the different types of cells the body uses to protect itself , killer T cells have an unique role: they can detect and neutralize cells that have been become dangerous for the organism – for example , cells which are cancerous or hijacked by viruses . In a healthy organism , T cells circulate through the body in an inactivated state . When a disease emerges , receptors at the surface of the cells can detect elements coming from harmful agents; this stimulation then triggers a molecular cascade inside the T cell that leads to activation . This system is relatively simple , pairing a finite number of receptors with a limited set of internal components . At the same time , the activity of T cells is finely regulated , and their activation tightly controlled: they must kill enough cells to stop the illness without causing excess damage . How this is accomplished is still unclear . A T cell can recognize harmful agents that bind its receptors with differing strengths , but how this variability in stimulation strength affects the signaling processes within the cell is still poorly understood . To investigate this question , Ma et al . used an approach called mass cytometry and analyzed the internal processes of mouse killer T cells receiving different strengths of stimulation . This investigation revealed little change in the patterns of signaling in response to signals of different strength . Instead , what differed was the proportion of T cells that became activated , and how fast this process took place: stronger stimulations led to a larger population of killer T cells being activated more rapidly . Overall , this work sheds light on how killer T cells fine-tune their response to illness using only a simple system to control their activation .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "immunology", "and", "inflammation" ]	2020	Stimulation strength controls the rate of initiation but not the molecular organisation of TCR-induced signalling
The PIWI-interacting RNA ( piRNA ) pathway is a small RNA-based immune system that controls the expression of transposons and maintains genome integrity in animal gonads . In Drosophila , piRNA-guided silencing is achieved , in part , via co-transcriptional repression of transposons by Piwi . This depends on Panoramix ( Panx ) ; however , precisely how an RNA binding event silences transcription remains to be determined . Here we show that Nuclear Export Factor 2 ( Nxf2 ) and its co-factor , Nxt1 , form a complex with Panx and are required for co-transcriptional silencing of transposons in somatic and germline cells of the ovary . Tethering of Nxf2 or Nxt1 to RNA results in silencing of target loci and the concomitant accumulation of repressive chromatin marks . Nxf2 and Panx proteins are mutually required for proper localization and stability . We mapped the protein domains crucial for the Nxf2/Panx complex formation and show that the amino-terminal portion of Panx is sufficient to induce transcriptional silencing . The piRNA pathway is a small RNA-based immune system that represses transposable elements in animal gonadal tissues ( Czech et al . , 2018; Ozata et al . , 2019 ) . At the core of this pathway are PIWI-clade Argonaute proteins that are guided by 23-30nt piRNA partners to silence transposon targets via two main mechanisms . In Drosophila , Aubergine and Argonaute 3 enforce post-transcriptional gene silencing ( PTGS ) via direct cleavage of transposon mRNAs in the cytoplasm ( Brennecke et al . , 2007; Gunawardane et al . , 2007 ) . Piwi , in contrast , operates in the nucleus where it instructs the co-transcriptional gene silencing ( TGS ) of transposon insertions ( Brennecke et al . , 2007; Klenov et al . , 2011; Sienski et al . , 2012 ) . Mutations that compromise TGS result in severe loss of transposon control , despite normal piRNA levels ( Dönertas et al . , 2013; Le Thomas et al . , 2013; Muerdter et al . , 2013; Ohtani et al . , 2013; Rozhkov et al . , 2013; Sienski et al . , 2015; Sienski et al . , 2012; Yu et al . , 2015 ) . Piwi , in complex with piRNAs , detects nascent transposon RNAs arising from active insertions and directs the silencing of these loci . Target silencing is achieved via recruitment of histone modifying enzymes that deposit repressive chromatin marks , mainly trimethylation of Lysine 9 on Histone 3 ( H3K9me3 ) ( Iwasaki et al . , 2016; Klenov et al . , 2014; Le Thomas et al . , 2013; Rozhkov et al . , 2013; Sienski et al . , 2012; Wang and Elgin , 2011 ) . Panoramix ( Panx ) is a key TGS effector , acting downstream of Piwi at the interface between the piRNA pathway and the general chromatin silencing machinery ( Sienski et al . , 2015; Yu et al . , 2015 ) . Strikingly , RNA-mediated recruitment of Panx , but not Piwi , to a locus is sufficient to trigger its epigenetic silencing , thus placing Panx at a critical node of the TGS mechanism . Downstream of Panx , the concerted action of dLsd1/Su ( var ) 3–3 and Eggless/dSETDB1 erases H3K4me2 and concomitantly deposits H3K9me3 , followed by chromatin compaction via Heterochromatin Protein 1a ( HP1a/Su ( var ) 205 ) ( Czech et al . , 2013; Iwasaki et al . , 2016; Rangan et al . , 2011; Sienski et al . , 2015; Wang and Elgin , 2011; Yu et al . , 2015 ) . Precisely how Panx recruits these histone modifying enzymes and what other factors participate in this process remains an outstanding question . Here we show that Panx coopts elements of the nuclear RNA export machinery to trigger transcriptional silencing . Panx is part of a complex that also contains Nuclear Export Factor 2 ( Nxf2 ) and Nxt1/p15 . Panx and Nxf2 are interdependent for their protein stability . nxf2 mutants show strong de-repression of Piwi-regulated transposons and severe loss of H3K9me3 at affected loci , similarly to panx mutants . We find that the amino-terminus of Panx delivers the critical silencing signal , as it is necessary and sufficient to trigger the deposition of repressive chromatin marks if tethered to a reporter construct , while its carboxyl-terminal region is involved in the interaction with Nxf2 . Nxf2 is closely related to the mRNA export factor Nxf1 , which also interacts and functions with Nxt1 ( Fribourg et al . , 2001; Herold et al . , 2001; Herold et al . , 2000 ) . Thus , our findings reveal that the evolution of transposon defense mechanisms involved exaptation of the nuclear RNA export machinery . To identify proteins associated with Panx , we immunoprecipitated a GFP-Panx fusion protein expressed from its endogenous promoter ( Handler et al . , 2013; Figure 1—figure supplement 1A ) from ovary lysates and identified co-purifying proteins by quantitative mass spectrometry . Three proteins showed the strongest enrichment and significance: Panx , Nxf2 and Nxt1 ( Figure 1A , Figure 1—source data 1 ) . Nxf2 is a homolog of the general messenger RNA ( mRNA ) export factor Nxf1 but was reported previously as being dispensable for canonical mRNA transport in S2 cells ( Herold et al . , 2001; Herold et al . , 2003 ) . Nxf2 contains all domains present in the family defined by Nxf1 , namely an amino-terminal region ( NTR ) , an RNA-binding domain ( RBD ) , leucine-rich repeats ( LRR ) , the NTF2-like domain ( NTF2 ) and a Ubiquitin-associated ( UBA ) domain ( Figure 1B and Figure 1—figure supplement 1B ) ( Herold et al . , 2001 ) . While the NTR , LRRs and RBD are typically involved in cargo binding , the NTF2 and UBA domains mediate binding to the Nuclear Pore Complex ( NPC ) and are required for Nxf1-mediated RNA export ( Braun et al . , 2001; Fribourg et al . , 2001 ) . Nxt1 , also known as p15 , is a co-factor of Nxf1 responsible for interaction with the NPC , specifically through the NTF2-fold ( Fribourg et al . , 2001; Lévesque et al . , 2001 ) . Interestingly , Nxt1 was also reported to interact with Nxf2 ( Herold et al . , 2001; Herold et al . , 2000 ) . Both Nxf2 and Nxt1 were previously identified in screens for piRNA-guided silencing in somatic and germline cells , and their depletion resulted in female sterility ( Czech et al . , 2013; Handler et al . , 2013; Muerdter et al . , 2013 ) . Contrary to expectations based upon previous findings ( Sienski et al . , 2015; Yu et al . , 2015 ) , we saw no enrichment for Piwi by mass spectrometry ( Figure 1A ) , results that are consistent with another recent study ( Batki et al . , 2019 ) . However , co-immunoprecipitation experiments detected weak but reproducible interactions between Piwi and Panx , Nxf2 , and Nxt1 , but not with a negative control ( Figure 1—figure supplement 1D ) , suggesting that low amounts of transposon substrates in unperturbed cells and/or transient associations might push Piwi below the limit of detection by less sensitive approaches . Using CRISPR/Cas9 , we generated flies that express a GFP-Nxf2 fusion protein from the endogenous nxf2 locus . GFP-Nxf2 is expressed in follicle and germline cells of the ovary and localizes predominantly to nuclei ( Figure 1C and Figure 1—figure supplement 1C ) . Mass spectrometry of GFP-Nxf2-associated proteins identified Panx , Nxf2 , and Nxt1 , as binding partners ( Figure 1D , Figure 1—source data 2 ) , implying the existence of a complex containing these three factors , which we named the Panx-induced co-transcriptional silencing ( PICTS ) complex . We therefore generated two nxf2 mutant alleles , nxf2F10* and nxf2∆1* , which harbor premature stop codons that disrupt the nxf2 open reading frame from amino acid 10 onwards ( Figure 1B and Figure 1—figure supplement 1E ) . Trans-heterozygous mutants were female sterile , with fewer eggs laid and none hatching ( Figure 1—figure supplement 1F ) . nxf2 mutants were severely compromised in the repression of soma- and germline-specific transposons , in a highly similar manner to panx mutants ( Figure 1E ) , with no change in piRNA levels or Piwi localization , despite compromised silencing ( Figure 1—figure supplement 1G–H ) . To assess the specificity of the impact of nxf2 mutations on the transcriptome , we performed RNA-seq from total RNA of heterozygote and mutant ovaries , using panx mutants for comparison . As reported previously , the expression of protein-coding genes was not generally affected in nxf2 mutants , with only 16 out of 7252 ( r2 = 0 . 963 ) being changed more than 4-fold ( Figure 1F ) ( Herold et al . , 2001; Herold et al . , 2003 ) . In contrast , 28 out of 60 transposon families ( that were above the expression threshold of 1 rpm ) were de-repressed by more than 4-fold in nxf2 mutants . Similar results were obtained for panx mutants: 16 out of 60 transposons were de-repressed and only 6 out of 7252 genes mis-regulated ( r2 = 0 . 991 ) . ChIP-seq for the H3K9me3 mark from nxf2 mutant ovaries showed reduced methylation levels at the same transposon families that were de-repressed according to RNA-seq , such as Het-A , while randomly chosen genomic intervals were not changed ( Figure 1—figure supplement 2A–B ) . Ovarian somatic cells ( OSCs ) , cultured in vitro , express a functional piRNA-guided co-transcriptional gene silencing machinery and provide a convenient context for mechanistic studies ( Saito et al . , 2009 ) . RNA-seq from OSCs depleted of Piwi , Panx or Nxf2 showed marked de-repression of soma-specific ( e . g . mdg1 , gypsy , 297 ) and intermediate transposon families ( e . g . blood ) when compared to control cells treated with GFP siRNAs ( Figure 1G left; Figure 1—figure supplement 2C ) . The de-repression of transposon families strongly correlates with the reduction in H3K9me3 levels mapped over their consensus sequences in ChIP-seq samples generated from the same knockdowns ( Figure 1G right ) . In contrast , no major changes in H3K9me3 were detected over transposons that show no de-repression in these cells upon piwi , panx , or nxf2 knockdown . We next focused on a set of 233 individual , Piwi-regulated genomic transposon insertions in OSCs ( see Materials and methods for details ) . This enabled analysis of chromatin states on individual loci , including flanking regions , rather than averaging contributions over a consensus sequence ( Figure 1—figure supplement 2D–H ) . Piwi depletion resulted in the accumulation of H3K4me2 at these loci and spreading of the mark , indicative of active transcription , beyond the transposon into downstream regions ( Figure 1—figure supplement 2D , F ) , similar to earlier reports ( Dönertas et al . , 2013; Klenov et al . , 2014; Sienski et al . , 2012 ) . Knockdown of panx or nxf2 showed similar , though less pronounced , effects . H3K9me3 marks were strongly reduced upon Piwi depletion , with panx and nxf2 knockdowns showing a similar but milder impact ( Figure 1—figure supplement 2E , G ) . H3K4me2 spreading typically correlates with increased RNA output and a decrease in H3K9me3 levels , as evident for a euchromatic gypsy insertion located within an intron of the 5' UTR of the gene ex on chromosome 2L ( Figure 1—figure supplement 2H ) . Proteins that form complexes are often interdependent for either localization or stability , and there are abundant examples of such interactions in the piRNA pathway ( Dönertas et al . , 2013; Ohtani et al . , 2013 ) . To test for such dependencies among TGS factors , we depleted Piwi , Panx or Nxf2 in germ cells of flies expressing GFP-Nxf2 ( Figure 2A ) or GFP-Panx ( Figure 2B ) . Germline knockdown of piwi had no effect on the localization of either Nxf2 or Panx . Depletion of Panx , however , led to a pronounced loss of nuclear GFP-Nxf2 in nurse cell nuclei ( Figure 2A ) . The reciprocal was also true , with GFP-Panx nuclear signal being reduced upon nxf2 knockdown in nurse cells ( Figure 2B ) . Similarly , the individual depletion of Panx or Nxf2 in follicle cells resulted in a reduction of both proteins ( Figure 2—figure supplement 1A–B ) . To assess whether the observed reduction reflects protein stability rather than mislocalization , we performed western blots on ovarian lysates . Panx protein level was strongly reduced in nxf2 mutant ovaries ( Figure 2C ) and GFP-Nxf2 signal was completely lost in homozygous panx mutants ( Figure 2D ) . Of note , mRNA levels of Nxf2 and Panx were not affected when the other factor was mutated ( Figure 1F ) , implying regulation at the protein level . Considered together , the localization and stability of Nxf2 and Panx are interdependent in vivo . To map the domains of Nxf2 and Panx that mediate their interaction , we expressed various combinations of full-length , truncated , or mutant proteins in S2 cells or in OSCs where native protein expression had been reduced by RNAi ( Figure 3A ) . Interactions were tested by co-immunoprecipitation and western blot analyses , and the subcellular localization analyzed by immunofluorescence staining . In OSCs , the ability of each construct to rescue transposon de-repression was monitored by qPCR . Full-length Nxf2 and Panx robustly co-immunoprecipitated ( Figure 3B ) and co-localized in S2 cell nuclei ( Figure 3C ) . Removing the carboxy-terminal half of Panx ( Panx-∆C ) yielded a protein that remained nuclear , while co-expressed Nxf2 remained largely cytoplasmic , and these proteins no longer formed a complex ( Figure 3B–D ) . When expressed alone in S2 cells , Nxf2 remained predominantly cytoplasmic ( ZsGreen-HA in Figure 3C ) , suggesting that interaction with Panx is necessary for nuclear localization of Nxf2 . Panx-∆N , in contrast , retained the ability to interact with Nxf2 but failed to localize to the nucleus ( Figure 3B–D ) . Strikingly , enforced localization of Panx-∆N to the nucleus also restored nuclear localization of Nxf2 ( Figure 3D and Figure 3—figure supplement 1C–D ) . Deleting only either the coiled coil domain #2 or C-terminal region , which together make up most of the carboxy-terminal half of Panx , reduced co-purification with Nxf2 ( Figure 3D and Figure 3—figure supplement 1C ) , with neither construct being able to rescue mdg1 repression upon panx knockdown ( Figure 3E ) . Overall , these results suggest that the N-terminal part of Panx carries its nuclear localization signal that aids proper localization of Nxf2 via interaction with the Panx C-terminal region . To probe the regions of Nxf2 that are essential for its function , we expressed , in the presence of full-length Panx , Nxf2 proteins that lacked either the regions required for RNA cargo binding or the region essential for its association with the NPC ( Braun et al . , 2001; Fribourg et al . , 2001; Herold et al . , 2001 ) ( Figure 3A and Figure 3—figure supplement 2A left ) . Nxf2-∆NPC failed to co-purify with Panx , and this was accompanied by an increased cytoplasmic protein localization ( Figure 3B–D ) . In contrast , deleting the cargo-binding region of Nxf2 had less impact on its co-purification with Panx or the nuclear localization of either protein ( Figure 3B–D ) . This mutant was able to interact with Panx-∆N but not Panx-∆C , as expected ( Figure 3F ) . Mutants of Nxf2 , which had individual domains removed , uniformly failed to rescue nxf2 knockdown in OSCs ( Figure 3—figure supplement 2D ) , yet all but Nxf2-∆UBA still co-purified with full-length Panx ( Figure 3D and Figure 3—figure supplement 2B–C ) . We also generated point mutants within the UBA domain , altering 2–4 highly conserved amino acids at a time ( Figure 3—figure supplement 2A right ) . UBA mutant #1 , showed a phenotype similar to the domain deletion with reduced binding to Panx , increased cytoplasmic localization , and failure to rescue loss of Nxf2 ( Figure 3B–E and Figure 3—figure supplement 2E–F ) . Thus , the interaction of Nxf2 and Panx relies on an intact UBA domain and requires the carboxy-terminal portion of Panx . The NTF2-like fold was previously shown to mediate the interaction of NXF proteins with Nxt1 ( Herold et al . , 2000; Kerkow et al . , 2012; Suyama et al . , 2000 ) . To probe a potential requirement for Nxt1 in silencing , we generated NTF2 domain point mutants in residues involved in the interaction with Nxt1 ( Kerkow et al . , 2012 ) . All four Nxf2-NTF2 point mutants localized predominantly to the nucleus and co-precipitated quantities of Panx comparable to the full-length control ( Figure 3D and Figure 3—figure supplement 2E–F ) . Yet , three mutants ( #1 , #2 , and #3 ) failed to rescue transposon de-repression upon depletion of Nxf2 ( Figure 3E ) , pointing to an involvement of Nxt1 in silencing . Indeed , reduced amounts of Nxt1 were recovered with the NTF2 mutants #1 and #2 , while the NTF2 mutant #4 , which rescued transposon expression ( Figure 3E ) , as well as UBA mutants #1 and #2 showed levels comparable to full-length Nxf2 ( Figure 3G and Figure 3—figure supplement 2G ) . Artificial tethering of Panx to nascent RNA or DNA was previously shown to result in co-transcriptional silencing and the concurrent accumulation of repressive chromatin marks ( Sienski et al . , 2015; Yu et al . , 2015 ) . To test whether Nxf2 could induce TGS , we created an integrated sensor comprising the Drosophila simulans ubiquitin promoter driving an HA-tagged ZsGreen transcript with 9 BoxB sites in its 3' UTR in OSCs ( Figure 4A ) . As expected from previous studies ( Sienski et al . , 2015; Yu et al . , 2015 ) , expression of λN-Piwi did not lead to reduced RNA or protein levels ( Figure 4B and Figure 4—figure supplement 1A ) , although it did localize to nuclei ( Figure 4—figure supplement 1B ) . Tethering of λN-Panx , in contrast , resulted in robust repression of sensor RNA and protein signals , as reported ( Sienski et al . , 2015; Yu et al . , 2015 ) . λN-Nxf2 caused an even stronger reduction of RNA and protein expression from the reporter ( Figure 4B ) . FISH experiments supported consistent repression by Panx and Nxf2 ( Figure 4—figure supplement 1C ) . Tethering of λN-Nxt1 also induced reporter repression ( Figure 4B ) . Strikingly , upon tethering of Nxt1 , Nxf2 or Panx , we observed increased levels of H3K9me3 over the reporter ( Figure 4C ) . These data suggest that Nxf2 and Nxt1 , along with Panx , act as key effectors of co-transcriptional silencing and are each sufficient to recruit the downstream silencing machinery . We also tested whether Nxf2 could silence artificial targets upon tethering to DNA rather than nascent transcripts . Our sensor construct carried 8 LacO sites upstream of the D . sim . ubiquitin promoter , which drives the expression of HA-ZsGreen ( Figure 4D ) . While tethering of LacI-Piwi did not affect sensor expression , LacI-Panx resulted in robust reductions in both RNA and protein levels ( Figure 4E and Figure 4—figure supplement 1D ) , as previously reported ( Sienski et al . , 2015 ) . LacI-Nxf2 also silenced the reporter , reducing both mRNA and protein output , albeit to a lesser extent than LacI-Panx tethering . Surprisingly , LacI-Nxt1 was unable to silence the sensor construct . Repression by tethered Panx or Nxf2 resulted in a striking decrease in H3K4me2 marks over the transcribed parts ( i . e . promoter and ZsGreen coding region ) , while the remainder of the reporter showed low read coverage and little difference in the prevalence of the mark ( Figure 4—figure supplement 1E ) . Conversely , H3K9me3 increased upon repression ( Figure 4F ) . Of note , the entire reporter sequence ( except a small gap around the LacO site and the 3' UTR where the mappability is poor ) was prominently decorated with H3K9me3 , suggesting spreading of this chromatin mark following initial silencing . The data presented above identified functional elements within Nxf2 and Panx that are required for proper localization , interaction , and Piwi-dependent transcriptional gene silencing . We next examined the ability of Panx and Nxf2 mutants to silence our artificial DNA reporters , thus bypassing Piwi-piRNA dependent target recognition . Neither LacI-Nxf2-∆Cargo nor LacI-Panx-∆N , which were predicted to interact with their full-length partner in OSCs ( Figure 3B–D ) , were able to repress the sensor ( Figure 4G ) . Yet , LacI-Panx-∆C , which could not interact with Nxf2 , substantially reduced RNA and protein expression from the sensor , with its effects as robust as upon DNA tethering of LacI-HP1a ( Figure 4G ) . This suggests that the amino-terminal part of Panx is necessary and sufficient to enforce silencing of an artificial reporter independent of its interaction with Nxf2 . Our data identify Nxf2 and Nxt1 as critical mediators of co-transcriptional gene silencing , acting in concert with Panx to repress loci in response to Piwi-piRNA target engagement ( Figure 4H ) . The emerging model for piRNA-dependent silencing implies that target recognition by Piwi is necessary to recruit the PICTS complex onto the appropriate nascent RNA targets . Difficulties in detecting stable interactions between Piwi and PICTS components in vivo may arise from a requirement for Piwi target engagement to licence it for recruitment of silencing complexes , as has been suggested previously ( Sienski et al . , 2015; Yu et al . , 2015 ) . The same mechanism may underlie the difficulties experienced in observing Piwi on its target loci by ChIP ( Marinov et al . , 2015 ) . We find that Panx and Nxf2 are interdependent for their protein stability and proper subcellular localization , underscoring the fact that correct assembly of the PICTS complex is essential for TGS , while the silencing capacity , per se , resides in Panx . Of note , previous work reported a partial destabilization of Nxf2 in cells depleted of Nxt1 ( Herold et al . , 2001 ) , potentially extending the interdependency to all three proteins . RIP-seq experiments from unperturbed cells found transposon RNAs enriched only with Panx , as reported ( Sienski et al . , 2015 ) , but not with Nxf2 ( Figure 4—figure supplement 1F ) , possibly due to low substrate availability combined with an insensitive assay . These results are consistent with another recent report that did not detect transposon enrichment in Nxf2 CLIP-seq from wild-type cells ( Batki et al . , 2019 ) . However , two other studies identified transposon mRNA association with Nxf2 in CLIP-seq experiments upon depletion of the previously described TGS factor , Mael ( Zhao et al . , 2019 ) , or by using a stable cell line and depletion of endogenous Nxf2 ( Murano et al . , 2019 ) . Considered together , these data suggest that Nxf2 might be important for stabilizing the binding of Panx to nascent RNAs . However , precisely how Nxf2 executes this function remains to be fully elucidated . Of note , Murano and colleagues find that Panx interacts with Nxf2 , Piwi , Mael and Arx ( Murano et al . , 2019 ) , which could imply that other TGS factors come into contact with the PICTS complex , although the relationship between these factors and PICTS requires further investigation . Mutational analyses suggest that Panx and Nxf2 must normally bind Nxt1 to carry out transposon repression . Direct recruitment of any of the PICTS complex components to RNA reporters results in robust chromatin silencing . Upon tethering to DNA , however , Panx induces potent TGS , whereas Nxf2 leads to less prominent effects and Nxt1 shows no silencing capability in our assays . Interestingly , recruitment of the amino-terminal part of Panx alone is necessary and sufficient to induce reporter repression , pinpointing this domain of Panx as harboring the silencing effector function . Future investigations will be crucial to uncover the molecular mechanism by which the Panx amino terminus instructs the downstream chromatin silencing machinery . Our work , and that of others ( Batki et al . , 2019; Murano et al . , 2019; Zhao et al . , 2019 ) indicates that piRNA-guided co-transcriptional silencing of transposons has coopted several components of the RNA export machinery , namely Nxf2 and Nxt1 . Of the four NXF proteins present in flies , only two have thus far been characterized . Interestingly , while Nxf1 , acting along with Nxt1 , is crucial for canonical mRNA export ( Braun et al . , 2001; Fribourg et al . , 2001; Herold et al . , 2001; Wilkie et al . , 2001 ) , Nxf2 has been coopted by the piRNA pathway and functions in co-transcriptional gene silencing . Nxf3 , which also is required for transposon repression in germ cells ( Czech et al . , 2013 ) , is emerging as being critical for the export of piRNA precursors generated from dual-strand clusters in the germline ( ElMaghraby et al . , 2019; Kneuss et al . , 2019 . The role of Nxf4 , whose expression is testis-specific , is yet to be established . This remarkable functional diversity of NXF family members correlates with tissue-specific expression patterns , and seems conserved in mammals ( Yang et al . , 2001 ) . However , deciphering how each achieves substrate specificity will be critical to understanding how these homologs can be exclusively dedicated to different targets and confer different fates upon the RNAs that they bind . Importantly , the fate of the nascent transcript that is detected by Piwi and instructed by PICTS for silencing remains unclear . One hypothesis is that instead of being exported , these targets undergo degradation by the nuclear exosome . Such mechanism would be contrary to yeast , where the TREX complex subunit Mlo3 was shown to oppose siRNA-mediated heterochromatin formation at gene loci ( Yu et al . , 2018 ) , and suggests that different lineages have evolved different silencing mechanisms . In any case , it is possible that a single transcript from a locus that is marked for silencing might pose a lesser threat than an unsilenced locus and , therefore , not be capable of exerting evolutionary pressure for the detemination of its fate . All flies were kept at 25°C . Flies carrying a BAC transgene expressing GFP-Panx were generated by the Brennecke lab ( Handler et al . , 2013 ) . Panx frameshift mutants panxM1 and panxM4 were described earlier ( Yu et al . , 2015 ) . The GFP-Nxf2 fusion knock-in and nxf2 frameshift mutations ( nxf2[F10] and nxf2[∆1] ) were generated for this study ( see below ) . Control w1118 flies were a gift from the University of Cambridge Department of Genetics Fly Facility . For knockdowns we used a stock containing the Dcr2 transgene and a nos-GAL4 driver ( described in Czech et al . , 2013 ) and dsRNA lines from the VDRC ( panxKK102702 , nxf2KK101676 , piwiKK101658 ) . Fertility of the nxf2 and panx mutant females was scored by crossing ten freshly hatched females to five w1118 males and counting the number of eggs laid in 12 hr periods and pupae that developed after 7 days . Frameshift mutant alleles of nxf2 were generated by injecting pCFD4 ( addgene plasmid # 49411; Port et al . , 2014 ) containing two gRNAs against Nxf2 ( generated by Gibson assembly ) into embryos expressing vas-Cas9 ( Bloomington stock 51323 ) . To generate GFP-Nxf2 fusion knock-in flies , homology arms of approximately 1 kb were cloned into pUC19 by Gibson assembly and co-injected with pCFD3 ( addgene plasmid # 49410; Port et al . , 2014 ) containing a single guide RNA into embryos expressing vas-Cas9 ( # 51323 , Bloomington stock center ) . Microinjection and fly stock generation was carried out by the University of Cambridge Department of Genetics Fly Facility . Mutant and knock-in flies were identified by genotyping PCRs and confirmed by sanger sequencing . Ovaries from ~170 GFP-Panx , GFP-Nxf2 and control flies ( 3–5 days old ) were dissected in ice-cold PBS and lysed in 300 μl of CoIP Lysis Buffer ( 20 mM Tris-HCl pH 7 . 5 , 150 mM NaCl , 2 mM MgCl2 , 10% glycerol , 1 mM DTT , 0 . 1 mM PMSF , 0 . 2% NP-40 supplemented with complete protease inhibitors [Roche] ) and homogenized using a motorized pestle . Lysates were cleared for 5 min at 16000 g and the residual pellet re-extracted with the same procedure . GFP-tagged proteins were immunoprecipitated by incubation with 30 µl of GFP-Trap magnetic beads ( Chromotek ) for 3 hr at 4°C on a tube rotator . The beads were washed 6x with Lysis Buffer and 2x with 100 mM Ammonium Bicarbonate , before TMT-labelling followed by quantitative Mass Spectrometry . TMT chemical isobaric labeling were performed as described ( Papachristou et al . , 2018 ) . Raw data were processed in Proteome Discoverer 2 . 1 software ( Thermo Fisher Scientific ) using the SequestHT search engine . The data were searched against a database derived from FlyBase ( ‘dmel-all-translation-r6 . 24’ ) at a 1% spectrum level FDR criteria using Percolator ( University of Washington ) . For the SequestHT node the following parameters were included: Precursor mass tolerance 20 ppm and fragment mass tolerance 0 . 5 Da . Dynamic modifications were oxidation of M ( +15 . 995 Da ) , deamidation of N , Q ( +0 . 984 Da ) and static modifications were TMT6plex at any N-Terminus and K ( +229 . 163 Da ) . The consensus workflow included S/N calculation for TMT intensities and only unique peptides identified with high confidence ( FDR < 0 . 01 ) were considered for quantification . Downstream data analysis was performed on R using the qPLEXanalyzer package ( https://doi . org/10 . 5281/zenodo . 1237825 ) as described ( Papachristou et al . , 2018 ) . Only proteins with more than one unique peptide were considered . Drosophila Ovarian Somatic Cells ( OSCs ) were a gift from Mikiko Siomi and were cultured at 26°C in Shields and Sang M3 Insect Medium ( Sigma Aldrich ) supplemented with 0 . 6 mg/ml Glutathione , 10% FBS , 10 mU/ml insulin and 10% fly extract ( purchased from DGRC ) as described ( Niki et al . , 2006; Saito , 2014; Saito et al . , 2009 ) . Cell identity was authenticated by whole genome DNA sequencing in-house . Gibco Drosophila Schneider 2 ( S2 ) cells were purchased from Thermo Fisher Scientific ( catalog number R69007 ) and were grown at 26°C in Schneider's Drosophila Media ( Gibco ) supplemented with 10% heat-inactivated FBS . Cell identity was characterized by Thermo Fisher Scientific through isozyme and karyotype analysis ( see product description ) . OSCs and S2 cells tested negative for mycoplasma contamination in-house . Knockdowns ( all siRNA sequences are given in Supplementary file 1 ) and transfections in OSCs were carried out as previously described ( Saito , 2014 ) . In short , for knockdown experiments 10 × 106 cells were nucleofected with 200 pmol annealed siRNAs using the Amaxa Cell Line Nucleofector Kit V ( Lonza , program T-029 ) . After 48 hr , 10 × 106 cells were nucleofected again with 200 pmol of the same siRNA and allowed to grow for an additional 48 hr before further processing . For rescue experiments , 5 µg of rescue construct plasmid were added to the second knockdown solution . OSCs were transfected with 10 μg of plasmid using Xfect ( Clontech ) , according to manufacturer’s instruction . S2 cells were transfected with 2 μg of plasmid using Effectene ( Qiagen ) , according to manufacturer’s instructions . S2 cells or OSCs were transfected with 3xFLAG- and HA-tagged constructs ( wild-type and mutants ) . Cells were harvested 48 hr after transfection and lysed in 250 μl of CoIP Lysis Buffer ( Pierce ) supplemented with Complete protease inhibitors ( Roche ) . 200 μg of proteins for each sample were diluted to 1 ml with CoIP Lysis Buffer and the 3xFLAG-tagged bait was immunoprecipitated by incubation with 20 μl of anti-FLAG M2 Magnetic Beads ( Sigma M8823 ) for 2 hr at 4°C on a tube rotator . The beads were washed 3 × 15 min with TBS supplemented with protease inhibitors . Beads were then resuspended in 2x NuPAGE LDS Sample Buffer ( Thermo Fisher Scientific ) without reducing agent and boiled for 3 min at 90°C to elute immunoprecipitated proteins . IPs , unbound fractions and input fractions were diluted to 1x NuPAGE LDS Sample Buffer concentration and reducing agent was added . Samples were boiled at 90°C for 10 min before separating proteins as described below . Protein concentration was measured using a Direct Detect Infrared Spectrometer ( Merck ) . 20 µg of proteins were separated on a NuPAGE 4–12% Bis-Tris gel ( Thermo Fisher Scientific ) . Proteins were transferred with an iBLot2 device ( Invitrogen ) on a nitrocellulose membrane and blocked for 1 hr in 1x Licor TBS Blocking Buffer ( Licor ) . Primary antibodies were incubated over night at 4°C . Licor secondary antibodies were incubated for 45 min at room temperature ( RT ) and images acquired with an Odyssey CLx scanner ( Licor ) . The following antibodies were used: anti-HA ( ab9110 ) , anti-FLAG ( Sigma #F1804 ) , anti-GFP ( ab13970 ) , anti-Piwi ( described in Brennecke et al . , 2007 ) , anti-Nxt1 ( Herold et al . , 2001 ) , anti-Histone H3 ( ab10799 ) , anti-Tubulin ( ab18251 ) , mouse anti-Panx ( Sienski et al . , 2015 ) , IRDye 680RD Donkey anti-Rabbit IgG ( H + L ) ( Licor ) , IRDye 800CW Donkey anti-Mouse IgG ( H + L ) ( Licor ) , IRDye 800CW Goat anti-Rat IgG ( H + L ) ( Licor ) . Fly ovaries were dissected in ice-cold PBS and fixed in 4% paraformaldehyde ( PFA ) at RT for 15 min . After two quick rinses in PBS with Triton at 0 . 3% ( PBS-Tr ) , samples were permeabilized with 3 × 10 min washes with PBS-Tr . Samples were then blocked in PBS-Tr with 1% BSA for 2 hr at RT and then incubated overnight at 4°C with primary antibodies in PBS-Tr and 1% BSA . The next day , samples were washed 3 × 10 min at RT in PBS-Tr and incubated overnight at 4°C with secondary antibodies in PBS-Tr and 1% BSA . The next day , samples were washed 4 × 10 min in PBS-Tr at RT and DAPI ( Thermo Fisher Scientific #D1306 ) was added during the third wash . After 2 × 5 min washes in PBS , samples were mounted on slides with ProLong Diamond Antifade Mountant ( Thermo Fisher Scientific #P36961 ) and imaged on a Leica SP8 confocal microscope ( 63x and 100x Oil objective ) . The following antibodies were used: chicken anti-GFP ( ab13970 ) , rabbit anti-Piwi ( described in Brennecke et al . , 2007 ) , mouse anti-Aub ( Senti et al . , 2015 ) , anti-Rabbit-555 ( Thermo Fisher ) , anti-Mouse-647 ( Thermo Fisher ) , anti-Chicken-647 ( Abcam ) . Cells were plated one day in advance on Fibronectin- or Concanavalin A- coated coverslips ( for OSCs and S2 cells , respectively ) , fixed for 15 min in 4% PFA , permeabilized for 10 min in PBS with 0 . 2% Triton ( PBST ) and blocked for 30 min in PBS , 0 . 1% Tween-20% and 1% BSA . Primary antibodies were diluted in PBS , 0 . 1% Tween-20% and 0 . 1% BSA and incubated overnight at 4°C . After 3 × 5 min washes in PBST , secondary antibodies were incubated for 1 hr at RT . After 3 × 5 min washes in PBST , DAPI was incubated for 10 min at RT , washed two times and the coverslips were mounted using ProLong Diamond Antifade Mountant ( Thermo Fisher Scientific #P36961 ) and imaged on a Leica SP8 confocal microscope ( 100x Oil objective ) . The following antibodies were used: anti-Lamin ( Developmental Studies Hybridoma Bank ADL67 . 10 ) , anti-HA ( ab9111 ) , anti-FLAG ( Cell Signaling Technology 14793S ) , anti-chicken-488 ( Abcam ) , anti-Rabbit-555 ( Thermo Fisher ) , anti-Mouse-647 ( Thermo Fisher ) . RNA FISH was performed with Hybridization Chain Reaction ( HCR ) , similar as reported ( Ang and Yung , 2016; Choi et al . , 2014 ) . OSCs were fixed for 15 min in 4% PFA , washed 2 × 5 min with PBS and permeabilized for at least 24 hr in 70% ethanol at −20°C . Ethanol was removed and slides washed twice for 5 min in 2x Saline-Sodium Citrate buffer ( SSC ) . Priming for hybridization was done by incubating for 10 min in 15% formamide in 2x SSC . HCR probes were diluted to 1 nM each in hybridization buffer ( 15% formamide , 10% dextran sulfate in 2x SSC ) and incubated overnight at 37°C in a humidified chamber . Excess probes were removed by rinsing twice in 2x SSC and washing once in 30% formamide for 10 min at 37°C . HCR hairpins conjugated to AlexaFluor-647 ( IDT ) were heat-denatured and diluted to 120 nM in 5x SSC and 0 . 1% Tween-20 . HCR amplification was carried out for 2 hr at RT in the dark and washed 3 × 10 min with 5x SSC and 0 . 1% Tween-20 . Nuclei were stained with DAPI for 10 min , followed by 3 × 10 min washes in 5x SSC . Slides were mounted with ProLong Diamond Antifade Mountant ( Thermo Fisher Scientific ) and imaged on a Leica SP8 confocal microscope ( 100x Oil objective ) . The sequences of all probes are given in Supplementary file 1 . Intensity plot profiles across individual egg chambers were acquired in Fiji ( lines displayed ) . Intensity values for each channel were averaged over 10 pixels and adjusted as a percentage of the highest value along the profile . A threshold of 30% DAPI intensity was set to define nuclei . Individual egg chambers used for analysis are displayed for each channel with inverted LUT . For RNA tethering , OSCs with a stable integration of the sensor plasmid ( pDsimUbi-HA-ZsGreen-NLuc-9xBoxB ) were generated in the lab . 4 × 106 cells were nucleofected with 5 μg of plasmid expressing λN-3xFLAG-tagged constructs , as described above . After 48 hr , 4 × 106 cells were nucleofected again with 5 μg of the same plasmid and allowed to grow for an additional 48 hr before the relative expression of the sensor was analyzed . For DNA tethering , OSCs were transiently transfected with 8xLacO-pDsimUbi-HA-ZsGreen sensor plasmid and LacI-3xFLAG fusion constructs . Cells were allowed to grow for 72 hr before the relative expression of the sensor was determined . Ovaries from 120 to 150 adult flies were dissected in ice-cold PBS , collected in 1 . 5 ml Bioruptor Microtubes ( Diagenode #C30010016 ) , and immediately frozen at −80°C . Samples were crosslinked in 1 ml A1 buffer ( 60 mM KCl , 15 mM NaCl , 15 mM HEPES pH 7 . 6 , 4 mM MgCl2 , 0 . 5% Triton X-100 , 0 . 5 mM dithiothreitol ( DTT ) , 10 mM sodium butyrate and complete EDTA-free protease inhibitor cocktail [Roche #04693159001] ) , in the presence of 1 . 8% formaldehyde . Samples were homogenized with a micropestle for 2 min and incubated for a total time of 15 min at RT on a rotating wheel . Crosslinking was stopped by adding 225 mM glycine followed by incubation for 5 min on a rotating wheel . The homogenate was centrifuged for 5 min at 4 , 000 g at 4°C . The supernatant was discarded , and the nuclear pellet was washed twice in 1 ml A1 buffer and once in 1 ml of A2 buffer ( 140 mM NaCl , 15 mM HEPES pH 7 . 6 , 1 mM EDTA , 0 . 5 mM EGTA , 1% Triton X-100 , 0 . 5 mM DTT , 0 . 1% sodium deoxycholate , 10 mM sodium butyrate and complete mini EDTA-free protease inhibitor cocktail ) at 4°C . Nuclei were then resuspended in 100 μl A2 buffer with 1% SDS and 0 . 5% N-laurosylsarcosine and incubated for 2 hr at 4°C with agitation at 1 , 500 rpm . Chromatin was sonicated using a Bioruptor Pico ( Diagenode #B01060010 ) for 16 cycles of 30 s on/30 s off . Sheared chromatin size peaked at 150 bp . After sonication and 5 min high-speed centrifugation at 4°C , fragmented chromatin was recovered in the supernatant and the final volume was raised to 1 ml in A2 buffer with 0 . 1% SDS . 50 µl of the diluted samples were used as DNA input control , in a final volume of 200 µl of A2 buffer with 0 . 1% SDS . Chromatin for IP was precleared by addition of 15 µl of Protein A/G Magnetic Beads ( Thermo Fisher Scientific ) suspension followed by overnight incubation at 4°C . Beads were removed by centrifugation , and anti-H3K9me3 ( Active Motif #39161 ) antibody was added ( 1:200 dilution ) to 5 µg of chromatin and incubated for 4 hr at 4°C on a rotating wheel . 50 µl of Protein A/G Magnetic Beads were added , and incubation was continued overnight at 4°C . Antibody-protein complexes were washed 4 times in A3 ( A2 +0 . 05% SDS ) buffer and twice in 1 mM EDTA , 10 mM Tris ( pH 8 ) buffer for 5 min at 4°C on a rotating wheel . Chromatin was eluted from the beads in 200 μl of 10 mM EDTA , 1% SDS , 50 mM Tris ( pH 8 ) for 30 min with agitation at 1 , 500 rpm and then reverse-crosslinked overnight at 65°C , together with the input DNA . IP and input samples were treated with 2 µl of Proteinase K ( Thermo Fisher Scientific #EO0491 ) for 3 hr at 56°C . DNA was purified using the MinElute PCR purification Kit ( Thermo Fisher Scientific ) , according to manufacturer’s instructions , and resuspended in 30 μl water . Recovered DNA was quantified with Qubit 4 Fluorometer ( Thermo Fisher Scientific ) and analysed with Agilent Bioanalyzer 2100 High Sensitivity DNA Chip ( Agilent ) . DNA libraries were prepared with NEBNext Ultra II DNA Library Prep Kit for Illumina ( NEB ) , according to manufacturer’s instructions . DNA libraries were quantified with KAPA Library Quantification Kit for Illumina ( Kapa Biosystems ) and deep-sequenced with Illumina HiSeq 4000 ( Illumina ) . For ChIP from OSCs we adapted a protocol by Schmidt and colleagues ( Schmidt et al . , 2009 ) . In short , 10 × 106 OSCs were crosslinked in 1% formaldehyde for 10 min . Crosslinking was quenched by addition of glycine solution , followed by three washes in ice-cold PBS . Crosslinked cells were either snap-frozen in liquid nitrogen and stored at −80°C or processed immediately . Cells were resuspended in 1 ml buffer LB1 ( 50 mM HEPES-KOH pH 7 . 5 , 140 mM NaCl , 1 mM EDTA , 10% glycerol , 0 . 5% Igepal CA-630 , 0 . 25% Triton-X 100 , EDTA-free protease inhibitor cocktail [Roche] ) and incubated on ice for 10 min while inverting several times . Cells were centrifuged for 5 min at 2 , 000 g at 4°C . Supernatant was discarded and pellet resuspended in 1 ml LB2 10 mM Tris-HCL pH 8 . 0 , 200 mM NaCl , 1 mM EDTA , 0 . 5 M EGTA , EDTA-free protease inhibitor cocktail ) . Cells were incubated on ice for 5 min and centrifuged again . Isolated nuclei were resuspended in 300 µl sonication buffer LB3 ( 10 mM Tris-HCL pH 8 , 100 mM NaCl , 1 mM EDTA , 0 . 5 mM EGTA , 0 . 1% Na-Deoxycholate , 0 . 5% N-lauroylsarcosine , EDTA-free protease inhibitor cocktail ) and transferred in 1 . 5 ml Bioruptor Microtubes ( Diagenode ) . Chromatin was sonicated using a Bioruptor Pico ( Diagenode ) for 16 cycles of 30 s on/30 s off . Sheared chromatin size peaked at 150 bp . The lysate was cleared by high-speed centrifugation at 4°C . 100 µl Protein A Dynabeads ( Thermo Fisher Scientific ) were incubated with 5 µl H3K9me3 ( Active Motif #39161 ) or H3K4me2 antibody ( Millipore # 07–030 ) over night at 4°C while rotating . The cleared lysate was split in two equal fractions and a 5 µl input fraction was saved for further processing . Lysate volumes were adjusted to 300 µl with LB3 and Triton-X 100 was added to a final concentration of 1% . Lysates were incubated with either H3K9me3 or H3K4me2 coated beads over night at 4°C while rotating . Washing , reverse-crosslinking , DNA purification and library preparation was done as described above ( ChIP-seq from ovaries ) . Cell pellets or fly ovaries were lysed in 1 ml Trizol and RNA was extracted using RNeasy mini prep column ( Qiagen ) , according to manufacturer’s instructions . 1 µg of total RNA was treated with DNAseI ( Thermo Fisher Scientific ) , according to manufacturer’s instructions . Reverse transcription was performed with Superscript III First Strand Synthesis Kit ( Thermo Fisher Scientific ) , using oligo ( dT ) 20 primers , according to the manufacturer’s instructions . Real-time PCR ( qPCR ) experiments were performed with a QuantStudio Real-Time PCR Light Cycler ( Thermo Fisher Scientific ) . Transposon levels were quantified using the ∆∆CT method ( Livak and Schmittgen , 2001 ) , normalized to rp49 and fold changes were calculated relative to the indicated controls . All oligonucleotide sequences are given in Supplementary file 1 . Ovaries from ~100 GFP-Panx or GFP-Nxf2 flies ( 3–5 days old ) were dissected in ice-cold PBS and fixed with 0 . 1% PFA for 20 min , followed by quenching with equal volumes of 125 mM Glycine . Fixed ovaries were lysed in 200 μl of RIPA Buffer ( supplemented with complete protease inhibitors ( Roche ) and RNasin Plus 40 U/ml ) and homogenized using a motorized pestle . Lysates were incubated 3 min at 37°C with 4 µl of Turbo DNase , incubated 20 min at 4°C on a tube rotator and sonicated with a Bioruptor Pico ( 3 cycles of 30 s on/30 s off ) . Lysates were pre-cleared using 40 µl of Pierce Protein A/G beads for 1 hr at 4°C and GFP-tagged proteins were immunoprecipitated by incubation with 50 µl of GFP-Trap magnetic agarose beads ( Chromotek ) overnight at 4°C . An aliquot of pre-cleared input lysate was saved for RNA isolation and library preparation . Following 3 washes in 150 mM KCl , 25 mM Tris ( pH 7 . 5 ) , 5 mM EDTA , 0 . 5% NP40 , 0 . 5 mM DTT ( supplemented with protease inhibitors and RNasin Plus 1:1000 ) , IP and input samples were reverse crosslinked in 1x Reverse Crosslinking buffer ( PBS , 2% Nlauroyl sarcosine , 10 mM EDTA , 5 mM DTT ) and Proteinase K . RNA isolation was performed using Trizol and 100 ng of input or IP RNA were used for library preparation using the SMARTer stranded RNA-seq Kit ( Clontech ) . DNA libraries were quantified with KAPA Library Quantification Kit for Illumina ( Kapa Biosystems ) and deep-sequenced with Illumina HiSeq 4000 ( Illumina ) . Small RNA libraries were generated as described previously ( Jayaprakash et al . , 2011 ) . Briefly , 18- to 29-nt-long small RNAs were purified by PAGE from 10 µg of total ovarian RNA . Next , the 3' adapter ( containing four random nucleotides at the 5' end ) was ligated overnight using T4 RNA ligase 2 , truncated KQ ( NEB ) . Following recovery of the products by PAGE purification , the 5' adapter ( containing four random nucleotides at the 3' end ) was ligated to the small RNAs using T4 RNA ligase ( Abcam ) for 1 hr . Small RNAs containing both adapters were recovered by PAGE purification , reverse transcribed and PCR amplified prior quantification using the Library Quantification Kit for Illumina ( Kapa Biosystems ) and sequenced on an Illumina HiSeq 4000 ( Illumina ) . All adapter sequences are given in Supplementary file 1 . 1 µg of total RNA was used as input material for library preparation . The NEBNext Poly ( A ) mRNA magnetic Isolation Module ( NEB ) was used to isolate poly ( A ) RNAs . Libraries were generated with the NEBNext Ultra Directional RNA Library Prep kit for Illumina ( NEB ) according to manufacturer's instructions . The pooled libraries were quantified with KAPA Library Quantification Kit for Illumina ( Kapa Biosystems ) and sequenced on an Illumina HiSeq 4000 ( Illumina ) . Raw fastq files generated by Illumina sequencing were analysed by a pipeline developed in-house . In short , the first and last base of each 50 bp read were removed using fastx trimmer ( http://hannonlab . cshl . edu/fastx_toolkit/ ) . RIP-seq reads were first aligned against rRNAs and mapped reads discarded . High-quality reads were aligned to the Drosophila melanogaster genome release 6 ( dm6 ) downloaded from Flybase using STAR ( Dobin et al . , 2013 ) . For transposon-wide analysis , genome multi-mapping reads were randomly assigned to one location using option '--outFilterMultimapNmax 1000 --outMultimapperOrder Random' and non-mapping reads were removed . Alignment files were then converted back to fastq format with samtools ( Li et al . , 2009 ) and re-aligned to the transposon consensus sequences allowing multi-mappers that were assigned to a random position . Generated bam alignment files were indexed using samtools index . For genome-wide analyses , multi-mapping reads were removed to ensure unique locations of reads . Normalization was achieved by calculating rpm ( reads per million ) using the deepTools2 bamCoverage function ( Ramírez et al . , 2016 ) with 10 bp bin sizes . The scaling factor for transposon mapping reads was calculated from reads that aligned to transposon consensus sequences relative to genome aligned reads . Reads mapping to genes were counted with htseq ( Anders et al . , 2015 ) and transposon derived reads were calculated using a custom script ( available with this article as Source code 1 ) . Metaplots flanking euchromatic insertion sites and transposon coverage plots were calculated by deepTools2 with bin sizes of 10 bp and 50 bp , respectively . Stranded RNA libraries were trimmed , aligned and indexed as described above . Alignment files were split in sense and antisense reads using samtools view . Normalization of the split alignment files as well as feature counting was performed as described above . For transposon expression analysis only sense reads were considered . Differential expression analysis was performed using a custom build R script ( available with this article as Source code 2 ) . Adapters from raw small RNA fastq files were clipped with fastx_clipper ( adapter sequence AGATCGGAAGAGCACACGTCTGAACTCCAGTCA ) keeping only reads with at least 23 bp length . Then the first and last four bases were trimmed using seqtk ( https://github . com/lh3/seqtk ) . Alignment and normalization were performed as described above . Only high-quality small RNA reads with a length between 23 and 29 bp were used for further analysis of piRNA profiles . piRNA distribution was calculated and plotted in R . For piRNA coverage plots over TEs , only the 5’ position of reads was plotted . The locations of euchromatic transposon insertions in OSCs were derived from Sienski et al . ( 2012 ) and updated to dm6 genome release coordinates using the UCSC liftOver tool . Transposon consensus sequences were downloaded from Flybase . Mappability tracks for dm6 with 50 bp resolution were calculated as described ( Derrien et al . , 2012 ) . Piwi-dependent OSC insertions were defined by comparing H3K9me3 signal intensities of siRNA-mediated knockdowns for gfp and piwi . Signal was counted by htseq using a customized GTF file including the locations of all euchromatic TE insertions in OSCs and reads were normalized to rpm . TE insertions were annotated as Piwi-dependent if the ratio of normalized signal intensity of GFP knockdown versus Piwi knockdown was higher than 2 . Random genomic windows for box plots of H3K9me3 ChIP-seq data were calculated using BEDtools’ random function ( Quinlan and Hall , 2010 ) with bin size 5 , 000 bp , bin number 1000 and random seed number 800 . 100 random windows were chosen ( number 200–300 ) and analysed along with ChIP-seq data for de-repressed TEs and those not affected . Welch two sample t-test was applied for statistics . Metaplots of euchromatic TE insertions as well as TE coverage plots for RNA-seq and ChIP-seq data were generated with deepTools2 and Adobe Illustrator . Scatterplots for differentially expressed transposons and genes were generated with R package ggplot2 . Heatmaps were calculated with deepTools2 and data plotted in R . For scatterplots , only TEs and genes with a scaled read count larger than 1 ( rpm >1 ) were used in the analysis and included in plots . Statistical analysis applied to qPCR data sets was calculated by unpaired ( two sample ) t Test . The number of biological replicates is indicated in the figure legends . Statistical analysis applied to data sets displayed as box plots ( Figure 1—figure supplement 2A ) was calculated by Welch two sample t-test . Sequencing data reported in this paper has been deposited in Gene Expression Omnibus under ID code GSE121661 . Mass Spectrometry data has been deposited to PRIDE Archive under ID code PXD011415 .	For an organism to form and grow correctly , it must rely on the genetic information it has received from its parents . DNA , however , is full of elements called transposons that can disrupt this information by moving around the genome and inserting themselves into genes . Changes caused by these invading elements can lead to devastating effects , such as cancer and cell death . To shield their DNA from harm , organisms have evolved regulatory machineries to recognize and correct alterations that may be damaging . One way cells can protect their DNA is by silencing disruptive transposons using small molecules known as piRNAs . These protective molecules detect transposons as soon as they are active and recruit other proteins to switch them off . However , questions still remain about how specific proteins recruited by piRNAs are involved in this process . In flies , a protein called Panoramix ( Panx ) is known to trigger transposon silencing , but how it does this is still unclear . Now , Fabry , Ciabrelli , Munafò et al . set out to investigate how Panx silences active transposons in the ovaries of flies and whether other proteins are involved . Like so many other proteins , Panx was found not to work alone but to form a complex with two other proteins , called Nxf2 and Nxt1 . The experiments showed that all three components of the complex , named PICTS , are critical for transposon control in flies , but Panx is the engine that drives the machine . Panx and Nxf2 were found to stabilize each other , and together with co-factor Nxt1 place a mark on the genome at the point where the transposon emerges , effectively switching it off . Notably , although Nxf2 and Nxt1 are part of a family of proteins that export molecules from the nucleus , both these factors appear to have been repurposed to silence transposable elements within the genome . This work expands our understanding of how cells employ regulatory machineries , like the PICTS complex , to guard against disruptive genetic changes . These mechanisms are often conserved throughout evolution , and the findings presented here may help identify ways to counteract harmful changes caused by transposons in other organisms , including humans . However , more work would be required to deepen our knowledge of how these processes work .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression" ]	2019	piRNA-guided co-transcriptional silencing coopts nuclear export factors
Mitochondrial dysfunction is a hallmark of aging , and underlies the development of many diseases . Cells maintain mitochondrial homeostasis through a number of pathways that remodel the mitochondrial proteome or alter mitochondrial content during times of stress or metabolic adaptation . Here , using yeast as a model system , we identify a new mitochondrial degradation system that remodels the mitochondrial proteome of aged cells . Unlike many common mitochondrial degradation pathways , this system selectively removes a subset of membrane proteins from the mitochondrial inner and outer membranes , while leaving the remainder of the organelle intact . Selective removal of preexisting proteins is achieved by sorting into a mitochondrial-derived compartment , or MDC , followed by release through mitochondrial fission and elimination by autophagy . Formation of MDCs requires the import receptors Tom70/71 , and failure to form these structures exacerbates preexisting mitochondrial dysfunction , suggesting that the MDC pathway provides protection to mitochondria in times of stress . Mitochondria play a central role in cellular metabolism . Metabolic pathways that occur within mitochondria include the TCA cycle , oxidative phosphorylation , amino acid metabolism , and biosynthesis of lipids , heme , and iron-sulfur clusters ( Rutter and Hughes , 2015 ) . Because so many vital processes take place within mitochondria , cells have evolved multiple mechanisms to maintain mitochondrial homeostasis under the diverse array of physiological states that a cell can exist in . These mechanisms are important for maintaining mitochondrial function during development , cellular specialization , aging and environmental challenges of nutrient availability or toxic stresses . Failure to maintain mitochondrial homeostasis under these conditions contributes to the development of numerous age-associated and metabolic disorders ( Nunnari and Suomalainen , 2012 ) . Many of the adaptive responses made by mitochondria are manifested through the composition and activity of the ~1000 mitochondrial proteins , which are regulated at multiple points from transcription to degradation ( Bohovych et al . , 2015; Fox , 2012 ) . In the area of mitochondrial protein degradation , cells are equipped with numerous proteolytic pathways that function to maintain mitochondrial homeostasis ( Anand et al . , 2013 ) . Some of these pathways target individual mitochondrial proteins , whereas others act more broadly and remove large portions of the organelle . Examples of the former include AAA+-ATPase proteases within all mitochondrial compartments that promote the degradation of unassembled protein complex subunits , oxidatively damaged proteins , and tail-anchored membrane proteins mistargeted to the mitochondrial outer membrane ( Chen et al . , 2014; Gerdes et al . , 2012; Okreglak and Walter , 2014 ) . There are also proteasome-dependent pathways that promote the turnover of proteins in the mitochondrial outer membrane and within internal mitochondrial compartments ( Taylor and Rutter , 2011 ) . The best-studied pathway for destruction of large portions of mitochondria is through selective autophagy , or mitophagy ( Youle and Narendra , 2011 ) . In metazoa , the PINK1/Parkin pathway serves as a paradigm of mitophagy ( Pickrell and Youle , 2015 ) . PINK1 is a mitochondrial-localized kinase that is rapidly turned over in healthy mitochondria , but becomes stabilized on the surface of mitochondria with a reduced mitochondrial membrane potential , or △Ψ ( Narendra et al . , 2010 ) . Upon stabilization , PINK1 recruits the E3 ligase Parkin , which ubiquitinates a number of mitochondrial surface proteins to promote fragmentation and destruction of dysfunctional mitochondria in the lysosome by autophagy ( Narendra et al . , 2008 ) . Importantly , all the general machinery of autophagy must be active for mitophagy to occur . The apparent specificity comes in tagging mitochondria for autophagy . The budding yeast , S . cerevisiae , contains no obvious sequence homologs of PINK1 and Parkin , but there are multiple reports of mitophagy occurring in yeast when cells are metabolically challenged or when △Ψ is compromised by genetic mutation or chemical treatment ( Kanki et al . , 2011 ) . The most clearly defined version of mitophagy in budding yeast requires a mitochondrial outer membrane protein , Atg32 , to link fragments of mitochondria to the autophagy machinery for degradation ( Kanki et al . , 2009b; Okamoto et al . , 2009 ) . Unlike the PINK1/Parkin pathway , the Atg32 pathway does not appear to respond to loss of △Ψ and does not utilize ubiquitin tagging for mitochondrial destruction . Instead , expression of Atg32 on the surface of mitochondria promotes turnover of mitochondria , primarily during times of regulated metabolic remodeling such as during the transition from robust growth to a starved or stationary phase-like state . In this regard , Atg32-dependent mitophagy may be a functional homologue to the tagging of mitochondria by the NIX protein during reticulocyte maturation ( Novak et al . , 2010 ) . NIX is proposed to serve as a receptor on mitochondria to mediate mitophagy in this developmental process . More recently , a second mode of eliminating large portions of mitochondria was discovered in mammalian cells . It involves the formation of mitochondrial-derived vesicles ( MDVs ) , which deliver oxidized mitochondrial proteins to the lysosome for degradation ( Soubannier et al . , 2012a ) . PINK1/Parkin are also needed for MDV formation and degradation , but interestingly the core autophagy machinery is not required ( McLelland et al . , 2014 ) . This has led to the speculation that MDVs may provide an early wave of mitochondrial protein quality control , which if unsuccessful in re-achieving cellular homeostasis , is followed by 'full blown' mitophagy ( Sugiura et al . , 2014 ) . A key area in which many of these pathways function to promote mitochondrial homeostasis is during the aging process . There has long been a strong association between cellular aging and mitochondrial dysfunction ( Gonzalez-Freire et al . , 2015 ) . However , there is an incomplete understanding of how these degradation systems contribute to promoting mitochondrial health in aging organisms . While mutations in PINK1/Parkin lead to early onset of neurodegeneration ( Kitada et al . , 1998; Valente et al . , 2004 ) , it is not yet clear what events trigger this dependence on the PINK1/Parkin pathway during the aging process . In fact , much of what is known about mitophagy and MDVs is based primarily upon acute stresses placed on mitochondria by chemical challenges or severe genetic perturbations ( e . g . treatments that cause global ROS damage or complete loss of △Ψ ) ( Sugiura et al . , 2014 ) . Replicative aging of S . cerevisiae , defined by the number of times an individual yeast cell produces a daughter cell , functions as a model system for understanding fundamental aspects of cellular aging , including age-dependent mitochondrial dysfunction ( Breitenbach et al . , 2014; Steinkraus et al . , 2008; Wasko and Kaeberlein , 2014 ) . We and others have shown that aged yeast cells have reduced △Ψ , decreased mitochondrial import , and altered mitochondrial morphology that transitions from a typical tubular shape to one that is highly fragmented and amorphous with increasing age ( Hughes and Gottschling , 2012; Lam et al . , 2011; McFaline-Figueroa et al . , 2011; Scheckhuber et al . , 2007 ) . This change in mitochondrial structure and function is driven at least in part by changes in the pH of the yeast lysosome-like vacuole that occur earlier in the yeast cell’s age ( Hughes and Gottschling , 2012 ) , as well as asymmetric partitioning of healthy mitochondria to daughter cells ( McFaline-Figueroa et al . , 2011 ) . In the present study , we followed the fate of these age-associated changes in mitochondria to further our understanding of how cells maintain mitochondrial homeostasis during the course of aging . As described here , our studies led to the discovery of a new mitochondrial protein degradation pathway that selectively remodels the mitochondrial proteome to maintain optimum mitochondrial function . As described above , autophagy systems play an important role in maintaining mitochondrial health across a variety of organisms . Therefore we tested whether autophagy played a role in maintaining mitochondrial function in aged yeast . Specifically , we examined aged cells for the presence of a mitochondrial outer membrane protein , Tom70-GFP , within vacuoles . No GFP signal was detected in the vacuole of wild-type cells across a wide range of ages ( data not shown ) . However , vacuolar proteases rapidly degrade proteins taken up by autophagy and thus frequently prevent protein detection in the vacuole . Therefore , we re-assessed the presence of vacuolar Tom70-GFP in aged cells lacking PEP4 , a gene encoding a master vacuolar protease that is required for turnover of autophagosomes in the vacuole ( Klionsky et al . , 1992; Takeshige et al . , 1992 ) . In young cells , which had functional mitochondria , no Tom70-GFP was detected in vacuoles by fluorescence microscopy ( Figure 1A ) . However , vacuoles in 70% of aged cells contained Tom70-GFP ( marked by white arrows in Figure 1A ) . Tom70-GFP in the vacuole appeared as a body exhibiting constant Brownian motion within the boundary of the vacuole membrane ( data not shown ) , which is characteristic of autophagosomes that cannot be broken down in pep4△ cells ( Takeshige et al . , 1992 ) . 10 . 7554/eLife . 13943 . 003Figure 1 . Mitochondrial proteins are degraded by autophagy in aged cells . ( A ) Tom70-GFP is degraded in the vacuole by autophagy in middle-aged cells . Wild-type ( WT ) and the indicated mutant cells expressing Tom70-GFP and the vacuole marker Vph1-mCherry were aged and visualized by fluorescence microscopy . Images depict wild-type cells , and the presence of Tom70-GFP in the vacuole ( white arrow ) of young , middle-aged , and old cells was scored for each strain . All strains including wild type are PEP4-deficient ( pep4△ ) . N = 30 . In all figures , young cells have undergone 0–3 divisions , middle-aged cells 7–12 divisions , and old cells >17 divisions . Divisions are scored by counting bud scars visualized with calcolfuor ( Calc ) . ( B ) Representative images of old ATG5- ( atg5△ ) and DNM1-deficient ( dnm1△ ) cells from ( A ) with fragmented vacuole morphology . DOI: http://dx . doi . org/10 . 7554/eLife . 13943 . 003 To test if the appearance of Tom70-GFP within the vacuole was autophagy-dependent , Tom70-GFP localization was examined in aged cells lacking ATG5 , a gene essential for all forms of autophagy ( Feng et al . , 2014 ) . Aged atg5△ pep4△ cells contained no Tom70-GFP in the vacuole , which indicated that Tom70-GFP normally entered the vacuole via autophagy ( Figure 1A ) . In addition to the core autophagy machinery , some forms of mitochondrial autophagy require the mitochondrial fission machinery ( Muller et al . , 2015 ) . Delivery of Tom70-GFP to the vacuole in aged cells also required the mitochondrial fission machinery , as this process was inhibited in cells lacking DNM1 , which encodes a conserved GTPase required for mitochondrial fission ( Figure 1A ) ( Bleazard et al . , 1999 ) . Lastly , starvation-induced mitophagy in yeast relies on Atg32 , a receptor on the mitochondrial surface that links mitochondria to the autophagy machinery for degradation ( Kanki et al . , 2009b; Okamoto et al . , 2009 ) . However , Tom70 degradation in aged cells is independent of Atg32; aged cells lacking ATG32 delivered Tom70-GFP to the vacuole at the same level as in wild-type cells ( Figure 1A ) . These results suggest we identified an autophagy-dependent pathway for degrading Tom70 in mitochondria from aging yeast cells that requires mitochondrial fission and is distinct from the previously characterized Atg32-dependent pathway . While examining the kinetics of Tom70-GFP vacuolar delivery in aged cells , we noticed that the presence of Tom70 in the vacuole peaked in middle-aged cells , and then declined in old cells ( Figure 1A ) . This decline coincided with a highly fragmented vacuole , depicted in the third panel of Figure 1A . For reasons that are unclear , vacuoles increase in size with age and then can become severely fragmented in very old cells ( Figure 1A ) ( Lee et al . , 2012 ) . Interestingly , in all cells with a severely fragmented vacuole , Tom70-GFP appears in small vesicle-like structures in the cytoplasm ( Figure 1A and Video 1 ) . These structures are not cytosolically-localized autophagosomes , because they are still present in old cells lacking ATG5 ( Figure 1B and Video 2 ) . However , their formation does require the mitochondrial fission GTPase DNM1 ( Figure 1B and Video 3 ) . Thus , although mitochondrial protein destruction is activated in middle-aged cells , this autophagy-dependent degradation appears compromised in very old yeast , leading to production of Dnm1-dependent vesicle-like structures . 10 . 7554/eLife . 13943 . 004Video 1 . 3D reconstruction of mitochondria and vacuoles in old wild-type cells . A representative 3D reconstruction of an old wild-type cell with the same characteristics as those depicted in Figure 1A showing small mitochondrial vesicle-like fragments ( green , marked with Tom70-GFP ) outside of the severely fragmented vacuole ( red , marked with Vph1-mCherry ) . To permit visualization of the vacuole lumen , the vacuole isosurface rendering becomes 60% transparent in the middle of the movie . Budscars ( blue , calcofluor ) at the beginning of the movie indicate the cell’s old age . DOI: http://dx . doi . org/10 . 7554/eLife . 13943 . 00410 . 7554/eLife . 13943 . 005Video 2 . 3D reconstruction of mitochondria and vacuoles in old atg5△ cells . A representative 3D reconstruction of an old atg5△ cell with the same characteristics as those depicted in Figure 1B showing small mitochondrial vesicle-like fragments ( green , marked with Tom70-GFP ) outside of the severely fragmented vacuole ( red , marked with Vph1-mCherry ) . To permit visualization of the vacuole lumen , the vacuole isosurface rendering becomes 60% transparent in the middle of the movie . Budscars ( blue , calcofluor ) at the beginning of the movie indicate the cell’s old age . DOI: http://dx . doi . org/10 . 7554/eLife . 13943 . 00510 . 7554/eLife . 13943 . 006Video 3 . 3D reconstruction of mitochondria and vacuoles in old dnm1△ cells . A representative 3D reconstruction of an old dnm1△ cell with the same characteristics as those depicted in Figure 1B showing the presence of mitochondria ( green , marked with Tom70-GFP ) , but the absence of small mitochondrial vesicle-like fragments outside of the severely fragmented vacuole ( red , marked with Vph1-mCherry ) . To permit visualization of the vacuole lumen , the vacuole isosurface rendering becomes 60% transparent in the middle of the movie . Budscars ( blue , calcofluor ) at the beginning of the movie indicate the cell’s old age . DOI: http://dx . doi . org/10 . 7554/eLife . 13943 . 006 We previously showed that mitochondrial dysfunction in aged cells is caused by disruption of a metabolic relationship between mitochondria and vacuoles ( Hughes and Gottschling , 2012 ) . Vacuoles are acidified by the Vacuolar-H+-ATPase ( V-ATPase ) ( Kane , 2006 ) , and the proton gradient generated by this protein complex is required for amino acid storage within the vacuole lumen ( Klionsky et al . , 1990 ) . Loss of vacuole acidity in aged cells causes mitochondrial dysfunction through an undefined mechanism that likely involves altered storage of cellular amino acids ( Hughes and Gottschling , 2012 ) . To test if loss of vacuole acidity triggers autophagy-dependent mitochondrial protein degradation , we took advantage of the fact that treatment of young cells with concanamycin A ( conc A ) , a specific inhibitor of the V-ATPase ( Drose et al . , 1993 ) , recapitulates age-associated changes in mitochondria ( Figure 2A ) ( Hughes and Gottschling , 2012 ) . Consistent with our previous findings , treatment with conc A caused an immediate loss of vacuolar acidity , followed by a decline in mitochondrial △Ψ within 30 min as measured by microscopy ( Figure 2B ) using the common mitochondrial membrane potential fluorescent dye DiOC6 ( Pringle et al . , 1989 ) , or by flow cytometry ( Figure 2—figure supplement 1 ) using DiOC6 and another mitochondrial membrane potential dye , TMRM ( Scaduto and Grotyohann , 1999 ) . Similar to our observations in aged cells , treatment of young cells with conc A caused accumulation of Tom70-GFP-containing bodies within the vacuole ( Figure 2C , white arrows ) . Because no Tom70-GFP was present in the vacuoles of untreated pep4△ cells , we conclude that loss of vacuolar acidity triggers activation of this pathway . Additionally , in contrast to old cells , it was not necessary to delete PEP4 to observe Tom70-GFP within the vacuole of conc A treated cells . This is likely because inhibition of the V-ATPase with conc A creates an environment within the lumen that prevents autophagosome breakdown ( Nakamura et al . , 1997 ) . The appearance of Tom70-GFP in the conc A treated vacuole occurred within three hours after mitochondrial depolarization , and was preceded by the formation of a single bright focus that contained Tom70-GFP , but appeared somewhat distinct from the rest of the mitochondria ( Figure 2C , white * ) . As in old cells , delivery of Tom70-GFP to the vacuole was dependent on both the autophagy ( Atg5 ) and mitochondrial fission machineries ( Dnm1 ) , but independent of Atg32 ( Figure 2C ) . Further supporting the role of autophagy in this process , delivery of Tom70-GFP to the vacuole also required the vacuole homotypic fusion protein Vam3 , which is essential for fusion of autophagosomes with the vacuole ( Figure 2C ) ( Darsow et al . , 1997 ) . 10 . 7554/eLife . 13943 . 007Figure 2 . Loss of vacuole function triggers mitochondrial protein degradation . ( A ) Schematic illustration showing that loss of vacuolar acidity ( 2 ) through aging or concanamycin A ( conc A ) -mediated inhibition of the Vacuolar H+-ATPase ( 1 ) leads to loss of mitochondrial function ( 3 ) through an unknown mechanism . ( B ) Loss of vacuolar acidity causes rapid mitochondrial depolarization . Wild-type cells expressing Tom70-mCherry were treated with concanamycin A for the indicated time ( hr ) and stained with DiOC6 as an indicator of △Ψ . N = 30 . ( C ) Loss of vacuolar acidity activates autophagy-dependent Tom70-GFP degradation . Wild-type ( WT ) and the indicated mutant cells expressing Tom70-GFP and Vph1-mCherry were treated with concanamycin A for the indicated time ( hr ) . The presence of Tom70-GFP in the vacuole ( white arrow ) was scored for each strain and time point . N = 50 . * indicates MDC . ( D ) Tom70-GFP was monitored for autophagy-dependent degradation using a GFP-cleavage assay in wild-type ( WT ) and the indicated mutant cells treated with concanamycin A ( ConcA ) for the indicated time ( hr ) . Whole-cell extracts from the treated cells were subjected to immunoblot analysis with anti-GFP antibody . The use of conc A as an inducer potentially limited the amount of GFP cleavage in the vacuole . Consequently , the exposure time of the free GFP immunoblot is 20 times longer than the exposure of the immunoblot with full length Tom70-GFP . DOI: http://dx . doi . org/10 . 7554/eLife . 13943 . 00710 . 7554/eLife . 13943 . 008Figure 2—figure supplement 1 . Concanamycin a treatment causes loss of mitochondrial membrane potential . Loss of vacuolar acidity causes mitochondrial depolarization . Wild-type cells were treated with concanamycin A for 4 hr , stained with mitochondrial membrane potential fluorescent dyes DiOC6 ( A ) or TMRM ( B ) , and analyzed by flow cytometry . FACS profiles and bar graphs showing median fluorescence intensity are shown for each stain . N > 10 , 000 cells for each . DOI: http://dx . doi . org/10 . 7554/eLife . 13943 . 008 In addition to the microscopy-based assays , turnover of GFP-tagged proteins by autophagy can also be detected by monitoring release of free GFP from the epitope-tagged protein in the vacuole using immunoblot analysis ( Kanki et al . , 2009a ) . Once released by vacuolar proteases , GFP is stable in the vacuole lumen . The release of GFP from Tom70-GFP was monitored by immunoblot analysis in cells treated with conc A , and the results mirrored what was observed in the microscopy-based assay ( Figure 2D ) . Treatment of cells for six hours caused release of GFP from full length Tom70-GFP ( Figure 2D ) . GFP cleavage required the vacuolar protease PEP4 , the autophagy and mitochondrial fission machineries , and was independent of ATG32 ( Figure 2D ) . Collectively , these results suggest that disruption of the mitochondrial-vacuole relationship by direct V-ATPase inhibition activates autophagy-dependent mitochondrial protein degradation . The timing of Tom70-GFP destruction shortly after mitochondrial depolarization raised the possibility that loss of mitochondrial membrane potential may activate this pathway downstream of changes in vacuole acidity . To test this , we analyzed Tom70-GFP vacuolar delivery in cells treated with two common mitochondrial depolarizing agents , the electron transport chain inhibitor antimycin A , and the proton ionophore carbonyl cyanide-4- ( trifluoromethoxy ) phenylhydrazone ( FCCP ) . Although treatment of cells with these inhibitors caused mitochondrial depolarization equal to or greater than conc A ( Figure 3A and B ) , we did not observe Tom70-GFP delivery to the vacuole at any point during treatment with these compounds ( Figure 3C ) . Instead , mitochondria coalesced in FCCP treated cells , and were largely unchanged with antimycin A treatment ( Figure 3C ) . These results suggest that a decrease in mitochondrial membrane potential is not sufficient to trigger Tom70-GFP degradation . 10 . 7554/eLife . 13943 . 009Figure 3 . Mitochondrial protein degradation is not triggered by loss of mitochondrial membrane potential or oxidative stress . Antimycin A and FCCP cause mitochondrial depolarization . Wild-type cells were treated with Antimycin A ( A ) or FCCP ( B ) for 4 hr , stained with the mitochondrial membrane potential fluorescent dye DiOC6 , and analyzed by flow cytometry . FACS profiles and bar graphs showing median fluorescence intensity are shown for each treatment . N > 10 , 000 cells for each . ( C ) Loss of mitochondrial membrane potential or oxidative stress does not activate autophagy-dependent Tom70-GFP degradation . Wild-type cells expressing Tom70-GFP were treated with the indicated compound for 4 hr . The presence of Tom70-GFP in the vacuole was scored for each strain and time point . N = 50 . Representative images showing mitochondrial aggregation and fragmentation in FCCP and hydrogen peroxide ( H2O2 ) treated cells are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 13943 . 009 In addition to mitochondrial depolarization , loss of vacuolar acidity also causes oxidative stress in cells through an undefined mechanism ( Milgrom et al . , 2007 ) . Therefore , we also tested whether oxidants trigger activation of this pathway . Treatment of cells with hydrogen peroxide caused mitochondrial fragmentation , but like FCCP and antimycin A , did not lead to the appearance of Tom70-GFP within the vacuole ( Figure 3C ) . This result suggests that oxidative stress is also not sufficient to activate this pathway , and that loss of vacuolar acidity triggers mitochondrial protein degradation through an unknown mechanism . During the analysis of mitochondrial protein degradation in young cells , we noticed a distinct structure containing Tom70-GFP that formed prior to the appearance of Tom70-GFP foci within the vacuole ( Figure 2C , white * ) . We refer to this structure as the mitochondrial-derived compartment ( MDC ) . MDCs appeared attached to mitochondria as a single focus with greater signal intensity than the rest of the organelle . These structures were also present in middle-aged cells undergoing mitochondrial degradation ( Figure 4A ) , and the timing of their appearance suggested that MDCs are an intermediate in this process . 10 . 7554/eLife . 13943 . 010Figure 4 . The mitochondrial-Derived compartment ( MDC ) is an intermediate step in mitochondrial protein degradation . ( A ) Aging induces MDC formation . Middle-aged cells expressing Tom70-GFP and Vph1-mCherry were scored by fluorescence microscopy for the presence of mitochondrial-derived compartment ( MDC ) structures ( white arrow ) . N = 30 . ( B ) Loss of vacuolar acidity triggers MDC formation . Wild-type ( WT ) and the indicated mutant cells expressing Tom70-GFP and Vph1-mCherry were treated with concanamycin A for the indicated time ( hr ) . The presence of MDCs ( white arrow ) was scored for each strain and time point . N = 50 . ( C ) Mitochondrial fission , but not autophagy , is required for MDC release . Representative images of MDCs ( white arrow ) in DNM1- ( dnm1△ ) and ATG5-deficient ( atg5△ ) cells from ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13943 . 01010 . 7554/eLife . 13943 . 011Figure 4—figure supplement 1 . Further support for the role of fission proteins in MDC release from the mitochondria . ( A ) The fission protein Fis1 is required for MDC release from mitochondria . fis1△ cells expressing Tom70-GFP were treated with concanamycin A for the indicated time ( h ) . The presence of Tom70-GFP in the vacuole was scored . N = 50 . Representative images of untreated and treated cells are shown . White arrow indicates MDC . 82% of 4 hr treated cells had an MDC . ( B ) Dnm1-GFP foci localize at or around the MDC . Representative maximum intensity projection images showing colocalization of Dnm1-GFP foci with MDCs ( marked with Tom70-mCherry ) in wild-type cells treated with concanamycin A for the indicated time ( h ) . White arrow indicates MDC with nearby Dnm1-GFP foci . 88% of cells exhibit this colocalization phenotype . N = 50 . DOI: http://dx . doi . org/10 . 7554/eLife . 13943 . 01110 . 7554/eLife . 13943 . 012Figure 4—figure supplement 2 . Further support for the role of autophagy in Tom70-GFP degradation . MDCs localize near the pre-autophagosomal structure ( PAS ) prior to vacuole entry . Representative maximum intensity projection images showing colocalization of PAS marker GFP-Atg8 with MDCs ( marked with Tom70-mCherry ) in wild-type cells treated with concanamycin A for the indicated time ( hr ) . White arrow indicates an MDC in close proximity to PAS on the vacuole surface . 68% of cells exhibit this colocalization phenotype . N = 50 . DOI: http://dx . doi . org/10 . 7554/eLife . 13943 . 012 To further analyze the kinetics of MDC formation during the degradation process , MDCs were monitored during a timecourse of conc A treatment . In wild-type cells , MDCs appeared shortly after mitochondrial depolarization , and were present in a large percentage of cells within two hours after conc A addition ( Figure 4B ) . Cells generally contained only a single MDC , and it often localized near the vacuolar membrane prior to the appearance of Tom70-GFP inside the vacuole lumen ( Figure 4B , white arrows ) . MDCs formed independently of both autophagy and mitochondrial fission ( Figure 4B ) . However , although fission-defective dnm1∆ cells formed MDCs normally at two hours after conc A addition , at later timepoints , MDCs grew very large in these cells and were not released from mitochondria to the vacuole ( Figure 4C , white arrows ) . This suggests that mitochondrial membrane fission activity is required for release of MDCs for autophagic degradation . In support of this conclusion , Fis1 , an essential component of the mitochondrial fission machinery that recruits Dnm1 to mitochondrial membrane ( Mozdy et al . , 2000 ) , is also required for release of MDCs from mitochondria ( Figure 4—figure supplement 1A ) . Moreover , foci of Dnm1-GFP , representing sites of mitochondrial fission , localize near MDCs in 88% of cells ( Figure 4—figure supplement 1B ) . Collectively , these data suggest that the mitochondrial fission machinery is required for MDC release . Like the fission mutants , autophagy-deficient mutants ( atg5△ ) formed normal MDCs after two hours of conc A treatment ( Figure 4C , white arrows ) , suggesting that autophagy is not required for MDC formation . However , at later timepoints , Tom70-GFP did not enter the vacuole in these mutants , and MDCs were less apparent . Instead , Tom70-GFP was present in small vesicle-like structures ( Figure 4C ) that looked similar to those that formed in very old cells ( Figure 1A ) . Further supporting the role of autophagy in MDC degradation , MDCs in 68% of cells localized at or near the preautophagosomal structure ( PAS ) at the vacuole membrane , which is the site of autophagosome formation and target engulfment in yeast ( Figure 4—figure supplement 2 ) ( Feng et al . , 2014 ) . Taken together , these results suggest that the MDC is an intermediate step on the way to Tom70-GFP degradation in the vacuole by autophagy , and that its release from the mitochondrial surface relies on mitochondrial fission . At this point , we also cannot rule out other alternative fates for the MDC , such as delivery to other cellular compartments . During characterization of the MDC pathway in young cells , we found that unlike Tom70 , the inner mitochondrial membrane protein Tim50-GFP was excluded from MDCs and not targeted for vacuolar degradation by this pathway ( Figure 5A and B ) . Similar specificity was apparent in aged cells ( Figure 5C ) , suggesting that MDCs are cargo-selective . 10 . 7554/eLife . 13943 . 013Figure 5 . Select mitochondrial proteins are incorporated into MDCs . ( A ) The inner mitochondrial membrane protein Tim50 is excluded from MDC-dependent degradation . Wild-type cells expressing Tom70-mCherry and Tim50-GFP were treated with concanamycin A for the indicated time and representative images showing Tim50 exclusion from the MDC ( 2 hr , white arrow ) and vacuole ( 4 hr , white arrow ) are shown . 100% of cells show this phenotype of Tim50 exclusion . N = 50 . ( B ) Tom70-GFP and Tim50-GFP were monitored for autophagy-dependent degradation using a GFP-cleavage assay in wild-type ( WT ) cells expressing either Tom70-GFP or Tim50-GFP treated with concanamycin A ( ConcA ) for the indicated time ( hr ) . Whole-cell extracts from the treated cells were subjected to immunoblot analysis with anti-GFP antibody . As in Figure 2D , the use of conc A as an inducer limited the amount of GFP cleavage in the vacuole . Consequently , the exposure time of the free GFP immunoblot is ~20 times longer than the exposure of the immunoblot with full-length GFP-tagged proteins . ( C ) Tim50 is excluded from MDCs in middle-aged cells . Wild-type cells expressing Tom70-mCherry and Tim50-GFP were aged and representative images showing Tim50 exclusion from the MDC ( white arrow ) are shown . 100% of cells show this phenotype of Tim50 exclusion . N = 30 . ( D ) Mitochondrial outer membrane proteins and inner membrane carrier proteins localize to MDCs . Wild-type cells expressing Tom70-mCherry and the indicated C-terminal GFP fusion proteins were treated with concanamycin A for 2 hr and protein inclusion in MDCs ( white arrows ) was assessed . GFP-tagged marker proteins represent mitochondrial outer membrane ( OM ) , inner membrane carrier proteins ( IM ) , and matrix proteins ( M ) . 100% of cells show the phenotypes in the representative images . N = 50 . ( E ) Tom20 is excluded from MDCs . Wild-type cells expressing Tom70-mCherry and Tom20-GFP were treated with concanamycin A for the indicated time and representative images showing Tom20 exclusion from the MDC ( 2 hr , white arrow ) and vacuole ( 4 hr , white arrow ) are shown . 100% of cells show this phenotype of Tom20 exclusion . N = 50 . DOI: http://dx . doi . org/10 . 7554/eLife . 13943 . 01310 . 7554/eLife . 13943 . 014Figure 5—figure supplement 1 . MDCs do not contain other major organelles . Marker proteins of other major organelles do not localize to MDCs . Wild-type cells expressing Tom70-mCherry and the indicated GFP-tagged organelle marker protein were treated with concanamycin A for 2 hr and representative images showing GFP exclusion from the MDC ( white arrow ) are shown . Organelle marker proteins are Sec63 for endoplasmic reticulum ( ER ) , Erg6 for lipid droplets ( LD ) , Pex11 for peroxisomes ( perox ) , Vrg4 for early golgi , and Sec7 for late golgi . DOI: http://dx . doi . org/10 . 7554/eLife . 13943 . 014 To understand the extent of cargo-selectivity , a microscopy-based screen was performed to identify the proteins that are incorporated into MDCs . For the screen , we created a collection of yeast strains coexpressing Tom70-mCherry and any protein of interest fused to GFP by crossing a strain containing Tom70-mCherry to the yeast GFP strain collection ( Huh et al . , 2003 ) . Strains containing GFP-tagged mitochondrial-localized proteins were screened after conc A treatment in a similar manner to the Tim50-GFP/Tom70-mCherry strain in Figure 5A . In total , 469 mitochondrial proteins were examined ( Supplementary file 1 ) and 304 of the proteins were detectable by microscopy . Of those , 26 localized to MDCs and were degraded in the vacuole ( Table 1 ) , indicating that this pathway exhibits narrow substrate specificity . Although the MDC incorporated a relatively small number of the total proteins examined , it is likely specific for mitochondrial proteins , because markers of other major organelles were not co-localized with MDCs ( Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 13943 . 015Table 1 . MDC Substrates . DOI: http://dx . doi . org/10 . 7554/eLife . 13943 . 015GeneORFMitochondrial LocalizationaNCA2YPR155COuter membraneALO1YML086COuter membraneUBP16YPL072WOuter membraneMFB1YDR219COuter membraneTCD2YKL027WOuter membraneTCD1YHR003COuter membraneMCY1YGR012wOuter membraneMSP1YGR028WOuter membraneTOM70YNL121COuter membraneTOM71YHR117WOuter membraneYPR098COuter membraneOM14YBR230COuter membraneSEN15YMR059WOuter membranePTH2YBL057COuter membraneMCP1YOR228COuter membraneSCM4YGR049WOuter membraneMIR1bYJR077CInner membraneCTP1bYBR291CInner membraneDIC1bYLR348CInner membraneOAC1bYKL120WInner membraneMTM1bYGR257CInner membraneYMC2bYBR104WInner membraneYHM2bYMR241WInner membraneCOX7YMR256CInner membraneYML007C-AUnknownECM19YLR390WUnknowna , Localization information obtained from SGD ( www . yeastgenome . org ) b , Mitochondrial carrier protein family member Mitochondrial proteins broadly localize to four distinct mitochondrial subdomains: the outer membrane , inner membrane , inner membrane space , and matrix ( Fox , 2012 ) . There was subdomain specificity in mitochondrial proteins that were targeted to the MDC . Representative examples of substrates and non-substrates from various mitochondrial subcompartments are shown in Figure 5D . Nearly all integral- and peripherally-associated mitochondrial outer membrane proteins , as well as a subclass of mitochondrial inner membrane proteins belonging to the mitochondrial carrier family localized to the MDC . In contrast , mitochondrial matrix proteins and inner membrane proteins that were not part of the carrier family were excluded . Because they were below the detection limit of the assay , it was not possible to ascertain whether mitochondrial inner membrane space proteins or β-barrel proteins of the outer membrane were present in MDCs . The MDC substrate specificity followed classic lines of mitochondrial import ( Schmidt et al . , 2010 ) . Many matrix and inner membrane-localized mitochondrial proteins contain a mitochondrial targeting sequence that is removed upon import into the mitochondria . After being translated in the cytoplasm , this class of presequence-containing proteins generally binds to the mitochondrial surface receptor Tom20 , and then are routed through the TOM complex of the outer membrane and the Tim23 complex of the inner membrane to their final destination . Proteins that are normally imported via this pathway were excluded from MDCs ( Figure 5D and Supplementary file 1 ) . In contrast to presequence-containing proteins , integral inner membrane carrier proteins of the metabolite carrier family and alpha-helical outer membrane proteins lack a cleavable mitochondrial presequence ( Harbauer et al . , 2014 ) . After translation , these proteins bind to the Tom70 mitochondrial surface receptor , and then are targeted to the inner membrane through the Tim22 complex or to the outer membrane through the Mim1 pathway ( Becker et al . , 2011 ) . All detectable proteins that rely on Tom70-associated import pathways were targeted to MDCs , including Tom70 itself ( Table 1 ) . In further support of the distinction between the Tom20 and Tom70 import pathways , Tom20 was excluded from MDCs ( Figure 5E ) , even though it is an integral outer membrane protein ( Schneider et al . , 1991 ) . Even Cox7 , which is not a carrier protein , but lacks a cleavable presequence and thus likely utilizes Tom70 for mitochondrial import was targeted to MDCs ( Table 1 ) ( Calder and McEwen , 1990; Mihara and Blobel , 1980 ) . Thus , the MDC pathway only degraded a subset of mitochondrial proteins , and was specific for non-presequence containing proteins that rely on the Tom70 surface receptor for import . Because MDC substrates were specifically confined to Tom70-dependent import pathways , we wondered if the MDC was an aggregate of newly synthesized proteins that could not be imported into dysfunctional mitochondria . If this were true , proteins incorporated into MDCs would be newly synthesized , not preexisting mitochondrial proteins . To test this , old and newly synthesized versions of proteins were monitored for their incorporation into the same MDC using the recombination induced tag exchange ( RITE ) system ( Verzijlbergen et al . , 2010 ) . As illustrated in Figure 6A , fusion of the RITE cassette to a protein of interest allows rapid and permanent switching of epitope tags through an estradiol-inducible Cre/loxP based mechanism . With this system , old protein synthesized before the switch is GFP tagged , while all subsequently synthesized protein is mRFP tagged . The RITE cassette was fused to Tom70 and Oac1 , outer and inner membrane MDC substrates , respectively . Cells expressing these proteins were treated with conc A to induced MDC formation , along with estradiol to switch tags on the proteins from GFP to mRFP . Both mRFP and GFP versions of each protein were found in the MDC ( Figure 6B ) , suggesting that MDCs incorporate old , preexisting proteins , and are not specific to unimported mitochondrial substrates . Because the mRFP-tagged protein targeted to MDCs may be synthesized prior to their formation , we cannot determine with this assay whether newly synthesized mitochondrial proteins are also targeted to this compartment . However , further supporting the idea that MDCs incorporate proteins that preexist within mitochondria , MDCs still formed in the presence of cycloheximide , which effectively prohibited synthesis of the newly synthesized version of both RITE-tagged Tom70 and Oac1 ( Figure 6C ) . These results suggest that preexisting mitochondrial proteins are segregated into MDCs prior to degradation by autophagy . 10 . 7554/eLife . 13943 . 016Figure 6 . MDCs sequester preexisting mitochondrial proteins . ( A ) Schematic of the recombination induced tag exchange ( RITE ) system . In untreated cells , the RITE-tagged ORF is expressed with a C-terminal HA-GFP fusion ( old protein ) . Treatment with estradiol induces Cre-EBD dependent recombination between LoxP sites creating a new C-terminal T7-mRFP fusion ( new protein ) . ( B ) Preexisting protein is incorporated into MDCs . Cells expressing either RITE-tagged Tom70 or Oac1 were treated with estradiol and concanamycin A simultaneously for 3 hr to induce epitope tag exchange and MDC formation . Cells were visualized with fluorescence microscopy for the presence of preexisting ( old ) or newly synthesized ( new ) protein in the MDC ( white arrows ) . 100% of the cells show the represented phenotype . N= 50 . ( C ) MDCs ( white arrows ) form in the absence of new protein synthesis . Cells expressing RITE-tagged Tom70 or Oac1 were treated as in ( B ) with the addition of cycloheximide to inhibit synthesis of new T7-RFP tagged proteins . 100% of the cells show the represented phenotype . N= 50 . DOI: http://dx . doi . org/10 . 7554/eLife . 13943 . 016 Because the proteins degraded by the MDC pathway all rely on the mitochondrial import receptor Tom70 for import into mitochondria , we wondered whether Tom70 is required for MDC formation beyond its role in mitochondrial import ( Sollner et al . , 1990 ) . To test this , we monitored the appearance of the MDC substrate Cox7-GFP ( Table 1 ) in the vacuole of conc A treated cells . Cox7-GFP acts exactly the same as Tom70-GFP upon concanamycin A treatment , entering MDCs at 2 hr and getting delivered to the vacuole after four hours of treatment ( Figure 7—figure supplement 1A–B ) . Despite the fact that Tom70 [and its paralog Tom71 ( Schlossmann et al . , 1996 ) ] are likely required for import of Cox7 into the mitochondria ( Schlossmann et al . , 1996; Schmidt et al . , 2010 ) , a large amount of Cox7 still localized to mitochondria in both tom70△ and tom70△ tom71△ cells ( Figure 7A ) . This import likely results from compensation by import receptors Tom20 and Tom22 ( Lithgow et al . , 1994; Ramage et al . , 1993; Steger et al . , 1990 ) . Nevertheless , deletion of TOM70 alone severely impaired conc A induced vacuolar delivery of Cox7 ( Figure 7B ) , and loss of both Tom70 and 71 provided complete inhibition ( Figure 7B ) . 10 . 7554/eLife . 13943 . 017Figure 7 . The mitochondrial import receptors Tom70 and Tom71 are required for MDC formation . ( A ) The MDC substrate Cox7 localizes to mitochondria lacking TOM70 and 71 . Wild-type ( WT ) and the indicated mutant strains expressing the inner membrane protein Cox7-GFP were visualized by fluorescence microscopy . ( B ) Tom70 and 71 are required for vacuole delivery of Cox7-GFP . Quantification of Cox7-GFP in vacuoles of wild-type ( WT ) and the indicated mutant strains treated with concanamycin A ( conc A ) for 4 hr . Data represents percentage of cells with Cox7-GFP in the vacuole . Error bars represent standard deviation of 3 replicates . N = 100 for each replicate . ( C ) Tom70 and 71 are required for MDC formation . Quantification of Cox7-GFP containing MDCs in fission deficient strains lacking the indicated genes treated with concanamycin A ( conc A ) for 2 hr . Data represents percentage of cells with Cox7-GFP in MDCs . Error bars represent standard deviation of 3 replicates . N = 100 per replicate . ( D ) Failure to form MDC exacerbates membrane potential loss . Median fluorescence intensity of mitochondrial dye TMRM in wild-type ( WT ) and the indicated mutant strains treated with and without conc A for 4 hr as measured by flow cytometry . Median fluorescence intensity of a population of cells is presented as a percentage of the WT untreated sample ( which is set at 100 ) . N > 20 , 000 cells . Error bars represent standard deviation of three independent replicates . p<0 . 01 , *p<0 . 001 , multiple comparison one-way anova test . ( E ) Model of the MDC pathway . Loss of vacuole function caused by aging or other mechanisms produces an unknown signal ( Step 1 ) that triggers MDC formation ( Step 2 ) . Select mitochondrial inner and outer membrane proteins are incorporated into MDCs , which are subsequently released from mitochondria by the fission GTPase Dnm1 ( Step 3 ) . MDCs are then engulfed by autophagosomes ( Step 4 ) , and delivered to the vacuole for degradation by autophagy ( Step 5 ) . It is currently not clear if the mitochondrial inner membrane ( dashed line ) is incorporated into MDCs . DOI: http://dx . doi . org/10 . 7554/eLife . 13943 . 01710 . 7554/eLife . 13943 . 018Figure 7—figure supplement 1 . Further support for the role of Tom70 and 71 in MDC formation . ( A ) Cox7-GFP functions as an MDC substrate . Wild-type cells expressing Tom70-mCherry and Cox7-GFP were treated with concanamycin A for 2 hr and representative images showing Cox7 and Tom70 in an MDC ( white arrow ) are shown . ( B ) Cox7-GFP gets delivered to the vacuole upon concanamycin A treatment . Wild-type cells expressing Tom70-mCherry and Cox7-GFP were treated with concanamycin A for the indicated time and representative images showing Cox7 and Tom70 within autophagosomes in the vacuole ( white arrow ) are shown . ( C ) Representative FACS profiles of wild-type ( WT ) and the indicated mutant strains grown in the absence or presence of concanamycin A for 4 hr . The median fluorescence intensity of these plots ( plus two others not shown here ) is presented in Figure 7D . ( D ) An alternative representation of the data from ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13943 . 018 To determine at which step in MDC-mediated degradation Tom70 and Tom71 function , we quantified the amount of Cox7-GFP present in MDCs from a conc A treated fission deficient strain ( dnm1∆ ) . Loss of TOM70 alone severely blocked MDC formation , and the loss of both TOM70 and TOM71 was completely inhibitory ( Figure 7C ) . Collectively , these results suggest that in addition to their function as mitochondrial import receptors , Tom70 and Tom71 are also required for formation of MDCs and the destruction of mitochondrial proteins residing within them . Because MDC formation coincided with depletion of the mitochondrial membrane potential ( Hughes and Gottschling , 2012 ) , we wondered whether MDC formation could protect mitochondria against further membrane potential loss . To test this , we monitored mitochondrial membrane potential by flow cytometry in conc A treated cells using TMRM . As expected , wild-type cells showed a reduction in TMRM staining after conc A treatment ( Figure 7D and Figure 7—figure supplement 1C–D ) . In the absence of conc A , cells lacking TOM70 alone or both TOM70 and TOM71 had a similar membrane potential as wild-type cells . However , upon treatment with conc A , these cells showed a greater reduction in TMRM staining ( Figure 7D and Figure 7—figure supplement 1C–D ) . At this time , we cannot rule out the possibility that further depletion of membrane potential in these strains results from a function of Tom70/71 unrelated to MDC formation . However , because these mutant strains cannot form MDCs under these conditions , it is possible that the sequestration of mitochondrial membrane proteins by MDCs may protect against mitochondrial depolarization caused by changes in vacuolar function . Using yeast as a model , we have identified a new form of autophagy-dependent mitochondrial protein degradation that is activated in aged cells undergoing vacuole-induced mitochondrial dysfunction . This pathway specifically destroys a subset of the mitochondrial membrane proteome through a series of steps outlined in Figure 7E . First , loss of acidity in the vacuole of aged or young cells triggers formation of a mitochondrial derived compartment , or MDC , through an unknown signal . Preexisting mitochondrial outer and inner membrane proteins that rely on Tom70 for import into the mitochondria are sequestered into the MDC through a mechanism that requires the import receptors Tom70/71 . After formation , the entire MDC or portions of it are released from the mitochondria by a process that requires the mitochondrial fission machinery , and subsequently delivered to the vacuole by autophagy . Our data suggests that MDC formation helps protect mitochondria from vacuole-induced stress , as cells that cannot form MDCs exhibit a greater loss of membrane potential in response to disruption of vacuolar acidity ( Figure 7D ) . A collapse of MDC-mediated autophagy occurs in very old yeast cells , leading to Dnm1-dependent formation of small Tom70-containing vesicle-like structures in the cytoplasm ( Figure 1A ) . Interestingly , deletion of DNM1 has been reported to extend lifespan in yeast ( Scheckhuber et al . , 2007 ) , raising the possibility that these Tom70-containing mitochondrial fragments may contribute to cell toxicity or death in aged yeast . The discovery of the MDC pathway raises new questions about the function of this system and how it relates to other known forms of mitochondrial protein degradation ( Anand et al . , 2013 ) . The MDC pathway appears mechanistically distinct from other autophagy-dependent systems such as PINK1/Parkin- and Atg32-dependent mitophagy ( Kanki et al . , 2011; Youle and Narendra , 2011 ) . Both of these pathways are thought to degrade whole portions of mitochondria , whereas MDCs selectively destroy a subset of mitochondrial membrane associated proteins . In this light , MDCs appear to have more in common with mitochondrial-derived vesicles , or MDVs , an autophagy-independent form of mitochondrial quality control thus far only identified in mammals ( Sugiura et al . , 2014 ) . Several different classes of MDVs have been described to date , and all of them selectively incorporate a subset of the mitochondrial proteome for delivery to the peroxisome or lysosome ( Neuspiel et al . , 2008; Soubannier et al . , 2012a ) . However , MDCs also differ from MDVs . Although the complete substrate specificity of MDVs is not well defined , it is clearly different than MDCs ( Neuspiel et al . , 2008; Soubannier et al . , 2012b ) . Additionally , unlike MDCs , release of MDVs from mitochondria does not require the mitochondrial fission machinery ( Neuspiel et al . , 2008 ) . Interestingly , recent studies have suggested that the PINK1/Parkin- and Atg32-mitophagy pathways might also act in substrate selective manners like MDCs and MDVs ( Abeliovich et al . , 2013; Chan et al . , 2011; Vincow et al . , 2013 ) . This raises the intriguing possibility that substrate selectivity may be a common theme of many different mitochondrial protein degradation systems . What is the function of the MDC pathway ? The answer to this question may lie in the nature of the substrates targeted for degradation through MDCs . By screening through the mitochondrial proteome , we found that membrane proteins lacking defined mitochondrial-targeting presequences are sequestered into MDCs . These proteins require the mitochondrial surface receptor Tom70 for import into the mitochondria ( Schmidt et al . , 2010 ) . The largest class of these Tom70-dependent import substrates is the mitochondrial carrier proteins , an evolutionarily conserved family of nutrient transporters that facilitate exchange of metabolites across the mitochondrial inner membrane ( Palmieri and Pierri , 2010 ) . Yeast contain ~35 members of this family in their inner membrane , which have affinity for many different metabolites including amino acids , nucleotides , metals , and TCA cycle intermediates . Interestingly , MDC formation is triggered by inhibiting vacuolar function . One important function of the vacuole is storage of nutrients ( Klionsky et al . , 1990 ) , and we previously showed that mitochondrial failure in response to loss of vacuole acidity results from impaired storage of nutrients such as amino acids within the vacuole lumen ( Hughes and Gottschling , 2012 ) . This raises the possibility that the purpose of the MDC pathway is to sequester mitochondrial nutrient transporters in response to cytoplasmic nutrient overload ( Wellen and Thompson , 2010 ) , perhaps to prevent unregulated nutrient influx into the mitochondria . Alternatively , the MDC pathway may degrade nutrient transporters to adjust mitochondrial metabolism towards a state preferable to survive vacuole impairment . In considering the MDC in nutrient transporter destruction , it is worth noting the similarity of MDCs to a recently characterized vacuole-derived compartment which functions in the turnover of a vacuole nutrient transporter ( Li et al . , 2015 ) . These compartments form from the vacuole membrane in response to cellular nutrient fluctuations , and have been shown to sequester a vacuolar lysine transporter away from the rest of the vacuolar membrane . They look morphologically similar to MDCs , but interestingly , are destroyed after release from the vacuole membrane by the multivesicular body pathway , not autophagy . Whether these vacuole-derived compartments are related in any way to MDCs is currently unclear . However , given the intimate metabolic relationship between the vacuole and mitochondria ( Rutter and Hughes , 2015 ) , and the recent identification of a physical tether ( vCLAMP ) between these organelles ( Elbaz-Alon et al . , 2014; Honscher et al . , 2014 ) , it will be important to explore the relationship between MDCs and vacuole-derived compartments ( Li et al . , 2015 ) , as well as to determine the role of the MDC in vacuole-mitochondria crosstalk . Another unanswered question is how are MDCs formed ? At this point , we know that the MDC contains both inner and outer mitochondrial membrane-associated proteins , and that proteins incorporated into MDCs preexist within mitochondria . This suggests a sorting and segregation system exists in the inner and outer mitochondrial membranes . The nature of how that might occur is unclear , but the import receptor Tom70/71 plays a role in the formation of the MDC . This suggests that these proteins have roles beyond their function in mitochondrial import . Consistent with this idea , Tom70/71 are known to be required for recruiting an F-box protein Mfb1 to the mitochondrial surface ( Kondo-Okamoto et al . , 2008 ) , and were recently shown to form an ER-mitochondrial tether with the ER sterol-transport protein Ltc1/Lam6 ( Murley et al . , 2015 ) . It will ultimately be important to determine whether MDC’s contain a single or double membrane , which along with identifying more MDC formation machinery will begin to shape our understanding of how proteins might be incorporated into these structures . At the moment , we cannot rule out the possibility that MDCs are formed through fusion of small mitochondrial-derived vesicles that are currently undetectable in our assay . Finally , it will be interesting to determine how MDCs are linked to the autophagy machinery for degradation . Autophagy can selectively degrade a number of organelles , cellular substructures , and protein aggregates ( Stolz et al . , 2014 ) . In mammals , the degradation of these structures is often facilitated by ubiquitination of target molecules to be degraded ( Kirkin et al . , 2009 ) . Ubiquitin tags are recognized by adaptor proteins such as p62 , which bind both ubiquitin and the autophagy protein Atg8/LC3 to link target molecules to the autophagy machinery ( Pankiv et al . , 2007 ) . However , yeast homologs of p62 lack ubiquitin-binding domains , and until very recently , it was unclear if yeast could target proteins for autophagy-dependent degradation through ubiquitin tagging ( Rogov et al . , 2014 ) . However , a recent study identified the conserved CUET protein Cue5 as an ubiquitin/Atg8-binding adaptor protein that facilitates autophagy-dependent turnover of ubiquitylated protein aggregates ( Lu et al . , 2014 ) . It remains to be seen whether ubiquitylation and the CUET protein family play any role in the MDC pathway . The discovery of the MDC pathway adds to a growing list of diverse systems for mitochondrial protein degradation . Understanding how these pathways function in concert with one another to maintain mitochondrial integrity in the face of various cellular stresses will ultimately be important for combating mitochondria-associated disease . All yeast strains are derivatives of S . cerevisiae S288c ( Brachmann et al . , 1998 ) and are listed in Supplementary file 2 . Strains were created by one step PCR-mediated gene replacement and epitope tagging using standard techniques ( Brachmann et al . , 1998; Sheff and Thorn , 2004 ) . Oligos to construct strains are listed in Supplementary file 3 . Plasmid templates for tagging and knockout construction were from the previously described pRS , pKT , and pBSseries of vectors ( Brachmann et al . , 1998; Shaner et al . , 2004; Sheff and Thorn , 2004 ) . pBS34 was obtained from the Yeast Resource Center at the University of Washington with permission from Roger Tsien . RITE tagged strains were created using the previously described template plasmid pVL015 ( Verzijlbergen et al . , 2010 ) . A collection of yeast strains expressing Tom70-mCherry/any protein-GFP was created by crossing a Tom70-mCherry query strain ( UCC4997 , see Supplementary file 2 ) to the yeast GFP strain collection ( Huh et al . , 2003 ) using a Biomek Robot and standard techniques for high-throughput strain construction ( Tong and Boone , 2006 ) . Strains were maintained and used for screening as diploids with both Tom70-mCherry and the GFP-fused proteins of interest in a heterozygous state . The final genotype of all strains in the collection is: MATa/MATα his3△1/his3△1 leu2△0/leu2△0 ura3△0/ura3△0 met15△0/+ lys2△0/+ anygene-GFP-His3MX/+ TOM70-mCherry-KanMX/+ . Strains used in Figure 5D ( Tcd2 , Oac1 , and Ilv2 ) , as well as Figure 5—figure supplement 1 ( Sec63 , Erg6 , Pex11 , Vrg4 , and Sec7 ) and Figure 7—figure supplement 1A–B ( Cox7 ) are from this strain collection . The GFP-ATG8 reporter strain used in Figure 4—figure supplement 2 expresses an extra copy of GFP-ATG8 from a GPD promoter integrated into an empty region of chromosome I ( between 199456 and 199457 ) . This strain was created by transformation and insertion of NotI-digested plasmid pAG306-GPD-eGFP-ATG8 chr I , which is described below . pAG306-GPD-eGFP-ATG8 chr I is a plasmid that can be integrated into an empty region of yeast chromosome I after digestion with the restriction enzyme NotI . The plasmid expresses GFP-ATG8 from the constitutive GPD promoter . pAG306-GPD-eGFP-ATG8 chr I was constructed in two steps . First , we created pAG306-GPD-eGFP-ccdB chr I , a plasmid for high expression of N-terminal GFP fusion constructs from the GPD promoter that can be integrated into chromosome I ( 199456–199457 ) after NotI digestion . We generated pAG306-GPD-eGFP-ccdB chr I by ligation of a SmaI-digested fusion PCR product that contained two ~500 base pair regions of chromosome I flanking a NotI site into AatII-digested pAG306-GPD-eGFP-ccdB ( Addgene plasmid 14308 ) ( Alberti et al . , 2007 ) . We generated the fusion PCR product using oligos ChrI PartB SmaI F and ChrI PartA SmaI R to amplify two templates generated by PCR of yeast genomic DNA using oligo pairs ChrI PartA NotI F and ChrI PartA SmaI R , and ChrI PartB SmaI F and ChrI PartB NotI R , respectively . Second , we inserted ATG8 into pAG306-GPD-eGFP-ccdB chr I from donor Gateway plasmid pDONR201-ATG8 ( HIP ID ScCD00011665 ) using LR clonase according to manufacturer’s instructions ( Thermo Fisher Scientific , Waltham , MA ) ( Hu et al . , 2007 ) . As previously described ( Hughes and Gottschling , 2012 ) , cells were grown exponentially for 15 hr to a max density of 5 x 106 cells/ml before the start of all aging and MDC assays . This period of overnight log-phase growth was carried out to ensure vacuolar and mitochondrial uniformity across the cell population . Cells were cultured in YEPD ( 1% yeast extract , 2% peptone , 2% glucose ) for all experiments . Yeast Complete ( YC ) medium used during construction of the Tom70-mCherry GFP strain collection was previously described ( Tong and Boone , 2006; van Leeuwen and Gottschling , 2002 ) . Concanamycin A ( Sigma-Aldrich , St . Louis , MO ) was added to cultures at a final concentration of 500 nM as indicated in figure legends . In the RITE tag experiments , cycloheximide ( Sigma-Aldrich ) was added at a final concentration of 50 μg/ml , and β-estradiol at 1 μM . In Figure 3 , FCCP ( Sigma-Aldrich ) , Antimycin A ( Sigma-Aldrich ) , and hydrogen peroxide ( Sigma-Aldrich ) were added at to cultures at final concentrations of 10 μM , 20 μg/ml , and 3 mM , respectively . For MDC assays , overnight log-phase cell cultures were grown in the presence or absence of conc A for the indicated time in the figure legends ( 0–6 hr ) . Cells were harvested by centrifugation , and imaged live in all experiments . The number of cells with MDCs or Tom70-GFP in vacuole-localized autophagosomes was quantified in each experiment at the appropriate timepoint . For screening of the Tom70-mCherry/GFP collection to identify MDC substrates , all strains were grown in batches of 20 following the same procedure used for all other MDC assays . For aging experiments , we used the Mother Enrichment Program ( MEP ) ( Lindstrom and Gottschling , 2009 ) coupled to biotin/streptavidin purification to isolate cells of different replicative ages for microscopy analysis . Biotin labeling and purification of MEP cells was carried out exactly as previously described ( Hughes and Gottschling , 2012 ) . Briefly , to attach biotin to the cell surface , we washed 2 . 5 x 107 cells from a 15 hr YEPD log-phase culture twice in phosphate buffered saline , pH 7 . 4 ( PBS ) and resuspended in PBS with 3 mg/ml Sulfa-NHS-LC-Biotin ( Thermo Fisher Scientific ) at a final concentration of 2 . 5 x 107 cells/ml . Cells were incubated for 30 min at room temperature ( RT ) , followed by two washes in PBS and one in YEPD . Biotinylated cells were resuspended in 10 ml of YEPD at 2 . 5 x 106 cells/ml and recovered with shaking for 2 hr at 30°C . These cells were used to seed cultures at a density of 2 x 104 biotinylated cells/ml in YEPD for aging experiments . To initiate the MEP aging program , β-estradiol ( 1 μM ) was added to cultures and cells were grown at 30° for an appropriate time to obtain cells of a desired age ( ~1 hr for young cells , 12 hr for middle-aged , and 24 hr for old cells ) . Cell densities never exceeded 4 x 106 cells/ml . 1 x 108 total cells were harvested for purification and microscopy analysis at each timepoint . For purification after aging , cells were washed twice with PBS , and then resuspended in 500 μl of PBS at a density of 2 x 108 cells/ml . Cells were then incubated for 30 min at RT with 25 μl of streptavidin-coated magnetic beads ( MicroMACS , Miltenyi Biotec , Germany ) . Cells were then washed twice in PBS , resuspended in 8 ml of PBS , and loaded onto a LS MACS column ( Miltenyi Biotec ) that had been equilibrated with 5 ml of PBS . Cells on the column were washed twice with 8 ml of PBS . Columns were then removed from the magnetic field and aged cells were eluted by gravity flow with 8 ml of PBS . Cells were centrifuged to concentrate them for microscopy analysis . 3 , 3’-dihexyloxacarbocyanine iodide ( DiOC6 ) ( Thermo Fisher Scientific ) staining was carried out as previously described ( Hughes and Gottschling , 2012 ) . Briefly , 2 x 106 log-phase cells were washed once in 10 mM HEPES , pH 7 . 6 + 5% glucose and then resuspended in 1 ml of the same buffer containing 175 nM DiOC6 . Cells were then incubated for 15 min at RT , followed by two washes with 10 mM HEPES , pH 7 . 6 + 5% glucose . Cells were resuspended in 10 mM HEPES , pH 7 . 6 + 5% glucose for imaging . Tetramethylrhodamine methyl ester ( TMRM ) ( Thermo Fisher Scientific ) staining was carried out exactly as DiOC6 staining , except that cells were incubated with 50 nM TMRM . For aging experiments , cell age was determined by calcofluor ( Sigma-Aldrich ) staining of bud scars . For this analysis , 5 μg/ml calcofluor was included in the first post-staining wash step prior to imaging . For each experiment , cells were grouped into 3 categories based on age range: Young ( 0–4 budscars ) ; middle-aged ( 7–12 ) ; and old ( >17 ) . For fluorescence microscopy analysis in all figures except those noted below , cells were visualized under 60X oil magnification using a Nikon Eclipse E800 with the appropriate filter set . Images were acquired with a CoolSNAP HQ2 CCD camera ( Photometrics , Tucson , AZ ) and quantified and processed using Metamorph version 7 . 1 . 1 . 0 imaging software . For DiOC6 experiments , cells were scored as reduced if they exhibited at least a 2-fold decrease in mean fluorescence intensity . Microscopy analysis in Figures 3 , 7 , and Figure 4—figure supplements 1B and 2 was carried out using a 100X oil objective on a Zeiss AxioImager M2 using the appropriate filter sets . Images were acquired with an Axiocam 506 mono camera , and processed using Zen imaging software ( Zeiss , Germany ) . Images in Figure 4—figure supplements 1B and 2 represent maximum intensity projections of 6–10 step Z-stacks . For Videos 1–3 showing Tom70-GFP and Vph1-mCherry in old yeast cells , Z-stack images were acquired with a DeltaVision Elite imaging system ( GE Healthcare , United Kingdom ) using a 100X , 1 . 4 NA oil immersion lens and a CoolSNAP HQ2 CCD camera ( Photometrics ) . Images were deconvolved with SoftWoRx 6 . 5 . 2 image analysis software ( GE Healthcare ) and images of the vacuole and mitochondria were IsoSurface rendered in Imaris 8 . 2 . 0 software ( Bitplane , Switzerland ) . After DiOC6 or TMRM staining , cells were analyzed on a BD LSRFortessa X-20 equipped with the appropriate filter sets . At least 10 , 000 events were analyzed for each sample . Statistical analysis for Figure 7D was conducted using Graphpad Prism software . 2 x 107 log-phase cells were resuspended in 100 μl H2O . An equal volume of 0 . 2 M NaOH was added to the cell suspension , and cells were incubated 5 min at room temperature . Samples were then centrifuged at 20 , 000 x g for 10 min at 4°C . Pellets were resuspended in SDS-lysis buffer ( 10 mM Tris-HCl , pH 6 . 8 , 100 mM NaCl , 1% SDS , 1 mM EDTA , and 1 mM EGTA ) containing protease inhibitors ( leupeptin , pepstatin , PMSF , and aprotinin ) for western blot analysis . Immunoblotting was carried out exactly as previously described ( Hughes and Gottschling , 2012 ) . Anti-GFP primary antibody was from Roche ( Switzerland ) ( #11814460001 ) , and secondary HRP-conjugated antibodies from Jackson Immunoresearch ( West Grove , PA ) .	Our cells contain compartments called mitochondria , which provide energy and serve as a home for many metabolic pathways that are critical for life . Changes in mitochondrial activity can contribute to aging and to the development of several age-associated diseases . However , our cells contain systems that detect changes in the performance of mitochondria and can act to re-establish a healthy state . These systems respond to stress or changes in metabolism by degrading particular proteins in the mitochondria . This enables damaged or unwanted proteins to be replaced with new proteins , but it is not clear how these mitochondrial defense systems work to keep mitochondria healthy as an organism ages . Budding yeast is a good model in which to study the aging process because the performance of this yeast’s mitochondria change in a characteristic way as the yeast ages . Previous studies have shown that these changes are caused by alterations in another cell structure called the lysosome , which can store nutrients and is where many proteins are degraded . Here , Hughes et al . used yeast cells to investigate how mitochondrial defense systems operate during aging . The experiments reveal an entirely new mitochondrial protein degradation system that helps to keep the mitochondria healthy as the yeast cells age . This process is rapidly triggered by changes in lysosome activity and results in certain proteins in each mitochondrion being sorted into a small compartment made from a part of the mitochondrion . This compartment is released from the mitochondrion and then travels to the lysosome where the proteins are destroyed . Inhibiting the formation of these compartments results in mitochondria being more sensitive to cellular stress . The next steps following on from this work are to find out exactly what role this mitochondrial defense pathway plays in cells and why it targets only a small set of all the proteins found in mitochondria .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2016	Selective sorting and destruction of mitochondrial membrane proteins in aged yeast
The gating ring-forming RCK domain regulates channel gating in response to various cellular chemical stimuli in eukaryotic Slo channel families and the majority of ligand-gated prokaryotic K+ channels and transporters . Here we present structural and functional studies of a dual RCK-containing , multi-ligand gated K+ channel from Geobacter sulfurreducens , named GsuK . We demonstrate that ADP and NAD+ activate the GsuK channel , whereas Ca2+ serves as an allosteric inhibitor . Multiple crystal structures elucidate the structural basis of multi-ligand gating in GsuK , and also reveal a unique ion conduction pore with segmented inner helices . Structural comparison leads us to propose a novel pore opening mechanics that is distinct from other K+ channels . Ligand-gated K+ channels open and close in response to various cellular chemical stimuli . The majority of prokaryotic ligand-gated K+ channels , as well as eukaryotic Slo channel families ( Slo1 or BK , Slo2 and Slo3 ) ( Salkoff et al . , 2006 ) , have one or two copies of a conserved C-terminal intracellular ligand-binding RCK ( regulating the conductance of K+ ) domain ( Jiang et al . , 2001; Jiang et al . , 2002a; Kuo et al . , 2005 ) ( Figure 1A ) . RCK domains are also ubiquitously distributed in the bacterial K+ uptake ( Trk or Ktr systems ) ( Schlösser et al . , 1993; Nakamura et al . , 1998 ) and efflux machinery ( Kef systems ) ( Bakker et al . , 1987; Munro et al . , 1991 ) . The wide distribution of RCK domains in K+ channels and transporters highlights their importance in regulating K+ transport across the cell membrane . 10 . 7554/eLife . 00184 . 003Figure 1 . RCK-Regulated K+ Channels . ( A ) Topology of RCK-regulated K+ channels . ( B ) Functional assembly of single RCK-containing channels such as MthK . ( C ) Functional assembly of double RCK-containing channels such as GsuK in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 00184 . 003 RCK domains associate as a dimer , ( Jiang et al . , 2001; Roosild et al . , 2002; Dong et al . , 2005 ) , which serves as the basic building block for the quaternary structural assembly in both K+ channels and transporters . As demonstrated in the structure of MthK , a Ca2+-gated K+ channel from Methanobacterium thermoautotrophicum , four RCK dimers assemble into an octameric gating ring in a functional channel tetramer ( Jiang et al . , 2002a ) . The same quaternary complex is also observed in the K+ transporter systems ( Albright et al . , 2006 ) , indicating that a gating ring of eight RCK domains is the functional assembly for both channels and transporters . Most prokaryotic RCK-containing K+ channels have only one copy of the RCK domain on each subunit and the formation of an octameric gating ring requires the co-expression of their cytosolic RCK domain via an alternative internal translation start site on the same gene ( Jiang et al . , 2002a ) . The eukaryotic Slo K+ channel families and a subset of prokaryotic K+ channels already contain two tandem RCK domains on each subunit and therefore the gating ring assembly in these channels no longer requires the co-expression of an isolated RCK domain ( Figure 1B , C ) . The formation of the gating ring provides a platform for diverse allosteric ligand regulation among RCK-containing channels , and the gating rings of some channels are susceptible to multiple cellular stimuli . For example , the RCK domains of BK channels have multiple divalent cation ( Ca2+ and Mg2+ ) binding sites for channel activation and can be modulated by phosphorylation and heme binding , and so on ( Zhang et al . , 2001; Shi et al . , 2002; Xia et al . , 2002; Bao et al . , 2004; Zeng et al . , 2005; Hou et al . , 2009 ) . The structural information of RCK-containing channels has been limited to the low resolution single-RCK MthK channel in a Ca2+-bound , open conformation ( Jiang et al . , 2002a ) or the isolated cytosolic RCK domains from other K+ channels and transporters ( Roosild et al . , 2002; Albright et al . , 2006; Ye et al . , 2006; Wu et al . , 2010; Yuan et al . , 2010; Yuan et al . , 2011 ) . Due to the resolution limit and poor crystal packing , the ion conduction pore of MthK was poorly defined and the linkers between the pore and the gating ring , which are essential for coupling the gating ring conformational change to the pore opening and closing , could not be resolved in the structure . In this study , we present structural and functional studies of a novel a two-transmembrane , RCK-regulated K+ channel , GsuK , named after the bacterium Geobacter sulfurreducens from which this K+ channel was cloned . Each GsuK subunit contains two tandem RCK domains , reminiscent of Slo K+ channels . We demonstrate that GsuK is a nucleotide-activated and Ca2+-deactivated K+ channel and reveal the structural mechanism of multi-ligand gating of this double-RCK K+ channel and a distinct pore opening mechanics . Our study of GsuK started with the structure determination of its two tandem RCK domains , labeled as RCK1 and RCK2 . The two intracellular ligand binding RCK domains form a bi-lobed structure equivalent to the MthK RCK dimer; each lobe consists of the N-terminal two-thirds of the RCK domain ( βA to αF ) and adopts a Rossmann-fold ( Rossmann et al . , 1974 ) ( Figures 2 and 3 ) . While the secondary structural elements of the C-terminal subdomains are similar between GsuK and MthK , their tertiary structural arrangements are quite distinct . In MthK , the N-terminal lobes and the C-terminal subdomains of the RCK dimer are connected by interlocking helix-turn-helix motifs ( αF-turn-αG ) , which provide extensive dimerization interactions at the so-called flexible interface ( Figure 3B ) . In GsuK , the equivalent αG helix is absent in RCK1 and becomes a shorter helix with a different orientation in RCK2 , resulting in swapped and loosely packed C-terminal subdomains ( Figure 3A ) . Four GsuK intracellular subunits assemble into a gating ring containing eight RCK domains through inter-subunit interactions at the assembly interfaces ( Figure 4A , B ) , with the N-terminal Rossmann-folded lobe of each RCK forming the core of the gating ring and the C-terminal subdomain loosely associating with the core of the gating ring on the periphery . 10 . 7554/eLife . 00184 . 004Figure 2 . Sequence and secondary structure comparison between GsuK and MthK . For comparative purposes , the secondary structural elements of each GsuK RCK domain are labeled following the same nomenclature used for MthK . A duplicate copy of MthK RCK is used in the alignment with GsuK RCK2 . DOI: http://dx . doi . org/10 . 7554/eLife . 00184 . 00410 . 7554/eLife . 00184 . 005Figure 3 . Structure of the GsuK intracellular subunit . ( A ) Stereoviews of GsuK intracellular subunit . RCK1 and RCK2 are colored green and orange , respectively . Ca2+ and Zn2+ ions are shown as red and silver spheres , respectively . The same color representations are used in all figures . ( B ) Stereoviews of MthK RCK dimer . The N-terminal lobes and the C-terminal subdomains are circled in RCK2 of GsuK and in one of the RCK subunits of MthK . DOI: http://dx . doi . org/10 . 7554/eLife . 00184 . 00510 . 7554/eLife . 00184 . 006Figure 4 . Structure of the GsuK gating ring . ( A ) Stereo representation of the GsuK gating ring viewed from the top . Arrows indicate the inter-subunit assembly interface . ( B ) Stereo view of the symmetrical MthK gating ring in the open state . ( C ) Dimension of the GsuK gating ring viewed from top ( left ) and bottom ( right ) . The diagonal distance is measured between the Ca atoms of Gly131 , which is the starting residue of the RCK1 . Red square marks the size of the central hole . ( D ) The position of linkers between the gating ring and the ion conduction pore in GsuK ( left ) and BK channel ( right ) . The linkers are in ball-and-stick representation and the gating rings are shown as surface rendered representation . The short N-terminal four-helix bundle on top of the GsuK gating ring is shown as green ribbons . DOI: http://dx . doi . org/10 . 7554/eLife . 00184 . 006 As the GsuK gating ring is formed by two different sets of RCK domains , its top and bottom halves are not twofold symmetrical , as seen in the MthK gating ring . The pore-connecting top half of the GsuK gating ring is in a contracted form similar to the closed MthK whereas the bottom half is more expanded ( Figure 4C ) , suggesting that the structure likely represents a closed conformation . Furthermore , each subunit also contains a small fragment of the pore-lining inner helix at its N-terminus , which forms a short four-helix bundle atop the center of the gating ring and creates a constriction point at the intracellular end of the pore that would occlude the passage of hydrated K+ ions ( Figure 4A , D ) . The same closed gate is also observed in the structures of the full-length channel as discussed later . The helix bundle is tethered to RCK1 by linkers in an extended configuration , ensuring a tight coupling between the gating ring conformational change and pore opening at the intracellular gate ( Figure 4D ) . Despite low sequence similarity , the position and structural features of this linker are similar to that observed in the BK channel gating ring ( Wu et al . , 2010 ) ( Figure 4D ) . The RCK2 domain of GsuK contains the conserved GxGxxG…D/E sequence motif for nucleotide binding in Rossmann-folded protein ( Bellamacina , 1996 ) ( Figure 2 ) . Indeed , electron density modeled as AMP was observed at the predicted nucleotide binding site in the GsuK gating ring structure ( Figure 5A ) . As no nucleotides were added during protein purification or crystallization , the bound nucleotide is likely from the E . coli cells used for protein expression . Although modeled as AMP , the electron density could actually be from other adenine-containing nucleotide whose AMP moiety is well structured while the rest is mobile . This is the case in the structure of the nucleotide binding RCK domain from a K+ transporter , in which only the AMP portion of a bound NAD+ can be defined ( Roosild et al . , 2002; Albright et al . , 2006 ) . Supporting this hypothesis , our functional characterization using the 86Rb flux assay demonstrated that ADP and NAD+ are the likely ligands for GsuK as discussed below . 10 . 7554/eLife . 00184 . 007Figure 5 . Ion and ligand binding in GsuK . ( A ) Structure of the nucleotide-binding site on RCK2 . The electron density ( blue mesh , contoured at 3σ ) from Fo–Fc omit map is modeled as AMP . Purple spheres represent the Cα atoms of glycine residues from the conserved nucleotide binding motif . ( B ) Local structure of the Zn2+ ( silver sphere ) binding site on RCK2 with Fo–Fc ion omit map ( blue mesh ) contoured at 9σ . ( C ) Stereoview of the inter-subunit interactions at the assembly interface . Side chains of hydrophobic residues are shown as cyan sticks . Residues that chelate the Ca2+ ion with backbone carbonyl oxygen atoms are labeled in red . The electron density of Ca2+ ( red sphere ) from Fo–Fc ion omit map is contoured at 5 . 5σ . ( D ) The inter-subunit interface and Ca2+ bowl ( red loop ) of the BK channel . DOI: http://dx . doi . org/10 . 7554/eLife . 00184 . 007 Two bound metal ions were observed in each GsuK subunit . One is identified as Zn2+ as it is chelated by His ( His359 and His391 ) and Cys , ( Cys364 and Cys388 ) in RCK2 ( Figure 5B ) with ion coordination chemistry and the local structure resembling a zinc-finger motif; fluorescence scanning of the crystal at the synchrotron also confirmed the presence of Zn2+ in the crystal . Whether Zn2+ plays any functional role is still unclear and warrants further study . The second bound ion , identified as Ca2+ based on ligand chemistry and functional assays , is positioned at the inter-subunit assembly interface , a location reminiscent of the Ca2+ bowl in BK channel ( Figure 5C , D ) . The six oxygen ligands , two of which are backbone carbonyl oxygen atoms , come from Thr183 , Asn210 and Thr214 of RCK1 and Glu449 , Asn450 and Gln453 of RCK2 from the neighboring subunit ( Figure 5C ) . Unlike that in MthK or BK , Ca2+ serves as an allosteric inhibitor in GsuK whose binding stabilizes the closed gating ring and deactivates the channel as confirmed by single channel electrophysiology and the full-length channel structure . As the crystallization conditions contain neither Zn2+ nor Ca2+ , both ions are likely from the E . coli cells or trace amounts of contaminants in the solutions used for protein purification and crystallization . We utilized the 86Rb flux assay initially to identify the potential nucleotide ligands for GsuK ( ‘Materials and methods’ ) . In this assay , various nucleotides at a concentration of 1 mM were added individually to GsuK-containing liposomes loaded with high [K+] followed by mixing with the flux buffer containing radioactive 86Rb . The effect of individual nucleotide on 86Rb influx into the liposome was monitored by measuring intraliposomal radioactivity levels 5 min after initial mixing . Among those nucleotides tested , guanidine nucleotides had no effect on 86Rb influx as compared to the control liposomes absent of nucleotide , while ADP and NAD+ led to about a 4–5-fold increase in intraliposomal radioactivity ( Figure 6A ) . Interestingly , other adenine-containing nucleotides such as ATP , AMP , or NADH had no obvious effect on 86Rb influx . 10 . 7554/eLife . 00184 . 008Figure 6 . Functional analysis of the GsuK channel . ( A ) 86Rb+ flux assays of GsuK-containing liposomes in the presence of various nucleotides . Data are averages of two measurements and normalized against the control sample without nucleotide . ( B ) – ( D ) Single channel traces and I–V curves of the wild-type channel and L97D mutant . Currents were recorded using giant liposome patch clamping with 150 mM NaCl and 150 mM KCl in the pipette and bath solutions , respectively . ( E ) Sample traces of wild-type channel in the presence and absence of intracellular Ca2+ ( left ) and the plot of [Ca2+]-dependent single channel open probability of wild-type GsuK and L97D mutant ( right ) . Both pipette and bath solutions contain symmetrical 150 mM KCl . Data for wild-type channel are fitted to the Hill equation with K1/2 of 197 uM and Hill coefficient n = 2 . 3 . Data are mean ± SEM of seven measurements . ( F ) Sample traces of partially deactivated GsuK channel in the presence and absence of 2 mM ADP . Shown on the right is the plot of [Ca2+]-dependent single channel open probability of wild-type GsuK in the presence of 2 mM various adenine-containing nucleotide . Both pipette and bath solutions contain symmetrical 150 mM KCl . Data are fitted to the Hill equation with K1/2 = 210 μM and n = 2 for AMP , K1/2 = 350 μM and n = 1 . 4 for ADP , and K1/2 = 370 μM and n = 1 . 3 for NAD+ . Data are mean ± SEM of five measurements . DOI: http://dx . doi . org/10 . 7554/eLife . 00184 . 008 Giant liposome patching was employed to assay the functional properties of both wild-type GsuK and L97D pore mutant , a mutation that enhances the channel conductance and open probability ( Shi et al . , 2011 ) , and also yields better diffracting crystals ( ‘Materials and methods’ ) . In the absence of intracellular Ca2+ , both the wild-type and mutant channels exhibit high single channel activity with an open probability ( Po ) of about 0 . 9 ( Figure 6B–D ) . The L97D mutation resulted in a significantly higher single channel conductance than the wild-type channel . Both channels are weakly K+ selective , with a reversal potential of about −40 mV under bi-ionic conditions , equivalent to a permeability ratio ( PNa/PK ) of about 0 . 2 ( Figure 6D ) . The weak K+ selectivity could be partly attributed to the presence of Phe instead of Tyr in the selectivity filter as demonstrated in a recent selectivity study on K+ channel in which replacing Tyr with Phe in the filter results in a decreased K+ selectivity ( Sauer et al . , 2011 ) . The presence of Ca2+ at the intracellular side decreases the channel open probability but not single channel conductance ( Figure 6E ) , whereas the presence of Ca2+ at the extracellular side has no obvious effect on channel open probability ( ‘Materials and methods’ ) . These results , combined with the observation of Ca2+ binding at the gating ring assembly interface , suggest that the reduction in channel activity by Ca2+ is a result of a gating rather than pore blocking effect . The wild-type channel exhibits cooperative Ca2+ deactivation with a K1/2 of 197 μM and Hill coefficient of 2 . 3 . The L97D mutant , on the other hand , mitigates Ca2+ deactivation and maintains a fairly high open probability ( Po ∼ 0 . 4 ) even in the presence of 3 mM Ca2+ , suggesting that this inner helix mutation favors the pore in an open conformation and thereby weakens the inhibitory gating effect of Ca2+ . Similar pore opening effect was also observed with equivalent mutation ( A88D ) in MthK ( Shi et al . , 2011 ) . The gating effects of nucleotide ligands identified by the flux assay were also assessed on the wild-type channel . Although the gating effect is not as profound as Ca2+ , both NAD+ and ADP can increase the channel open probability , and their activation effect is more obvious under a Ca2+-deactivated state ( >300 μM [Ca2+] ) and less so at lower [Ca2+] where the channel are already highly active ( Figure 6F ) . Consistent with the flux assay , such an activation effect is not observed with other adenine-containing nucleotides such as AMP . Two full-length channel structures , wild-type channel and L97D mutant , were determined at 3 . 7 Å and 2 . 6 Å , respectively ( Table 1 ) . Both channels share a similar overall structure and , therefore , the higher resolution L97D mutant is used here for the description of the overall full-length channel structure ( Figure 7A ) . The L97D mutant channel crystals are of the space group C2 with unit cell dimensions of a = 232 . 9 Å , b = 111 . 7 Å , c = 164 . 1 Å and β = 134 . 5° , and contain four channel subunits in an asymmetric unit . These subunits do not belong to the same channel tetramer and , instead , are divided into two half channels , which participate in the formation of two channel tetramers with their crystallographic twofold related partners . The gating ring of the full-length channel tetramer adopts a similar structure to the isolated gating ring , indicating a closed conformation . Both Zn2+ and Ca2+ are present in the full-length channel structure , but no clear density for nucleotide is observed . 10 . 7554/eLife . 00184 . 009Table 1 . Data collection and refinement statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 00184 . 009Data CollectionIntracellular subunitWild typeL97D mutantL97D mutant /ADPL97D mutant /NADSpace groupI222C2C2C2C2Cell dimensions: a , b , c ( Å ) 110 . 6 , 161 . 7 , 310 . 1235 . 0 , 108 . 4 , 165 . 8232 . 9 , 111 . 7 , 164 . 1234 . 3 , 111 . 4 , 164 . 7232 . 5 , 111 . 1 , 164 . 6α , β , γ ( ° ) 90 , 90 , 9090 , 135 . 0 , 9090 , 134 . 5 , 9090 , 134 . 9 , 9090 , 134 . 8 , 90Wavelength ( Å ) 0 . 97921 . 03320 . 97860 . 97920 . 9792Resolution ( Å ) 50 . 0 − 3 . 050 . 0 − 3 . 750 . 0 − 2 . 650 . 0 − 2 . 850 . 0 − 3 . 2Measured reflections309 , 562123 , 263303 , 788225 , 169141 , 834Unique reflections55 , 43527 , 53987 , 30572 , 26047 , 223Redundancy5 . 64 . 53 . 53 . 13 . 0Completeness ( % , highest shell ) 99 . 2 ( 93 . 6 ) 86 . 7 ( 47 . 4 ) 94 . 7 ( 57 . 8 ) 96 . 4 ( 88 . 9 ) 95 . 6 ( 92 . 5 ) Mean I/σI ( highest shell ) 35 . 9 ( 1 . 8 ) 19 . 0 ( 1 . 0 ) 18 . 4 ( 1 . 0 ) 15 . 6 ( 1 . 0 ) 14 . 7 ( 1 . 0 ) Rsym ( % , highest shell ) †6 . 0 ( 68 . 3 ) 7 . 8 ( 82 . 9 ) 6 . 4 ( 51 . 6 ) 7 . 0 ( 86 . 7 ) 7 . 1 ( 79 . 2 ) RefinementResolution ( Å ) 50 . 0 − 3 . 050 . 0 − 3 . 730 . 0 − 2 . 650 . 0 − 2 . 850 − 3 . 2No . of reflections \|F\|>0 σF54 , 22427 , 45887 , 24271 , 57047 , 059R-factor/R-free ( % ) ‡22 . 8/26 . 226 . 1/29 . 320 . 3/24 . 921 . 3/25 . 523 . 0/27 . 0No . of protein atoms13 , 84014 , 24614 , 23514 , 24414 , 255No . of solvent atoms0039127350No . of ions ( K+/Ca2+/Zn2+ ) 0/4/412/4/412/4/411/4/412/4/4No . of ligands4 AMP004 ADP4 NADRmsd bond lengths ( Å ) §0 . 0090 . 0060 . 0040 . 0050 . 003Rmsd bond angles ( ° ) 1 . 2431 . 0530 . 6660 . 9020 . 723Redundancy = total measurements/unique reflections . †Rsym = Σ\|Ii − <Ii>\|/ΣIi , where <Ii> is the average intensity of symmetry equivalent reflections . ‡R factor = Σ\|F ( obs ) − F ( cal ) \|/ΣF ( obs ) , 5% of the data were used in the Rfree calculations . §Rmsd = root-mean-square deviation . Numbers in parentheses are statistics for highest resolution shell . 10 . 7554/eLife . 00184 . 010Figure 7 . Structure of the full-length GsuK channel . ( A ) Stereoview of full-length GsuK channel with L97D mutation . The transmembrane helices are shown as blue cylinders and the gating ring is in ribbon representation with RCK1 in green and RCK2 in orange . Subdomains from the front and back subunits are disordered and absent in the structure . ( B ) Comparison of the translational distances between the gating ring and the membrane-spanning pore in GsuK ( left , L97D mutant ) and MthK ( right ) . ( C ) Comparison of the relative orientation between the gating ring ( ribbon representation ) and ion conduction pore ( cylinder representation ) in GsuK ( left ) and MthK ( right ) . Only one subunit from each channel is colored . Both channels are superimposed using the pore region and viewed from the extracellular side . Arrows connect the central fourfold axis ( square ) to the starting residue ( Cα atoms of Gly131 in GsuK and Arg114 in MthK ) of the first RCK domains , indicating the approximate direction of the linker . DOI: http://dx . doi . org/10 . 7554/eLife . 00184 . 010 The full-length GsuK structure exhibits two major differences in relative position between the gating ring and the pore as compared to MthK . First , GsuK has a long inner helix that is seven residues ( about two helical turns ) longer than that of MthK , resulting in the attached gating ring being further away from the membrane ( Figure 7B ) . Second , GsuK and MthK adopt different relative orientations between the gating ring and the pore ( Figure 7C ) . With their pores superimposed , the gating rings of MthK and GsuK have about a 50-degree rotation relative to each other about the central axis . This difference in relative orientation could contribute to the different pore opening mechanics between GsuK and MthK as discussed later . While having the same structure at the selectivity filter region , the membrane-spanning pore of GsuK has several unique structural features as compared to other K+ channels . First , instead of forming a single straight helix , the long inner helix of GsuK is segmented into three parts , labeled TM2A , 2B and 2C , respectively ( Figure 8A ) . The break between TM2A and 2B is at a position near the helix-breaking PVP region of Kv channels ( Long et al . , 2007 ) . Second , rather than forming a bundle crossing right below the central cavity as seen in the closed KcsA structure ( Doyle et al . , 1998; Zhou et al . , 2001 ) , the four inner helices of GsuK are more parallel to the central pore axis and generate an elongated , water-filled vestibule that spans twice the length of the KcsA cavity ( Figure 8A ) . Third , the second inner helix break reorients the TM2C segment towards the central axis , forming a constriction at the very end in a channel tetramer and pinching shut the pore at residue Leu117 ( Figure 8A , B ) . 10 . 7554/eLife . 00184 . 011Figure 8 . The ion conduction pore of GsuK . ( A ) Structural comparison between the GsuK pore and KcsA . K+ ions in the filter are shown as green spheres . Grey surface representation illustrates the space of the central cavity . GsuK inner helix ( TM2 ) is segmented into three parts labeled as 2A , 2B and 2C , respectively . ( B ) Zoomed-in view of the GsuK intracellular gate with the surrounding charged residues drawn as sticks . Leu117 side chains are shown in CPK models . ( C ) Superimposition of the ion conduction pores from the wild-type ( magenta ) and the L97D mutant ( blue ) channels . ( D ) View of the superimposition from the intracellular side . The intracellular gate remains closed in the L97D mutant . DOI: http://dx . doi . org/10 . 7554/eLife . 00184 . 011 The closed gate is tethered to the RCK gating ring through the extended linkers as was seen in the isolated gating ring structure . A stretch of charged residues , mainly positive ones , are clustered at the end of the inner helix and cuff around the closed gate ( Figure 8B ) . These positively charged residues , also observed in some RCK-containing Slo channels , could potentially participate in channel gating by interacting with lipids similar to the mechanism in inwardly rectified K+ channels ( Hansen et al . , 2011; Whorton and MacKinnon , 2011 ) . Despite the similar gating ring structures , the ion conduction pores of the wild-type and L97D mutant channels exhibit obvious differences along the inner helix . As shown in the pore superimposition ( Figure 8C , D ) , the inner helices of the wild-type and mutant channels diverge at the glycine gating hinge ( Gly92 ) , but converge at the C-terminal end of the inner helix where the bundle crossing forms . In the L97D mutant , a noticeable helix bend occurs at Gly92 ( hinge 1 ) , whereas a sharp turn occurs between TM2B and 2C ( hinge 2 ) in the wild-type channel inner helix . To reveal the structural basis of nucleotide activation , we also co-crystallized the L97D mutant with ADP and NAD+ and determined the complex structures at 2 . 8 and 3 . 2 Å , respectively ( Table 1 ) . The overall structure of the nucleotide-bound L97D mutant is similar to that of the apo state , with the gating ring in a Ca2+-bound , closed conformation . In the ADP-bound structure , the AMP moiety of the ligand can be unambiguously defined in all four subunits within the asymmetric unit at the same position as was observed in the isolated gating ring structure ( Figure 9A ) . The electron density for the β-phosphate group , although not as well resolved as the rest due to higher mobility , points to two possible configurations . In one subunit , the ADP is in an extended trans configuration , whereas in the other subunits the ADP is in a cis configuration with respect to the phosphoester bond . In the cis configuration , the β-phosphate makes a sharp turn and inserts itself into a pocket formed by the loops from βA-to-αA and from βD-to-αD on the second RCK domain ( Figure 9A , B ) . The pocket is located at the base of the flexible interface and analogous to the Ca2+ binding site in the MthK RCK ( Figure 9C ) , suggesting that ADP binding in the cis configuration represents an activating state and promotes the gating ring conformational change at the flexible interface similar to Ca2+ binding in MthK . The pocket is large enough to accommodate one phosphate group and is in a position only accessible by the β-phosphate , explaining the ligand specificity for ADP but not ATP or AMP . 10 . 7554/eLife . 00184 . 012Figure 9 . ADP binding in GsuK . ( A ) Bound ADP in trans ( left , non-active ) and cis ( right , active ) configurations . The electron density is from Fo–Fc omit map contoured at 3 . 0σ . ( B ) Stereoview of the cis-ADP activation site from inside the gating ring . RCK1′ and RCK2′ are from the neighboring subunits . Dotted oval indicates the location of the gate and the central arrow indicates the direction of the central axis of the pore . ( C ) Local structure comparison between the ADP activation site in GsuK and the Ca2+ activation site in MthK . DOI: http://dx . doi . org/10 . 7554/eLife . 00184 . 012 The NAD+-bound structure has lower ligand occupancy , likely due to a lower ligand affinity in the Ca2+-bound closed gating ring . Nevertheless , partially occupied nucleotide is clearly visible in the Fo–Fc omit map in two of the subunits ( Figure 10A ) . In particular , the ADP and nicotinamide moieties of NAD+ can be properly defined in the electron density map as both are engaged in direct interactions with the protein , whereas the bridging ribose in between is flexible . The position of the nicotinamide group suggests a different activating mechanism between NAD+ and ADP . The NAD+ nicotinamide group is inserted beneath the N-terminal end of the crossover αF helix in RCK1 ( Figure 10A , B ) , at the hinge of the flexible interface between βF and αF where the gating ring conformational change occurs ( Ye et al . , 2006 ) . The strategic position and the positive charge suggest that the nicotinamide serves as the activation group of NAD+ and works as a lever whose insertion promotes the hinged motion at the flexible interface towards the open conformation in GsuK . One plausible explanation for the specificity for NAD+ but not NADH is that the tightly-spaced binding pocket permits the insertion of a flat pyridine ring from the nicotinamide of NAD+ but excludes the puckered dihydropyridine ring from NADH . 10 . 7554/eLife . 00184 . 013Figure 10 . NAD+ binding in GsuK . ( A ) NAD+ binding in GsuK with Fo–Fc omit map contoured at 3 . 0σ . ( B ) Stereoview of the NAD+ activation site from inside the gating ring . DOI: http://dx . doi . org/10 . 7554/eLife . 00184 . 013 The structure of GsuK provides an excellent model system for understanding the structural basis of multi-ligand regulation of the RCK gating ring as commonly seen in the eukaryotic Slo channels . Ca2+ stabilizes the closed gating ring by binding at the inter-subunit assembly interfaces and deactivates the GsuK channel . Ca2+ deactivation has a dominant effect on GsuK gating as the removal of Ca2+ increases the channel open probability regardless of the presence or absence of nucleotides . The gating effect of nucleotide binding is less profound than Ca2+ as it only moderately enhances the channel open probability under a Ca2+-deactivated state . While both NAD+ and ADP anchor their adenine rings at the same site on RCK2 by using the conserved nucleotide binding motif , they utilize different functional groups and activate the channel at different sites . In addition to nucleotide and Ca2+ , the GsuK RCK2 also has a high affinity site for Zn2+ ion whose functional role , if any , is unclear . Despite differences in ligands , there is a convergence of the ligand activation sites between GsuK and other RCK-regulated channels . Albeit with opposite gating effect , the Ca2+ binding site of GsuK is analogous to the position of the Ca2+-bowl in BK ( Wu et al . , 2010; Yuan et al . , 2010 ) ; and the ADP activation site in GsuK is equivalent to the Ca2+ binding site in MthK ( Jiang et al . , 2002a; Dong et al . , 2005 ) . In addition , many RCK domains from channels or transporters contain the conserved nucleotide binding motif at the same position , indicating a common site for adenine-containing nucleotide ligand . GsuK has its intracellular gate at the end of the inner helix , distal from the canonical bundle crossing seen in KcsA , allowing for direct coupling to the gating ring through the extended linker . Although the intracellular gates of both the wild-type and L97D mutant channels are closed in the structures , the structural difference between their ion conduction pores suggests a distinct pore opening mechanism for GsuK . The L97D mutant appears to promote inner helix bending at the glycine gating hinge ( Gly92 ) and the direction of the helix bend is parallel with the orientation of the linker between the pore and the gating ring . Its gating ring , however , is still in the Ca2+-bound , closed conformation , which locks the intracellular gate closed at the end of the inner helix and prevents the concurrent movement of the TM2C with TM2B ( Figure 8C , D ) . Consequently , the sharp turn between TM2B and 2C in the wild-type channel inner helix becomes straightened in the L97D mutant . These structural differences , along with the electrophysiological observation that the L97D mutant favors pore opening and has lower sensitivity to Ca2+ deactivation , suggest that the L97D structure is in an intermediate state in which the pore inner helices are undergoing conformational changes towards opening but the channel's intracellular gate remains shut , deactivated by Ca2+ . With the gating ring constriction alleviated , we expect that the inner helix of the open pore undergoes a similar bending movement at the glycine hinge with TM2B and 2C moving concurrently . A working model for the open pore can therefore be generated by applying the inner helices bending observed at Gly92 of L97D mutant onto the wild-type pore and allowing TM2B and 2C to move as a rigid body ( Figure 11A ) . In this open pore model , Leu117 is rotated away from the ion permeation pathway , yielding a larger entrance at the gate . 10 . 7554/eLife . 00184 . 014Figure 11 . Proposed pore opening mechanics of GsuK . ( A ) Working model of the GsuK intracellular gate from closed ( magenta ) to open ( green ) . Arrows indicate the direction of inner helix movement from closed to open state . Leu117 side chains are shown as sticks . ( B ) Pore opening mechanics of MthK . KcsA is used as the closed model for MthK . DOI: http://dx . doi . org/10 . 7554/eLife . 00184 . 014 The proposed pore opening mechanics of GsuK is distinct from other K+ channels . The GsuK inner helix has two different hinge points in response to external stimuli , allowing the pore to adopt an open , closed , and Ca2+-deactivated states . The GusK pore dilates open in a very different direction as compared to MthK or Kv channels ( Jiang et al . , 2002b; Long et al . , 2005; Ye et al . , 2010 ) ( Figure 11A , B ) . This reversed dilation of pore opening between GsuK and MthK can be attributed to the differences in the relative orientation between the gating ring and the pore as shown in Figure 7C . As current structural studies of RCK-regulated gating suggest that the expansion of the gating ring upon activation is mechanically coupled to the pore opening ( Ye et al . , 2006; Yuan et al . , 2011 ) , then the inner helix movement in different directions is expected between GsuK and MthK . Although protein samples were prepared in nominal Ca2+-free conditions , the gating rings in all GsuK crystal structures are in a Ca2+ bound , closed conformation , suggesting a relatively high Ca2+ binding affinity in GsuK . The likely source of Ca2+ is contamination of the chemicals used for protein purification and crystallization . Single channel recordings , however , showed a lower apparent affinity of Ca2+ with a K1/2 of about 200 μM . One possible explanation for this discrepancy is that the protein crystals were grown in a detergent environment whereas channel recording was performed in a lipid membrane . As lipids are known to be important for membrane protein stability , their presence may be necessary to stabilize the GsuK pore in an open state and , therefore , give rise to a lower efficacy of Ca2+ deactivation . A possible region in GsuK for lipid interactions is the stretch of positively charged residues surrounding the closed gate at the carboxyl terminus of the inner helix , which could potentially interact with the phospholipid head groups and influence channel gating . Certain resemblances between GsuK and Slo channels are noteworthy . Like GsuK , the majority of Slo2 . 1 and Slo3 channels have the same TVGFG signature sequence , and a study of mouse Slo3 showed a K+ selectivity similar to that of GsuK ( Schreiber et al . , 1998 ) . The Slo2 inner helix contains a conserved Pro at the position equivalent to the helix breaking Leu97 of GsuK , suggesting a similar inner helix break . NAD+ has also been shown in a recent study to modulate Na+ activation in Slo2 by binding at the same nucleotide site in RCK2 ( Tamsett et al . , 2009 ) . Furthermore , the relative orientation between the transmembrane pore and the gating ring in Slo channels , and consequently pore opening mechanism , may have a closer resemblance to GsuK than MthK . One piece of evidence is that the acidic Asp90 residue in the S0–S1 loop of the BK channel transmembrane pore has been shown to interact with the Glu374/Glu399 Mg2+ site from the neighboring subunits ( Yang et al . , 2008 ) . Docking of the Kv channel pore onto the BK gating ring using MthK as a model would position such interaction within the same subunit ( Wu et al . , 2010 ) , but is otherwise possible if using GsuK as a structural model . The GsuK gene from Geobacter sulfurreducens was initially subcloned into the pQE 70 vector using Sph I and Bgl II restriction sites , and contains a thrombin cleavage site between the C-terminal His-tag and the channel . The N-terminus of GsuK was modified for the expression of both the full-length channel and its intracellular subunit . The construct for full-length channel starts at residue Ala9 and contains five extra amino acids ( MQRGS ) at the N-terminus . The construct for the expression of the GsuK intracellular subunit starts at residue Tyr10 and contains two extra amino acids ( MQ ) at the N-terminus . Interestingly , although both constructs are very similar , the expression of the latter one starts from Met107 instead of the first Met , producing only the intracellular subunit . Despite the lack of the membrane-spanning segments , solubilization and purification of the GsuK intracellular subunit still requires the presence of detergent at a concentration above the CMC ( critical micelle concentration ) . Both constructs were expressed in E . coli BL21 ( DE3 ) cell lines by induction ( at A600 ∼ 0 . 8 ) with 0 . 4 mM isopropyl-β-d-thiogalactopyranoside ( IPTG ) at 37°C for 3–4 hr . Cells were harvested and lysed in a solution of 50 mM Tris–HCl , pH 8 , 250 mM KCl and protease inhibitors including leupeptin , pepstatin , aprotinin and PMSF ( Sigma-Aldrich , St . Louis , MO ) . Expressed proteins were then extracted from the cell lysate for 3 hr at room temperature in the above solution by adding 40 mM n-decyl-β-d-maltoside ( DM ) . The detergent-solubilized proteins were loaded on a Talon Co2+ affinity column ( Clontech , Mountain View , CA ) equilibrated with 50 mM Tris–HCl , pH 8 . 0 , 250 mM KCl and 4 mM DM . In-gel digestion was performed by incubating the protein-bound Co2+ resin with thrombin ( 1 unit per liter of bacterial culture ) at 4°C overnight to remove the His-tag and released the proteins into the equilibration solution . After elution , proteins were further purified and buffer exchanged on a Superdex-200 ( 10/30 ) gel filtration column in a solution of 20 mM CHES , pH 9 . 0 , 150 mM KSCN , 0 . 1 mg/ml E . coli polar lipid , 2 mM DTT and 4 mM DM for the full-length GsuK channel , and in a solution of 50 mM Tris–HCl , pH 8 . 0 , 250 mM KCl , 2 mM DTT and 4 mM LDAO for the GsuK intracellular subunit . Both proteins elute at a position corresponding to the size of a tetramer . Purified GsuK intracellular subunit was concentrated to approximately 8 mg/ml using an Amicon Ultra centrifugal filtration device ( 50 kDa MW cutoff ) and crystallized at 20°C using the sitting drop vapor diffusion method by mixing equal volumes of concentrated protein and well solution containing 20–23% PEG3350 , 120 mM KCl , 80 mM NaNO3 , 1% glycerol and 100 mM Bis-Tris propane , pH 8 . 5 . The crystals were cryo-protected by slowly supplementing the crystallization drops with extra 20% PEG400 and flash frozen in liquid nitrogen . Two mutations , E52A and Q77E or R , were introduced to the pore region of the full-length channel in order to obtain diffracting crystals . Both mutations did not have any observable effect on channel function as tested in single channel electrophysiology , and therefore this full-length channel is considered as wild-type in this study . The crystal quality was further improved by supplementing the buffer solutions with E . coli polar lipids during protein purification . The protein was purified in DM using the same procedure as described above and concentrated to approximately 6 mg/ml for crystallization at 20°C with well solution containing 13–18 % PEG3350 , 250–500 mM KSCN , and 100 mM CHES , pH 9 . 0 . The additional L97D mutation gave rise to better diffracting full-length channel crystals under the same crystallization condition . This L97D mutant was also used for co-crystallization with nucleotides where ADP or NAD+ was added to the protein solution to a final concentration of 1 mM before crystallization trials . All full-length channel crystals were cryo-protected by slowly increasing the PEG3350 concentration in the crystallization drops to 20% followed by a supplement of 20% PEG400 . X-ray data were collected at the Advanced Photon Source ( APS ) Beamlines 19-ID and 21-ID , and at the Advanced Light Source ( ALS ) of the Lawrence Berkeley Laboratory ( LBL ) beamline 8 . 2 . 1 . Data processing and scaling were performed using the HKL2000 software ( Otwinowski and Minor , 1997 ) . Crystals of the GsuK intracellular subunit are of space group I222 with unit cell dimensions of a = 110 . 6 Å , b = 161 . 7 Å , c = 310 . 1 Å , and α = β = γ = 90° , and contains four subunits , which form a gating ring in the asymmetric unit . The structure was determined by molecular replacement method using the open MthK gating ring ( PDB ID: 1LNQ ) as the search model followed by repeated cycles of model building with XtalView ( McRee , 1999 ) and refinement with REFMAC ( Collaborative Computational Project , 1994 ) . The final model was refined to 3 . 0 Å with Rwork of 22 . 8% and Rfree of 26 . 2% ( Table 1 ) and contained residues from 110 to 564 of each subunit . The full-length GsuK crystals are of space group C2 and contain four subunits in an asymmetric unit . The four subunits do not belong to the same channel tetramer , but instead participate in the formation of two channel tetramers with their crystallographic twofold related partners . The structure was determined by molecular replacement method using half of the GsuK gating ring structure ( two intracellular subunits ) as the search model , followed by repeated cycles of model building in Coot ( Emsley and Cowtan , 2004 ) and refinement with PHENIX ( Adams et al . , 2010 ) . The final models for the wild-type channel and L97D mutant were refined to 3 . 7 Å and 2 . 6 Å , respectively . In the models of all full-length channel structures , two of the subunits contain residues 17 to 564 whereas the subdomains of the other two subunits ( residues 262 to 349 and residues 481 to 564 ) are disordered . Detailed data collection and refinement statistics are listed in Table 1 . All structure figures were generated in PyMOL ( DeLano , 2002 ) . The cavity space within the ion conduction pore was analyzed in HOLLOW ( Ho and Gruswitz , 2008 ) . The full-length channel proteins purified in DM was reconstituted into lipid vesicles composed of a 3:1 ratio of 1-palmitoyl-2-oleoyl-phosphatidylethanolamine ( POPE ) and 1-palmitoyl-2-oleoyl-phosphatidyl glycerol ( POPG ) ( Avanti Polar Lipids , Alabaster , Al ) as described ( Heginbotham et al . , 1999; Alam et al . , 2007 ) using a dialysis solution containing 10 mM HEPES , pH 7 . 4 and 450 mM KCl . The reconstituted liposome samples were kept at −80°C in 100 µl aliquots . For 86Rb flux assays , a protein/lipid ratio of 10 μg/mg was used in the reconstitution . The 86Rb flux assay was performed following the same procedures as described ( Heginbotham et al . , 1998 ) . Liposomes were thawed and sonicated in a bath sonicator for 30 s before the assay . To remove extra-liposomal KCl , samples were passed through a pre-spun Sephadex G-50 fine gel filtration column ( 1 . 5 ml bed volume in a 5 ml disposable spin column ) swollen in 450 mM Sorbitol and 10 mM HEPES , pH 7 . 4 . Each tested nucleotide was added to a 30 μl aliquot of the liposomes collected after the buffer exchange step followed by the addition of 56 μl 86Rb flux buffer ( 450 mM Sorbitol , 10 mM HEPES , pH 7 . 4 , 50 μM KCl , and 5 μM 86RbCl ) . The final concentration of the nucleotide in the reaction mixture is 1 mM . After 5 min , this reaction mixture was passed through another pre-spun gel filtration column as described above to eliminate extraliposomal 86Rb . The final eluate was mixed with 10 ml scintillation cocktail and its radioactivity measured in a scintillation counter . The radioactivity of each sample was normalized against the control sample in which no nucleotide was added . Initial single channel recordings using lipid bilayers or giant liposome patching with vesicles reconstituted with the GsuK channel failed to detect channel activity . Low open probability and small single channel conductance can both contribute to the lack of channel activity in electrophysiological recording . Our recent study of the MthK channel demonstrated that mutations at Ala88 on the MthK inner helix can have dramatic effects on single channel conductance and open probability—replacing Ala88 with a larger hydrophobic residue such as Leu significantly reduces both whereas the opposite effect is seen with a negatively charged Asp residue ( Shi et al . , 2011 ) . The equivalent residue in GsuK is Leu97 , which we reasoned could potentially be the cause of low channel activity . Furthermore , Ca2+ binding at the assembly interface of GsuK also implies a potential gating role . To this end , we introduced an L97D mutation to enhance the channel conductance and/or open probability and also performed single channel recordings in the presence and absence of Ca2+ . As the results showed , the Ca2+ ion initially presented in the patching solutions is the main cause of low channel activity . For single channel recordings , a protein/lipid ratio 0 . 5–2 μg/mg was used in the reconstitution . Giant liposome was obtained by air drying 2–3 µl of liposome sample on a clean cover slip overnight at 4°C followed by rehydration in bath solution at room temperature . Patch pipettes were pulled from Borosilicate glass ( Harvard Apparatus , Holliston , MA ) to a resistance of 8–12 MΩ upon filled with the pipette solution containing 150 mM NaCl ( for recordings shown in Figure 6B , C ) or KCl ( for recordings shown in Figure 6E , F ) , 1 mM CaCl2 , and 10 mM HEPES , pH 7 . 4 buffered with KOH or NaOH . The standard bath solution contains 150 mM KCl , 10 mM HEPES , 1 mM EGTA , pH 7 . 4 buffered with KOH . A giga seal ( >10 GΩ ) was obtained by gentle suction when the patch pipette attached to the giant liposome . To get a single layer of membrane in the patch , the pipette was pulled away from the giant liposome and the tip was exposed to air for 1–2 s . Membrane voltage was controlled and current recorded using an Axopatch 200B amplifier with a Digidata 1322A converter ( Axon Instruments , Union City , CA ) . Currents were low-pass filtered at 1 kHz and sampled at 20 kHz . Only patches containing a single channel with Po>0 . 85 in the absence of Ca2+ and having its intracellular side facing the bath solution were used for further experiments . In the study of intracellular Ca2+ deactivation , various concentrations of Ca2+ were added to the bath solution . The free Ca2+ concentration in the range of 0–100 µM was controlled by mixing 1 mM EGTA with an appropriate amount of CaCl2 calculated using the software MAXCHELATOR ( http://maxchelator . stanford . edu ) . No EGTA was added in the bath solution for [Ca2+] above 100 μM . Most recordings were performed with symmetrical KCl except for the selectivity measurements in which KCl in the pipette solution was replaced by 150 mM NaCl . As the channel in liposome patch has its intracellular ligand binding gating ring facing the bath solution , the positive ( outward ) current is defined as the cation movement from the bath solution to the pipette . The presence of 1 mM CaCl2 in the pipette solution ( extracellular to the channel ) has no effect on channel open probability both in negative and positive voltages , which rules out the possibility of a slow blockade of the pore by Ca2+ .	Most cells are surrounded by a semipermeable membrane , and although this membrane allows very few molecules to pass through it , cells can use transmembrane proteins to overcome this barrier . Some of these proteins import glucose , amino acids and other nutrients into the cell , while others transport ions into or out of the cell . Ion transport across the cell membrane is essential for a wide variety of biological processes , including signal transduction and the generation of electrical impulses in nerve cells . The pores that allow ions to travel through the cell membrane are known as ion channels , and most channels allow only one type of ion—usually sodium , calcium or potassium ( K+ ) ions—to pass through them . There are many different types of ion channels and they are classified according to the type of ion they allow to pass through them , and by the gating mechanism that is used to open and close the channel . For example , ligand-gated K+ channels facilitate the passage of potassium ions and are opened and closed by ligands binding and unbinding to and from the channel . Most K+ channels are made up of four identical subunits , and in the majority of ligand-gated K+ channels in prokaryotes , each of these subunits will have one or two ligand-binding RCK domains ( where RCK stands for regulating the conductance of K+ ) . This is also true for some K+ channels in eukaryotes . While it is known that RCK domains are responsible for regulating the transport of potassium ions across the cell membranes of diverse organisms , little is known about the structure or gating mechanisms of K+ channels that are gated by more than one ligand . Kong et al . have studied a ligand-gated K+ channel called GsuK that has two RCK domains per subunit and is found in the bacterium G . sulfurreducens . They found that the opening process was mediated by a ligand that contains adenine , such as NAD+ or ADP , and the channel was closed by the presence of calcium ions . And by determining multiple crystal structures , Kong et al . were able to understand , from a structural point of view , how these ligands regulate this channel , and to propose a gating mechanism that is distinct from the mechanisms that are known to control other potassium channels .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2012	Distinct gating mechanisms revealed by the structures of a multi-ligand gated K+ channel
Perceptual decisions are classically thought to depend mainly on the stimulus characteristics , probability and associated reward . However , in many cases , the motor response is considered to be a neutral output channel that only reflects the upstream decision . Contrary to this view , we show that perceptual decisions can be recursively influenced by the physical resistance applied to the response . When participants reported the direction of the visual motion by left or right manual reaching movement with different resistances , their reports were biased towards the direction associated with less effortful option . Repeated exposure to such resistance on hand during perceptual judgements also biased subsequent judgements using voice , indicating that effector-dependent motor costs not only biases the report at the stage of motor response , but also changed how the sensory inputs are transformed into decisions . This demonstrates that the cost to act can influence our decisions beyond the context of the specific action . In laboratory experiments , participants are often asked to make decisions that are purely based on the features of the sensory input – a process that we refer to here as perceptual decision-making . However , in many of our daily situations , decisions are made in a behavioural context , in which the action that follow our decisions can differ dramatically in terms of required physical effort ( or the motor cost ) . For example , in the orchard , one may aim to pick the reddest-looking apple from the tree . Some of the apples may be hanging high-up on the tree , which will require more effort to pick compared to other fruits hanging on the lower branch . In such situations , does the difference in the motor cost between the options influence the decision of which fruit to pick ? If so , is such influence a result of serial integration between the perceptual decision ( i . e . decision based on the visual feature ) and the motor decision ( i . e . decision for action selection to avoid the effortful action ) at the output stage , or is the perceptual decision itself is affected by the cost on the downstream action ? It has been shown that physical effort is used in motor planning ( Huang et al . , 2012; Izawa et al . , 2008 ) , and the physical effort to obtain a reward can influence behavioural decisions ( Prévost et al . , 2010; Hosokawa et al . , 2013 ) . Moreover , the uncertainty in perceptual decisions is transmitted to the motor system , influencing the parameters of action control ( de Lange et al . , 2013 ) . However , it remains unclear whether the motor cost is simply integrated with the perceptual decision to optimise the expected utility ( Burk et al . , 2014; Cos et al . , 2014 ) , or whether the preceding experience of unequal motor costs can recursively influence the perceptual decision itself . Here , we show that manipulating the motor response cost for arm movements during a visual motion discrimination task changes not only the decision when responding with the arm , but also when reporting the perceptual decisions verbally . First , in Experiment 1 , we examined if the decision of the visual motion direction can be biased when one of the two responses requires more effort . Ten right-handed participants observed a moving random-dot stimulus and made decisions about the direction of motion ( leftward or rightward ) ( Britten et al . , 1992 ) . Participants held two robotic manipulanda , one in each hand . They indicated their decision by either moving their left hand ( indicating leftward decision ) or right hand ( rightward decision , Figure 1A ) . In the baseline phase , the resistance for moving the manipulanda was the same for both hands ( velocity-dependent resistance: 0 . 10 Ns/cm ) . In the subsequent induction phase , the resistance for the left hand increased by a small amount each time the participant moved the left hand ( 0 . 0008 Ns/cm; Figure 1B ) . Because the change was gradual , most of the participants did not report becoming aware of the increased motor cost when asked afterwards , even though their left hand was eventually exposed to 1 . 8 times greater resistance than the right ( 0 . 18 Ns/cm for left , 0 . 10 Ns/cm for right; see Materials and methods and Figure 1—figure supplement 1 ) . This procedure was employed to minimise any cognitive strategy participants may use , such as explicitly avoiding the costly hand response regardless of the decision about the visual stimulus . In the test phase , participant then continued to perform the visual discrimination task under the accumulated asymmetry in manual resistance . We plotted the proportion of rightward judgement against different stimulus intensities , and determined the point of subjective equality ( PSE , the point at which participants judge 50% of the trials to go rightward ) for both the baseline and the test phase ( Figure 1D ) . If the increased physical resistance for expressing leftward judgements was incorporated into the decision , the proportion of ‘leftward’ judgements should decrease in the test phase compared to the baseline , resulting in the shift of PSE towards the left ( Figure 1D ) . As expected , the PSE shifted towards the left from baseline to test phase ( −4 . 33% , paired t-test ( 2-tailed ) : t9 = 2 . 43 , p=0 . 038 , d = 0 . 76 , 8/10 individuals showed the effect ) ( Figure 1D–E , Figure 1—source data 1 ) . This indicates that the participants started to avoid making motion direction decisions in which the response is costly . 10 . 7554/eLife . 18422 . 003Figure 1 . Setup of the experiment and the shift of PSE induced by the motor cost . ( A ) Participants made 15 cm reaching movement to the target with their hand ( left or right ) , in response to the perceived direction ( left or right ) of the random-dot motion . ( B ) In all the experiments , the baseline phase and the test phase was interleaved by the induction phase , in which the resistance for one of the manipulandum movement gradually increased . ( C ) In Experiment 3 , the baseline and the test phase included both manual and vocal motion discrimination , each being serially presented within a 10 trial block . ( D ) Fitted psychometric function to the probability of a response towards the right in the baseline ( blue ) and the test ( red ) phase of a representative participant ( Experiment 1 ) . Negative motion coherence value indicates the leftward motion ( with manual resistance ) , and positive towards the right ( without manual resistance ) . ( E ) Shift of PSE from the baseline in Experiments 1 , 2 , 3 and 4 . Negative value indicates the PSE shift towards the motion direction with resistance ( i . e . decreased judgements towards the motion direction having resistance in their manual response ) . Error bars indicate standard error of mean across participants . Data for Figure 1E is available as Figure 1—source data 1 . p<0 . 05 , p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 18422 . 00310 . 7554/eLife . 18422 . 004Figure 1—source data 1 . Individual PSE shift for Experiments 1–4 , which is the data summarised in Figure 1E . DOI: http://dx . doi . org/10 . 7554/eLife . 18422 . 00410 . 7554/eLife . 18422 . 005Figure 1—figure supplement 1 . Example of the peak resistive force increase profile in the study . The presented data is force data from Experiment 1 , averaged across trials and participants in each session . Here , each session contains 66 trials ( 33 trials for each left and right ) of reaching movements . Since the resistive force is provided in a velocity-dependent manner , the variance is reflecting the variability of the movement velocity across participants . DOI: http://dx . doi . org/10 . 7554/eLife . 18422 . 005 In Experiment 2 , we examined whether motor cost and visual features need to be directly associated , or whether simply gaining experience of one action being more effortful than the other is sufficient to bias subsequent decisions . The baseline and the test phase involved judging direction of visual dot motion , as in Experiment 1 . However , the induction phase was now replaced with a simple reaching movement , in which the participants moved their left or right hand according to a simple leftward or rightward arrow presented in the centre of the screen . As in Experiment 1 , the resistance for moving the left hand was gradually increased . The motor cost during the induction phase was not associated with any motion direction judgement; the participants were only exposed to the gradually increasing motor cost differences between the two hands . Again , PSE significantly shifted leftwards in the test phase ( Figure 1E , Figure 1—source data 1; baseline vs . test; mean −4 . 26% , paired t-test ( two-tailed ) ; t8 = 3 . 91 , p=0 . 005 , d = 1 . 3 , 8/9 individuals showed the effect ) . This indicates that the direct association of higher motor cost with a specific decision during the induction phase is not critical for inducing the bias . This may suggest that the ( implicit ) knowledge about the response costs is sufficient to recursively influence the decision . Alternatively , these results could indicate that the bias is only transiently induced during the test phase itself . In Experiments 1 and 2 , we showed that manual motor costs reliably bias decisions that involve these manual response . Where in the process of translating a stimulus into a response does this bias arise ? A simple model posits that decision-making occurs in three sequential stages ( Gold and Ding , 2013 ) . First , features of the sensory input are extracted and encoded as a sensory representation . Second , a categorical decision is made based on this sensory representation ( decision layer ) . Third , the output from the decision layer is transmitted to the relevant effector for the response ( Figure 2 ) . One possibility is that the motor cost only biases the decision layer in the context of the specific response ( Oliveira et al . , 2010 ) ( Figure 2A ) . In other words , the motor cost only influences decisions when the participant anticipates to perform the action associated with the motor cost , thus , the decision simply takes into account the upcoming motor costs . Alternatively , the repeated exposure to the manual motor cost may affect the perceptual decision about this type of stimulus in general , no matter which effector is used to make a response ( Figure 2B ) ( Bennur and Gold , 2011; Filimon et al . , 2013 ) . Finally , the motor cost could also directly bias the sensory representation ( Figure 2C ) , affecting the initial encoding of the information before it is transmitted to the decision layer . Only in the two latter scenarios , should the bias observed during the manual decisions generalise to decisions expressed with a different effector . In Experiment 3 , we therefore examined whether a hand-specific motor cost could also influence a visual judgement that used a vocal response . A manual to vocal transfer of the motor cost effect would indicate that the motor cost influences the decision about the visual stimulus itself ( i . e . perceptual decision ) , not the decision coupled with the effector selection ( i . e . motor decision ) . 10 . 7554/eLife . 18422 . 006Figure 2 . Schematic diagram illustrating the process of perceptual decision making , and the possible influence of the motor cost on the decision process . Perceptual decision making consists of three different processing stages . First , the features of the sensory input are extracted and encoded as in the sensory representation . Second , the perceptual ( categorical ) decision is made based on this sensory representation ( decision layer ) . Finally , the decision is transferred to the response effector . The motor cost asymmetry during the manual response can affect the perceptual decision making process in several different ways . ( A ) The motor cost for the manual response may only bias the decision layer that involves this response , but leave the decision layer for different response effectors unaffected . If this is the case , the bias observed during the manual response should not generalise to the verbal response . ( B ) The motor cost may bias the decision layer in general or ( C ) the sensory representation directly . In either of the latter two cases , the effect of motor cost should be also observable during the response using the different effector . DOI: http://dx . doi . org/10 . 7554/eLife . 18422 . 006 Fourteen new participants performed the visual motion discrimination as in Experiment 1 . In the induction phase , we gradually increased the resistance for one of the hands while participants performed manual decisions as in Experiment 1 . The resistance was increased for half of the participants ( 7 ) on the left hand , and for the other half on the right hand , accounting for any hand-dependent effects . To analyse these left and right resistance increase data together , we aligned the data depending on the side of the resistance applied by assigning negative motion coherence level to the motion direction associated with the direction of the resistance . During the manual task , participants moved their left or right hand according to their perceived motion direction . For the vocal task , participants indicated the direction of the motion by vocally responding ‘left’ or ‘right’ ( Figure 3—figure supplement 1A ) without moving their hands . During the baseline and the test phase , participants alternated between tasks: each 10 trials of manual judgements were followed by 10 trials of vocal judgements ( Figure 1C ) . This ‘top-up’ procedure is commonly used to assess the effect of sensory adaptation on the subsequent perceptual judgements ( Fujisaki et al . , 2004 ) . If the bias induced by the motor cost is affecting the decision regardless of the response effector , the vocal decision should be also biased towards the same direction as the manual decision . For the manual task , the result of Experiment 1 was replicated . The exposure to the resistance made the PSE to shift significantly away from the stimulus associated with the costlier movement ( baseline vs . test , mean: −7 . 06% , paired t-test:t13 = 2 . 94 , p=0 . 012 , d = 0 . 78 , 12/14 individuals showed the effect ) ( Figure 1E , Figure 1—source data 1 ) . More importantly , for the vocal task , judgement also shifted to the same direction as the manual task ( baseline vs . test , mean: −3 . 00% , paired t-test:t13 = 2 . 44 , p=0 . 030 , d = 0 . 65 , 12/14 individuals showed the effect ) ( Figure 1E , Figure 1—source data 1 ) , even though the motor cost for the vocal responses was not manipulated . Since the direction of manual motor cost was counterbalanced across participants , this finding cannot be explained by any time-dependent drift of the decision towards one of the directions . This result suggests that the bias induced by the manual motor cost transfers to decisions expressed with other effectors . Although in Experiment 3 , the response effector differed between the manual and the vocal task , the abstract response code ( ‘left’/ ‘right’ ) remained the same between the two tasks . Therefore , it is possible that the manual motor cost got associated with these semantic labels , but did not necessarily influence the stimulus-based perceptual decision itself . To test this possibility , in Experiment 4 , we again examined the manual-to-vocal transfer of effect caused by the motor cost , but this time varied not only the response effector , but also the response codes between the two tasks . Twelve new participants performed visual motion judgements , where in the baseline and the test phase , manual decisions and the vocal decisions alternated in a mini-block of 11 and 7 trials , respectively ( Figure 3A ) . The induction phase involved only the manual task , with gradually increasing left hand resistance . As in Experiment 3 , the manual task was a left-right motion discrimination task . The vocal task , however , was changed to the motion detection task . Participants were asked to detect a near threshold coherent motion by vocally responding ‘yes’ or ‘no’ . The to-be-detected motion direction ( left or right ) was instructed at random before each trial ( Figure 3—figure supplement 1B ) . Half of the trials included left or right coherent motion , and in the other half , the coherent motion was absent ( 0% coherence ) . 10 . 7554/eLife . 18422 . 007Figure 3 . Trial structure of Experiment 4 and the effect of preceding motor cost experience on participants' motion judgements . ( A ) In Experiment 4 , participants made vocal judgements to a yes-no motion detection task , and manual judgement to a left-right discrimination task . ( B ) : Shift of the criterion of motion detection from the baseline during the vocal judgement task in Experiment 4 ( d' data is presented in the Figure 3—Figure Supplement 1C ) . Negative value indicates the shift towards more conservative criterion for the motion detection . ( C ) PSE shift from the baseline condition in Experiment 3 , plotted against the number of trials from the preceding manual judgements . Negative value indicates the shift of PSE towards the motion direction with resistance ( i . e . decreased judgements towards the motion direction having resistance in their manual response ) . ( D ) Vocal motion detection criterion differences between the leftward ( with manual response resistance ) and rightward ( without resistance ) motion ( Experiment 4 ) . The difference is plotted against the number of trials from the preceding manual judgements . Negative value indicates a more conservative criterion for leftward than for rightward motion . Error bars indicate standard error of mean across participants . Data for Figure 3B–D is available as Figure 3—source data 1 . p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 18422 . 00710 . 7554/eLife . 18422 . 008Figure 3—source data 1 . Individual criterion shift of the vocal trials ( Experiment 4; summarised in Figure 3B ) , individual PSE shift for the vocal trials across different trials ( Experiment 3; summarised in Figure 3C ) and individual criterion shift for the vocal trials across different trials ( Experiment 4; summarised in Figure 3D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18422 . 00810 . 7554/eLife . 18422 . 009Figure 3—figure supplement 1 . Task instruction of Experiment 3 and 4 , and the d' data of Experiment 4 . ( A , B ) The task instruction of the vocal task in Experiment 3 ( A ) and 4 ( B ) , where the participants were asked to vocally discriminate the motion direction ( Experiment 3 ) , or asked to vocally detect the motion for the instructed direction ( Experiment 4 ) . ( C ) Shift of the d-prime ( sensitivity ) of motion detection from the baseline during the vocal judgement task ( Experiment 4 ) . ( D ) Difference in d’ for leftward and rightward motion direction in the vocal judgement of Experiment 4 , plotted against the number of trials from the preceding manual judgements . Positive value indicates the higher sensitivity for the rightward motion and the negative for the higher sensitivity to the leftward motion . Error bars indicate standard error of mean across participants . DOI: http://dx . doi . org/10 . 7554/eLife . 18422 . 009 For the manual task , a significant shift of PSE was observed again , reflecting the avoidance of the costly decision ( baseline vs . test , mean: −4 . 39% , paired t-test: t11 = 2 . 88 , p=0 . 015 , d = 0 . 75 , 9/12 individuals showed the effect ) ( Figure 1E , Figure 1—source data 1 ) . For the vocal task , participants’ judgement criterion for leftward motion detection became more conservative after being exposed to the manual motor cost , which was not the case for the rightward motion . The interaction between the phase of the experiment ( baseline/test ) and the visual motion direction ( left/right ) was significant , ( F1 , 11 = 6 . 76 , p=0 . 025 , η2=0 . 36 ) ( Figure 3B , Figure 3—source data 1 ) . Since the manual task required a left-right decision and the vocal task a yes-no decision , the abstract response code of these two task were different . Therefore , significant manual-to-vocal transfer cannot be simply explained by the motor cost inducing a bias for choosing a particular type of abstract response label . Instead , the results indicate that the motor cost influenced the perceptual decision – that is the decision based on the feature of the visual stimulus input – itself . Together , these results demonstrate that the motor cost on the downstream response can recursively change how the input visual stimulus is transformed into the decision; at the level of sensory representation or at the decision layer . In contrast to the criterion , the sensitivity ( d’ ) for the motion detection did not change for either visual motion direction ( F1 , 11 = 0 . 44 , p=0 . 52 , η2=0 . 04 ) ( Figure 3—figure supplement 1C ) . This indicates that the motor cost did not increase or decrease the signal to noise ratio ( gain ) of the motion signal to one specific direction . Until now , we have shown that the motor cost can bias the decision based on a visual stimulus independent from the response effector or abstract response code . Finally , using a model-based approach , we tried to elucidate the processing stage in which the motor cost could have influenced the decision . We analysed both reaction time and choice data of the manual tasks ( Experiments 1 , 2 , 3 and 4; n = 45 ) under the framework of diffusion decision model ( DDM ) ( Ratcliff and McKoon , 2008 ) . The DDM postulates that a decision variable temporally accumulates sensory evidence in favour of one decision ( by increasing its value ) or in favour of the alternative decision ( by decreasing its value ) . When the decision variable hits a certain threshold level ( decision bound ) , the decision is made and the response is triggered ( Ratcliff and McKoon , 2008; Palmer et al . , 2005 ) ( Figure 4—figure supplement 1A ) . Under this framework , we examined whether the source of the manual decision bias that transferred to the vocal decisions occurred in the sensory representation of the stimulus that is accumulated ( Figure 2C; sensory representation ) or in the decision bound that is used to make the decision ( Figure 2B; decision layer ) ( Ratcliff and McKoon , 2008; Palmer et al . , 2005; Hanks et al . , 2006 ) . If the bias is introduced at the sensory representation stage , it would increase the input signal ( the perceived motion coherence ) in the easier direction . We exclude the possibility that the motor cost made the sensory representation of the preferred direction more accurate ( increased gain of the signal only in one direction ) as we did not observe the discrimination sensitivity change ( JND: just noticeable difference ) between the baseline and the test condition across different experiments ( t44 = 0 . 26 , p=0 . 77 ) . Rather assumed that the motor cost would shift evidence accumulation towards the easier direction ( sensory evidence model; Figure 4—figure supplement 1B ) . With this bias , the decision variable would drift towards the preferred decision even in the absence of any coherent visual motion . Alternatively , we considered the possibility that the motor cost changed the decision bounds ( decision layer ) , that is , the amount of evidence required for each of the choices . This change can be parsimoniously modelled as shift of the starting point of the accumulation process , which will consequently change the distance from the starting point to each decision bound ( starting point model; Figure 4—figure supplement 1C ) . The sensory evidence and starting point models predict qualitatively similar pattern of choice probabilities ( i . e . bias towards the direction to avoid the motor cost ) , but different pattern of decision times for correct trials across different motion intensities ( Methods , Figure 4—figure supplement 1B–C ) . Therefore , by comparing whether which of the two models explains our data better ( Hanks et al . , 2006; Ding and Gold , 2012 ) ( see Materials and method ) , we may infer the source of the bias . Additionally to the starting point model and the sensory evidence model , we also fitted a model that allowed for both shifts simultaneously ( full model ) . This allows us to directly compare the effect of each parameter . Also , to check whether the starting point or the sensory evidence shift was necessary to explain the data in the first place , we also prepared a baseline model which we did not model the starting point and/or the evidence accumulation shift , but only modelled the difference in non-decision time ( baseline model: see Material and methods ) . Note that , since we did not record the reaction time of the vocal decisions , this analysis was restricted to model the bias during the manual decisions . First , we fit each model to the average group data and compared the BIC weights by converting the Bayesian Information Criterion ( BIC ) for each model ( Wagenmakers and Farrell , 2004 ) . We then repeated this process 10 , 000 times , each time drawing 45 participants from our sample with replacement to obtain an estimate of the reliability of our conclusion . The results ( Figure 4C–D , Figure 4—source data 1 , Table 1 ) clearly indicate that the starting point model explained the data substantially better than the other models . 10 . 7554/eLife . 18422 . 010Table 1 . BIC and BIC weights calculated for different DDM . BIC and the BIC weights for different DDM models . Values calculated using the group averaged data , and the 95% confidence interval is calculated from the 10 , 000 bootstrap resampling is presented . DOI: http://dx . doi . org/10 . 7554/eLife . 18422 . 010Starting point modelSensory evidence modelFull modelBaseline modelBIC ( 95% confidence interval ) averaged346 . 09360 . 37348 . 76432 . 88upper bound433 . 94447 . 89441 . 07547 . 49lower bound335 . 04348 . 64337 . 62401 . 31BIC weight ( 95% confidence interval ) averaged0 . 79110 . 00060 . 20830 . 0000upper bound0 . 86000 . 43030 . 74450 . 0000lower bound0 . 06000 . 00000 . 13360 . 000010 . 7554/eLife . 18422 . 011Figure 4 . DDM parameter estimates and the fitted psychometric and chronometric function . ( A , B ) Histogram of individual starting point shift ( A ) and the evidence accumulation shift ( B ) calculated from the DDM ( full model ) . Black dotted line indicates the 0% point ( i . e . no effect ) , and the red dotted line indicates the median of the distribution ( i . e . amount of shift ) . Significant rightward shift of the starting point was observed ( median: 5 . 6% ) , whereas not for the evidence accumulation shift ( median: 1 . 39% ) . ( C , D ) Fit of DDM to the choice ( C ) and the decision time ( D ) data averaged across participants ( see Materials and methods and Figure 4—figure supplement 1 Panel C ) . Data for Figure 4A–D is available as Figure 4—source data 1 . p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 18422 . 01110 . 7554/eLife . 18422 . 012Figure 4—source data 1 . Individual starting point shift in the test phase from the baseline phase during the manual trials , estimated from the full-model DDM ( summarised in Figure 4A ) , individual evidence accumulation shift in the test phase from the baseline phase during the manual trials , estimated from the full-model DDM ( summarised in Figure 4B ) , data points consisting the psychometric function estimated from the DDM ( starting point model ) using the group averaged data ( summarised in Figure 4C ) and data points consisting the chronometric function estimated from the DDM ( starting point model ) using the group averaged data ( summarised in Figure 4D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18422 . 01210 . 7554/eLife . 18422 . 013Figure 4—figure supplement 1 . Schematic diagram explaining the drift diffusion model ( DDM ) and the simulated choice and decision time data . ( A ) DDM model postulates that a decision is transformed into action when the evidence favouring one of the choices has been accumulated to a certain threshold level ( decision bound ) ( left panel ) . The model makes a prediction about the pattern of choice probability and the decision time in respect to the strength of the motion signal ( right panel ) . For the baseline , the starting point of the evidence accumulation is set to 0 . ( B ) When there is more sensory evidence in favour of rightward motion ( red line ) , the drift speed for the rightward decision will increase ( left panel ) , and left would decrease . As a result , a rightward decision becomes more likely ( shift of psychometric function ) and the decision time pattern generally shifts to the left , showing a tendency to respond faster for the rightward motion ( right panel ) . If the motor cost influences the sensory representation ( Figure 2C ) , we would expect this pattern of results ( sensory representation model ) . ( C ) A shift in the starting value of the accumulation process induces a prior bias towards a rightward decision , decreasing the required amount of evidence for rightward decision compared to the left ( left panel ) . This will again bias the decision to favour the rightward decision . Instead of shifting the pattern of decision times to the left ( as in B ) , the starting point model predicts an additional offset to the rightward and leftward decision time; shorter for the rightward decision and longer for the leftward decision . In this model , the bias therefore arises from a change in the decision layer transforming the sensory representation into the decision ( Figure 2B ) , while the sensory evidence itself is not changing . DOI: http://dx . doi . org/10 . 7554/eLife . 18422 . 01310 . 7554/eLife . 18422 . 014Figure 4—figure supplement 2 . Change of the correct RT , error RT and the correct rate from the baseline to the test phase . ( A ) simulated data from the starting point model , ( B ) simulated data for the evidence accumulation model , ( C ) actual experimental data ( n = 45 ) . Left and right indicates the stimulus motion direction . The left decision was associated with the increased resistance during the test phase . Simulation parameters are based on the estimated parameters from the actual data ( Table 2 ) . Error bars in C indicate standard error of mean across participants . DOI: http://dx . doi . org/10 . 7554/eLife . 18422 . 014 Second , we compared the model parameter of the full model fitted to each participants’ individual data . Consistent with superior fit of the starting point model , we found a significant shift of the starting point ( median; 5 . 6% , signed rank test: z44 = 2 . 50 , p=0 . 01 , d = 0 . 32; Figure 4A , Figure 4—source data 1 ) , but no significant change in the evidence accumulation ( median; 1 . 38% , signed rank test: z44 = 1 . 15 , p=0 . 25 , d = 0 . 21; Figure 4B , Figure 4—source data 1 ) . Therefore , our data suggest that the motor cost biased the decisions by changing the decision layer ( starting point ) that transforms the input signal into the decision . The DDM also contains a parameter that captures the motor response time that is independent from the decision ( non-decision time; see Material and methods ) . In the baseline phase , there was no significant difference between the non-decision time of the hands ( 462 . 8 ms vs . 468 . 8 ms , paired t-test: t44 = 0 . 94 , p=0 . 34 , d = 0 . 08 ) . However , in the test phase , although the non-decision time became shorter for both the hands without resistance ( dTA: −7 . 69 ms ) and the hand with resistance ( dTB: −27 . 7 ms ) , the decrease was larger for the hand with resistance ( paired t-test: t44 = 2 . 69 , p=0 . 01 , d = 0 . 64 ) . This finding resembles a previous report ( Ding and Gold , 2012 ) , which demonstrated that the stimulation of the caudate neuron with visual motion directional tuning biased monkey’s motion direction judgements towards the neuron’s tuned direction , but at the same time decreased the estimated non-decision time for the response ( eye movement ) to the opposite ( non-stimulated ) direction . While the DDM models were only fit to the choice probabilities ( psychometric function ) and the RT function of the correct trials ( chronometric function , see Material and methods ) , we also checked whether the models could predict the patterns of RTs on error trials . For this purpose , we simulated individual trials using the estimated group parameters based on the starting point and the sensory evidence models ( Table 2 ) . For the correct trials , both model simulations showed similar tendencies; the RT reduced for the non-costly motion stimulus compared to the costly stimulus ( Figure 4—figure supplement 2A–B left panel ) . The pattern of the error trial RT differed between the two simulations . For the starting point model , error RTs were shorter for the costly motions ( non-costly decision ) , whereas the pattern was opposite for the sensory evidence model ( Figure 4—figure supplement 2A , B ) . This is because , for the former , the distance between the starting point and the non-costly decision bound decreases , whereas for the latter , the drift rate increases towards the error decision direction for the costly stimulus ( i . e . non-costly decision ) ( Mulder et al . , 2012 ) . The pattern of RTs for the experimental data ( Figure 4—figure supplement 2C ) was qualitatively similar to that of the starting point model . Therefore , the general pattern of error RTs supported our claim that the motor cost induced a starting point shift . 10 . 7554/eLife . 18422 . 015Table 2 . Parameter estimates for data from Experiment 1-4 for the starting point and sensory evidence model . *Parameters for the simulations for the error RTs were chosen to be these fitted parameters . DOI: http://dx . doi . org/10 . 7554/eLife . 18422 . 015DDM parameterskABT01 ( dTA ) T02 ( dTB ) spdcohstarting point model0 . 2911 . 7612 . 02460 ( −9 ) 459 ( −28 ) −1 . 620sensory evidence model0 . 2911 . 5912 . 20460 ( −17 ) 458 ( −22 ) 04 . 00 We showed that participants incorporate the cost of the response into the perceptual decision , and flexibly changes the way of interpreting the sensory environment . To further investigate the temporal dynamics of these flexible changes , we examined how the induced bias developed over the course of the 10 ( Experiment 3 ) or 7 ( Experiment 4 ) trials of vocal decisions following a series of manual decision trials ( see detail in the Materials and methods ) . In Experiment 3 , on average , PSE shift was slightly stronger for the vocal trials that immediately followed the manual task ( mean shift of PSE; −3 . 51% for the first five trials ( initial point of Figure 3C ) , >−3% for the rest , Figure 3—source data 1 ) , although this time dependence did not reach significance ( F5 , 65=0 . 60 , p=0 . 75 ) . In Experiment 4 , the strength of bias ( criterion shift ) significantly decayed depending on the number of trials from the manual task , showing a stronger bias in the first 4 out of 7 trials ( initial two time points of the curve . Figure 3D , Figure 3—source data 1 , F4 , 44 = 2 . 70 p=0 . 042 , η2=0 . 2 ) . These results indicate that the biasing effect may be relatively short-lived . However , the time scale of the retention is comparable to common perceptual adaptations , such as motion aftereffects . Mather et al . ( Mather et al . , 2008 ) shows that on maximum , the motion aftereffect lasts for 10 ~ 15 s . A single trial of our task takes at least , 3 ~ 4 s , so our effect lasted for 9 ~ 16 s . Therefore , our results indicate that , in the absence of any further confirmatory evidence of asymmetric response costs for the decision , the brain readapts relatively quickly to the new situation , resembling other examples of spontaneous decay in perceptual and motor adaptation phenomena ( Mather et al . , 2008; Webster , 2011; Smith et al . , 2006 ) . This shows that while perceptual decisions can be updated relatively quickly and flexibly , they do exhibit a substantial memory of past motor costs . In this study , we showed that visual motion direction decisions can be biased by the cost of the action that is used to report the decision ( the motor cost ) . Moreover , we demonstrated that the motor cost indeed affects the decision about the input stimulus identity , and not only the decision about which action to select . Previous behavioural studies have shown that the perceptual decisions can be biased by changing the frequency ( expectation ) of stimulus presentation , or by manipulating the response-reward association for the correct/incorrect decisions ( Mulder et al . , 2012; Whiteley and Sahani , 2008 ) . Here , we demonstrate that motor costs associated with the response can also bias the perceptual decisions , even when the response is made with a completely different effector that is not associated with increased motor cost ( here , verbal instead of manual responses ) . Therefore , our study provides evidence that the cost on the response for perceptual decisions , which has been regarded as downstream output channel of the decision , can recursively influence the decision of the input stimulus itself . In other words , the observed bias not only reflects the serial integration of the transient motor cost into the ongoing decision ( Burk et al . , 2014 ) , but represents a more global change in the way of transforming the sensory input to the decision , by taking the prior experience of the motor cost into account ( Makin et al . , 2010 ) . Congruent with our results , a recent study has shown that asymmetric biomechanical costs induces a bias into the decisions , and that this bias cannot be explained by strategically choosing the easier option when perceptual uncertainty is high ( Marcos et al . , 2015 ) . The critical contribution of our study is to show that this influence on the decision process is not limited to the judgements involving the asymmetric motor cost , but generalised to judgement using vocal responses without motor cost manipulation . The present study parsimoniously shows that the motor cost influence is not simply due to the bias of decision at the motor preparation/execution stage . Our DDM analysis readily explains the effect of motor cost as the change in the required amount of input evidence ( i . e . shifts of the starting point ) for the decision . Even though only fitted to the reaction times for correct trials ( Palmer et al . , 2005 ) , this model also correctly predicted the RT pattern for error trials . These findings are consistent with previous literature that shows that increasing the presentation frequency or amount of reward for one of the choice biases decision in a way that can be best modelled as a starting point shift ( Mulder et al . , 2012 ) . Indeed , it has been suggested that shifting the starting point of accumulation process is the optimal solution to account for such contextual changes ( Bogacz et al . , 2006; Simen et al . , 2009 ) . It should be noted , however , that alternative models involving collapsing bounds may perform better in situations in which the stimulus strength varies randomly ( Hanks et al . , 2011; Tajima et al . , 2016 , but also see Hawkins et al . , 2015 ) . Electrophysiological studies have shown that the electrical stimulation of parietal or basal ganglia neurons can bias perceptual decisions ( Hanks et al . , 2006; Ding and Gold , 2012 ) . These effects were explained by shifts of the starting point in the DDM framework ( or equivalently the decision bounds ) , thus the change in the decision layer of the perceptual task . Therefore , these brain regions are likely candidate neuronal substrates where the motor cost interacts with the sensory input to bias the perceptual decision . Neurons in the lateral intraparietal area ( LIP ) code the feature that is relevant for the visual decision , independent of response type ( Bennur and Gold , 2011 ) . Experiments 3 and 4 similarly suggested motor-induced changes for the perceptual decision is independent of the effector used for response . The subcortical network in the basal ganglia has been suggested to represent the cost of action or the ‘vigour’ of movement initiation ( Mazzoni et al . , 2007 ) . Prolonged exposure to altered motor costs during perceptual decisions may similarly change the response properties of these areas , altering how the system judges the sensory evidence from the environment . How does our current finding relate to the existing theories of perceptual decision making ? One of the recent theories is the intentional framework ( Shadlen et al . , 2008 ) . This framework posits that decisions and actions are tightly coupled , with each decision maker separately accumulating the sensory evidence until the threshold level for the specific action is reached . In this scenario , any decision bias induced by imposing a motor cost to a specific action would not transfer to a decision performed by a different action , as there is no explicit communication between the multiple decision makers . Thus , our results indicate that perceptual decisions are either made centrally by a high-order process that is common across different actions ( Filimon et al . , 2013; Sun and Landy , 2016 ) , or at least that different local decision makers exhibit a certain degree of mutual dependency , such as a shared cost ( value ) of the input stimuli ( external environment ) . In conclusion , we demonstrate that the motor cost involved in responding to a visual classification task is integrated into the perceptual decision process . Our everyday perceptual decisions seem to be solely based on the incoming sensory input . They may be , however , influenced by the preceding history of physical cost of responding to such input . The cost of our own actions , learned through the life-long experience of interacting with the environment , may partly define how we make perceptual decisions of our surroundings . A total of 52 participants ( Experiment 1: 12 ( 6 females ) , Experiment 2: 10 ( 5 females ) , Experiment 3: 16 ( 8 females ) , Experiment 4: 14 ( 5 females ) ; with ages ranging from 18 to 38 years ( M = 25 . 5 ) participated in the study . All had normal or corrected-to-normal vision , were right-handed and naive regarding the experimental purpose . None of them declared any history of neurological diseases . All participants gave informed written consent , and all procedures were approved by the UCL ethics committee . No statistical test was run to determine sample size a priori . The chosen sample sizes are similar to those in previous publications related to perceptual decisions ( Hagura et al . , 2012; Rahnev et al . , 2011 ) . Furthermore , we replicated the result of Experiment 1 in the subsequent Experiments 2 , 3 and 4 using the similar sample sizes . Participants were seated comfortably in front of a virtual environment setup , which has been described in more detail previously ( Reichenbach et al . , 2013 ) . The visual stimulus was presented on the display , which was mounted 7 cm above the mirror . The mirror was mounted horizontally above the manipulanda , preventing direct vision of the hands but allowed participants to view a visual scene on the monitor . During the task , participants leaned slightly forward with their forehead supported by a forehead rest , maintaining the distance from the eye and the mirror constant ( 25 cm ) . As a result , the viewing distance from the eye to the monitor was 32 cm . The chair was placed at the position where the participants could most comfortably perform the reaching movement using the manipulanda . Depending on the judgement of the visual stimulus ( see below ) , they made 15 cm straight reaching movements while holding onto a robotic manipulandum ( update rate 1 kHz , recording of position and force data at 200 Hz ) using their left or right hand ( Figure 1A ) . The hand positions were represented by white circles ( cursors , 0 . 3 cm diameter ) located vertically above the real positions of the hands . The movement were executed from a starting box ( unfilled white squares , 0 . 5 cm size , 6 cm to the left and right from body midline ) to a target box ( unfilled white squares , 1 cm size ) . In the centre of the screen , random-dot motion stimulus was presented ( Britten et al . , 1992 ) ( Figure 1A ) . In a 9 deg diameter circular aperture , dots were presented in a density of 1 . 7 dot/deg2 . The speed of the dots was 10 deg/s . For each trial , 0% , 3 . 2% , 6 . 4% , 12 . 8% , 25 . 6% , or 51 . 2% of the dots moved coherently to the left or the right . All other dots moved in a random direction , picked for each dot separately between 0 and 360 deg . There were three phases in the main task; the baseline phase , induction phase and the test phase . Participants performed the same motion direction judgement throughout the experiment , but the resistive force they were exposed during each of the hand movement was different between the phases . Each phase consisted of 5 , 15 , 5 blocks of trials , respectively . Each block contained 66 trials , and each the 11 movement direction x coherence level combinations was repeated six times in each block . The resistive force ( f ) was velocity dependent , calculated as from the equation;[fx;fy]= −α[vx;vy] where v denotes for the movement velocity , and α denotes for the coefficient of the viscosity ( Ncm/s ) . Here , negative value indicates the force against the movement direction . In the baseline phase , the coefficient was set to 0 . 10 Ns/cm for movements of either hand . In the induction phase , the resistance increased by 0 . 0008 Ns/cm for each left hand movement . The strength increased until the coefficient reached 0 . 18 ( Ns/cm ) , and this value remained for the rest of the induction phase and the test phase ( Figure 1B; upper panel ) . Progression of the actual peak value of the resistive force ( N ) through the experiment is presented in Figure 1—figure supplement 1 . Our aim of gradually increasing the resistance force was to make the resistance implicit as possible to the participants , avoiding any cognitive strategy to be involved when performing the task . After the experiment , we assessed their awareness using three questions . We first asked participants 'whether they had realised any change in the task during the experiment' , and then more explicitly ‘whether they had realised that the resistance increased for either of the hand’ . Only the participants clearly stating ‘no’ to both of the questions were included in the analysis . Two participants who clearly realised the increased left hand resistance ( answer ‘yes’ to both of the questions ) were excluded from the analysis . The same procedure was adopted for the remaining experiments; 1 , 2 , and 2 participants were excluded from the analysis of Experiments 2–4 , respectively . Movement onset was defined as the point when the movement velocity exceeded 2 . 5 cm/s . Reaction time was defined as the time elapsed between the onset of the visual stimulus to movement onset . Reaction times smaller than 100 ms or larger than 850 ms were excluded from the analysis , since the former decision is unlikely to be based on the visual motion , and the latter is likely to be made after the stimulus disappearance . Movement end was defined as movement velocity falling below 2 . 5 cm/s . If both hands moved , the hand with the larger movement amplitude was taken as the participant’s decision ( leftward or rightward ) on that trial . Probability of detecting a slight movement for the non-judged hand was 1 . 9% of all the trials across four experiments . When the movement amplitude of each judged and the non-judged hand was calculated , in Experiment 1 , the average amplitude was 15 . 4 cm and 3 . 2 cm , respectively . For the movement of the non-judged hand , the movement amplitude that exceeded 2SD from the mean movement amplitude was five trials for one participant , one trial for two participants and 0 trial for the rest of the participants . We confirmed that excluding these few trials did not affect any of the subsequent analysis performed on this data . Same confirmation was also done for Experiment 2 ( judged; 13 . 93 cm , non-judged; 0 . 23 cm; above 2SD: three trials for one participant , two trials for one participant , one trial for one participant ) , Experiment 3 ( judged; 15 . 5 cm , non-judged; 2 . 1 cm; above 2SD: 0 trial for all of the participants ) and Experiment 4 ( judged; 15 . 3 cm , non-judged; 2 . 6 cm; above 2SD: 0 trial for all the participants ) . For each participant , the percentage of ‘right’ judgement responses for each visual motion coherence level was calculated . Logistic regression was used to describe the function of participant’s response against the motion strength . The point of subjective equality ( PSE ) , i . e . the motion coherence level at which the participant answered ‘rightward’ in 50% of the trials was estimated from each regression . This was done independently for the baseline and the test phase , and the PSEs between the two phases were compared using paired t-test ( two-tailed ) . As for all other statistical comparisons , Levene’s test was used to confirm the equality of variance before performing this statistical comparison . We additionally confirmed that all the results from the above parametric tests can be replicated by the non-parametric Wilcoxon’s signed rank test , which shows that our data is not biased by the particular statistical test used to assess the results . The experiment was largely similar to Experiment 1 , except that during the induction phase , reaching movements were not associated with any visual motion judgements . Participants were required to make a left or right hand reaching , according to the arrow presented in the centre of the screen , which pointed either to the left or to the right . Otherwise , the procedure was equivalent to Experiment 1 . The structure of the experiment was similar to Experiments 1 and 2; where baseline phase was followed by an induction phase , and finally with the test phase . In the induction phase , the resistance for one of the hands slowly increased while the participants performed the manual perceptual decision about the dot motion ( moving left hand for leftward motion , and right hand for right motion ) . The resistance increased on the left hand for half of the participants ( 7 ) and on the right hand for the other half , aimed to account for any hand-dependent effects . For the dot motion , 10 different strengths ( i . e . coherence level of the motion ) were used ( ±3 . 2% , ±6 . 4% , ±12 . 8% , ±25 . 6% , ±51 . 2%; negative value indicates leftward motion , and the positive for the rightward motion ) . The induction phase involved 14 blocks of 60 trials each . During the baseline and the test phase , participants alternated between responding to the visual motion manually ( manual task ) or vocally ( vocal task ) . During the manual task participants moved their left or right hand according to their perceived motion direction . For the vocal task , participants were asked to indicate the direction vocally without moving their hands . The vocal task started with a tone . After 1500 ms , a random-dot motion stimulus was presented for 500 ms . Participants were asked to judge whether they perceived a motion direction towards the left or to the right , by vocally answering ‘left’ or ‘right’ . Their response was recorded by the experimenter . Each 10 trials of manual judgements were followed by 10 trials of vocal judgements ( mini-block; Figure 1C , Figure 3—figure supplement 1A ) . Within a mini-block , the manual and the vocal tasks were presented serially , and this structure was repeated four times within a block ( in total , 80 trials per block ) . Participants performed four blocks each for the baseline phase and the test phase . The analysis of the manual task was similar to the above experiments . To analyse left and right resistance increase data together , we aligned the data depending on the side of the resistance applied , by assigning negative motion coherence level to the motion direction associated with the direction of the resistance . This is equivalent to converting the right resistance increase data to the left resistance increase data; which was the case for Experiments 1 and 2 . The vocal task was analysed similarly to the manual task , in which the PSE between the baseline and the test phase was compared . To examine the time-dependent effect of the manual motor cost onto the vocal decision , we analysed the vocal task data depending on the number of trials from the last manual trials . To obtain enough trials for the analysis , we calculated the PSE using the first five trials of a mini-block of 10 vocal trials . This procedure was repeated by shifting the window , resulting in six analysis ranges ( 1st–5th , 2nd–6th , 3rd–7th , 4th–8th , 5th–9th , 6th–10th trials ) . Finally , PSE of the vocal task during the baseline phase was subtracted from these six values . These values will indicate the change in the strength of the influence of manual motor cost on the vocal decisions over time . We performed a one-way ANOVA to examine this temporal change . The procedure was generally similar to Experiment 3 , with the main difference that the vocal task was a motion detection task , rather than a motion discrimination task as used for the manual task . Also , as in Experiments 1 and 2 , the resistance increased for the left hand during the induction phase . The vocal task started with a tone , and the to-be-detected motion direction ( left or right ) was instructed ( Figure 3A , Figure 3—figure supplement 1B ) . After 1500 ms , a random-dot motion stimulus was presented for 500 ms . The stimulus included either the near threshold level coherent motion towards the instructed direction or a 0% coherent random-dot motion . Participants were asked to judge whether they perceived a coherent motion towards the instructed direction or not , by vocally answering ‘yes’ or ‘no’ . Their response was recorded by the experimenter . The strength of the motion coherence was defined individually before the experiment to approximately match 75% correct rate , causing the percentage correct rate to be between 65% and 85% during the baseline phase of the task . The aim of the vocal task was to examine whether the bias induced by the motor cost would transfer to the judgement using the different response effector . Additionally , the task was designed such that the abstract response code of the manual task ( left-right ) would be unrelated that of the verbal task ( yes-no ) . Therefore , the performance of the vocal task could not be biased by the manual task through its commonality of the effector or response code . Eleven trials of manual task were followed by seven trials of vocal task ( Figure 3A ) . This combination of manual/vocal task mini-block was repeated for four times in a block . Therefore , the manual and the vocal tasks were performed serially , similar to Experiment 3 . Participants performed six blocks , during each of the baseline phase and the test phase . The induction phase was similar to that of Experiment 1 , and contained only the manual tasks , 13 blocks of 66 trials each . The analysis of manual task was identical to the above experiments . For the vocal task , responses for left motion trials and right motion trials were analysed independently . Any trial in which a hand movement was detected during vocal task was excluded from the analysis . For each motion direction , the sensitivity ( d’ ) and the bias ( criterion; C ) were calculated using signal detection theory ( Macmillan and Creelman , 2005 ) . Difference of these measures between the baseline and the test phase were compared between the leftward and the rightward motion using two-way ANOVA ( phase ( 2 ) x motion direction ( 2 ) ) . We found that the criterion ( C ) for the leftward motion became more conservative after exposed to increased resistance on the left hand ( Figure 3B ) . As in Experiment 3 , we also examined whether the strength of this effect decayed as a function of the number of trial since the last manual trial . We calculated the d’ and C for both leftward and rightward motion using the first three trials of the vocal judgement in each mini-block ( 3trials x 24 mini-blocks = 72 trials ) . We repeated this procedure by shifting the window by one trial , resulting in the five analysis ranges 1st~3rd , 2nd~4th , 3rd~5th , 4th~6th , 5th~7th trial . Then , the values calculated similarly for the baseline ( d’ and C , each for leftward and rightward motions ) were subtracted , to calculate the change from the baseline condition . Finally , to test for a difference in bias for left and right motions , the left values were subtracted from the right values [i . e . negative values indicate less sensitive ( or more conservative ) judgements for leftward motion] . Intuitively , this shows how the leftward or rightward bias of d’ and the C changes over time as the temporal ( trial ) distance increases from the preceding manual judgement trials . We performed a one-way ANOVA to examine the temporal change of the left-right bias during the vocal task . Data of the manual judgement task from Experiments 1 , 2 , 3 and 4 ( n = 45 ) were re-analysed together under a framework of Diffusion Decision Model ( DDM ) ( Ratcliff and McKoon , 2008 ) , to examine the possible source of the decision bias; whether it is ( 1 ) increasing the sensory evidence favouring one of the decision ( change in the sensory representation; Figure 2C , Figure 4—figure supplement 1B ) , or ( 2 ) shifting the starting point of the evidence accumulation process more near to one of the decision bounds ( equivalent to changing distance to each of the decision bounds ) ( change in the decision layer; Figure 2B , Figure 4—figure supplement 1C ) . For this , reaction time and the choice data of both baseline and the test phase was simultaneously modelled with the DDM , and the estimated parameters were evaluated ( Palmer et al . , 2005; Hanks et al . , 2006; Ding and Gold , 2012 ) . Since we did not obtain the reaction time for the vocal trials , only the manual decision trials across different experiments were analysed . We analysed only the data for non-zero motion coherence level ( Experiment 3 did not have 0 coherence level condition ) and for the RTs for the correct decision trials , which are established to be well explained by the DDM ( Palmer et al . , 2005; Shadlen et al . , 2006 ) . The sign of the data from the participants having resistance on the right hand was flipped ( Experiment 3 ) , allowing the data to be analysed together with the left hand resistance increased participants . For the baseline phase , the model had five basic parameters; A , B , k , T01 and T02 . In this framework , momentary motion evidence is drawn randomly from a Gaussian distribution N ( μ , 1 ) , where μ is calculated as a motion strength ( coherence level: Coh ) scaled by the parameter k: μ= k×Coh . Decision is transformed into action when the accumulated momentary motion evidence reaches either of the decision bound; A ( right decision ) or –B ( left decision ) . Here , leftward decision is the one with the higher resistance for the response . Decision time is defined as the elapsed time between the stimulus onset and the time when the evidence reached either of the decision bound ( Figure 4—figure supplement 1A ) . Reaction time is the sum of decision time and the non-decision time ( T01 for a left and T02 for a right judgement ) , where the non-decision time is a pure action processing time that is assumed not to depend on the amount of the sensory evidence . The expected value of rightward judgements across different coherence levels can be calculated as ( Palmer et al . , 2005 ) :e2μB−1e2μB−e−2μA . The average decision time for the rightward motion decision can be described as:A+Bμcoth⁡ ( μ ( A+B ) ) − Bμcoth⁡ ( μB ) , and for the leftward motion decision as:A+Bμcoth⁡ ( μ ( A+B ) ) −Aμcoth⁡ ( μA ) To explain the change in decision bias observed between the baseline and test phase within the same model , additional parameters that describe the change in the parameters across two phases ( baseline and test ) were added to the above five base parameters ( delta parameters ) . Three different models with different delta parameter settings were generated . In the first model ( sensory evidence model ) , the motor cost changed the sensory evidence by changing the motion coherence by dcoh . Thus , the motion strength in the test phase was μ= k× ( Coh+dcoh ) . Since we know that the effect of motor cost does not change the discrimination sensitivity ( just noticeable difference: JND ) , but changes only the PSE ( Figure 3C ) , change in the sensory evidence is modelled as addition to the input stimulus ( +dcoh ) , rather than as the change in the gain itself ( direct change of k ) . In the second model ( starting point model ) , parameter that indicates the shift of the starting point of the accumulation processes ( sp ) was added , which will consequently change the amount of evidence required for each decision . Equivalently , we can think of this parameter as a shift in the two decision bounds to [A-sp] and [–B-sp] , leaving the distance between the two bounds fixed . In the final model ( full model ) , both coherence level change ( dcoh ) and the starting point shift ( sp ) were added as additional parameters . In all models across the three models , we also modelled the difference in the non-decision time between the baseline and the test phase . There were 840 ( Experiment 3 ) ~990 ( Experiment 1 ) trials of reaching movement between the baseline and the test phase , and the reaction time is decreased in the test phase compared to the baseline phase regardless of the motion coherence level ( F ( 1 , 35 ) =11 . 95 , p=0 . 0015 , η2=0 . 255 ) . We assume that this was due to the reduction of the non-decision time induced by the repetition of the reaching movement . To account for this , we added an additional parameter modelling the decrease of the non-decision time across the two hands . Since such reduction of the non-decision time may differ between the left and the right hand , the difference was modelled separately for the right ( dTA ) and the left ( dTB ) . Therefore , the non-decision time for the test phase was modelled as T01- dTA and T02- dTB for right and left , respectively ( same model as ref 17 ) . As a result , the three DDM models consisted of 8 ( sensory evidence model; five basic parameters + dTA + dTB + dcoh ) , 8 ( starting point model; five basic parameters + dTA + dTB + sv ) , and 9 ( full model; five basic parameters + dTA + dTB + sv + dcoh ) parameters , respectively . In addition to these three experimental models , we also prepared a baseline model , in which we fit the baseline and the test phase data only with the delta parameter of non-decision times ( seven parameter baseline model; five basic parameters + dTA + dTB ) . The DDM we used in this study is the most basic one proposed by Palmer et al . ( 2005 ) . This simple version of the DDM predicts the choice probabilities ( psychometric function ) and the mean RT function ( chronometric function ) of the correct trials . Therefore , this model is sufficient to distinguish between the models of interest – a change in starting point of evidence accumulation ( Figure 4—figure supplement 1B–C ) . A number of extensions to the DDM framework have been proposed to explain the full RT distributions of correct and incorrect trials using trial-by-trial variability of the drift rate ( Ratcliff and McKoon , 2008 ) or by incorporating the time-dependent decision bounds ( Drugowitsch et al . , 2012 ) . While these extensions are important , they do not change the primary predictions regarding the mean RT and choice probabilities under the two models . For the sake of parsimony , we therefore use the simpler model here . The group-averaged reaction time and choice data of the experiments was first fitted by each of the four models ( three models + baseline model ) , by searching the parameters that minimised the negative log likelihood of the fit ( maximum likelihood estimate ) . We used the group-average data , as each individual had a limited number of trials , and the noise level was rather high . This can induce a bias towards more complex models , as it can over fit the noise . Using group-average data strongly attenuates this effect ( Donchin et al . , 2003; Thoroughman and Shadmehr , 2000 ) . To obtain estimates of the reliability of the group-average fit , we resampled the data 10 , 000 times across participants with replacement , and fit the model to each of the averaged resampled data ( Efron , 1979 ) . To select the best model to explain the data from the above four , the Bayesian Information Criterion ( BIC ) ( Schwarz , 1978 ) was calculated for each model , BIC=−2log⁡L+ αlog ( n ) where logL denotes for log likelihood of the fitted function , α for number of parameters used for the fit and n for number of data points in the sample . The latter term in the BIC equation penalises the number of parameters used for the fit . Therefore , smallest BIC among the three models will indicate the most parsimonious model . To compare the explanative power between each model in an intuitive way , we converted the BIC values to the BIC ( Schwartz ) weights ( Wagenmakers and Farrell , 2004 ) , which expresses the explanatory power of BIC values into ratios among the candidate models . w ( i ) =exp⁡{−1/2△BIC ( i ) }∑k=1K ( exp⁡{−1/2△BIC ( k ) } ) where K is the number of models used to explain the data , △BIC ( i ) is the difference in BIC from the model with the smallest ( best ) BIC . The descriptive statistics ( averaged and the 95% confidence interval of the 10 , 000 bootstrap ) of BIC value and the BIC weight distributions are summarised in Table 1 . We also estimated the delta parameters ( dcoh , sp ) of the full model for each individual – thereby avoiding possible biases in the parameter estimates when using averaged data ( Estes and Maddox , 2005 ) . The parameters were statistically tested against zero ( no significant change in the test phase compared with the baseline phase ) using a Wilcoxon’s signed rank test . The impact of sp depends of the distance between the two decision bounds . Therefore , we normalised the individual starting point shift ( sp ) by the estimated distance between the two decision bounds ( sp/[A + B] ) . The parameters of the DDM models were estimated using the proportion of correct decisions and the RT data for correct trials . To test whether this model could also capture the pattern of error RTs , we simulated single trial data from the starting point and the sensory evidence models ( 10 , 000 times for each stimulus strength per condition ) , using the parameters estimated from the group data ( Table 2 ) . In both models , the leftward judgements is costlier in the test phase . For each simulation , the RT difference between the baseline and the test phase for both correct and error trials was calculated , separately for the leftward and rightward motion stimulus . We also calculated the difference in the correct rates . Then we compared these patterns with the actual experimental data analysed in a same way .	Imagine you are in an orchard , trying to decide which of the many apples to pick . On what do you base your decision ? Most research into this type of decision-making has focused on how the brain uses visual information – about features such as colour , size and shape – to make a choice . But what about the effort required to obtain the apple ? Does an apple at the top of the tree look more or less tempting than the low-hanging fruit ? To answer this kind of question , Hagura et al . asked volunteers to decide whether dots on a screen were moving to the left or to the right . The volunteers indicated their choice by moving one of two levers . If they thought the dots were moving to the right , they moved a lever in their right hand . If they thought the dots were moving to the left , they moved a lever in their left hand . What the volunteers did not know , however , is that one of the levers was slightly heavier and therefore harder to move than the other . Hagura et al . found that the volunteers biased their decisions away from the direction that would require the most effort . If the right-hand lever was heavier , the volunteers decided that dots with ambiguous motion were moving to the left . Those for whom the left-hand lever was heavier felt that the same dots were moving to the right . The participants showed this bias despite failing to notice that the levers had different weights . Moreover , they continued to show the bias even when subsequently asked to simply say their answers rather than use the levers . These results indicate that the effort required to act on a decision can influence the decision itself . The fact that participants were biased even when responding verbally , and despite being unaware that the levers differed in weight , suggests that they were not deliberately choosing the easier option . Instead , the cost to act changed how they perceived the stimuli themselves . The findings also suggest that it might be possible to help people make better decisions by designing environments in which less favourable options require more effort .	[ "Abstract", "Introduction", "Results", "Discussion", "Material", "and", "methods" ]	[ "neuroscience" ]	2017	Perceptual decisions are biased by the cost to act
Dendritic cells residing in the skin represent a large family of antigen-presenting cells , ranging from long-lived Langerhans cells ( LC ) in the epidermis to various distinct classical dendritic cell subsets in the dermis . Through genetic fate mapping analysis and single-cell RNA-sequencing , we have identified a novel separate population of LC-independent CD207+CD326+ LClike cells in the dermis that homed at a slow rate to the lymph nodes ( LNs ) . These LClike cells are long-lived and radio-resistant but , unlike LCs , they are gradually replenished by bone marrow-derived precursors under steady state . LClike cells together with cDC1s are the main migratory CD207+CD326+ cell fractions present in the LN and not , as currently assumed , LCs , which are barely detectable , if at all . Cutaneous tolerance to haptens depends on LClike cells , whereas LCs suppress effector CD8+ T-cell functions and inflammation locally in the skin during contact hypersensitivity . These findings bring new insights into the dynamism of cutaneous dendritic cells and their function opening novel avenues in the development of treatments to cure inflammatory skin disorders . In 1868 , Paul Langerhans described a novel dendritic-shaped , non-pigmentary cell population in the epidermis ( Langerhans , 1868 ) . These so-called Langerhans cells ( LCs ) were first classified as cellular members of the nervous system , due to their morphological similarity with neurons . It was not until the 1980s when it became clear that this peculiar epidermal cell fraction with its potent antigen presentation activity belonged to the dendritic cell ( DC ) family ( Romani and Schuler , 1989; Schuler and Steinman , 1985 ) . Despite the fact that LCs share many features with DCs , they are generally considered as epidermal tissue-resident macrophages , mainly due to their dependence on CSF1 , their embryonic origin and local self-maintenance ( Wynn et al . , 2013 ) , although a conventional ‘macrophage signature’ ( e . g . CD16/32 , CD64 , and MerTK expression ) is missing ( Gautier et al . , 2012 ) . LCs can sense invading pathogens and initiate an intrinsic maturation process that drives their migration out of the epidermis ( Romani et al . , 2001 ) . As such , LCs have been regarded as a prototype antigen-presenting cell ( APC ) ( Nagao et al . , 2009 ) that can , after antigen capture , migrate to the draining lymph nodes ( LNs ) to initiate an immune response by priming naïve LN-resident T-cells ( Romani et al . , 2003 ) . Antigen presentation can , however , occur in skin-draining LNs independently of LCs ( Henri et al . , 2010 ) . In fact , the skin hosts several other distinct dermal DC subpopulations ( Henri et al . , 2010; Kissenpfennig et al . , 2005 ) , the presence of which complicates the analysis of the cellular contribution to skin immune responses , such as contact hypersensitivity ( CHS ) . Consequently , the paradigm of ‘who is doing what’ ( i . e . epidermal LCs versus dermal DC counterparts ) is still controversial ( Bennett et al . , 2005; Bobr et al . , 2010; Bursch et al . , 2007; Clausen and Stoitzner , 2015; Kaplan et al . , 2005; Noordegraaf et al . , 2010; West and Bennett , 2017 ) . Here , we demonstrate that under steady-state conditions , LCs most likely do not exit the skin , or if so , in very low numbers . Through a combined use of genetic fate mapping and novel inducible LC-ablating mouse models , we show that the originally described LN LC fraction is actually an independent LClike cell population that originates from the dermis , not from the epidermis . These LClike cells are ontogenitically different from LCs and are replaced over time by bone marrow ( BM ) -derived cells with slow kinetics before trafficking to the LN . The skin and the skin-draining LNs contain several distinct DC subpopulations . To delineate migratory LCs and dermal DCs , we profiled DC subsets in the epidermis , dermis , and skin-draining LNs . In the epidermis , we confirmed that CD326+CD207+ LCs are predominantly found within the CD11bhiF4/80hi fraction ( Nagao et al . , 2009; Valladeau et al . , 2000; Figure 1A ) . In the dermis , we found a fraction of CD11bhiF4/80hi cells that co-expressed CD326 and CD207 ( Figure 1B , upper panel ) . These cells could be immigrated LCs , although we cannot exclude a contamination from the epidermis during the isolation procedure . As expected , the remaining dermal CD11bhiF4/80hi cells were CD326-CD207- tissue-resident macrophages ( Sheng et al . , 2015; Tamoutounour et al . , 2013 ) . Dermal DCs were localized in the F4/80int and CD11chiMHCII+ DC fraction , which we could separate into three subpopulations based on CD103 and CD11b expression: CD103+CD11b- ( defined as cDC1 ) , CD103-CD11blow , and CD103-CD11bhi ( defined as CD11bhi ) . CD103+CD11b- DCs but not CD103-CD11bhi DCs co-expressed CD326 and CD207 . We could also divide the CD103-CD11blow subpopulation into CD326-CD207- ( defined as triple negative [TN] ) and CD326+CD207+ ( defined as LClike ) fractions ( Figure 1B , right panel ) . The signal regulatory protein α ( Sirpa ) , a myeloid cell-specific receptor , was expressed on dermal LCs , LClike , TN , and CD11bhi DCs , while , as expected , dermal CD103+ DCs lacked this receptor ( Figure 1—figure supplement 1A ) , result validated also by the single-cell RNA-sequencing ( scRNA-seq ) analysis shown later ( Figure 3—figure supplement 1 ) . In terms of costimulatory receptors , dermal LCs and LClike cells express similar levels of CD80 and CD86 , whereas the remaining DC subpopulations display lower levels ( Figure 1—figure supplement 1A ) . To track the corresponding migratory DCs in the skin-draining LNs , we first gated on CD11cint-hiMHCIIhi cells , which represent the migratory DC fraction ( Sheng et al . , 2017 ) . Similar to our findings in the dermis , CD11b and CD103 labelling separated the migratory DCs into CD103+CD11b- ( cDC1 ) , CD103-CD11bhi ( CD11bhi ) , and CD103-CD11blow cells ( Figure 1C ) . The CD103-CD11blow cells could be further separated into two fractions: CD326-CD207- ( TN ) and CD326+CD207+ ( LClike ) subpopulations ( Figure 1C ) . Notably , we did not detect the bona fide epidermal and dermal LCs showing the original F4/80hiCD11bhi phenotype in the LN ( Figure 1C , right , lower panel ) . In agreement with previous work ( Henri et al . , 2010 ) , the CD11bhi DC fraction represented the largest DC subpopulation in the dermis , whereas in the LN all four DC subpopulations ( CD11bhi , cDC1 , TN , and LClike ) were almost equally represented ( Figure 1D ) . To confirm our observation , we took advantage of the Lang-EGFP mouse to trace directly EGFP-expressing CD207+ cells in all three tissues ( epidermis , dermis , and skin-draining LN ) . Clearly epidermal EGFP+ cells were co-expressing high levels of CD326 and F4/80 ( Figure 1—figure supplement 2 , upper panel ) . In the dermis , two main EGFP+ cells’ populations were detectable: one with lower levels of CD326 expressing xCR1 , typical marker for cDC1 and a second population co-expressing EGFP and CD326 was further separated into F4/80hi and F4/80low fractions ( Figure 1—figure supplement 2 , middle panel ) . Differently to the skin , EGFP expression in the LN was weaker and was restricted to CD326low and CD326hi cells: CD326low cells represent resident ( CD11chiMHCIIint ) DCs and CD326hi cells the migratory ( CD11cintMHCIIhi ) fraction . With respect to resident DCs , EGFP+ cells were only detectable in xCR1+ cells but not in the CD11b+ fraction . On the contrary , the migratory DCs were further subdivided into CD326hiEGFP+/low xCR1+ cDC1 and F4/80low LClike cells . No CD326hiEGFP+/low F4/80hi cells were detectable in the LN ( Figure 1—figure supplement 2 , lower panel ) . Because we detected no phenotypic F4/80hi LCs in the LNs , we hypothesized that the cutaneous DCs en route to the LN were not derived from epidermal LCs , but rather from distinct dermal CD11bhi , cDC1 , TN , and F4/80low LClike DC populations . This analysis cannot exclude , however , the possibility that the migrating LCs might change their phenotype as demonstrated previously ( Schuler and Steinman , 1985 ) . Since LC and LClike cells co-exist together in the dermis , we aimed to investigate their relationship and respective gene signature by scRNA-seq analysis . Unsupervised clustering and uniform manifold approximation and projection ( UMAP ) were performed on 9605 enriched cells isolated from the dermis of ears obtained from five mice . The origin of distinct CD45+ and CD45- dermal cell subpopulations are visualized in a colour-coded UMAP plot ( Figure 2A ) . Nine different cell clusters could be broadly identified by unsupervised clustering and classified as follows: ( 1 ) LC , ( 2 ) LClike , ( 3 ) mast cells/neutrophils , ( 4 ) DC/monocytes , ( 5 ) macrophages , ( 6 ) lymphocytes 1 , ( 7 ) lymphocytes 2 , ( 8 ) mesenchymal cells , and ( 9 ) epithelial cells . Conventional DCs , monocytes , and other myeloid-related signature genes , such as Zbtb46 ( DCs ) , Xcr1 and Clec9a ( cDC1 ) , Siglech ( plasmacytoid DC ) , Ly6c and Ccr2 ( monocytes ) , Gata2 and Fcer1a ( mast cells ) , and Ly6g ( neutrophils ) , are mainly detectable in the DC/mono and mast cell/neutrophil clusters ( 3–4 ) and are mainly absent or weakly expressed in the LC/LClike clusters ( 1–2 ) ( Figure 2B , C and Figure 2—figure supplement 1 ) . Cd207 and Cd326 expressing cells are detected in LC ( 1 ) , LClike ( 2 ) , as well as in DC/monocyte cluster ( 4 ) , which confirms the presence of three distinct CD207+CD326+ dermal subpopulations observed by flow cytometry ( Figure 1B ) . Cd207 and Cd326 expressing cells detected in the cluster 4 are co-expressing Clec9a , Xcr1 , Irf8 hence they represent the cDC1s ( Figure 2B and Figure 2—figure supplement 1 ) . Cd207 and Cd326 expressing cells in clusters 1 ( LC ) and 2 ( LClike ) share many of the previously reported LC signature genes ( e . g . Cd11c , Adgre1 , Cd74 , Mafb , Pu . 1 , Csf1r , Tgfbr1 ) ( Figure 2C and Figure 2—figure supplement 1 ) , but several other genes are differentially expressed in LClike cells ( e . g . Tgfbr2 , Sylt3 , Col27a1 , Fernt2 , Spry2 ) or in LC cells ( e . g . Cd209a , Agpat4 , Birc3 , Dusp16 , Gdpd3 , Ly75 and Ppfibp2 ) , respectively ( Figure 2C , D ) . To further elucidate the relationship between different dermal DC and macrophage populations , a developmental trajectory was obtained from a UMAP analysis specifically obtained from clusters 1 , 2 , 4 , and 5 shown in Figure 2 ( Figure 3A ) . Clearly there is a close relationship between LC and LClike as well as between macrophages/monocytes and CD11b+/TN DCs , whereas cDC1s are identified as a separate independent cell cluster ( Figure 3A and B ) . Furthermore , detailed transcription factor ( TF ) analysis revealed that LC and LClike cells share many TFs , some equally expressed ( Mafb , Irf4 , Irf8 ) , some higher expressed in LC ( Pu . 1 ) , and some more elevated in LClike ( Stat3 , Runx2 , Runx3 , Id2 , Klf4 , Maf ) ( Figure 3C and Figure 3—figure supplement 1A ) . Zbtb46 , a TF selectively expressed on classical DCs , is expressed on both LC and LClike cells but not as high as on classical DCs ( cDC1 , CD11b+ , and TN ) ( Figure 3—figure supplement 1A ) . Interestingly , Zeb2 , a specific cDC2 TF , is only weakly expressed on LC and LClike cells ( Figure 3—figure supplement 1A ) . Furthermore , LC-related genes , such as Mmp9 , Cd9 , Mfge8 , Cldn1 ( Ferrer et al . , 2019; Miyasaka et al . , 2004; Ratzinger et al . , 2002; Zimmerli and Hauser , 2007 ) , are elevated in dermal LCs and weakly expressed in dermal LClike cells ( Figure 3D ) , whereas no substantial differences are detected in expression of Sirpa , Ccr7 , and Cd24a ( Figure 3E , Figure 3—figure supplement 1B ) . The reduced expression of Adgre1 ( F4/80 ) in LClike cells ( Figure 3E ) , re-confirms the downregulation of F4/80 surface expression on this cell type observed in our previous flow cytometry analysis ( Figure 1 ) . Interestingly , both β and γ chains of the IL-2R ( Il2rb and Il2rg ) , previously reported to be expressed in DCs ( Zelante et al . , 2012 ) , are highly expressed in LClike cells , whereas LCs are the dermal cells expressing the lowest levels among the different DCs and macrophages subpopulations ( Figure 3—figure supplement 1C ) . In summary , the unsupervised clustering of single cells obtained from dermis suggests that LC and LClike cells are two independent cell fractions and distinct from CD207+CD326+ cDC1s as well as from cDC2 CD11b+ and TN DCs , as already shown in conventional flow cytometry analysis ( Figure 1B ) . Fate mapping experiments have shown that epidermal LCs derived partially from primitive yolk sac progenitors ( Hoeffel et al . , 2012; Sheng et al . , 2015 ) ; therefore , the developmental origin of LCs is distinct from conventional DCs and resembled more microglia . To study in detail a possible yolk sac origin of distinct cutaneous LC and DC subpopulations , a single injection of tamoxifen ( TAM ) was given to E7 . 5 pregnant KitMerCreMer/R26 mice ( Figure 4A ) . Three months later , the epidermis , dermis , and brain ( microglia as positive control ) were collected and isolated cells were then analysed for YFP expression . As previously reported , microglia , the prototype yolk sac-derived macrophage , were strongly labelled ( ~40% ) ( Figure 4B and F ) . However , about 12% of epidermal LCs were YFP labelled , confirming their partial yolk sac origin ( Figure 4C and F ) . In comparison , the dermal LC counterparts showed a similar labelling profile ( ~10% ) , whereas the remaining dermal DC subpopulations ( LClike , cDC1 , CD11bhi , and TN ) showed a significantly lower 5% YFP signal , very likely , attributed to small spillover of labelling in the HSCs ( haematopoietic stem cells ) ( Figure 4D and F ) . Therefore , YS only contributed to LCs but not to dermal LClike cells . Low YFP labelling was also obtained in all migratory LN DC counterparts ( Figure 4E and F ) . LCs are the only cell type from the DC family that originate from self-renewing radio-resistant embryonic precursors ( Merad et al . , 2002 ) ; other DC subpopulations are short-lived and constantly replenished by BM progenitors ( Kissenpfennig et al . , 2005 ) . To delineate the radio-resistant properties of the newly identified LClike cells , we generated BM chimeric mice by transplanting congenic CD45 . 1+ mouse BM cells into irradiated CD45 . 2+ recipients ( Figure 5A ) . We then analysed the CD45 . 1+/CD45 . 2+ ratio in different skin-related DC subpopulations 1 or 4 months after reconstitution . In the epidermis and dermis , LCs were mostly CD45 . 2+ , and thus retained their host origins due to local self-renewal ( Figure 5B , C ) . By contrast , dermal cDC1 , TN , and CD11bhi DCs exhibited a wholly CD45 . 1+ phenotype after just 1 month following reconstitution; this finding means that they are fully BM-derived . Only LClike cells showed a mixed contribution from both CD45 . 2+ host and CD45 . 1+ donor cells . In fact , after 1 month following reconstitution , only a minority ( ~10% ) of LClike cells were replenished by CD45 . 1+ cells; this percentage increased to ~50% by 4 months after reconstitution ( Figure 5B , C ) . In skin-draining LNs , we found that cDC1 , TN , and CD11bhi cells were mostly derived from donor CD45 . 1+ BM cells , excluding their origins from the radio-resistant LC population . Comparable to its dermal counterpart , only the LClike cell fraction was split into donor CD45 . 1+ and host CD45 . 2+ cells , respectively ( Figure 5B , C ) . In addition , the contribution of CD45 . 1+ donor cells increased over time , from ~10% after 1 month to ~45% after 4 months . This unique temporal replacement suggests a dual origin for LClike cells , distinguishing this DC fraction from both conventional long-lived radio-resistant self-renewing LCs and short-lived BM-derived DCs . To allow high-resolution and unbiased data-driven dissection of skin DC subpopulations in the reconstituted chimeric mice , we performed a UMAP analysis of flow cytometry data . Both CD45 . 1+ and CD45 . 2+ LClike cells were clearly visible and clustered separately , but in close proximity ( Figure 5D ) . Using this dimensional reduction algorithm , we detected that CD11c+MHCIIhi dermal DC subpopulations could be grouped into five separate clusters: cDC1 , TN , CD11bhi , and two LClike cell clusters ( BM-derived CD45 . 1+ and resident CD45 . 2+ ) . To investigate the molecular relationship between the resident LClike cell population and the BM-derived LClike cells , we performed RNA-seq on LN LClike cells isolated from chimeric mice ( CD45 . 1+ donor BM cells into CD45 . 2+ recipient mice ) . Unsupervised hierarchical clustering ( Euclidean distance , complete linkage ) and principal component analysis ( PCA ) analysis revealed that both CD45 . 1+ and CD45 . 2+ LClike cells clustered closely together ( not shown ) , with ~85% of their gene expression overlapping ( Figure 5E ) . The high level of similarity between resident and BM-derived LClike fractions indicates that the microenvironment , and not the cellular origin , seems to determine the LClike cell identity . BM chimeras require full body irradiation , which can damage the local skin microenvironment and attract BM-derived newcomers . This irradiation could , therefore , complicate the analysis of skin-resident cell homeostatic turnover kinetics . To circumvent this issue , we performed a fate mapping study under steady-state conditions using KitMerCreMer/R26 fate mapping mice . These mice allow for the turnover rates of cell populations derived from BM precursors to be estimated ( Sheng et al . , 2015 ) . We performed our analyses at different time points ( 1 , 4 , and 8 months ) after TAM injection to ensure a sufficiently long time frame to monitor populations that turn over slowly ( Figure 6A ) . In the epidermis , LCs showed minimal YFP labelling over the entire 8-month chase period; this finding was expected as these cells are not replaced by BM-derived cells ( Figure 6B and C , left panel ) . Similarly in the dermis , CD11bhiF4/80hiCD326+CD207+ cells showed minimal labelling from 1 to 8 months ( Figure 6B and C , middle panel ) . We propose that this fraction most likely represents immigrant LCs in the dermis . cDC1 , TN , and CD11bhi DCs , however , were fully labelled with YFP after just 1 month and the labelling was maintained for the remaining 8 months . This finding is consistent with the fast turnover rate identified for these three DC subsets . By contrast , LClike cells gradually accumulated the label from 10% to 60% over the 8-month chase period , supporting that dermis-resident LClike DCs are replaced slowly by BM progenitors . In the skin-draining LNs , all DC subsets behaved similarly to their dermal counterparts ( Figure 6B and C , right panel ) . Briefly , cDC1 , TN , and CD11bhi DCs showed a fast turnover by reaching plateau level of labelling after 1 month while LClike cells demonstrated a slow turnover rate over the 8-month chase period . Cell cycle analysis was performed for dermal LC , LClike , and different DC subsets based on scRNA-seq ( Figure 6—figure supplement 1 ) . The LClike subset exhibited higher proliferating capability than LC and other DC subsets , consistent with the previous findings that LCs are dividing extremely slow ( Ginhoux and Merad , 2010 ) and conventional DCs do not proliferate in tissues and mainly depend on their BM progenitors for expansion ( Liu and Nussenzweig , 2010 ) . Overall , we showed that LClike cells displayed slower turnover kinetics than other DC subpopulations and higher proliferating capability to refill the emigration gap . To interrogate the relationship between LC and LClike cells , we exploited a novel DC-SIGN-DTR transgenic mouse strain ( Figure 7—figure supplement 1A ) , which allowed us to deplete epidermal and dermal LCs without affecting the LClike cell pool ( Figure 7 ) . Although not detectable by flow cytometry on the cell surface ( Figure 1—figure supplement 1B ) , we measured DC-SIGN ( or CD209a ) specific mRNA levels in murine LCs as well as in CD11bhi DCs , the latter already known to express this receptor ( Cheong et al . , 2010 ) , whereas cDC1 , TN DCs , or LClike cells were negative ( Figure 7—figure supplement 1B ) , a result which was corroborated by the scRNA analysis of dermal cells ( Figure 2 ) . Since DC-SIGN expression has never been reported for LCs , quantitative PCR ( qPCR ) analysis was performed to detect the expression of DTR ( known as human heparin-binding EGF-like growth factor [HBEGF] ) in LCs obtained from DC-SIGN-DTR mice . Accordingly , HBEGF mRNA was detected only in LC isolated from epidermis of DC-SIGN DTR mice and was absent in LCs obtained from WT mice ( Figure 7—figure supplement 1C ) . Thus , the decrease in LCs observed after diphteria toxin ( DT ) injection was achieved due to the high sensitivity of the DT-DTR system ( Ruedl and Jung , 2018 ) , although no DC-SIGN was measurable on the cell surface of LCs from DC-SIGN DTR+ mice . To exclude a downregulation of the DC-SIGN receptor in LCs upon maturation , we sorted epidermal LCs , cultured them overnight with GM-CSF and LPS and compared by qPCR the Cd209a expression between unstimulated and stimulated epidermal LC fractions . Clearly no Cd209a downregulation was observed in activated LCs ( Figure 7—figure supplement 1C ) ; therefore , an eventual transition from maturing dermal DC-SIGNpos LCs to dermal DC-SIGNneg LClike cells can be excluded . We established short and long depletion protocols ( Figure 7A ) to capture even potentially very slowly migrating ‘LC derivatives’ ( Bursch et al . , 2007 ) . In the DT-treated DC-SIGN DTR mice , LCs were efficiently depleted in both the epidermis and dermis by the short-term and long-term depletion protocols ( Figure 7A–C ) . We also found that cells in the CD11bhi cell fraction were affected by the DT treatment; this was particularly evident during the short-term depletion protocol , in which the cell numbers were reduced by ~80% ( Figure 7C ) . Importantly , cDC1 , TN , and LClike cell numbers were unaffected and thus were comparable between DT-injected WT and DC-SIGN mouse strains . These results strongly support the independency of LClike cells from classical bona fide epidermal LCs . To further confirm that LClike cells represent a distinct cell lineage from LCs , we crossed DC-SIGN DTR mice with a KitMerCreMer/R26 fate mapping mouse , which would enable us to trace BM-derived cells in absence of LC . We treated these mice ( DC-SIGN DTR-KitMerCreMer/R26 ) with TAM and then injected them with DT for 5 weeks to maintain long-term LC depletion ( Figure 7D ) . Although epidermal LCs were absent over the whole period , the YFP labelling profiles of skin-derived LN DC subsets , including the LClike fraction , were comparable between DT-injected DC-SIGN DTR+-KitMerCreMer/R26 and DC-SIGN DTRneg-KitMerCreMer/R26 mice ( Figure 7E , F ) . These data show that in absence of LC , the replenishment of LClike cells by BM-derived cells is not affected . We next investigated the specific local contribution of LCs during CHS . Therefore , DT-injected WT and DC-SIGN DTR mice were sensitized with 0 . 5% 2 , 4-dinitrofluorobenzene ( DNFB ) and challenged at day 5 with 0 . 2% DNFB to induce a CHS reaction . The ear swelling was subsequently monitored over 12 days . In another group of mice , ears were processed 3 days post challenge for qPCR analysis as well as processed for cell isolation ( Figure 8A ) . In absence of LCs , clearly a pronounced increased ear swelling persisted over 10 days ( Figure 8B ) , a phenotype which was already reported in other LC-ablating transgenic mouse strains such as human langerin-DTA ( hu-DTA ) and human langerin-DTR ( huDTR ) mice ( Kaplan et al . , 2005; Bobr et al . , 2010 ) . Proinflammatory cytokines such as IL-6 , TNF-α , and IL-1β were clearly significantly upregulated in ears of DT-treated DC-SIGN DTR mice ( Figure 8C , upper panel ) . VEGFα , a biomarker for keratinocyte impairment ( Canavese et al . , 2010 , Bae et al . , 2015 ) , was also clearly augmented in absence of LCs ( Figure 8C , lower panel ) . In addition , higher IFN-γ levels were measured in ears lacking LCs ( Figure 8C , lower panel ) , values that correlated with an augmented CD8+ T-cell-dependent IFN-γ response observed by flow cytometry analysis ( Figure 8D , Figure 8—figure supplement 1 ) . In accordance with previously published data ( Igyarto et al . , 2009 ) , no major difference was observed in activated dermal ICOS+ Tregs in presence or absence of LCs ( Figure 8D , right panel , Figure 8—figure supplement 1A ) . In our final assays , we aimed to determine the contribution of distinct skin-resident DC subpopulations to the induction of tolerance to CHS . Here , we injected a series of different DTR mice , including CD207-DTR , Clec9A-DTR , and DC-SIGN DTR mice , with DT to deplete the target cells over the whole period of tolerization , sensitization , and final challenge ( Figure 8E ) . Of note , each DTR mouse strain shows a different LC/DC depletion profile: DT-treated CD207-DTR mice lack all epidermal and dermal CD207 expressing cells ( LC , LClike , and cDC1 ) ( Figure 8—figure supplement 2A–C ) , Clec9-DTR mice lack cDC1s ( Figure 8—figure supplement 2A , D and E ) , and DC-SIGN lack epidermal and dermal LCs and CD207-CD326-CD11bhi DCs ( Figure 7 ) . We next established an experimental 2 , 4-dinitrothiocyanobenzene hapten ( DNTB ) -mediated tolerance mouse model . Here , we induced immune tolerance against the strong contact sensitizer DNFB by epicutaneous application of an innocuous DNTB . We applied DNTB to the shaved abdomen of DT-injected WT , CD207-DTR , Clec9A-DTR , and DC-SIGN DTR mice 7 days prior to subsequent DNFB sensitization ( Figure 8E ) . As a positive control , we established a group of WT mice that was not treated and tolerized with DNTB . When the mice were ear-challenged with DNFB , we found that the non-tolerized WT mice developed robust CHS , as evidence by ear swelling that increased over time upon DNFB challenge . CD207-DTR mice were clearly not tolerized against DNFB , since they showed robust ear inflammation over time . Such ear swelling was not observed in DNTB-tolerized WT , Clec9A-DTR , and DC-SIGN-DTR mice ( Figure 8F ) . CHS is conferred by cytotoxic T-cell-mediated skin inflammation induced by exposure to strong contact sensitizers , such as DNFB . We thus analysed activated IFN-γ-producing CD8+ cytotoxic T-cells in the skin-draining LNs by flow cytometry on day 5 post DNFB sensitization . In line with the observed ear swelling profile , we found that a low IFN-γ-producing CD8+ cytotoxic T-cell response was restricted to DNTB-tolerized WT , Clec9A-DTR , and DC-SIGN-DTR mice . This response was comparatively enhanced in CD207-DTR and WT mice ( Figure 8G , Figure 8—figure supplement 2B ) . The diminished level of ear inflammation in DNTB-tolerized WT , Clec9A-DTR , and DC-SIGN-DTR mice correlated with an increased presence of activated ICOS+CD4+FOXP3+ Tregs in the LN , which exhibit suppressive activity in CHS to DNFB ( Vocanson et al . , 2010 ) . Remarkably , we did not observe this phenotype in DNTB-tolerized CD207-DTR or non-tolerized WT mice ( Figure 8H , Figure 8—figure supplement 2B ) . Because of this clear difference between CD207-DTR mice ( that lack CD207+ LCs/LC-like cells and CD207+ DC1s ) and DC-SIGN DTR mice ( that lack only CD207+ LCs ) and Clec9A-DTR mice ( that lack only CD207+ cDC1 ) , we conclude that only LClike cells critically contribute to tolerance induction . Epidermal LCs are the only APCs localized in the epidermis . These cells were recently re-defined as ‘macrophages in DC clothing’ due to their unique ontogeny , and self-renewing and radio-resistant characteristics ( Doebel et al . , 2017 ) . By contrast , there are multiple DC and macrophage subpopulations that reside in the dermis ( Tamoutounour et al . , 2013 ) . Although these dermal DCs share some common markers with LCs ( such as langerin [CD207] and EpCAM [CD326] ) , they constitute a distinct cell lineage on the basis of their developmental origins and cytokine requirements ( Bursch et al . , 2007; Ginhoux et al . , 2007; Poulin et al . , 2007 ) . Three CD207+ DC subpopulations have been described in the skin-draining LN: two subpopulations are skin-derived and one subpopulation originates from the BM ( Bursch et al . , 2007; Douillard et al . , 2005; Henri et al . , 2001; Romani et al . , 2010 ) . Due to this diverse skin-resident DC network , it became evident that not only LCs but other skin-derived DCs might be involved either in tolerance or immune response induction in draining LNs . Although it is commonly believed that the journey of an LC starts from the epidermis and ends in the skin-draining LN after a transit through the dermis in steady state , we found that it is in fact their look-alike counterparts , LClike cells , that migrate to the draining LNs . Our new insight was gained by re-analysing established mouse strains ( KitMerCreMer/R26 mice ) and exploiting newly generated transgenic mouse strains ( DC-SIGN-DTR mice and DC-SIGN-DTR- KitMerCreMer/R26 fate mapping mice ) , which allowed us to visualize , with increasing resolution , the in vivo dynamics of skin-resident DCs under steady state . We first characterized and re-defined different DC/LC subsets in the dermis by flow cytometry and scRNA-seq analysis , which delineated classical F4/80hi LC and four different DC subsets , namely cDC1 , TN DCs , CD11bhi DCs , and an unappreciated CD11blowF4/80low LClike cell fraction . With the exception of classical F4/80hi LCs , we found all of these cells in the migratory CD11cintMHCIIhi DC fraction of the skin-draining LN . This finding suggests that the majority of migratory CD326+CD207+ DCs are CD103+ cDC1 and CD103- LClike cells and not classical CD11bhiF4/80hi LCs which are hardly seen in the LN if not at all . Corroborating evidence for differential migratory behaviours among different skin DCs was provided by real-time intravital two-photon microscopy . Under steady-state conditions , due to the structural integrity of the basement membrane , epidermal LCs are sessile with static and almost immobile dendrites . In contrast , dermal DC subpopulations are actively crawling through the dermal interstitial space at high velocity even in absence of inflammation suggesting that continuous migration to LN is a steady-state property of dermal DCs and not epidermal LCs ( Ng et al . , 2008; Shklovskaya et al . , 2011 ) . Our analysis of dermal DCs is in full agreement with recent published data by Henri et al . , 2010 . The DC family in the dermis was likewise disentangled in five subpopulations: two subsets lacking the expression of CD207 ( CD207-CD11b- [TN] and CD207-CD11b+ [CD11bhi] ) and three expressing CD207 ( CD11bintCD207++ mLCs , CD11blow/-CD207+CD103+ [cDC1] , and CD11blowCD207+CD103- [LClike] ) ( Henri et al . , 2010 ) . Similarly , cutaneous LNs were distinguished in five analogous subpopulations including mLCs , which were defined for their characteristics in radioresistance and not for the expression of classical LC markers ( CD11bhi and F4/80hi ) . Henri et al . speculated that LN LCs downregulated CD11b and F4/80 expression ( Henri et al . , 2010 ) and therefore these markers lost their discriminatory power to segregate distinct CD207+ cells in the LN . To circumvent the ‘complication’ of the potential shift in phenotype , we adopted an alternative approach based on genetic fate mapping analyses which allowed to trace cell lineages between distinct LC and DC subpopulations avoiding lethal irradiation and generation of chimeric mice . First , our E7 . 5 embryo ‘labelling strategy’ demonstrated that only LCs are partially yolk sac-derived ( Sheng et al . , 2015 ) , but not the other migratory DCs’ subpopulations , inclusive of LClike cells . Second , our detailed analyses of the fate mapping kinetics revealed that the radio-resistant and radiosensitive CD11blowCD207+CD103- subpopulations described by Henri et al . represented instead a truly homogeneous radio-resistant LClike subpopulation , which is gradually replaced over time by BM-derived progenitors . Furthermore , we corroborated their ‘LC independency’ , since long-term absence of LCs did not affect the numbers of LClike cells in our DC-SIGN DTR mouse model . We also ruled out a possible downregulation of DC-SIGN during LC maturation which excludes the transition of a dermal DC-SIGNpos LC to a DC-SIGNneg LClike cell . Accordingly , our analysis delineated only four , and not five , LN migratory DC subpopulations , excluding LCs . The phenotypes , transcription profiles , and cytokine requirements of dermal cDC1 , TN , and CD11bhi DCs have been extensively described ( reviewed in Clausen and Stoitzner , 2015 ) ; however , there has been comparatively less attention given to the LClike subpopulation . Unsupervised clustering of scRNA-seq trascriptome data of dermal cells indicated that LC and LClike cells , although sharing some common myeloid cell markers and TFs , are two independent cell fractions and clearly distinct from macrophages and the other skin DC subpopulations . However , unlike LCs , which are BM-independent , radio-resistant , and self-renewing ( Ghigo et al . , 2013; Hoeffel et al . , 2012 ) , LClike cells represent a radio-resistant population that is progressively replaced postnatally by BM-derived precursors . Similar to resident macrophages in tissues , such as skin , gut , kidney , and heart , LClike cells have a dual origin involving both embryonic ( but not yolk sack-like LCs ) and adult haematopoiesis ( Molawi et al . , 2014; Sheng et al . , 2015; Soncin et al . , 2018 ) . Unlike other skin DC subpopulations , which are short-lived and exhibit a high turnover rate , we show that fetal-derived LClike cells are long-lived , show higher proliferation rates than conventional DCs , and are replaced very slowly by BM-derived cells . These fetal-derived and BM-derived LClike cells co-exist together in adult tissue , and although derived from different origins , they show high similarity . This finding suggests that it is the local tissue microenvironment and not the cellular origin that shapes their final identity . The existence of an LC-independent radio-resistant dermal DC fraction was previously observed in other study that described the presence of an in situ proliferating , radio-resistant dermal DC subpopulation not only in the murine but also in human dermis ( Bogunovic et al . , 2006 ) . Similarly , a langerin-expressing dermal myeloid CD1+ DC fraction unrelated to XCR1+ DCs and LCs has been reported in human dermis ( Bigley et al . , 2015 ) . It is likely that these cells are the LClike cells described here and further studies will be needed to analyse the potential relationship between murine CD207+ LClike cells and human myeloid CD207+ DC counterparts . Although all dermal DCs migrate into LNs in a CCR7-dependent fashion ( Förster et al . , 1999 ) , LClike cells seem to migrate at slower rate than other DCs under steady-state conditions . Similar slow trafficking dynamics was originally attributed to LCs ( Bursch et al . , 2007; Ruedl et al . , 2000 ) but we now strongly believe that these previously reported slow migratory cells are in fact LClike cells . To further rule out the possibility that epidermal F4/80hiCD11bhi LC downregulate CD11b and F4/80 and turn into F4/80lowCD11blow LClike cells in the dermis , we exploited a novel DC-SIGN DTR transgenic mouse strain where LCs , but not LClike cells , could be ablated . Even long-term depletion ( 6 weeks ) of epidermal and dermal LCs had no effect on the numbers of LClike cells in the dermis and LNs while maintaining their LClike YFP-labelling profile in the absence of epidermal LCs in DC-SIGN DTR-KitMerCreMer/R26 mice . The contribution of LCs to hapten sensitization is still controversial and matter of debate since the phenotype observed in different LC depleting mouse models is ranging from reduced to exaggerated CHS reactivity ( Kaplan et al . , 2005; Bennett et al . , 2005; Noordegraaf et al . , 2010 , Bobr et al . , 2010; Clausen and Stoitzner , 2015 ) . The depletion of LCs , in the herein presented novel DC-SIGN DTR transgenic mouse line , resulted in enhanced CHS , a similar phenotype also reported in hu-DTA and huDTR mice ( Kaplan et al . , 2005; Bobr et al . , 2010 ) . The Tregs-independent LC-mediated suppressive effect is restricted locally to the skin since several proinflammatory cytokines ( IL-6 , TNF-α , and IL-1β ) as well as effector IFN-γ CD8+ T-cells are enhanced at the challenged cutaneous side in absence of LCs . Interestingly VEGF-α , recently reported as a novel biomarker for pathological cutaneous alterations such as psoriasis ( Canavese et al . , 2010 , Bae et al . , 2015 ) , is highly upregulated in LC-deficient mice which indicates an involvement of LCs in controlling keratinocytes in their VEGF production . In summary , our data corroborate the view that LCs data corroborate the view that LCs are one of the main immune regulators within the skin ( West and Bennett , 2017 ) . The role of different skin-resident DCs in triggering tolerance is also widely debated . Several studies have demonstrated that LCs are crucial for inducing tolerance ( Igyarto et al . , 2009; Kautz-Neu et al . , 2011; King et al . , 2015; Price et al . , 2015; Shklovskaya et al . , 2011 ) by stimulating Tregs and anergizing CD8+ T-cells . This effect has been shown , for example , in an experimental CHS model ( Gomez de Agüero et al . , 2012 ) . Conversely , others have demonstrated the specific contribution of dermal DCs into maintaining skin tolerance ( Azukizawa et al . , 2011; Tordesillas et al . , 2018 ) . Given that we only detected LClike cells and not bona fide LCs in the draining LNs , we readdressed this issue in a cutaneous tolerance model towards the weak contact allergen , DNTB . We used our novel inducible DC-SIGN DTR mice , in which LCs but not dermal or LN LClike cells can be ablated , together with other CD207+CD326+ LC/DCs depleting DTR strains ( inducible CD207-DTR and Clec9A-DTR mice ) . Deletion of LCs or cDC1 had no effect on tolerance induction , while deletion of both LCs and LClike cells prevented tolerance induction . This finding demonstrates that migratory LClike cells , which are capable of reaching the draining LNs , are responsible for the expansion of ICOS+CD4+FOXP3+ Tregs and the suppression of a cytotoxic T-cell response . In summary , our genetic fate mapping approach , used to delineate the complex skin DC network , does not support the established paradigm of LCs as being the main ‘prototype’ migrating APCs to draining LNs under homeostatic conditions ( Wilson and Villadangos , 2004 ) . We propose , rather , that LCs at steady state , similar to other tissue-resident macrophages , are sessile and act locally in the skin , whereas dermal LClike cells assume many of the functions previously attributed to LCs . The identification of this novel migratory dermal LClike subpopulation opens new avenues and approaches in the development of treatments to cure diseases such as contact allergic dermatitis and other inflammatory skin disorders like psoriasis . C57BL/6J and B6 . SJL-Ptprca Pepcb/BoyJ ( B6 CD45 . 1 ) were obtained from The Jackson Laboratory ( USA ) . KitMerCreMer/Rosa26-LSL-eYFP ( called KitMerCreMer/R26 ) and Clec9A-DTR mice were generated as previously described ( Piva et al . , 2012; Sheng et al . , 2015 ) . KitMerCreMer/R26 mice were backcrossed with DC-SIGN-DTR mice to obtain DC-SIGN-DTR-KitMerCreMer/R26 mice . B6 . 129S2-Cd207tm2Mal/J mice were bred and housed at the Malaghan Institute of Medical Research ( Wellington , New Zealand ) . CD207-DTR mice were obtained from the Singapore Immunology Network ( SIgN; A*Star , Singapore ) . DC-SIGN DTR mice were generated as follows: the IRES-DTR fusion gene was inserted into the 3'-UTR region of the DC-SIGN gene locus on BAC RP24-306K4; the gene targeting vector was then retrieved from the modified BAC ( Figure 7—figure supplement 1A ) . The gene targeting vector was linearized and electroporated into Balb/C embryonic stem ( ES ) cells and correctly recombined ES colonies were selected by PCR . Gene targeted ES cells were injected into C57BL/6 blastocysts and transferred into the oviduct of a pseudo-pregnant mother . F0 male chimera mice were mated with F1 Balb/C females to obtain F1 Balb/C DC-SIGN DTR mice; these mice were then backcrossed to C57BL/6 for 12 generations to generate C57BL/6 DC-SIGN DTR mouse used in this study . All mice , with the exception of B6 . 129S2-Cd207tm2Mal/J mice , were bred and maintained in the specific pathogen-free animal facility of the Nanyang Technological University ( Singapore ) . All studies involving mice in Singapore were carried out in strict accordance with the recommendations of the National Advisory Committee for Laboratory Animal Research and all protocols were approved by the Institutional Animal Care and Use Committee of the Nanyang Technological University ( ARF-SBS/NIE A-0133; A-0257; A0126 , A-18081 ) . For animal work performed in New Zealand , experimental protocols were approved by the Victoria University of Wellington Animal Ethics Committee and performed in accordance with institutional guidelines . KitMerCreMer/R26 and DC-SIGN-DTR- KitMerCreMer/R26 fate mapping mice were used to monitor the turnover rates of distinct skin-related DC subpopulation subsets . Upon TAM injection , the YFP label will be induced in all c-kit-expressing cells , predominantly residing in the BM , and these cells will retain the YFP label once they left the BM and seeded into the periphery . Each mouse was administered 4 mg TAM ( Sigma-Aldrich , St . Louis , MO , USA ) for 5 consecutive days by oral gavage for adult labelling , as previously described ( Sheng et al . , 2015 ) . Pregnant mice ( E7 . 5 ) were injected once with 16 mg TAM for embryo labelling . DC-SIGN-DTRpos and DC-SIGN-DTRneg mice were injected intraperitoneally ( i . p . ) with 20 ng/g DT ( Sigma-Aldrich ) to deplete DC-SIGN-expressing cells . Two different DT injection protocols were used ( Figure 7A ) . For the short-term depletion protocol , mice were injected i . p . at day −2 and −1 before collection of tissues . For the long-term protocol , DT was injected once a week over 5 weeks prior tissue collection . Chimeric mice were generated by irradiating recipient C57BL/6 or DC-SIGN-DTR mice ( CD45 . 2+ ) with two doses of 550 cGy , 4 hr apart . Then , 106 B6 . Ly5 . 1 ( CD45 . 1+ ) BM cells were injected intravenously ( i . v . ) , 24 hr after treatment . The mice were allowed to recover from 1 to 4 months before analysis . Mouse ears were cut and separated into dorsal and ventral halves using fine forceps . Both the dorsal and ventral halves ( with the epidermis side facing upwards ) were incubated for 1 hr at 37°C in 1 ml IMDM ( Thermo Fisher Scientific , Waltham , MA , USA ) medium containing 1 U/ml Dispase II ( Thermo Fisher Scientific ) . The epidermis and dermis were separated using fine forceps , cut into small pieces and digested for another 1 hr at 37°C in 1 mg/ml Collagenase D ( Roche , Basel , Switzerland ) . To obtain single-cell suspensions , the digested tissue was passed through a 40 mm cell strainer . To process skin-draining auricular LNs , the dissected LNs were minced and incubated in 1 mg/ml collagenase D for 60 min at 37°C . The following antibodies were used: anti-mouse CD45 ( 30-F11 ) , anti-mouse CD11b ( M1/70 ) , anti-mouse F4/80 ( BM8 ) , anti-mouse Ly6c ( HK1 . 4 ) , anti-mouse CD11c ( N418 ) , anti-mouse I-A/I-E ( M5/114 . 15 . 2 ) , anti-mouse CD103 ( 2E7 ) , anti-mouse CD326 ( G8 . 8 ) , anti-mouse CD207 ( 4C7 ) , anti-mouse CD45 . 1 ( A20 ) , anti-mouse CD45 . 2 ( 104 ) . They were purchased all from Biolegend ( San Diego , CA , USA ) . Anti-mouse CD45 microbeads from Milteny ( Bergisch Gladbach , Germany ) . All antibodies were used for extracellular stainings with the exception of the anti-CD207 Ab which was used for intracellular labelling after have fixed and permeabilized the cells with 2% paraformaldehyde and 0 . 05% saponin , respectively . Single-cell epidermal , dermal , or LN tissue suspensions were pre-incubated with 10 mg/ml anti-Fc receptor antibody ( 2 . 4G2 ) on ice for 20 min . Then , the suspensions were further incubated with fluorochrome-labelled antibodies at 4°C for 20 min , before being washed and re-suspended in PBS/2% FCS for analysis on a five-laser flow cytometer ( LSR Fortessa; BD Bioscience , San Jose , CA , USA ) . The data were analysed with FlowJo software ( TreeStar , Ashland , OR , USA ) and UMAP analysis was performed using the FlowJo UMAP plugin . Immune cells were enriched using anti-mouse CD45 microbeads from dermal single-cell suspension . Briefly , enriched CD45+ dermal cells were loaded into chromium microfluidic chips with v3 chemistry and barcoded with a 10× Chromium Controller ( 10X Genomics , Pleasanton , CA , USA ) . RNA from the barcoded cells was subsequently reverse-transcribed and sequencing libraries constructed with reagents from a Chromium Single Cell v3 reagent kit ( 10X Genomics ) according to the manufacturer’s instructions . Library sequencing was performed at Novogene Co . , Ltd ( Tianjin Novogene Technology Co . , Tianjin , China ) with Illumina HiSeq 2000 according to the manufacturer’s instructions ( Illumina , San Diego , CA , USA ) . FastQC was used to perform basic statistics on the quality of the raw reads . Raw reads were demultiplexed and mapped to the reference genome by 10X Genomics Cell Ranger pipeline using default parameters . All downstream single-cell analyses were performed using Cell Ranger and Seurat unless mentioned specifically . In brief , for each gene and each cell barcode ( filtered by Cell Ranger ) , unique molecule identifiers were counted to construct digital expression matrices . Secondary filtration for Seurat analysis: a gene with expression in more than three cells was considered as expressed and each cell was required to have at least 200 expressed genes . All mouse RNAs were analysed using an Agilent Bioanalyser ( Agilent , Santa Clara , CA , USA ) . The RNA Integrity Number ranged from 3 . 4 to 9 . 3 , with a median of 8 . 2 . cDNA libraries were prepared from a range of 18 , 24 . 2 , 68 , and 100 ng total RNA starting material using the Ovation Universal RNA-seq system . The length distribution of the cDNA libraries was monitored using a DNA High Sensitivity Reagent Kit on an Agilent Bioanalyser . All 11 samples were subjected to an indexed paired-end sequencing run of 2 × 100 bp on an Illumina Novaseq 6000 system ( Illumina , San Diego , CA , USA ) . The paired-end reads were trimmed with trim_galore1 ( option: -q 20 –stringency 5 –paired ) . The trimmed paired-end reads were mapped to the Mouse GRCm38/mm10 reference genome using the STAR2 ( version 2 . 6 . 0a ) alignment tool with multi-sample two-pass mapping . Mapped reads were summarized to the gene level using featureCounts3 in the subread4 software package ( version 1 . 4 . 6-p5 ) and with gene annotation from GENCODE release M19 . DESeq25 was used to analyse differentially expressed genes , and significant genes were identified with Benjamini-Hochberg adjusted p-values<0 . 05 . DESeq2 analysis was carried out in R version 3 . 5 . 2 . For functional analysis , hierarchical clustering based on Euclidean distance and complete linkage , was performed using the R ‘pheatmap’ package . PCA was performed using the R ‘prcomp’ package . The first two principal components were analysed on a multidimensional scatterplot that was created using the R ‘scatterplot 3D’ function . DC-SIGN-DTRneg and DC-SIGN-DTR+ mice were treated for 2 days with DT . Ears were collected and split into dorsal and ventral halves and subsequently incubated with 3 . 8% ammonium thiocyanate ( Sigma-Aldrich ) in PBS for 20 min at 37°C . Epidermal and dermal sheets were separated and fixed in ice-cold acetone for 15 min . Then , the epidermal sheets were pre-incubated with 10 mg/ml anti-Fc receptor antibody ( 2 . 4G2 ) on ice for 20 min and subsequently stained with FITC-labelled anti-MHC class II antibody for a further 30 min on ice for LC visualization . Ears were harvested and immediately homogenized in TRIzol reagent ( Thermo Fisher Scientific ) . Total RNA was subsequently purified using the RNAsimple Total RNA kit ( Tiangen Biotech Ltd , Beijing , China ) . Real-time PCR was performed according to the manufacturer’s instructions using the Primer design Precision FAST protocol ( Primerdesign Ltd , Cambridge , UK ) . F4/80hiCD326+ LCs were isolated from pooled murine epidermis sheets and purified by cell sorting ( purity >90% ) . 5 × 104 LCs were immediately used for RNA processing , the remaining 5 × 104 LCs were cultured in a 96-well round bottom plate for 16 hr in presence of 20 ng/ml GM-CSF and 2 hr/ml LPS and processed the next day for RNA isolation . WT and DC-SIGN DTR mice were injected with DT and 2 days later were sensitized with 1% DNFB dissolved in an acetone and olive oil mixture ( 4:1 , v/v ) . DT injection was repeated for 7 days every 3–4 days to maintain the LC pool ablated . The ears of both WT and DC-SIGN DTR mice were challenged with 0 . 5% DNFB . Ear swelling was measured daily for 12 consecutive days post challenge . Another mouse group was sacrifized at day 3 post challenge for qPCR analysis and for dermal T-cell response analysis . WT , CD207-DTR , Clec9A-DTR , and DC-SIGN DTR mice were injected with DT every 3–4 days over a period of 20 days to maintain the depletion of the target cells ( CD207-DTR: LC , cDC1 , and LClike cells; Clec9A-DTR: cDC1 and DC-SIGN-DTR: LC and CD11bhi cells ) . All mouse strains were tolerized with a 100 μl volume of 1% DNTB ( Sigma-Aldrich ) in an acetone and olive oil mixture ( AOO ) ( 4:1 , v/v ) , administered epicutaneously to the shaved abdomen . One group of WT mice was painted only with AOO as a control . After 7 days , all mouse strains were sensitized by skin painting the dorsal side of the ears with 0 . 5% DNFB ( Sigma-Aldrich ) ( 25 μl in AOO ) . Another group of mice was further ear-challenged 5 days later with 0 . 1% DNFB ( 4 μl in AOO ) , and ear swelling was measured using a digital caliper ( Mitutoyo , Kanagawa , Japan ) over the course of 6 days . In a second group of mice , the draining LNs were harvested at day 5 post tolerization/sensitization . To determine the capacity of CD8+ T-cells to secrete IFN-γ , isolated cells were stimulated in a round-bottom 96-well culture plate ( Corning , Corning , NY , USA ) with 10 ng/ml phorbol 12 , 13-dibutyrate ( PMA , Sigma-Aldrich ) and 1 mg/ml Ionomycin ( Sigma-Aldrich ) in complete IMDM for 3 hr followed by an additional 2 hr incubation with 10 μg/ml Brefeldin A ( Sigma-Aldrich ) at 37°C . The cells were then harvested and stained for CD3 and CD8 , fixed with 2% paraformaldehyde and permeabilized in 0 . 05% saponin ( Sigma-Aldrich ) before staining with anti-IFN-γ antibodies . To quantify activated Tregs , isolated cells were co-stained for CD4 and ICOS , fixed , and permeabilized using a Fix/Perm Buffer Set ( eBioscience ) before staining with an anti-Foxp3 antibody . The data represent the means ± SEM or SD , as indicated in the figure legends . GraphPad Prism software was used to display the data and for statistical analysis . Statistical tests were selected based on the appropriate assumptions with respect to data distribution and variance characteristics . All statistical tests are fully described in detail in the figure legends . Samples were analysed by two-tailed Student’s t-test to determine statistical differences between two groups . A two-way ANOVA with Bonferroni post-test was used to determine the differences between more than two groups . A p-value < 0 . 05 was considered to be statistically significant . The number of animals used per group is indicated in the figure legends as ‘n’ .	Our immune cells are constantly on guard to defend and protect us against invading pathogens , such as bacteria and viruses . Specialized immune cells , known as antigen-presenting cells , or APCs , have a key role in this process . They engulf invaders , chew them up , and travel to the closest local lymph node to stimulate other immune cells with small fragments of these pathogens . This ramps up the immune response to control infection and disease . APCs are a large and diverse family of immune cells , which includes dendritic cells and macrophages . Some APCs work as mobile surveillance units , travelling around the body to find new threats . Others embed themselves in particular organs and tissues , such as the skin , to provide local , on-the-spot surveillance . Langerhans cells are one of the main types of APC in the skin and are found in the thin outer layer of the epidermis . While it is commonly believed that Langerhans cells can move from the epidermis to the skin-draining lymph nodes , some seemingly contradictory evidence exists to suggest that this may not be the case . Now , Sheng et al . have investigated this issue by tracking APCs , including Langerhans cells , in the skin of mice . A powerful genetic cell labelling technique allowed them to track the movement of immune cells inside a living mouse . Sheng et al . found that majority of 'real' Langerhans cells did not leave the skin . Yet , a second lookalike cell that shared many of the same features of a Langerhans cell was found in the dermal layer of skin , and this cell could travel to local lymph nodes . Both the original and lookalike cells had distinct and separate roles in the skin . This research , which has uncovered a new type of Langerhans-like immune cell in the skin , may be extremely useful for developing new targeted therapies to boost immune responses during infection; or to suppress inappropriate immune activation that can lead to autoimmune diseases , such as psoriasis .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "immunology", "and", "inflammation" ]	2021	Fate mapping analysis reveals a novel murine dermal migratory Langerhans-like cell population
Eukaryotic mismatch repair ( MMR ) utilizes single-strand breaks as signals to target the strand to be repaired . DNA-bound PCNA is also presumed to direct MMR . The MMR capability must be limited to a post-replicative temporal window during which the signals are available . However , both identity of the signal ( s ) involved in the retention of this temporal window and the mechanism that maintains the MMR capability after DNA synthesis remain unclear . Using Xenopus egg extracts , we discovered a mechanism that ensures long-term retention of the MMR capability . We show that DNA-bound PCNA induces strand-specific MMR in the absence of strand discontinuities . Strikingly , MutSα inhibited PCNA unloading through its PCNA-interacting motif , thereby extending significantly the temporal window permissive to strand-specific MMR . Our data identify DNA-bound PCNA as the signal that enables strand discrimination after the disappearance of strand discontinuities , and uncover a novel role of MutSα in the retention of the post-replicative MMR capability . The evolutionarily conserved mismatch repair ( MMR ) system corrects replication errors post-replicatively to prevent their being fixed as mutations in the next round of complementary DNA synthesis ( Iyer et al . , 2006; Jiricny , 2013; Kunkel and Erie , 2015 ) . Since erroneously inserted nucleotides are present on newly synthesized DNA strands , identification of the newly synthesized strand is a critical step in MMR . By rectifying replication errors , MMR increases the replication fidelity by ~2 orders of magnitude in yeast ( Lujan et al . , 2014 ) . In humans , genetic or epigenetic inactivation of MMR genes elevates the risk of tumorigenesis in both hereditary and sporadic manners ( Jiricny , 2013 ) . In Escherichia coli , MMR distinguishes the parental and daughter strands mainly by using adenine methylation on GATC sites ( Lahue et al . , 1989; Iyer et al . , 2006 ) . Mismatches are recognized by the mismatch sensor MutS homodimer . MutS and the MutL homodimer then activate the latent nicking-endonuclease MutH , which cleaves the unmethylated strand at the hemi-methylated GATC sequences . MMR is possible from the time of hemi-methylated GATC generation by DNA synthesis until full methylation of the GATC sites . Maintaining this temporal window is critical for efficient MMR , because over-expression of the Dam methylase , by which full-methylation of the GATC sites is accelerated , significantly elevates the mutation frequency ( Herman and Modrich , 1981; Marinus et al . , 1984 ) . The E . coli MMR system can also correct replication errors through a methylation-independent mechanism , where strand discontinuities can substitute for GATC methylation both in vivo and in vitro ( Laengle-Rouault et al . , 1986; Lahue et al . , 1987; 1989 ) . Eukaryotic MMR is directed by strand discontinuities such as nicks or gaps in vitro ( Holmes et al . , 1990; Thomas et al . , 1991 ) . Two MutS heterodimers , MutSα ( Msh2-Msh6 ) and MutSβ ( Msh2-Msh3 ) recognize replication errors; MutSα has a biased preference for base-base mismatches and small insertion/deletion loops ( IDLs ) , while MutSβ preferentially recognizes large IDLs ( Iyer et al . , 2006; Jiricny , 2013; Kunkel and Erie , 2015 ) . After mismatch binding , MutSα/β are converted into closed ‘clamp-like’ forms , by which they can translocate along DNA . They then recruit the latent nicking-endonuclease MutLα ( Mlh1-Pms2 in vertebrates and Mlh1-Pms1 in yeast ) to initiate the removal of the error-carrying DNA strand . Two other eukaryotic MutL homologs , MutLβ ( Mlh1-Pms1 in vertebrates and Mlh1-Mlh2 in yeast ) and MutLγ ( Mlh1-Mlh3 ) are suggested to play minor roles in somatic MMR ( Jiricny , 2013; Campbell et al . , 2014 ) . Successful reconstitutions of eukaryotic MMR have shown that MutSα , or MutSβ , and MutLα , the Proliferating Cell Nuclear Antigen ( PCNA ) sliding clamp , the clamp loader Replication Factor C ( RFC ) , the Exo1 exonuclease , and the DNA synthesis components promote MMR when a strand discontinuity is present ( Genschel and Modrich , 2003; Dzantiev et al . , 2004; Constantin et al . , 2005; Zhang et al . , 2005 ) . A strand discontinuity , which can occur on either 5’- or 3’-side of the mismatch , activates MutLα through a MutSα- and PCNA-dependent mechanism to induce successive rounds of nicking on the error-carrying strand ( Kadyrov et al . , 2006; 2007; Pluciennik et al . , 2010 ) . Single-strand DNA termini such as 5’-ends of the Okazaki fragments serve as entry points for Exo1 and strand discrimination signals in vivo as well ( Pavlov et al . , 2003; Nick McElhinny et al . , 2010; Liberti et al . , 2013; Lujan et al . , 2014; Liu et al . , 2015 ) . Recent studies have revealed that ribonucleotides embedded by replicative DNA polymerases serve as strand signals for MMR in vitro and contribute to a sub-fraction of leading strand MMR in vivo , after converted into single-strand gaps through RNaseH2-dependent ribonucleotide excision repair ( Ghodgaonkar et al . , 2013; Lujan et al . , 2013 ) . PCNA has also been presumed to be the strand discrimination marker in eukaryotes ( Umar et al . , 1996; Chen et al . , 1999; Pavlov et al . , 2003; Dzantiev et al . , 2004; Kadyrov et al . , 2006; Pluciennik et al . , 2010; Hombauer et al . , 2011b; Peña-Diaz and Jiricny , 2012; Georgescu et al . , 2015; Kunkel and Erie , 2015 ) . PCNA is a ring-shaped homo-trimer that supports various DNA transactions including DNA replication and repair ( Georgescu et al . , 2015 ) . PCNA is loaded onto DNA from the template-primer junction by RFC , and likely unloaded by an RFC-like complex containing Elg1 after the completion of DNA synthesis ( Kubota et al . , 2013; 2015 ) . Since its DNA binding is asymmetric with respect to polarities of the parental and daughter strands , DNA-bound PCNA can hold information for the newly synthesized strand ( Bowman et al . , 2004; Georgescu et al . , 2015 ) . PCNA plays an essential role in an early MMR step that precedes degradation of the error-carrying strand ( Umar et al . , 1996 ) . PCNA loaded from a strand discontinuity induces strand-specific , mismatch- and MutSα-dependent activation of the MutLα endonuclease ( Kadyrov et al . , 2006; 2007; Pluciennik et al . , 2010 ) . When PCNA is loaded onto closed circular DNA without defined orientation , it induces unbiased nicking on both DNA strands ( Pluciennik et al . , 2010; 2013 ) . These findings have led to a proposal that orientation of DNA-bound PCNA is a critical determinant for the nicking specificity of MutLα . In addition to its proposed role in strand discrimination , PCNA is also involved in multiple steps of MMR . Numbers of PCNA mutants in yeast exhibit mutator phenotypes that are epistatic to MMR mutations ( Johnson et al . , 1996; Umar et al . , 1996; Chen et al . , 1999; Amin et al . , 2001; Lau et al . , 2002; Goellner et al . , 2014 ) . It interacts with MutSα/β through their PCNA-interacting peptide ( PIP ) motifs , which reside at the N-termini of Msh6 and Msh3 ( Clark et al . , 2000; Flores-Rozas et al . , 2000; Kleczkowska et al . , 2001 ) . In both cases , the PIP motifs and the mispair binding domains are connected through long linkers , which are predicted to be disordered in yeast ( Shell et al . , 2007 ) . Mutants of the PIP motifs in Msh6 and Msh3 show substantial but not complete reduction of the MMR activity , indicating that the PIP motif plays an important , yet assistive , non-essential role in MMR ( Clark et al . , 2000; Flores-Rozas et al . , 2000 ) . The PIP motif has been proposed to tether MutSα to the replication fork to assist its mismatch recognition ( Kleczkowska et al . , 2001; Hombauer et al . , 2011a; Haye and Gammie , 2015 ) . A recent study has shown that this motif in MutSα is important for a late MMR step ( s ) involving degradation of the error-carrying strand , especially by the Exo1-independent mechanism ( Goellner et al . , 2014 ) . These findings suggest that PCNA coordinates multiple reactions in MMR , yet the exact role of PCNA in MMR is still not clearly defined . An important remaining question in eukaryotic MMR is how ( and how long ) the strand discrimination signals are retained after DNA synthesis . It has been shown in yeast that MutSα must be present during S-phase to suppress mutations ( Hombauer et al . , 2011b ) . Therefore , the generation and retention of the strand signals for MMR should be intimately coupled with DNA replication . Since PCNA transiently remains on DNA after the completion of DNA synthesis , DNA-bound PCNA could mediate the coupling of MMR with DNA replication . However , because experimental evidence that DNA-bound PCNA induces strand-specific MMR in the absence of strand discontinuities is currently lacking , it is still uncertain whether PCNA by itself can mediate the coupling of MMR with DNA replication . In addition , how functional interaction between the MMR system and DNA-bound PCNA , which should be unloaded from DNA shortly after the completion of DNA synthesis , is ensured remains highly unclear . Here , we demonstrate that MutSα and PCNA play a critical role in maintenance of the MMR capability after DNA synthesis . We show that nucleoplasmic extract ( NPE ) of Xenopus eggs efficiently induces gap-directed , strand-specific MMR , whose capability remains even after sealing of the gap . We further show that DNA-bound PCNA induces strand-specific MMR in the absence of strand discontinuities . Strikingly , we found that MutSα attenuates unloading of PCNA and retains the MMR capability largely through its PIP motif . Our data thus identify PCNA as the eukaryotic strand discrimination molecule that retains the MMR capability after DNA synthesis , and delineate a critical role of MutSα in maintenance of the temporal window for eukaryotic MMR . The nucleoplasmic extract ( NPE ) , a highly concentrated nuclear protein extract of Xenopus eggs , has been used as a powerful in vitro model system for DNA repair reactions coupled with DNA synthesis ( Walter et al . , 1998; Räschle et al . , 2008; Olivera Harris et al . , 2015 ) . We exploited its extremely high capacity for DNA synthesis to study MMR . A plasmid carrying an A:C mismatch , which is a preferred MMR substrate in crude Xenopus egg extracts ( Varlet et al . , 1990 ) , and a 15-nucleotide ( nt ) single-strand gap was constructed ( Figure 1A ) . This plasmid was synthesized in vitro by second strand DNA synthesis after annealing of a primer carrying a 'C'-mismatch on single-stranded phagemid DNA corresponding to the 'A'-strand . Since NPE contains a high concentration of Geminin , a specific inhibitor for assembly of pre-replicative complexes , no DNA replication initiates when DNA is incubated directly in NPE , while DNA synthesis from existing 3’-termini is active ( Walter et al . , 1998 ) . NPE corrected the A:C mismatch by selectively editing the base on the gap-carrying strand ( Figure 1B–E , see 'mock' ) , more efficiently than conventional crude Xenopus egg extracts ( Figure 1—figure supplement 1 ) . As seen previously ( Olivera Harris et al . , 2015 ) , a gap could be placed on either 5’ or 3’ of the mismatch , and repair DNA synthesis occurred preferentially within a fragment containing the shorter path between the gap and the mismatch ( Figure 1—figure supplement 2 ) , indicating that NPE supports bidirectional MMR . Depletion of the Msh2-containing ( MutSα/β ) ( Figure 1B–C and Figure 1—figure supplement 3 ) or Mlh1-containing complexes ( MutLα/β/γ ) ( Figure 1D–E and Figure 1—figure supplement 3 ) abolished the repair reaction . Unlike other systems , a nick did not efficiently induce MMR ( Figure 1—figure supplement 4 ) . These results demonstrate that NPE efficiently recapitulates gap-directed , bi-directional MMR that is dependent on both MutSα/β and MutLα/β/γ . Some background repair observed in the absence of a gap could be due to either spontaneous base damages , which elicit base excision repair that in turn direct MMR ( Repmann et al . , 2015 ) , or strand breaks that have occurred during handling of the substrate . 10 . 7554/eLife . 15155 . 003Figure 1 . NPE promotes gap-directed MMR with efficient DNA synthesis at the gap site . ( A ) DNA substrates used in this study . The 3011-base pair plasmid DNA carries an A:C mispair ( pMM1AC ) , or an A:T base-pair ( pMM1AT ) at position 1 . Sequences surrounding an A:C mispair are designed so that the A:T or G:C repair product forms an XhoI or BamHI cleavage site , respectively . Two BbvCI nicking restriction enzyme sites were used to introduce a 15-nt single-strand gap . A PacI site was placed within the gap . If needed , a biotin-dT modification was introduced at position 1670 . ( B ) NPE was depleted using pre-immune antibodies ( lane 1 ) or a mixture of Msh2 and Msh6 antibodies ( lane 2 ) . Immunoblots of the NPE samples ( 0 . 05 μl each ) are shown . Orc2 served as a loading control . ( C ) Covalently closed ( lanes 1–3 and 10–12 ) , A-strand-gap ( 3’ to the mismatch ) -carrying ( lanes 4–6 and 13–15 ) or C-strand-gap ( 5’ to the mismatch ) -carrying pMM1AC ( lanes 7–9 and 16–18 ) were incubated in NPE described in ( B ) , and sampled at the indicated times . DNA was purified and digested with XmnI and either BamHI ( upper , A to G repair ) or XhoI ( lower , C to T repair ) . %repair was calculated based on the percentage of XhoI or BamHI sensitive DNA molecules . ( D ) NPE was depleted using pre-immune antibodies ( lane 1 ) or Mlh1 antibodies ( lane 2 ) . Immunoblots of the NPE samples ( 0 . 05 μl each ) are shown . ( E ) The MMR reaction in NPE described in ( D ) . DNA was digested with XmnI and either BamHI ( upper , A to G repair ) or XhoI ( lower , C to T repair ) . ( F ) Percentages of DNA synthesis at the gap , estimated by PacI sensitivity ( %Gap-filling synthesis ) , A to G repair ( %A→G repair ) , C to T repair ( %C→T repair ) and closed circular molecules ( %closed; Figure 1—figure supplement 5 ) , calculated from two independent experiments including the one described in Figure 1—figure supplement 6 , were plotted onto a graph . To calculate the %Gap-filling synthesis , DNA was digested with XmnI and PacI . The mean values were connected by lines . Error bars: ± 1 standard deviation ( SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 00310 . 7554/eLife . 15155 . 004Figure 1—figure supplement 1 . Comparison of the MMR efficiencies between in crude Xenopus egg extracts and in NPE . Homoduplex pMM1AT carrying no strand break ( lanes 1–3 and 13–15 ) or A-strand gap ( lanes 4–6 and 16–18 ) , or mismatch-containing pMM1AC carrying no strand break ( lanes 7–9 and 19–21 ) , or A-strand gap ( lanes 10–12 and 22–24 ) was incubated in High speed supernatants ( HSS; lanes 1–12 ) or NPE ( lanes 13–24 ) , and sampled at the indicated times . %repair was calculated as described in Figure 1C . Gap-directed A to G MMR was more efficient in NPE ( 67% ) than in HSS ( 34% ) , indicating that NPE has higher MMR activity than HSS . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 00410 . 7554/eLife . 15155 . 005Figure 1—figure supplement 2 . Detection of tracts of DNA repair synthesis during gap-directed MMR in NPE . Covalently closed ( lanes 1–3 ) , A-strand-gap-carrying ( lanes 4–6 ) or C-strand-gap-carrying pMM1AC ( lanes 7–9 ) was incubated in NPE , and sampled at the indicated times . DNA was purified and digested with XmnI and either BamHI ( upper , A to G repair ) or XhoI ( middle , C to T repair ) . To analyze the incorporation of radioactivity , DNA was digested with DrdI ( bottom ) . α-[32P]-dCTP was preferentially incorporated into the 1 kb fragment corresponding to the shorter path between the gap and the mismatch in both 5’-gap- and 3’-gap-directed MMR . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 00510 . 7554/eLife . 15155 . 006Figure 1—figure supplement 3 . Characterization of xMsh2 , xMsh6 and xMlh1 sera . 0 . 1 μl of NPE and the indicated amounts of recombinant proteins were separated using 4–15% SDS-PAGE and transferred onto PVDF membranes . Each membrane strip was probed with either the indicated antiserum or pre-immune serum ( PI ) from the same rabbit . The same exposure sets with PI and antisera are presented . Either xMsh2 ( Mr = 1 . 04 × 105 ) or xMsh6 ( Mr = 1 . 49 × 105 ) was detected as nearly a single band in NPE . Two major bands were detected with Mlh1 antibodies , and we concluded that the more quickly migrating band is xMlh1 , because its mobility on SDS-PAGE was closer to the expected molecular weight of Mlh1 ( Mr = 8 . 41 × 104 ) , it migrated slightly faster than recombinant xMlh1 tagged with 6×His-epitope through a 12 amino acid linker , and this band was specifically immunoprecipitated by xMlh1 antibodies ( see Figure 1D ) . ( * ) indicates cross-reacting band . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 00610 . 7554/eLife . 15155 . 007Figure 1—figure supplement 4 . Comparison of the MMR efficiencies between nick-carrying and gap-carrying substrates . Mismatch-containing pMM1AC carrying no strand break ( lanes 1–3 ) , A-strand nick carrying pMM2AC ( lanes 4–6 ) , or A-strand gap carrying pMM1AC ( lanes 7–9 ) was incubated in NPE , and sampled at the indicated times . %repair was calculated as described in Figure 1C . Gap-directed A to G MMR ( 62% ) was more efficient than nick-directed A to G MMR ( 22% ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 00710 . 7554/eLife . 15155 . 008Figure 1—figure supplement 5 . Kinetics of gap filling in NPE in the absence of a mismatch . ( A ) Calculation of the relative coefficient of SYBR Gold fluorescence for closed-form DNA over open-form DNA . A reaction mixture containing in vitro synthesized pMM0AT ( a pMM1 derivative that carries no BbvCI nicking site ) was split into two aliquots and incubated in the absence ( lane 1 ) or presence ( lane 2 ) of the nicking endonuclease Nb . BtsI , and separated by ethidium bromide containing agarose gel electrophoresis . The gel was stained with SYBR Gold nucleic acid stain and scanned with Typhoon FLA9000 using SYBR Gold specific setting ( a 473 nm excitation laser and a 510 nm Long pass filter ) . A representative result is shown . The relative fluorescence coefficient for closed-form DNA over open-form DNA ( c ) , which reflects difference in the saturating amount of SYBR Gold stain between open and closed form DNA , was calculated by the following formula . c=open ( +enzyme ) −open ( −enzyme ) closed ( −enzyme ) . The calculated c value was 1 . 15 ( standard deviation = 0 . 06 , n = 10 ) . Percentage of closed circular molecules ( %closed ) was calculated by the following formula . %closed=c × closed circular moleculesopen circular molecules+ ( c × closed circular molecules ) × 100 . ( B ) Kinetics of the gap-filling reaction in NPE on a homoduplex DNA . Gap-carrying homoduplex pMM1AT was incubated in NPE and sampled at the indicated times . %closed was calculated as described in ( A ) . ( * ) indicates linearized fragments produced by the nicking endonuclease . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 00810 . 7554/eLife . 15155 . 009Figure 1—figure supplement 6 . Kinetics of gap filling and MMR on a mismatch-carrying DNA in NPE . A-strand-gap-carrying pMM1AC was incubated in NPE and sampled at the indicated times . %closed , % Gap-filling synthesis and %repair were calculated as described in Figure 1—figure supplement 5 , Figure 1F , and C , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 009 On a substrate that has no mismatch , a single-strand gap was sealed within 2 min in NPE , resulting in quick accumulation of closed circular plasmids ( Figure 1—figure supplement 5 ) . To understand the temporal relationship between gap filling and gap-directed MMR , we compared the kinetics of DNA synthesis at the gap site and repair of the mismatch simultaneously on the same substrate . As shown in Figure 1F , DNA synthesis at the gap site ( estimated by digestion with PacI placed within the gap ) was mostly completed within 30 s . However , this DNA synthesis did not covalently close the mismatch-carrying plasmid , suggesting that new strand breaks are introduced on the DNA ( Figure 1F and Figure 1—figure supplement 6 ) . The A to G MMR gradually progressed until ~30 min , and shortly after the mismatch correction , closed circular molecules were accumulated . These observations suggest that , in this gap-directed MMR model , efficient DNA synthesis at the gap site precedes mismatch correction involving degradation of the error-carrying strand . As suggested previously ( Varlet et al . , 1996 ) , the 3’-terminus is likely used as a signal for MMR , rather than the initiation point of extensive strand removal . If the 3’-terminus generates a signal for strand discrimination , does the signal remain after the sealing of the gap , or is the 3’-terminus essential to initiate MMR ? To answer this question , we set up a stepwise incubation assay that separates gap filling and mismatch recognition . We first incubated the MMR substrates in a MutSα/β-depleted ( MutS-depleted ) NPE , in which the gap should be filled and ligated without mismatch recognition , and then supplied MutSα/β by addition of fresh NPE ( Figure 2A ) . In the MutS-depleted NPE , more than 80% of gap-carrying plasmids were converted into the closed-circular form within 1 . 5 min ( Figure 2B , lane 9 ) . However , ~40% of A to G MMR was still observed upon addition of fresh NPE at 2 min , at a time when at most , only ~14% strand discontinuities remained ( Figure 2B , lane 10 and Figure 2C , lane 14 ) . From 0 . 5 to 2 min , the MMR efficiency consistently exceeded the amounts of remaining strand discontinuities . These data suggest that the strand information derived from a gap is 'memorized' on DNA to induce subsequent MMR . 10 . 7554/eLife . 15155 . 010Figure 2 . Strand memory derived from a gap directs MMR after the completion of gap filling . ( A ) Schematic diagram of the stepwise incubation assay . pMM1AC carrying a 15-nt gap was incubated in a MutS-depleted NPE to fill the gap without recognizing the mismatch . Subsequently , fresh NPE was added to an aliquot of the reaction to initiate MMR by supplying MutSα/β . After 60 min incubation , DNA was purified and the direction and the efficiency of repair were examined . ( B ) Kinetics of the gap-filling reaction in MutS-depleted NPE . Covalently closed ( lanes 1–5 ) or A-strand-gap-carrying pMM1AC ( lanes 6–10 ) was incubated in MutS-depleted NPE and sampled at the indicated times . ( * ) indicates linear DNA produced by contaminating endonuclease activity in Nt . BbvCI . These linear molecules were excluded from the calculation of %closed . ( C ) Strand-specific MMR reaction after supplying MutSα/β . Aliquots were sampled at the indicated times , mixed with fresh NPE , and incubated for an additional 60 min . No repair was observed when the second NPE was omitted ( lanes 2 and 9 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 010 The gap-filling reaction in Xenopus egg extracts was PCNA-dependent ( Figure 3—figure supplement 1 ) . Therefore , at least one PCNA trimer must be involved in every gap filling , and it could retain the strand information after gap filling . If PCNA stores strand information for MMR , PCNA loaded on DNA in a specific orientation should direct MMR in NPE even in the absence of strand discontinuities . To test this idea , we established an in vitro PCNA-loading reaction using purified human RFC ( hRFC ) and PCNA ( hPCNA ) , which can substitute for Xenopus PCNA ( xPCNA ) in the MMR reaction ( Figure 3—figure supplement 2 ) . A mismatch plasmid carrying an A-strand nick was bound to Sepharose beads through a site-specific biotin modification ( see Figure 1A ) . hPCNA was loaded onto the immobilized plasmids in a nick-dependent , and therefore orientation-specific manner . The nick left on the DNA was sealed with the T4 DNA ligase . The hPCNA-DNA complex was then washed with 1 M KCl to remove hRFC , T4 ligase , and loosely associated hPCNA ( Figure 3A ) . 10 . 7554/eLife . 15155 . 011Figure 3 . DNA-bound PCNA bypasses the requirement of a gap for strand-specific MMR in NPE . ( A ) Schematic diagram of the assay . Singly biotinylated pMM2AC ( a pMM1 derivative carrying only one BbvCI site ) carrying a nick was bound to Sepharose beads and incubated with recombinant hPCNA and hRFC . The nick was then ligated , and the complex was washed with a buffer containing 1 M KCl . The hPCNA-DNA complex was incubated in NPE to test whether MMR occurs . ( B ) In vitro PCNA-loading reaction . Untreated DNA ( top ) , linearized DNA ( middle , by XmnI ) , and quantitative immunoblottings for hPCNA and hRfc2 ( bottom ) of samples from the in vitro PCNA-loading assay using covalently closed , and A-strand-nick-carrying pMM2AC are presented . ( * ) indicates linear DNA , which was excluded from the calculation of %closed . ( C ) The hPCNA-DNA complexes described in ( B ) were split into two portions to test whether PCNA encircles DNA ( C ) and whether MMR occurs upon incubation in NPE ( D ) . The one portion was treated with either control buffer , or buffer containing XmnI whose cleavage site is located 1382 bp away from the PCNA entry point . DNA from the reaction ( top ) , linearized DNA ( middle , by XmnI ) , and a hPCNA immunoblot ( bottom ) are shown . Since nick-carrying molecules were accumulated during incubation , the level of closed-circular molecules was lower than the original substrates shown in ( B ) . ( D ) The other potion of the hPCNA-DNA complexes described in ( B ) was incubated in NPE . The MMR efficiencies were calculated as described in Figure 1C . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 01110 . 7554/eLife . 15155 . 012Figure 3—figure supplement 1 . Requirement of PCNA for gap-filling and nick-ligation reactions in Xenopus egg extracts . ( A ) Characterization of xPCNA sera . 0 . 1 μl of NPE and 10 ng of recombinant xPCNA were separated using 4–15% SDS-PAGE and transferred onto PVDF membranes . Each membrane strip was probed with either the xPCNA or pre-immune serum ( PI ) from the same rabbit . The same exposure sets of PI and xPCNA antibodies are shown . xPCNA ( Mr = 2 . 88 × 104 as a monomer ) was detected as nearly a single band in NPE . ( * ) indicates cross-reacting band . ( B ) The efficiency of PCNA depletion from HSS . HSS was depleted with pre-immune ( lanes 1–4 ) or xPCNA antibodies ( lanes 5–9 ) , and supplemented with either control buffer ( lanes 4 and 8 ) or 0 . 5 μM recombinant xPCNA ( lane 9 ) . HSS ( 0 . 5 μl ) from each depletion rounds was separated by SDS-PAGE and probed with the indicated antibodies . Orc2 served as a loading control . ( C ) The gap-filling reaction in PCNA-depleted HSS . Gap-carrying homoduplex pMM1AT plasmids were incubated in PCNA-depleted HSS described in ( B ) and sampled at the indicated times . The efficiency of the gap-filling reaction was analyzed using ethidium bromide containing agarose gel electrophoresis . Depletion of PCNA inhibited appearance of closed-form plasmids even after 60 min ( lane 12 ) , indicating that PCNA is essential for either DNA synthesis at the gap site , flap-processing , or ligation at the gap in HSS . ( * ) indicates linearized fragments produced by the nicking endonuclease . ( D ) The gap-filling reaction in p21-peptide supplemented NPE . Because PCNA-depletion from NPE was not possible due to its high concentration , p21 peptides ( Mattock et al . 2001 ) were used to inhibit PCNA function in NPE . Gap-carrying homoduplex pMM1AT plasmids were incubated in NPE supplemented with either control buffer ( lanes 2–4 ) , 1 mg/ml wildtype ( lanes 5–7 ) , PIP mutant ( lanes 8–10 ) or jumbled p21 peptide ( lanes 11–13 ) , and sampled at the indicated times . The efficiency of gap filling was analyzed using ethidium bromide containing agarose gel electrophoresis . The wildtype but not the PIP mutant nor jumbled p21 peptide inhibited appearance of closed-form plasmids , suggesting that PCNA is required for either DNA synthesis at the gap site , flap-processing , or ligation at the gap in NPE . ( * ) indicates linearized fragments produced by the nicking endonuclease . ( E ) The depletion efficiency of PCNA from HSS . HSS was depleted using pre-immune ( lane 1 ) or xPCNA antibodies ( lane 2 ) . Immunoblots of the HSS samples ( 0 . 5 μl each ) are shown . Orc2 served as a loading control . ( * ) indicates cross-reacting band . ( F ) Comparison of nick ligation and gap filling in PCNA-depleted HSS . Nick-carrying homoduplex pMM2AT ( lanes 1–11 ) or gap-carrying homoduplex pMM1AT ( lanes 12–22 ) was incubated in either mock-treated or PCNA-depleted HSS described in ( E ) , and sampled at the indicated times . The efficiencies of accumulation of closed-circular molecules were analyzed using ethidium bromide containing agarose gel electrophoresis . Although gap filling was completely inhibited by PCNA-depletion , nick ligation was only partially inhibited , indicating that PCNA is only partially required for nick ligation in HSS . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 01210 . 7554/eLife . 15155 . 013Figure 3—figure supplement 2 . Characterization of recombinant hRFC and hPCNA . ( A ) 300 ng of recombinant hRfc1-5 purified from baculovirus-infected High Five insect cells was separated using SDS-PAGE and stained with Coomassie brilliant blue R-250 . ( B ) 1 μg each of recombinant xPCNA and hPCNA purified from E . coli was separated using SDS-PAGE and stained with Coomassie brilliant blue R-250 . ( C ) The depletion efficiency of PCNA from HSS . HSS was depleted using pre-immune ( lane 1 ) or xPCNA antibodies ( lane 2 ) . The HSS samples were separated using SDS-PAGE and probed with the indicated antibodies . Orc2 served as a loading control . ( D ) hPCNA can replace xPCNA in the MMR reaction in HSS . A-strand-gap-carrying pMM1AC plasmids were incubated in HSS described in ( C ) supplemented with either control buffer ( lanes 1–6 ) , 0 . 5 μM of xPCNA ( lanes 7–9 ) or hPCNA ( lanes 10–12 ) and sampled at the indicated times . %repair was calculated as described in Figure 1C . hPCNA , as well as xPCNA , restored the A to G MMR in xPCNA-depleted HSS , indicating that hPCNA can functionally replace xPCNA in MMR . Note that the overall MMR efficiency was weakened because of the rather harsh xPCNA depletion condition ( 4 round depletion ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 013 Using fluorescent-based quantitative western blotting , we estimated that 4–10 hPCNA trimers reproducibly loaded on the DNA in a nick- and ATP-dependent manner ( Figure 3B; see also Figure 4 ) . Ligation of the nick was ~80% efficient . The DNA-bound hPCNA topologically encircled DNA with free-sliding ability , because linearization of DNA resulted in dissociation of essentially all hPCNA ( Figure 3C ) . As PCNA was loaded from the A-strand nick in this case , A to G MMR was predicted . When the hPCNA-DNA complex was transferred into NPE , ~75% of the A bases were specifically repaired into G bases ( Figure 3D ) . Such a high repair efficiency could not be explained by the remaining nicks , which could contribute to at most 20% of MMR . 10 . 7554/eLife . 15155 . 014Figure 4 . The orientation of PCNA loading determines strand specificity of MMR in NPE . ( A ) In vitro PCNA-loading reaction with pMM2AC carrying no nick , an A-strand nick or a C-strand nick . Untreated DNA ( top ) , linearized DNA ( middle , by XmnI ) , and quantitative immunoblots ( bottom ) are presented . ( B ) The hPCNA-DNA complexes described in ( A ) were incubated in NPE . ( C ) The hPCNA-DNA complexes ( Figure 4—figure supplement 2 ) were incubated in mock-treated , MutS-depleted or MutL-depleted NPE described in Figure 4—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 01410 . 7554/eLife . 15155 . 015Figure 4—figure supplement 1 . Requirement of hPCNA and hRFC in PCNA-directed MMR in NPE . ( A ) Singly biotinylated pMM2AC , either covalently closed ( lanes 1–4 and 9–12 ) or A-strand nick carrying ( lanes 5–8 and 13–16 ) , was bound to Sepharose beads and incubated with either hPCNA ( lanes 2 , 6 , 10 and 14 ) , hRFC ( lanes 3 , 7 , 11 and 15 ) , or both ( lanes 4 , 8 , 12 , and 16 ) . Untreated DNA ( top ) , linearized DNA ( middle , by XmnI ) , and quantitative immunoblottings of hPCNA and hRfc2 ( bottom ) are shown . ( B ) The hPCNA-DNA complexes described in ( A ) were incubated in NPE . %repair was calculated as described in Figure 1C . Strand-specific A to G MMR was induced only when both hPCNA and hRFC were present in the loading reaction ( lanes 22–24 ) . ( C ) The efficiencies of MMR were calculated from three independent experiments including the one shown in ( B ) and plotted onto a graph . Error bars: ± 1 SD . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 01510 . 7554/eLife . 15155 . 016Figure 4—figure supplement 2 . Requirement of the MutS and MutL complexes in PCNA-directed MMR in NPE . ( A , B ) Depletion of the MutSα/β ( A ) or the MutLα/β/γ complexes ( B ) from NPE . NPE was depleted using pre-immune antibodies ( lane 1 ) , a mixture of Msh2 and Msh6 ( A; lane 2 ) , or Mlh1 antibodies ( B; lane 2 ) . The NPE samples were separated using SDS-PAGE and probed with the indicated antibodies . Orc2 served as a loading control . ( C ) In vitro PCNA-loading reaction . Untreated DNA ( top ) , linearized DNA ( middle; by XmnI ) , and quantitative immunoblottings of hPCNA and hRfc2 ( bottom ) from in vitro PCNA-loading assays used in Figure 4C are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 016 If PCNA directs strand-specific MMR , PCNA loaded onto DNA in the opposite orientation should invert the strand to be repaired . Critically , the hPCNA-DNA complex prepared with A-nick DNA directed ~60% of A to G MMR , and the complex prepared with C-nick DNA directed ~70% of C to T MMR in NPE ( Figure 4A and B ) . When either PCNA or RFC was omitted from the reaction , most of the strand-specific repair was inhibited ( Figure 4—figure supplement 1 ) . The PCNA-directed MMR reaction was dependent on both MutSα/β and MutLα/β/γ ( Figure 4C and Figure 4—figure supplement 2 ) . These results indicate that directional loading of hPCNA is sufficient to induce strand-specific MMR in the absence of strand discontinuities in NPE . The above results suggest that DNA synthesis at the gap site leaves PCNA on DNA as the strand memory , which can be subsequently used for MMR . Consistent with this idea , a significant portion of nick ligation was independent of PCNA , explaining why a nick did not induce MMR as efficiently as a gap in NPE ( Figure 3—figure supplement 1 ) . In the absence of MutSα/β , the strand information was gradually lost after the disappearance of strand discontinuities ( Figure 2 ) , suggesting that PCNA is unloaded from DNA within this time window . To examine PCNA dissociation in NPE , we loaded hPCNA on DNA , incubated it in NPE , and quantified the remaining hPCNA using an hPCNA specific monoclonal antibody that does not cross-react with xPCNA ( Figure 5—figure supplement 1 ) . This method distinguishes retention of hPCNA from sticking of xPCNA . In the mock-treated NPE , hPCNA was quickly dissociated from a homoduplex DNA within a few minutes , while dissociation was attenuated in the NPE depleted of xRfc3 , indicating that PCNA is removed from DNA mostly through an active unloading process requiring the Rfc3-containing complex ( es ) ( Figure 5 ) . Due to the unavailability of appropriate antibodies , the specific Rfc3-containing complex responsible for PCNA unloading could not be identified . 10 . 7554/eLife . 15155 . 017Figure 5 . Rfc3 is required for the unloading of PCNA from closed-circular DNA . ( A ) Depletion efficiency of xRfc3 from NPE . The depletion efficiency was estimated to be >98% . ( B ) hPCNA loaded onto immobilized pMM2AT ( carrying no mismatch ) was incubated in NPE described in ( A ) . The efficiency of nick ligation in the PCNA loading reaction was estimated to be ~89% . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 01710 . 7554/eLife . 15155 . 018Figure 5—figure supplement 1 . Specific detection of hPCNA by a monoclonal antibody . The hPCNA monoclonal antibody used in this study does not detect xPCNA . 10 ng each of recombinant xPCNA and hPCNA were separated using 4–15% SDS-PAGE . Recombinant proteins were visualized using silver nitrate or transferred onto PVDF membranes . The membrane strip was probed with hPCNA antibody . hPCNA monoclonal antibodies do not detect recombinant xPCNA even when the membrane was exposed for longer periods ( lanes 3–6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 018 Amount of DNA-bound PCNA should correlate with the MMR capability . In addition , since DNA-bound MutSα/β can interact with PCNA ( Flores-Rozas et al . , 2000; Bowers et al . , 2001; Lau and Kolodner , 2003; Iyer et al . , 2008 ) , MutSα/β might affect dynamics of PCNA retention . We therefore examined hPCNA retention and the MMR capability in the presence or absence of MutSα/β ( Figure 6A ) . In the MutS-depleted NPE , hPCNA was again quickly unloaded from mismatch-carrying DNA ( Figure 6B and Figure 6—figure supplement 1 ) . Concurrent with PCNA unloading , the PCNA-directed MMR capability rapidly dropped , leaving only ~10% of repairable molecules at 5 min ( Figure 6C , lane 5 ) . In contrast , in the MutL-depleted NPE ( containing endogenous MutSα/β but deficient for MMR ) , even at 30 min , more than one hPCNA trimer on an average still remained on DNA , and ~20% of mismatches were still repairable ( Figure 6C , lane 14 ) . The difference in PCNA retention and the kinetics of repair was not due to remaining nicks on the DNA , because neither MutS- nor MutL-depletion significantly affected the level of nick-carrying molecules ( Figure 6—figure supplement 2 , see also Figure 7 ) . These results suggest that mismatch-bound MutS complexes interfere with unloading of PCNA to retain strand information for MMR . 10 . 7554/eLife . 15155 . 019Figure 6 . The MutS complexes inhibit Rfc3-dependent PCNA unloading in NPE . ( A ) Schematic diagram of the stepwise incubation assay . After the in vitro PCNA-loading reaction , hPCNA-DNA complexes were incubated in either MutS- or MutL-depleted NPE . The samples were split into two portions , fresh NPE was added to one portion to measure MMR , and DNA-bound proteins were recovered from the other portion to quantify hPCNA . ( B ) hPCNA loaded onto immobilized pMM2AC carrying an A-strand nick was incubated in NPE depleted of either MutSα/β or MutLα/β/γ . The efficiency of nick ligation in the PCNA loading reaction was estimated to be ~72% . The amounts of DNA-bound hPCNA are shown . See Figure 6—figure supplement 1 for the depletion efficiencies ( MutS: >98% , MutL: >98% ) . ( C ) Kinetics of the PCNA-directed MMR capability in the presence or absence of MutSα/β . Aliquots were sampled from the reaction described in ( B ) and mixed with fresh NPE . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 01910 . 7554/eLife . 15155 . 020Figure 6—figure supplement 1 . Depletion efficiencies of Msh2 and Mlh1 from NPE . NPE was depleted using a mixture of Msh2 and Msh6 antibodies ( lane 1 ) or Mlh1 antibodies ( lane 2 ) . The NPE samples and a dilution series of NPE were separated using SDS-PAGE and probed with the indicated antibodies for the depletion efficiencies , which we estimated that >98% for both Msh2 and Mlh1 . Orc2 served as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 02010 . 7554/eLife . 15155 . 021Figure 6—figure supplement 2 . Effect of MutS- and MutL-depletion on the level of closed-circular DNA molecules . ( A ) NPE was depleted using a mixture of Msh2 and Msh6 antibodies ( lane 1 ) or Mlh1 antibodies ( lane 2 ) . The NPE samples and a dilution series of NPE were separated using SDS-PAGE and probed with the indicated antibodies for the depletion efficiencies , which we estimated >98% for both Msh2 and Mlh1 . Orc2 served as a loading control . ( * ) indicates cross-reacting band . ( B ) After the in vitro PCNA-loading reaction , the hPCNA-DNA complex ( lane 2 ) was incubated in either MutS- or MutL-depleted NPE described in ( A ) . Untreated DNA ( top ) , DNA-bound PCNA and Msh2 ( middle ) , and linearized DNA ( bottom , by XmnI ) are shown . Depletion of either MutS or MutL complex did not significantly affect the level of closed-circular molecules . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 02110 . 7554/eLife . 15155 . 022Figure 7 . The MutS complexes retain strand memory derived from a gap in NPE . ( A ) The gap-filling reaction of A-strand-gap-carrying pMM1AC in NPE described in Figure 7—figure supplement 1 . ( B ) The efficiencies of the gap-filling reaction were calculated from three independent experiments including the one described in ( A ) and plotted on a graph . The lines represent one-phase decay fitting of the data . See Table 1 for the fitting parameters . ( C ) Strand-specific MMR reaction after supplying fresh NPE . ( D ) The efficiencies of MMR were calculated and presented as described in ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 02210 . 7554/eLife . 15155 . 023Figure 7—figure supplement 1 . Depletion efficiencies of Msh2 and Mlh1 from NPE . NPE was depleted using a mixture of Msh2 , Msh6 , and Mlh1 antibodies ( lane 1 ) , a mixture of Msh2 and Msh6 antibodies ( lane 2 ) or Mlh1 antibodies ( lane 3 ) . The depletion efficiencies were estimated to be >99% for both Msh2 and Mlh1 . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 023 To test whether MutSα/β also maintain the MMR capability derived from a gap , we repeated the stepwise incubation experiments presented in Figure 2 , but used MutL-depleted NPE as the first NPE this time . At least 80% of the gaps were filled in all conditions at 2 min ( Figure 7A and B , and Figure 7—figure supplement 1 ) . In either MutS-depleted or MutS/MutL-depleted NPE , more than 40% of mismatches were still repairable upon addition of fresh NPE at 2 min , confirming the ~2 min strand memory seen in Figure 2 ( Figure 7C and D ) . Strikingly , in the MutL-depleted NPE , >60% of mismatches were repairable at 10 min , and ~35% were still repairable even at 30 min . Using one-phase exponential decay fitting , which could explain most data with reasonable quality ( R2 > 0 . 8 ) , we estimated the half-life of the strand memory in the absence of MutSα/β to be approximately 2 min , and to be approximately 40 min in the presence of MutSα/β ( Table 1 ) . These results strongly suggest that the MutSα/β-dependent mechanism maintains the strand information by inhibiting unloading of PCNA that is loaded during DNA synthesis at the gap site . 10 . 7554/eLife . 15155 . 024Table 1 . Fitting parameters for the gap-derived strand memory experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 024%gap [95% Confidence Interval]%A→G repair [95% Confidence Interval]t1/2 ( min ) %gap at 0 minR2t1/2 ( min ) %repair at 0 minR2ΔMutS/MutL0 . 60 [0 . 51−0 . 73]100 [96 . 7−103 . 3]0 . 991 . 87 [1 . 52−2 . 45]74 . 8 [68 . 6−80 . 7]0 . 97ΔMutS0 . 68 [0 . 58−0 . 81]100 [96 . 4−103 . 5]0 . 992 . 00 [1 . 76−2 . 31]82 . 8 [79 . 0−86 . 6]0 . 99ΔMutL0 . 75 [0 . 66−0 . 88]100 [96 . 5−103 . 5]0 . 9939 . 6 [12 . 3− +∞]72 . 8 [63 . 6−81 . 9]0 . 80Parameters for the one-phase exponential decay fitting of the data described in Figure 7B and D are presented . %gap: percentage of remaining gaps ( 100 - %closed ) , t1/2: half-life , R2: coefficient of determination . n = 3 . Curve fitting was carried out using the GraphPad Prism 6 software ( GraphPad Software , CA , USA ) To address whether the PIP motif located at the N-terminus of Msh6 is required for retention of DNA-loaded PCNA and strand information , we purified recombinant xMutSα harboring alanine substitutions in the PIP motif ( xMutSαPIP; Figure 8A and B ) . In good agreement with previous reports ( Kleczkowska et al . , 2001; Iyer et al . , 2008; Bowen et al . , 2013 ) , when no stepwise incubation was involved , the gap-directed MMR on the 3 . 0 kb substrate was largely restored by xMutSαPIP in MutS-depleted NPE ( Figure 8—figure supplement 1 ) , indicating that in this experimental setup the PIP function is not important for MMR . Mismatch binding of xMutSαPIP was not significantly compromised ( Figure 8—figure supplement 1 ) . However , xMutSαPIP was almost inert for PCNA retention . We repeated the PCNA-unloading assay using MutS/MutL-depleted NPE supplemented with xMutSαWT or xMutSαPIP ( Figure 8C ) . Although xMutSαWT attenuated PCNA unloading , xMutSαPIP did not exhibit detectable retention of PCNA on mismatched DNA , compared to the buffer control ( Figure 8D ) . In the gap-derived strand memory assay , xMutSαWT restored retention of strand memory in the MutS/MutL-depleted NPE without affecting the gap-filling reaction ( t1/2 = 25 . 5 min; Figure 9 , Figure 9—figure supplement 1 , and Table 2 ) . In contrast , only modest restoration of the retention of strand information was seen with xMutSαPIP ( t1/2 = 12 . 5 min ) . These results indicate that the PIP motif is required for inhibition of PCNA unloading and largely responsible for strand memory retention , yet a PIP-independent mechanism also contributes to the retention of strand information derived from a gap . 10 . 7554/eLife . 15155 . 025Figure 8 . The PIP-motif of xMutSα is important for retention of PCNA . ( A ) The primary structure of Msh6 and the mutation sites in the PIP motif are presented . ( B ) 1 μg each of purified xMutSα complexes were separated by SDS-PAGE and stained with Coomassie brilliant blue R-250 . ( C ) NPE was depleted of both MutS and MutL ( lanes 1–4 ) , MutS ( lane 5 ) , or MutL ( lane 6 ) and supplemented with either control buffer , 630 nM of xMutSαWT , xMutSαPIP , or xMutSαΔN . Immunoblots of each NPE are shown . ( D ) hPCNA loaded onto immobilized pMM2AC carrying an A-strand nick was incubated in NPE described in ( C ) . The efficiency of nick ligation in the PCNA loading reaction was estimated to be ~75% . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 02510 . 7554/eLife . 15155 . 026Figure 8—figure supplement 1 . Requirement of the PIP-containing N-terminal domain of MutSα in MMR and mismatch binding . ( A-C ) Effect of the MutSα mutant complexes on gap-directed MMR in NPE . ( A ) NPE was depleted with pre-immune antibodies ( lane 1 ) or a mixture of Msh2 and Msh6 antibodies ( lanes 2–5 ) , and supplemented with either control buffer ( lanes 1 and 2 ) , 630 nM of xMutSαWT ( lane 3 ) , xMutSαPIP ( lane 4 ) or xMutSαΔN ( lane 5 ) . The NPE samples were separated by SDS-PAGE and probed with the indicated antibodies . Orc2 served as a loading control . ( B ) The MMR assay in NPE . A-strand-gap-carrying pMM1AC were incubated in NPE described in ( A ) and sampled at the indicated times . %repair was calculated as described in Figure 1C . The xMutSαPIP and xMutSαΔN complexes nearly fully restored MMR in the MutS depleted NPE ( lanes 11–16 ) . ( C ) The efficiencies of MMR were calculated from three independent experiments including the one presented in ( B ) and plotted onto a graph . Error bars: ± 1 SD . ( D ) Dissociation constants of the MutSα mutants on mismatch-carrying DNA . The xMutSαWT , xMutSαPIP and xMutSαΔN complexes were flowed over a sensor chip coated with mismatch-carrying oligo DNA . The kinetics of association and dissociation of xMutSα were monitored by surface plasmon resonance using BIACORE 3000 ( GE Healthcare , Little Chalfont , UK ) . The KD of the xMutSαWT , xMutSαPIP and xMutSαΔN for the mismatched DNA were 2 . 2 nM , 4 . 4 nM and 3 . 6 nM , respectively ( n = 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 026 In yeast , the N-terminal region ( NTR ) of Msh6 not only carries a canonical PIP motif but also contains some cryptic PCNA binding sites ( Shell et al . , 2007 ) . To examine whether the NTR of Msh6 is responsible for the weak retention of strand memory , we purified xMutSα lacking the entire NTR of Msh6 ( Δ1–323; xMutSαΔN ) . As with xMutSαPIP , xMutSαΔN did not show significant defects in gap-directed MMR and in mismatch binding ( Figure 8—figure supplement 1 ) , and was essentially deficient in PCNA retention ( Figure 8C and D ) . Interestingly , the xMutSαΔN complex partially restored the strand memory reaction , with an estimated half-life of 12 . 3 min , although it did not affect the kinetics of the sealing of the gap in the first NPE ( Figure 9 and Table 2 ) . We concluded that the retention of strand memory is mediated largely by the PIP motif in MutSα , and there is a partial strand memory retention that is independent of the NTR of Msh6 . 10 . 7554/eLife . 15155 . 027Figure 9 . The PIP-motif of xMutSα is important for retention of the MMR capability derived from a gap . ( A ) Kinetics of the gap-filling reaction in MutS/MutL-depleted NPE containing 630 nM of xMutSαWT , xMutSαPIP , or xMutSαΔN . ( B ) The efficiencies of the gap-filling reaction were calculated from three independent experiments including the one shown in ( A ) and plotted onto a graph . Lines: one-phase decay fitting . See Table 2 for the fitting parameters . ( C ) Kinetics of MMR after supplying fresh NPE to the reaction described in ( A ) . Addition of neither MutS-depleted nor MutL-depleted NPE alone , but both restored MMR in MutS/MutL-depleted NPE ( lanes 2–4 ) , showing both MutSα/β and MutLα/β/γ were functionally depleted . ( D ) The efficiencies of MMR were calculated and presented as described in ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 02710 . 7554/eLife . 15155 . 028Figure 9—figure supplement 1 . Depletion efficiencies of Msh2 and Mlh1 from NPE . NPE was depleted using a mixture of Msh2 , Msh6 , and Mlh1 antibodies ( lanes 1–4 ) , a mixture of Msh2 and Msh6 antibodies ( lane 5 ) or Mlh1 antibodies ( lane 6 ) and supplemented with either control buffer , 630 nM of xMutSαWT , xMutSαPIP , or xMutSαΔN . The depletion efficiencies were estimated to be >98% for both Msh2 and Mlh1 . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 02810 . 7554/eLife . 15155 . 029Table 2 . Fitting parameters for the gap-derived strand memory experiment with MutSα mutantsDOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 029%gap [95% Confidence Interval]%A→G repair [95% Confidence Interval]t1/2 ( min ) %gap at 0 minR2t1/2 ( min ) %repair at 0 minR2+buffer1 . 44 [0 . 51−0 . 73]100 [95 . 2−104 . 7]0 . 992 . 94 [2 . 32−3 . 98]75 . 7 [69 . 2−82 . 2]0 . 97+xMutSαWT1 . 35 [0 . 58−0 . 81]100 [95 . 1−104 . 9]0 . 9925 . 5 [17 . 6−46 . 1]84 . 8 [80 . 5−89 . 2]0 . 98+xMutSαPIP1 . 48 [1 . 18−2 . 00]100 [95 . 4−104 . 6]0 . 9912 . 5 [10 . 9−14 . 7]81 . 9 [78 . 5−85 . 4]0 . 99+xMutSαΔN1 . 43 [1 . 14−1 . 93]100 [95 . 6−104 . 4]0 . 9912 . 3 [11 . 3−13 . 6]81 . 9 [79 . 8−84 . 0]1 . 00Parameters for the one-phase exponential decay fitting of the data described in Figure 9B and D are presented . %gap: percentage of remaining gaps ( 100 - %closed ) , t1/2: half-life , R2: coefficient of determination . n = 3 . Identification of the newly synthesized DNA strand is vital to post-replicative mismatch repair . Since DNA replication generates identical copies of parental DNA , the strand discrimination reaction in MMR necessarily requires direct or indirect interplay with the replication complex ( Wagner and Meselson , 1976 ) . Because of the asymmetric nature of its binding to DNA , its critical roles in an early step of MMR and in the activation of MutLα , PCNA is presumed to be the mediator of the interplay between the replication complex and MMR ( Umar et al . , 1996; Chen et al . , 1999; Pavlov et al . , 2003; Dzantiev et al . , 2004; Kadyrov et al . , 2006; Pluciennik et al . , 2010; Hombauer et al . , 2011b; Peña-Diaz and Jiricny , 2012; Georgescu et al . , 2015; Kunkel and Erie , 2015 ) . In this study , we provided experimental evidence that DNA-loaded PCNA indeed directs strand-specific MMR in the absence of strand discontinuities . We further discovered that MutSα inhibits PCNA unloading using its PIP motif on the N-terminus of Msh6 , and that this reaction significantly extends the temporal window during which MMR is possible . Our results thus uncovered a novel mechanism in which interplay between PCNA and MutSα maintains the post-replicative temporal window for MMR . NPE has been extensively used as a physiological model system that recapitulates various DNA repair reactions coupled with DNA synthesis ( Räschle et al . , 2008; Olivera Harris et al . , 2015 ) . As seen in other systems , we observed efficient bidirectional MMR in NPE ( frequently >80% efficient ) . Curiously , we noticed some difference between NPE and other systems . Both in reconstitution systems and in human cell extracts , 5’-nick directed MMR can occur in the absence of MutLα , whereas 3’-nick directed MMR depends on MutLα ( Drummond et al . , 1996; Genschel and Modrich , 2003; Dzantiev et al . , 2004; Constantin et al . , 2005; Bowen et al . , 2013 ) . In contrast , not only 3’-gap directed but also 5’-gap directed MMR was dependent on MutL complexes in NPE ( Figure 1B–E ) . Although this could possibly reflect difference in the 5’ to 3’ strand excision mechanism between species , we assume that the difference largely come from highly efficient DNA synthesis and ligation in NPE; quick sealing of a gap would necessitate MutLα-dependent incision for MMR even when a gap was placed on the 5’-side of the mismatch . Interestingly , DNA synthesis at the 3’-gap clearly precedes MMR ( Figure 1F ) , indicating that strand degradation does not progressively occur from the 3’-terminus of the gap . PCNA molecules loaded at the 3’ terminus , perhaps more than one trimer , could be used by both MMR and DNA polymerases without significantly interfering with each other ( see below ) . Consistently , a nick is suggested to function as a strand signal rather than an entry point for exonucleases in crude extracts of Xenopus eggs ( Varlet et al . , 1996 ) . Our stepwise incubation experiments demonstrated that MMR is possible even after the complete sealing of the gap ( Figure 2 ) . Three lines of evidence indicate that the molecule responsible for memorizing strand information is PCNA . First , directional loading of PCNA efficiently bypassed the need for strand discontinuities in strand-specific MMR ( Figure 3 and 4 ) . Second , the amount of DNA-bound PCNA correlated with the MMR capability ( Figure 6 ) . Third , MutSα maintained the MMR capability derived from a gap largely through the PIP motif ( Figure 9 ) . The fact that PCNA was essential for the gap-filling reaction in NPE also supports this conclusion , because at least one PCNA trimer must be involved in DNA synthesis at the gap . It is unlikely that RFC determines the strand specificity of MMR in our experiments , because amounts of RFC remained on DNA were insufficient to explain the efficiency of MMR . Based on the above evidence , we suggest that PCNA would function as the strand discrimination signal for MMR after the disappearance of local strand discontinuities during DNA replication . A series of seminal studies from the Modrich lab has led to the model in which DNA-bound PCNA activates the MutLα nicking endonuclease to initiate strand specific MMR ( Kadyrov et al . , 2006; 2007; Pluciennik et al . , 2010; 2013 ) . Our data are consistent with this prevailing model , since 1 ) DNA-bound PCNA induced strand-specific MMR in NPE , and 2 ) PCNA-directed MMR in NPE was completely dependent on the MutL complexes . As gap-directed MMR in NPE was bidirectional and preferentially degrades the DNA strand through the shortest path from the gap to the mismatch , we assume that the PCNA-directed MMR would function in the same manner . If so , the strand excision complex presumably built upon a specific face of PCNA would catalyze strand degradation in either direction depending on the relative orientation toward a mismatch . How this can be achieved is an important next question , and the system established here will be a useful tool to investigate this mechanism . However , since our plasmid substrates carry multiple PCNA molecules with free sliding ability , we currently do not exclude a possibility that PCNA-directed MMR catalyzes only a specific direction of strand excision , and PCNA molecules localized on a specific side toward a mismatch are preferentially used for MMR . Unexpectedly , the 'strand memory' experiments revealed that MutSα/β significantly extend the duration of strand memory . Our data clearly indicate that this retention involves inhibition of PCNA unloading . A recent study showed that DNA polymerase δ captures DNA-bound PCNA which is otherwise quickly unloaded by the clamp-unloading activity of RFC ( Hedglin et al . , 2013 ) . It would be possible that MutSα inhibits PCNA unloading in a similar manner; binding of the PIP motif onto PCNA may physically prevent the engagement of PCNA with RFC . Since in vivo PCNA unloading during S-phase largely depends on the Elg1-containing RFC-like complex ( Elg1-RLC ) both in humans and in yeast ( Kubota et al . , 2013; Lee et al . , 2013; Shiomi and Nishitani , 2013; Kubota et al . , 2015 ) , the PIP-motif in MutSα may also limit the access of Elg1-RLC to DNA-bound PCNA . Detailed understanding of the molecular mechanism of this inhibition of PCNA unloading must await reconstitution of this reaction in vitro . Interestingly , the PIP motif and the mispair-binding domain are connected by a long ( ~300 a . a . ) linker that is predicted to be disordered in yeast ( Shell et al . , 2007 ) . Although a significant portion of this linker in human MutSα seems to adopt a globular conformation ( Iyer et al . , 2008 ) , MutSα might be able to retain PCNA that is located far from MutSα using this long 'PIP-arm' , possibly even over nucleosomes . In this scenario , strand removal would initiate when MutLα reaches to PCNA that is retained by the PIP-arm of MutSα . This could be accomplished either by sliding of MutSα-MutLα complexes along the DNA contour , or through loading of multiple MutLα molecules on DNA ( Hombauer et al . , 2011a; Qiu et al . , 2015 ) . Interaction between PCNA that is retained by MutSα and MutLα may invoke a specific strand degradation pathway , because a recent study in yeast has suggested that recruitment or retention of PCNA by MutSα around mismatches activates a specific strand removal pathway that is independent of Exo1 ( Goellner et al . , 2014 ) . As seen for MutSα , we predict that MutSβ may inhibit PCNA unloading as well , since the NTR of yeast Msh6 and Msh3 are inter-exchangeable without significant deleterious effects ( Shell et al . , 2007 ) . However , because NTRs of MutSα and MutSβ likely possess different characteristics such as the PWWP domain found only in vertebrate Msh6 ( Clark et al . , 2007; Laguri et al . , 2008; Iyer et al . , 2010; Li et al . , 2013 ) , this point will need rigorous investigation . To our surprise , xMutSαPIP and xMutSαΔN retained partial strand memory . A possibility would be that , although our PCNA unloading experiment failed to detect attenuation of PCNA unloading with these mutants ( Figure 8D ) , they might have the ability to maintain DNA-bound PCNA that is below our detection limit . Considering the fact that we quantified PCNA after several washing steps , our quantification almost certainly underestimated the number of PCNA molecules on DNA . Alternatively , PCNA and/or a gap could induce structural alteration of MutSα ( and possibly other chromatin binding factors ) , to maintain the strand information . How MutSα retains strand memory independently of the NTR will be an interesting question for future studies . From the data presented in this work , we propose a model wherein two different modes of strand memory contribute to MMR ( Figure 10 ) . After the completion of local DNA synthesis , e . g . after ligation of Okazaki fragments , PCNA remains on DNA for a certain period until unloaded by an Rfc3-containing complex ( es ) ( t1/2 = ~2 min in NPE; Figure 10 , [2] ) . The MMR system utilizes such PCNA as strand memory . However , if PCNA is not available closely adjacent to mismatches , MutSα ( and possibly MutSβ also ) retains available PCNA with its long 'PIP-arm' , to maintain strand memory for subsequent MMR ( Figure 10 , [3] ) . This mode of memory survives for >30 min in NPE , and perhaps even longer in cells . In addition to the PIP-dependent tethering of MutSα/β to the replication fork ( Kleczkowska et al . , 2001; Hombauer et al . , 2011a; Haye and Gammie , 2015 ) , we suggest that the PIP motif would contribute to the replication fidelity through this mode of strand memory . Yeast studies have shown that the PIP motif has an assistive but not essential role in MMR ( Clark et al . , 2000; Flores-Rozas et al . , 2000 ) . The PIP-deficient MutSα complexes were proficient in repairing a mismatch in our in vitro MMR system ( Figure 8—figure supplement 1 ) , and similar observations were already reported both in human cell extracts and in a yeast in vitro reconstitution system ( Kleczkowska et al . , 2001; Iyer et al . , 2008; Bowen et al . , 2013 ) . Therefore , a simple MMR reaction involving a mismatch and a closely-placed strand-discrimination signal , either PCNA or a strand discontinuity , would not require retention of PCNA by MutSα . However , when a stepwise incubation , which mimics transient arrest of the MMR reaction , was involved , the PIP motif became important for MMR ( Figure 9 ) . We speculate that inhibition of PCNA unloading is needed for MMR only under specific situations , e . g . when a strand-discrimination signal is buried within chromatin . In such difficult-to-repair situations , successful MMR may rely on the maintenance of the strand signals by MutSα . Since the leading strand seems to have less PCNA than the lagging strand ( Yu et al . , 2014 ) , such situations could happen more frequently in the leading strand , in which MMR is less efficient than in the lagging strand ( Pavlov et al . , 2003; Lujan et al . , 2014 ) . An interesting possibility worth testing in the future would be that MutSα may take over PCNA from other factors such as DNA polymerases . Hijacking PCNA from polymerases may not be overly deleterious to DNA synthesis , because new PCNA loading at the 3’-terminus would quickly restore the PCNA-polymerase complex . Similar hijacking might occur on other PCNA interactors such as CAF-1 and FEN-1 , both of which are reported to show complex interplay with MutSα ( Kadyrova et al . , 2011; Schöpf et al . , 2012; Kadyrova et al . , 2015; Liu et al . , 2015 ) . On the leading strand that has less PCNA , such competition could be important for effective MMR . 10 . 7554/eLife . 15155 . 030Figure 10 . Two modes of strand memory maintain the MMR capability . Three mechanisms , including two 'strand-memory' mechanisms , may ensure eukaryotic MMR . ( 1 ) During ongoing DNA synthesis , strand discontinuities or polymerase-associated PCNA are directly used for strand discrimination . ( 2 ) After sealing of local strand discontinuities , PCNA that remains on DNA provides strand information until they are unloaded ( t1/2 = ~2 min in NPE; 'short-term' strand memory ) . ( 3 ) MutSα inhibits PCNA unloading to maintain the strand discrimination capability , largely through the PIP-motif on the NTR of Msh6 ( t1/2 = >30 min in NPE; 'long-term' strand memory ) . Black and blue lines represent the template and the newly synthesized DNA strands , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 15155 . 030 Much of our understanding of MMR has come from bacterial studies including E . coli , whose MMR utilizes adenine methylation as the strand discrimination signal . Although E . coli MMR seems to function with strand discontinuities occurring at the replication fork , GATC methylation plays a critical role in maintaining the temporal window permissive to MMR . Strand discontinuities are clearly strand discrimination signals in eukaryotic MMR in vitro ( Holmes et al . , 1990; Thomas et al . , 1991 ) . There is also concrete evidence that single-strand DNA termini such as 5’-ends of the Okazaki fragments or strand breaks generated by ribonucleotide excision repair contribute to MMR in vivo ( Pavlov et al . , 2003; Nick McElhinny et al . , 2010; Ghodgaonkar et al . , 2013; Liberti et al . , 2013; Lujan et al . , 2013; 2014; Liu et al . , 2015 ) . We suggest that , in addition to such DNA termini , two modes of strand memory , somewhat resembling short-term and long-term memories in neuroscience , would operate as functional parallels of GATC methylation in E . coli . Analogous reactions could operate in bacteria and archaea lacking the methyl-dependent MMR ( Pillon et al . , 2015 ) , and it would be interesting to study dynamics of the replication clamp in such organisms . Further investigation of the regulation of PCNA dynamics by MMR both in NPE and in vivo will lead to a more detailed understanding of complex interplay between the MMR system and the replication complex . In vitro synthesis of mismatch-carrying plasmids was performed essentially as described previously ( Higashi et al . , 2012 ) . A 5’-phosphorylated oligonucleotide ( 5’-CAGTAACATGGATCXCGAGATGCAGTACGGTCACC-3’; for homo-duplex , X=T , and for A:C mismatch , X=C ) was annealed to single-stranded DNA prepared by using the M13KO7 phage . To introduce a site-specific biotin modification , an additional oligonucleotide carrying a site-specific biotin-dT modification ( 5’-CGCCTTGATCGT[Biotin-dT]GGGAACCGGAGCTGAATGAAGC-3’ ) was also added . Second-strand DNA synthesis was performed by using T7 DNA polymerase ( New England Biolabs , MA , USA ) and T4 DNA ligase ( Nippongene , Tokyo , Japan ) . The mismatch-carrying DNA was treated with XhoI ( Takara Bio , Kusatsu , Japan ) to digest DNA whose mismatch base was edited by T7 DNA polymerase . The covalently closed products were purified with the Cesium Chloride density gradient ultracentrifugation . A site-specific gap was introduced as follows . In vitro synthesized pMM1 was doubly nicked with Nt . BbvCI ( for A-strand gap ) or Nb . BbvCI ( for C-strand gap ) at 37°C for 1 hr , purified , and incubated at 70°C for 20 min to dissociate the 15-nt fragment flanked by two BbvCI sites from parental DNA . The DNA was then immediately loaded on a Microspin S-400HR Column ( GE Healthcare , Little Chalfont , UK ) to remove the 15-nt fragment . A site-specific nick was introduced by treating in vitro synthesized pMM2 with Nt . BbvCI ( for A-strand nick ) or Nb . BbvCI ( for C-strand nick ) at 37°C for 1 hr . Singly biotinylated plasmid DNA was bound to Sepharose beads as described previously ( Higashi et al . , 2012 ) . All restriction enzymes were purchased from New England Biolabs unless otherwise indicated . Immobilized DNA ( 100 ng bound to 1 μl of Streptavidin-biotin-Sepharose beads ) was incubated in 2 vol of mHBS ( 10 mM Hepes-NaOH pH 7 . 5 , 0 . 05% Tween-20 , 10 mM MgCl2 , 200 μM EDTA , 150 mM NaCl ) , containing 50 mM phosphocreatine , 25 μg/ml creatine phosphokinase , 2 mM ATP , 400 μM DTT , 145 ng/μl hPCNA , and 2 . 2 ng/μl hRFC at 32°C for 10 min . All reagents were purchased from Sigma Aldrich ( MO , USA ) unless otherwise indicated . The hPCNA-DNA complexes were then washed thrice with mHBS . 1 vol of the DNA beads was then incubated in 4 vol of ligation buffer ( 66 mM Tris-HCl pH 7 . 4 , 6 . 6 mM MgCl2 , 10 mM DTT , 1 mM ATP ) containing 25 units/μl T4 DNA Ligase at 16°C for 90 min . This was followed by washing thrice with mHBS , once with Egg lysis buffer ( ELB; 10 mM Hepes-KOH pH 7 . 7 , 2 . 5 mM MgCl2 , 50 mM KCl ) containing 1 M KCl , and then once with ELB . To linearize DNA after PCNA loading , the hPCNA-DNA complexes were treated with either control buffer or buffer containing XmnI and washed thrice with mHBS . To recover DNA from NPE , the hPCNA-DNA complexes were washed once with ELB containing 0 . 2% Triton X-100 and once with ELB . NPE was supplemented with 2 mM adenosine-triphosphate ( ATP ) , 20 mM phosphocreatine , and 5 μg/ml creatine phosphokinase , pre-incubated at 22°C for 5 min , followed by addition of DNA substrates ( 20 ng/μl ) . A typical MMR reaction consists of 17 . 4 μl of NPE , 0 . 2 μl of 200 mM ATP , 0 . 4 μl of 1 M phosphocreatine , 0 . 02 μl of 5 mg/ml creatine phosphokinase , and 2 μl of DNA substrate ( 200 ng/μl in 10 mM Tris-HCl pH 7 . 4 , 1 mM EDTA ) . To monitor DNA synthesis , reaction mixtures were supplemented with 2 μCi of α-[32P]-dCTP ( PerkinElmer Japan , Tokyo , Japan ) before addition of the MMR substrates . For the stepwise incubation experiments , an equal volume of fresh NPE was added to the reaction mixture . The samples were incubated at 22°C , and aliquots of samples ( 1 . 5 ~ 3 μl for most experiments ) were stopped by addition of 100 μl of 1% SDS in 20 mM EDTA . DNA was purified by treatment with 50 μg/ml Proteinase K , extracted with Phenol/Chloroform , precipitated with ethanol , and dissolved in 10 mM Tris-HCl pH 7 . 4 , 1 mM EDTA containing 10 μg/ml RNase at 2 . 5 ng/μl . To analyze the MMR efficiency ( %repair ) , DNA synthesis ( %Gap-filling synthesis ) and the incorporation of radioactivity during MMR , 12 ng each of DNA was digested with following restriction enzymes: XmnI , BamHI-HF , XhoI , PacI or DrdI . After agarose gel electrophoresis , DNA was stained with SYBR Gold nucleic acid stain ( Life technologies , CA , USA ) , and scanned with Typhoon FLA9000 ( GE Healthcare ) . Signal intensities were quantified using ImageQuant TL software ( GE Healthcare ) . To analyze the gap-filling efficiency ( %closed ) , ethidium bromide containing agarose gel electrophoresis was performed to separate covalently closed DNA from gap/nick-carrying open-form DNA . The gel was stained with SYBR Gold nucleic acid stain , and then scanned by Typhoon FLA9000 using SYBR Gold specific setting ( 473 nm excitation laser and 510 nm Long pass filter ) . Calculation of %closed DNA is described in Figure 1—figure supplement 5 . To moniter DNA synthesis , DNA was separated by agarose gel , stained with SYBR Gold , photographed , vacuum-dried on filter paper , contacted on a phosphor imaging plate ( Fujifilm , Tokyo , Japan ) , and the 32P signals were scanned by Typhoon FLA9000 . All agarose gel electrophoresis was performed with 0 . 8% agarose gel and 0 . 5× Tris-borate-EDTA buffer . DNA samples were linearized by XmnI and separated using agarose gel electrophoresis , and quantified using known amounts of linear DNA as standards . Proteins recovered with the DNA-beads were separated by SDS-PAGE alongside with recombinant protein standards and probed with either hPCNA or hRfc2 antibodies . The number of hPCNA and hRFC on each plasmid ( bound PCNA [RFC] / plasmid ) were calculated by dividing the protein amounts from quantitative immunoblotting by bead-bound plasmid amounts calculated from linearized DNA . If 10 ng ( 5 . 2 × 10–15 mol ) DNA is loaded , 1 ng hPCNA ( Mr = 8 . 61 × 104 as a trimer , 1 . 2 × 10–14 mol ) and hRFC ( Mr = 2 . 87 × 105 as a complex , 3 . 5 × 10–15 mol ) correspond to binding of approximately 2 . 3 and 0 . 7 molecules per plasmid , respectively . Preparation of HSS and NPE was carried out as described previously ( Walter et al . , 1998; Lebofsky et al . , 2009 ) . In all experiments , plasmid DNA was incubated at 20 ng/μl concentration . The human Proliferating Cell Nuclear Antigen ( PCNA ) gene was amplified from pT7-PCNA ( Fukuda et al . , 1995 ) by PCR using primers 5’-GGAACATATGTTCGAGGCGCGCCTGGTCC-3’ and 5’-GGAAGGATCCCTAAGATCCTTCTTCATCCTCG-3’ , digested with NdeI and BamHI , and cloned into pET21a ( Merck Millipore , MA , USA ) , resulting in pET21a-hPCNA . The Xenopus laevis pcna gene was amplified from a Xenopus egg cDNA library ( a kind gift from Vladimir Joukov ) by PCR using primers 5’-GGAACATATGTTTGAGGCTCGCTTGGTGC-3’ and 5’-GGAAGGATCCTTAAGAAGCTTCTTCATCTTCAATCTTGG-3’ , digested with NdeI and BamHI , and cloned into pET21a , resulting in pET21a-xPCNA . The Xenopus laevis msh2 gene was amplified from the Xenopus egg cDNA library by two-step PCR using primers 5’-AAAGCAGGCTCCACCATGGCTGTGCAGCCCAAAGAGAAGTTG-3’ and 5’-ACAAGAAAGCTGGGTCTCCTGCAGGCAATCCCGTTTTGGTTCTGG-3’ , and then primers 5’-GGGGACAAGTTTGTACAAAAAAGCAGGCTCCAC-3’ and 5’-GGGGACCACTTTGTACAAGAAAGCTGGGTC-3’ , and cloned into pDONR201 ( Life technologies ) using the Gateway BP reaction , resulting in pDONR-xMSH2 . Cloning of Xenopus laevis msh6 gene was performed as follows . A BLAST search using the human Msh6 protein sequence identified a Xenopus laevis EST clone , TC357542 . Based on this sequence , the full-length msh6 cDNA was amplified from Xenopus egg cDNA by two-step PCR using primers 5’-AAAGCAGGCTCCACTCATATGTCTAAGCAAAAAACCCTCTTCAGCTTCTTCACC-3’ and 5’-ACAAGAAAGCTGGGTTGGTACCTTGGAGCAACTTCAGCCGCTTGTGG-3’ , and then primers 5’-GGGGACAAGTTTGTACAAAAAAGCAGGCTCCAC-3’ and 5’-GGGGACCACTTTGTACAAGAAAGCTGGGTC-3’ , and cloned into pDONR201 using the Gateway BP reaction , resulting in pDONR-xMSH6 . A FLAG tag was added to the C-terminus of xMsh6 by two-step PCR using primers 5’-AAACATATGTCTAAGCAAAAAACCCTCTTCAGC-3’ and 5’-GTCGTCCTTGTAGTCGGTACCGCCTTGGAGCAACTTCAGCCGC-3’ , and then primers 5’-AAACATATGTCTAAGCAAAAAACCCTCTTCAGC-3’ and 5’-GGAACCTGCAGGTTACTTGTCATCGTCGTCCTTGTAGTCGG-3’ , digested with NdeI and Sse8387I , and cloned into pDE1a , a derivative of the pDONR201 vector , resulting in pDE1a-xMSH6-FLAG . The xMsh6 PCNA-interacting peptide ( PIP ) motif mutant ( xMsh6PIP ) was constructed by two-step PCR using pDE1a-xMSH6-FLAG as the template and primers 5’-AAAACCCTCTTCAGCGCGGCGACCAAGTCTCCCCCTGTTTCC-3’ and 5’-CCGTCCCCTCCTTGACTGTACTG-3’ , and then primers 5’-AAACATATGTCTAAGGCGAAAACCCTCTTCAGC-3’ and 5’-CCGTCCCCTCCTTGACTGTACTG-3’ , digested with NdeI and XhoI , and cloned into pDE1a-xMSH6-FLAG , resulting in pDE1a-xMSH6PIP-FLAG . The xMsh6 N-terminal deletion mutant ( xMsh6ΔN ) was constructed by PCR using pDE1a-xMSH6-FLAG as the template and primers 5’-GGAACATATGTCTGCCCCTGAGTCATTTGAATCACAGGC-3’ and 5’-CCATGCGCCGACTTGTCTTGGC-3’ , digested with NdeI and XmnI , and cloned into pDE1a-xMSH6-FLAG , resulting in pDE1a-xMSH6ΔN-FLAG . The Xenopus laevis mlh1 gene was amplified from Xenopus egg cDNA by PCR using primers 5’-AAAGCAGGCTCCACCATGGCGGGAGTTATTCGGCGGCTGG-3’ and 5’-ACAAGAAAGCTGGGTCTCCTGCAGGGCACCTTTCAAACACTTTATATAAGTCGGG-3’ , then primers 5’-GGGGACAAGTTTGTACAAAAAAGCAGGCTCCAC-3’ and 5’-GGGGACCACTTTGTACAAGAAAGCTGGGTC-3’ , and cloned into pDONR201 using the Gateway BP reaction , resulting in pDONR-xMLH1 . All sequences were confirmed after each PCR step . Baculoviruses for expression of xMsh2 , xMsh6WT-FLAG , xMsh6PIP-FLAG and xMsh6ΔN-FLAG were constructed by transferring xMSH2 , xMSH6-FLAG , xMSH6PIP-FLAG and xMSH6ΔN-FLAG genes into BaculoDirect C-term Linear DNA ( Life technologies ) using the Gateway LR reaction . Construction of pMM0 , pMM1 and pMM2 was performed as follows: A synthetic linker prepared by annealing of 5’-phosphorylated oligonucleotides 5’-GAATTCAAGCTTAGTCTGTTCCATGTCATGCAAGATATCTTCAGTC-3’ , 5’-ACTGGGTGACCGTACTGCATCTCGAGATCCATGTTACTGCGTCAGT-3’ , 5’-CGCTAACAGTCACGAACTGCTGCAGGAATTCGTAC-3’ , 5’-GAATTCCTGCAGCAGTTCGTGACTGTTAGCGACTGACGCAGTAACA-3’ , 5’-TGGATCTCGAGATGCAGTACGGTCACCCAGTGACTGAAGATATCTT-3’ , and 5’-GCATGACATGGAACAGACTAAGCTTGAATTCAGCT-3’ , was inserted between the KpnI and SacI sites in pBluescriptII KS ( - ) ( Stratagene , CA , USA ) , resulting in pMM0 . A synthetic linker carrying two BbvCI sites prepared by the annealing of 5’-phosphorylated oligonucleotides 5’-GCTCCTCAGCTTAATTAACCTCAGC-3’ and 5’-AGCGCTGAGGTTAATTAAGCTGAGG-3’ was inserted into the BspQI site in pMM0 , resulting in pMM1 . A synthetic linker carrying one BbvCI site prepared by annealing of 5’-phosphorylated oligonucleotides 5’-GCTCCTCAGCATATGCCTCGC-3’ and 5’-AGCGCGAGGCATATGCTGAGG-3’ was inserted into the BspQI site in pMM0 , resulting in pMM2 . Purification of hPCNA was carried out essentially as described previously ( Fukuda et al . , 1995 ) , with minor modifications . Protein expression was induced in the Escherichia coli BL21 ( DE3 ) strain transformed with pET21a-hPCNA by addition of 0 . 1 mM Isopropyl β-D-1-thiogalactopyranoside ( IPTG ) to media for 20 hr at 20°C . hPCNA was purified using DEAE Sepharose Fast Flow , HiTrap Q-HP , Hi Load 16/60 Superdex 200 prep grade , and then MonoQ 5/50 GL ( GE Healthcare ) in this order . Recombinant xPCNA was expressed and purified using essentially the same method as hPCNA . Purification of the hRfc1-5 complex has been described previously ( Ohta et al . , 2002; Shiomi et al . , 2004 ) . Purification of xMutSα was performed as follows: Recombinant proteins were expressed by co-infecting Sf9 insect cells with xMsh2 and xMsh6WT-FLAG , xMsh6PIP-FLAG or xMsh6ΔN-FLAG baculoviruses at 28°C in Sf-900 II SFM medium ( Life Technologies ) supplemented with 2% ( v/v ) fetal bovine serum ( FBS ) . Cells were harvested , washed with Phosphate Buffered Saline ( PBS ) and frozen in liquid nitrogen . Cells were suspended in buffer S ( 25 mM Tris-HCl pH 7 . 4 , 250 mM NaCl , 5 mM 2-mercaptoethanol , 1 mM EDTA , 2 mM phenylmethylsulfonyl fluoride [PMSF] , 1 mM benzamidine and 1x cOmplete , EDTA-free [Roche life science , Penzberg , Germany] ) , and the lysates were centrifuged at 81 , 800 ×g ( 30 , 000 rpm ) for 30 min in a Beckman 50 . 2Ti rotor ( Beckman Coulter , CA , USA ) . Cleared lysates were passed through a DEAE Sepharose Fast Flow column , and then a FLAG-M2 agarose column ( Sigma Aldrich ) . The xMsh2-xMsh6-FLAG complexes were eluted from the FLAG-M2 resin using 50 μg/ml FLAG-peptide ( Sigma Aldrich ) in buffer S . The peak fractions were pooled and three-fold diluted with buffer A ( 25 mM Tris-HCl pH 7 . 4 , 5% glycerol , 5 mM 2-mercaptoethanol , 1 mM EDTA and 0 . 1x cOmplete , EDTA-free ) , loaded on a MonoQ 5/50 GL column , and bound proteins were eluted with a 0–1 M NaCl linear gradient in buffer A . Peak fractions were loaded on a Hi Load 16/60 Superdex 200 prep grade column and eluted with buffer A containing 0 . 1 M NaCl . Fractions whose molecular weight correspond to ~2 . 5 × 105 ( xMsh2: Mr = 1 . 04 × 105 , xMsh6-FLAG: Mr = 1 . 50 × 105 ) were pooled , concentrated using Amicon Ultra ( Merck Millipore ) , and frozen with liquid nitrogen as small aliquots . Rabbit xMsh2 antiserum was raised against N-terminally His-tagged and C-terminally Strep-II-tagged full-length Xenopus Msh2 expressed in E . coli . Rabbit xMsh6 antiserum was raised against peptide NH2-CNGSPEGLALHKRLKLLQ-COOH , corresponding to residues 1324–1340 of Xenopus Msh6 . Rabbit xMlh1 antiserum was raised against N-terminally His-tagged , full-length Xenopus Mlh1 expressed in E . coli . Rabbit xPCNA antiserum was raised against full-length Xenopus PCNA expressed in E . coli . Rabbit xRfc3 antiserum was raised against peptide NH2-CKKFMEDGLEAMMF-COOH , corresponding to residues 344–356 of Xenopus Rfc3 . All antibodies except for xMlh1 and xPCNA were affinity purified using corresponding antigens . Rabbit xOrc2 antiserum was a kind gift from Johannes Walter ( Vashee et al . , 2003 ) . hPCNA antibodies ( MBL International Corporation , Nagoya , Japan , #MH-12-3 ) and hRfc2 antibodies ( Abcam , Cambridge , UK , #ab3615 ) are commercially available . For immunoblotting , xMsh2 , xMsh6 , xMlh1 , xOrc2 , xPCNA and xRfc3 antibodies were used at 1:10 , 000 dilutions , hPCNA antibodies were used at a 1:2000 dilution , and hRfc2 antibodies was used at a 1:4000 dilution . HRP-conjugated Goat Rabbit IgG ( H+L ) antibodies ( Jackson ImmunoResearch , PA , USA , #111-035-003 ) , Goat Mouse IgG ( H+L ) antibodies ( #115-035-146 ) or Donkey Goat IgG ( H+L ) antibodies ( #705-035-003 ) were used at a 1:10 , 000 dilution as the secondary antibody for all immunoblottings except for quantitative immunoblottings for hPCNA , for which the same antibodies conjugated with Alexa Fluor 647 was used . To evaluate the depletion efficiency , 0 . 125 μl of NPE was loaded on SDS-PAGE , unless otherwise indicated . For xMsh2/xMsh6 double depletion from NPE , 2 μg of Msh2 IgG , 0 . 5 μg of Msh6 IgG , and 2 . 9 μl of Msh6 serum were bound to 1 μl of the recombinant protein A-Sepharose ( 'PAS' , GE Healthcare ) . For xMlh1 depletion , 3 vol of xMlh1 serum was bound to 1 vol of PAS . For xMsh2/xMsh6 and xMlh1 triple depletion , 2 μl of PAS bound 4 μg of Msh2 IgG , 1 μg of Msh6 IgG , and 5 . 8 μl of Msh6 serum was combined with 1 μl of PAS bound 3 μl of xMlh1 serum . For xRfc3 depletion from NPE , 5 μg of xRfc3 IgG was bound to 1 μl of PAS . To deplete NPE except for the depletion of xMsh2/xMsh6 and xMlh1 triple depletion , 0 . 2 vol of antibody-coupled PAS beads were incubated in 1 vol of extracts at 4°C for 1 hr , and the procedure was repeated once in all depletion experiments except for the depletion of xMsh2/xMsh6 or xRfc3 from NPE , for which the procedure was repeated twice . For xMsh2/xMsh6 and xMlh1 triple depletion from NPE , 0 . 3 vol of antibody-coupled PAS beads were incubated in 1 vol of extracts at 4°C for 1 hr , and the procedure was repeated twice . For xPCNA depletion from HSS , 3 vol of xPCNA serum was bound to 1 vol of PAS . A total of 0 . 2 vol of antibody-coupled PAS beads was incubated in 1 vol of HSS at 4°C for 1 hr , and the procedure was repeated thrice . In most cases , we depleted 40 ~ 60 μl of extracts for an experiment . Affinity of the xMutSα complexes to DNA was analyzed using BIACORE 3000 ( GE Healthcare ) ( Lee and Alani , 2006; Shell et al . , 2007 ) . A 42 bp biotin-conjugated oligo DNA 5’-GGGTGACCGTACTGCATCTCGAGATCCATGTTACTGCGTCAG-3’-[Biotin] was coupled to a Sensor Chip SA ( GE Healthcare ) until the SPR signal reached to ~100 response units . The sequence around the mismatch on pMM1AC was used to design the DNA substrate . A complementary oligonucleotide carrying a mismatch base 5’-CTGACGCAGTAACATGGATCCCGAGATGCAGTACGGTCACCC-3’ was then flowed in on the Sensor tip to obtain double stranded DNA substrates carrying an A:C mismatch . Various concentrations of xMutSαWT , xMutSαPIP and xMutSαΔN complexes were flowed over the sensor chip for 3 min at 20 μl/min in running buffer ( 20 mM Hepes-NaOH pH 7 . 5 , 100 mM NaCl , 1 mM DTT and 0 . 005% Tween-20 ) to analyze the association step , followed by running buffer containing no protein to monitor the dissociation step for 3 min . The signal from an empty flow cell was used for reference subtraction for all experiments . The chip surface was regenerated with 3 M NaCl for 1 min . Dissociation constants ( KD ) were calculated using the BIAevaluation software v4 . 1 . The GenBank accession numbers for sequences of Xenopus msh2 , msh6 and mlh1 mRNA reported in this paper are LC075519 , LC075520 , and LC075521 , respectively .	To pass on genetic information from one generation to the next , the DNA in a cell must be precisely copied . DNA is made of two strands and genetic information is encoded by sequences of molecules called bases in the strands . The bases from one strand form pairs with complementary bases on the other strand . However , errors in the copying process result in unmatched pairs of bases . Such errors are corrected by a repair system called mismatch repair . When DNA is copied , the two strands are separated and used as templates to make new complementary strands . This means that errors only arise on the new strands . Mismatch repair must therefore target the new strands to maintain the original information encoded by the template DNA . The repair needs to happen before the copying process is complete because the template strands and the new strands become indistinguishable afterwards . However , it is not clear how the two processes communicate with each other . Previous studies have identified a ring-shaped molecule called the replication clamp – which is essential for the copying process – as a prime candidate for the molecule responsible for this communication . This molecule binds to the DNA to promote the copying process , and afterwards it is removed from the DNA by other molecules . Furthermore , a group of proteins called the MutSα complex , which recognizes unmatched bases in DNA molecules , physically interacts with the replication clamp . Kawasoe et al . used eggs from African clawed frogs to study how the replication clamp connects the copying process and mismatch repair in more detail . The experiments show that when the replication clamp is bound to the DNA , it is able to direct mismatch repair to a specific DNA strand . When MutSα recognizes unmatched bases , it prevents the replication clamp from being removed from the DNA . By doing so , MutSα prevents the information about the new DNA strand from being lost until mismatch repair has taken place . These findings reveal new interactions between DNA copying and the correction of errors by mismatch repair . The next steps will be to understand how MutSα is able to keep the replication clamp on the DNA and to clarify its role in protecting DNA from gaining mutations .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "biochemistry", "and", "chemical", "biology" ]	2016	MutSα maintains the mismatch repair capability by inhibiting PCNA unloading
Working memory ( WM ) , the ability to actively hold information in memory over a delay period of seconds , is a fundamental constituent of cognition . Delay-period activity in sensory cortices has been observed in WM tasks , but whether and when the activity plays a functional role for memory maintenance remains unclear . Here , we investigated the causal role of auditory cortex ( AC ) for memory maintenance in mice performing an auditory WM task . Electrophysiological recordings revealed that AC neurons were active not only during the presentation of the auditory stimulus but also early in the delay period . Furthermore , optogenetic suppression of neural activity in AC during the stimulus epoch and early delay period impaired WM performance , whereas suppression later in the delay period did not . Thus , AC is essential for information encoding and maintenance in auditory WM task , especially during the early delay period . Working memory ( WM ) refers to the ability to actively hold information in memory over a time scale of seconds . It is a fundamental component of various cognitive functions ( Baddeley , 1992 ) . Previous studies have found that the prefrontal cortex ( PFC ) is crucial for WM because sustained neural activity was observed in PFC during the delay period of WM task ( Fuster and Alexander , 1971; Meyers et al . , 2012; Miller et al . , 1996; Romo et al . , 1999 ) and lesions to PFC produced profound WM deficits ( Petrides , 2000 ) . Besides , there is increasing evidence suggesting that sensory cortices are important components of the circuitry that underlies WM when the task requires short-term retention of sensory information ( Gayet et al . , 2018; Pasternak and Greenlee , 2005; Scimeca et al . , 2018 ) . For example , human fMRI studies had been successful in decoding WM information from the activity of sensory cortices , both visual ( Harrison and Tong , 2009; Serences et al . , 2009 ) and auditory ( Linke and Cusack , 2015; Linke et al . , 2011 ) . Electrophysiological recordings in monkeys had observed memory-related activity during the delay period in sensory cortices in visual ( Supèr et al . , 2001 ) , auditory ( Gottlieb et al . , 1989 ) , and somatosensory ( Zhou and Fuster , 1996; Zhou and Fuster , 2000 ) modalities . However , whether and when sensory cortices play a functional role in sensory memory maintenance in WM task remains unclear . The auditory cortex ( AC ) is known to be a major site for auditory information processing . Some recent studies have reported that AC neurons also play a role in a number of auditory behavior tasks ( Fritz et al . , 2003; Lee and Middlebrooks , 2011; Niwa et al . , 2012; Otazu et al . , 2009 ) . Furthermore , correlates of auditory WM have been reported in AC ( Bigelow et al . , 2014; Scott et al . , 2014 ) . Here , we delineate the functional role of AC in auditory WM . Traditional methods for perturbing neural activity such as surgical lesion , pharmacological inactivation , and tissue cooling techniques do not provide the temporal resolution required for delineating the functional role of AC in memory maintenance in WM task because they disrupt both the encoding of the sensory stimuli and their retention in WM . Furthermore , the activity could not be silenced rapidly enough to test when AC contributes to memory maintenance during the delay period . In the present study , we used a combination of optogenetics and electrophysiology methods to examine the functional role of AC in auditory WM . Because we were particularly interested to know when AC contributes to auditory memory maintenance , we applied optogenetic silencing at different time points across the delay period . We found that AC neurons exhibited elevated activity not only during the presentation of the auditory stimulus but also during the early delay period . Furthermore , optogenetic suppression of AC activity during the stimulus epoch and early delay period caused a reduction in WM performance , whereas suppression later in the delay period did not . These findings reveal a causal role of AC in encoding and maintaining the auditory information in the auditory WM task , especially early in the delay period . We trained head-fixed mice to perform an auditory delayed match to sample ( DMS ) task ( Figure 1a ) . In this task , water-restricted mice were presented a 200 ms auditory stimulus ( 3 kHz or 12 kHz tone ) as the sample , followed by a delay period ( 1 . 5 s ) and then a testing auditory stimulus , either matched or nonmatched to the sample . Licking within a response time window ( 1 s ) in the match trial was rewarded with water ( Figure 1b ) . Thus , mice had to remember the briefly presented auditory sample stimulus during the delay period . The behavioral performance declined with increasing duration of the delay period , indicating that the task required short-term memory processes ( Figure 1d ) . To examine the neural correlate of auditory WM in AC , we recorded the single-unit activity of AC by using tetrodes while mice were performing the auditory DMS task ( n = 13 mice , Figure 1e ) . A total of 915 neurons were recorded , of which 287 ( 31 . 4% ) exhibited activity related to the task ( compared with baseline activity , evaluated with a paired t test , at the p<0 . 05 level ) and were selected for further analysis . Responses from one typical AC neuron are shown in Figure 2a left . The neuron exhibited a phasic response during the auditory sample stimulus presentation . After the offset of the sample stimulus , the neuron continued to exhibit activity for 800 ms into the delay period . This response pattern was representative of our population ( Figure 2b ) . Figure 2c shows that more than 60% of AC neurons exhibited increased firing rate during the first 500 ms of the delay . The incidence decreased to ~30% during 500–800 ms of the delay . For the majority of these units ( ~90% ) , the elevated firing rate appeared in the first 800 ms of the delay period and did not persist any further . As shown in the example ( Figure 2a left ) and population ( Figure 2e ) , the firing rate of AC neurons showed selectivity of the preceding sample stimulus during both the auditory stimuli and early delay . To quantify the ability to discriminate between the two stimuli , we performed a receiver operating characteristic ( ROC ) analysis . The mean ROC values for the population of AC neurons increased rapidly after the onset of the auditory stimulus and retained above the value of 0 . 5 during the early delay ( <800 ms ) ( Figure 2g ) . The permutation test also showed that about 50% of neurons showed significant sample selectivity during the stimulus and early delay ( 500 ms after stimulus offset ) ( p<0 . 05 ) . The proportion decreased to 20–30% during the subsequent period of 500–800 ms and dropped to about 10% after 800 ms ( Figure 2h ) . These results indicated that AC neurons carried the auditory information during the early delay period in WM task . To determine whether this delay-period activity was related to the task or only reflected the continuation of the auditory response , we recorded 836 neurons during the passive presentation of the stimulus . 255 neurons responded to at least one auditory sample stimulus ( compared with baseline activity , paired t-test , p<0 . 05 ) . In the passive presentation condition , AC neurons exhibited robust responses and showed stimulus selectivity during the sample . They showed selectivity for a much shorter interval in the delay period , in contrast to results obtained during WM task ( Figure 2 ) . These results indicated that the continuation of neural activity into the delay period in WM task was not a passive process but was related to the animals’ behavior . We observed that after the offset of the 200 ms sample stimulus , the neuron continued to exhibit activity for 800 ms into the delay period , 1000 ms from the stimulus onset in total . To test whether the temporal dynamics of delay-period activity is relative to the stimulus onset or offset , we increased the stimulus duration to 300 ms and 400 ms and test whether the sustained activity would shift with later stimulus offset . The AC activity was recorded from a subgroup of mice ( n = 4 ) while performing the auditory WM task with the stimulus duration of 300 ms or 400 ms . The averaged population firing rates showed that the neurons exhibited phasic responses during the auditory sample stimulus presentation . After the offset of the sample stimulus , the neurons continued to exhibit elevated activity for 700 ms into the delay period in the 300 ms stimulus duration task and exhibit elevated activity for 600 ms into the delay period in the 400 ms stimulus duration task ( Figure 2—figure supplement 1 ) . Together with the result from the 200 ms sample stimulus duration task , which showed that the neurons continued to exhibit elevated activity for 800 ms into the delay period , all of these results showed that AC neurons exhibited elevated activity for 1000 ms from the stimulus onset , regardless of the duration of the sample stimulus . These results suggested that the temporal dynamics of the delay-period activity in AC might be relative to the stimulus onset rather than the offset . To directly test whether and when AC contributes to auditory WM , we suppressed the activity of AC pyramidal neurons transiently during the delay period with optogenetic methods . It was achieved through expressing the inhibitory halorhodopsin ( eNpHR3 . 0 ) in pyramidal neurons by injecting AAV-CaMKIIα-eNpHR3 . 0-eYFP into AC . The expression and functionality of NpHR were verified by immunostaining and optetrode recording ( Figure 3a , b ) . We applied optogenetic suppression during the delay period by randomly interleaving ‘laser ON’ and ‘laser OFF’ trials in the same behavioral session . Suppressing the activity of AC pyramidal neurons during the delay period caused a strong impairment in task performance ( t-test , p=1 . 1 × 10−5 , laser ON: 64 . 84 ± 1 . 59%; laser OFF: 77 . 98 ± 1 . 17% , n = 8 ) , with a substantial increase in false alarm rate ( t-test , p=2 . 63 × 10−5 , laser ON: 50 . 96 ± 2 . 65%; laser OFF: 31 . 66 ± 1 . 71% , n = 8 ) and a small decrease in hit rate ( t-test , p=0 . 018 , laser ON: 80 . 52 ± 1 . 74%; laser OFF: 87 . 69 ± 2 . 04% , n = 8 ) ( Figure 3c , d ) . Thus , the delay-period activities of AC pyramidal neurons were important for maintaining the auditory information after the auditory stimulus ceased . Next , we test the necessity of AC during the stimulus epoch . Optogenetic suppression of AC during the stimulus epoch dramatically reduced the animals’ ability to perform the task ( t-test , p=4 . 92 × 10−7 , laser ON: 61 . 71 ± 2 . 12%; laser OFF: 81 . 76 ± 1 . 3% , n = 9 ) , ( Figure 3—figure supplement 1 ) . This result indicated that AC was crucial for the initial encoding of auditory information in the WM task . For a genuine WM task , subjects should be able to perform beyond two cues . To further test the role of AC in auditory WM with more cues , a subgroup of mice ( n = 7 ) was tested in a four tones auditory WM task ( 3 kHz , 6 kHz , 12 kHz , 24 kHz ) . Again , optogenetic suppression of AC during the delay period resulted in substantial impairment in task performance ( t-test , p=0 . 025 , laser ON: 63 . 69 ± 2 . 58%; laser OFF: 73 . 48 ± 2 . 84% , n = 7 ) ( Figure 3e ) . Electrophysiology results showed that AC neurons continued to exhibit activity for 800 ms into the delay period , and these activities did not persist any further . We speculated that AC neurons might carry auditory information only during this early delay period . Therefore , optogenetic suppression of AC would be expected to affect WM only when applied during this early delay period . To test this idea , we divided the delay into two epochs , 300–800 ms and 800–1300 ms , and applied optogenetic suppression briefly within each epoch individually . Notably , optogenetic suppression during the epoch of 300–800 ms caused a strong reduction in behavioral performance ( t-test , p=0 . 01 , laser ON: 70 . 36 ± 2 . 5%; laser OFF: 80 . 51 ± 1 . 58% , n = 8 ) ( Figure 4a ) . Optogenetic suppression during the epoch of 800–1300 ms produced a smaller effect that was not significantly different from the laser-off condition ( t-test , p=0 . 34 , laser ON: 81 . 68 ± 1 . 72%; laser OFF: 83 . 91 ± 1 . 47% , n = 8 ) ( Figure 4b ) . To examine the temporal specificity of the effect of AC suppression more precisely , we used even shorter inactivation periods: 300–550 ms , 550–800 ms , 800–1050 ms , and 1050–1300 ms . Again , only inactivation during the 300–550 ms and 550–800 ms had an effect on behavioral performance ( 300–550 ms: t-test , p=0 . 016 , laser ON: 73 . 11 ± 2 . 23% , laser OFF: 79 . 38 ± 1 . 4% , n = 8; 550–800 ms: t-test , p<0 . 001 , laser ON: 73 . 72 ± 2 . 82% , laser OFF: 86 . 31 ± 1 . 13% , n = 8 ) , while the effect of inactivation after 800 ms was small and not statistically significant ( 800–1050 ms: t-test , p=0 . 38 , laser ON: 77 . 93 ± 2% , laser OFF: 80 . 68 ± 2 . 25% , n = 7; 1050–1300 ms: t-test , p=0 . 42 , laser ON: 84 . 97 ± 1 . 57% , laser OFF: 82 . 74 ± 2 . 19% , n = 8 ) ( Figure 4c–e ) . To further examine whether the temporal specificity of the AC suppression effect changes with increasing delay length , we increased the delay length to 3 s and 7 s . Again , only perturbation during 300–800 ms had a significant disruption on behavior ( 3 s delay , t-test , p=1 . 16 × 10−3 , laser ON: 71 . 69 ± 1 . 79% , laser OFF: 81 . 81 ± 1 . 73% , n = 8; 7 s delay , t-test , p=0 . 033 , laser ON: 72 . 8 ± 2 . 13% , laser OFF: 80 . 38 ± 2 . 32% , n = 7 ) ( Figure 5a , c ) . No obvious effects were seen when optogenetic suppression occurred after 800 ms ( 3 s delay , t-test , p=0 . 39 , laser ON: 82 . 53 ± 1 . 92% , laser OFF: 85 . 33 ± 2 . 51% , n = 8; 7 s delay , t-test , p=0 . 17 , laser ON: 81 . 25 ± 1 . 83% , laser OFF: 84 . 69 ± 1 . 49% , n = 7 ) ( Figure 5b , d ) . These results indicated that AC is critical for WM only during the early delay period . Active WM maintenance requires resistance to distractors presented during the delay period . To test this ability in mice , we added a noise distractor ( 20–20 , 000 Hz , 200 ms , 60 dB ) during the early ( 300–500 ms ) delay period of the WM task . Mice were first trained to perform the WM task , and then the noise distractor was added . Mice could quickly adapt to the WM task with noise distractor , despite the initial drop in performance . After 2 days of training , the performance on the third day of the WM task with noise distractor was no worse than that in the simple WM task ( t-test , p=0 . 247 , WM task with noise distractor: 79 . 55 ± 1 . 77%; simple WM task: 82 . 4 ± 1 . 58% , n = 9 ) ( Figure 6a ) . We then optogenetically suppressed the activity of AC pyramidal neurons after the noise distractor in the WM task . Optogenetic suppression of AC during the early delay period after the distractor ( 500–800 ms ) resulted in impairment in task performance ( t-test , p=0 . 017 , laser ON: 69 . 83 ± 2 . 8%; laser OFF: 77 . 32 ± 1 . 63% , n = 9 ) , with a substantial increase in false alarm rate ( t-test , p=0 . 016 , laser ON: 45 . 6 ± 4 . 48%; laser OFF: 33 . 49 ± 2 . 5% , n = 9 ) and no significant change in hit rate ( t-test , p=0 . 156 , laser ON: 85 . 29 ± 1 . 68%; laser OFF: 88 . 41 ± 1 . 26% , n = 9 ) ( Figure 6b ) . Thus , AC activity is important for active early maintenance of the auditory information in the face of noise distractor in the WM task . To study the functional specificity of optogenetic suppression , we designed an additional series of experiments . First , we trained another group of mice to perform a delayed go/no-go auditory discrimination task . The stimuli were the same as the DMS task , except that mice made a decision depending on the first stimulus ( Figure 7a ) . This task required auditory perception , the memory of the task-relevant information , but not auditory frequency information retention during the delay period ( Goard et al . , 2016; Kamigaki T and Kamigaki and Dan , 2017 ) . Laser illumination was applied after the sample stimulus and before the go signal , simulating the delay period in the DMS task . We found that laser illumination had no effect on the behavior ( t-test , p=0 . 335 , laser ON: 76 . 81 ± 2 . 01% , laser OFF: 79 . 9 ± 2 . 34% , n = 8 ) ( Figure 7a ) . Therefore , AC delay-period activity appeared to be more important in memory of the auditory cue but not the action plan . Second , mice were trained to perform a go/no-go auditory discrimination task . The stimuli were the same as the DMS task , except that mice made a decision based on the second stimulus ( Figure 7b ) . This task retained the general processes such as attention or expectation of the forthcoming second stimulus but did not require auditory WM during the delay period . Laser illumination was applied after the starting cue and before the sample stimulus , simulating the delay period in the DMS task . The result showed that laser illumination produced a smaller effect that was not significantly different from the laser-off condition ( Wilcoxon rank-sum test , p=0 . 536 , laser ON: 81 . 55 ± 3 . 1% , laser OFF: 84 . 71 ± 2 . 2% , n = 9 ) ( Figure 7b ) . The effect of AC suppression in go/no-go task is much smaller than that in the DMS task ( t-test , p=7 . 68×10−3 , Δperformance [laser ON-laser OFF] , go/no-go = −3 . 15 ± 2 . 5% , n = 9; DMS = -13 . 14 ± 1 . 98% , n = 8 ) . Therefore , AC delay-period activity appeared to be more critical in memory retention than other general processes such as attention or expectation of the forthcoming second stimulus . By measuring and manipulating the activity of neurons in AC during the auditory WM behavior , we explored the causal role of AC in WM . Experiments showed that AC neurons were active not only during the presentation of the auditory stimulus but also during the early delay period when no stimulus was presented . Further experiments showed that optogenetic suppression of neuronal activity in AC during the stimulus epoch and early delay period caused a substantial reduction in WM performance , whereas suppressing later in the delay period did not . These results indicated that although AC may not be involved in WM storage during the whole delay period , it is crucial for WM tasks in the initial encoding of auditory information , maintaining the memory trace for a limited time , and then transferring this information for further WM storage elsewhere . The role of sensory cortices in WM is debate . There is evidence supporting the idea that the same system involved in sensory processing also participates in retaining sensory information in WM task ( Gayet et al . , 2018; Pasternak and Greenlee , 2005; Scimeca et al . , 2018 ) . It is based on fMRI decoding and monkey neurophysiology studies showing that during the memory period neuronal activity in sensory cortices can be maintained and reflected the identity of the remembered stimulus ( Fuster and Jervey , 1981; Harrison and Tong , 2009; Mendoza-Halliday et al . , 2014 ) . Furthermore , perturbation of neural activity in sensory cortices can impair WM performance ( Colombo et al . , 1990; Harris et al . , 2002; Zhang et al . , 2019 ) . However , there are also studies showing that when the distractor is applied during the delay period , the sustained neural activity and WM decoding were no longer present in sensory cortices . Meanwhile , the presence of distractor had no impact on behavioral performance ( Bettencourt and Xu , 2016; Miller et al . , 1996 ) . This suggests that sensory cortices may not be essential for memory maintenance in WM task and WM-related activities observed in sensory cortices may largely reflect feedback signals indicative of the storage of WM content elsewhere in the brain , probably in the posterior parietal cortex and PFC ( Leavitt et al . , 2017; Xu , 2017; Xu , 2018 ) . Active WM maintenance requires resistance to distractors presented during the delay period ( Baddeley , 2012; Bettencourt and Xu , 2016; Miller et al . , 1996 ) . If the delay-period activity of a sensory cortex is essential for WM , the neural activity should be able to maintain information even following the distractor . With regard to the present study , if suppressing the delay-period activity of AC following the distractor impairs the performance , it will implicate AC in active information maintenance in WM task . As we expected , our result showed that suppressing the early delay-period activity of AC after the distractor resulted in performance impairment . Therefore , AC is essential for active WM maintenance even in the face of the distractor . The early delay-period activity observed in AC reflected the active memory maintenance , not the residual auditory trace . Thus , our present results support the idea that sensory cortices participate in sensory information maintenance in WM task . In the present study , we observed a temporal-specific effect of AC suppression during the delay period of WM task . Only optogenetic suppression of AC activity during the stimulus epoch and early delay period ( 300–800 ms ) caused a substantial reduction in WM performance . These results were consistent with the electrophysiology results showing that AC neurons exhibited elevated activity during stimulus presentation and the first 800 ms of the delay period . These findings agree with transcranial magnetic stimulation ( TMS ) studies in humans showing that disruption of primary somatosensory cortex functioning early in the delay period ( at 300 ms or 600 ms ) interfered with tactile WM performance . In contrast , TMS later in the delay did not ( Harris et al . , 2002 ) . Furthermore , the WM performance impairment induced by delay-period optogenetic suppression of AC cannot be attributed to direct disruption of auditory perception . Because AC suppression during the delay period of a delayed go/no-go auditory discrimination task that also required auditory sample stimulus perception had no effect on the behavior ( Figure 7a ) . It is also unlikely that the optogenetic suppression during the delay period impaired perception of the test stimulus . Because laser illumination during the delay period of 800–1300 ms , before the test stimulus , did not affect WM behavior ( Figure 4b ) . In contrast to the result of the DMS task , we found that suppression of AC activity during the delay period of a delayed go/no-go task did not affect behavior . There is an essential difference in the memory requirement for the two tasks . While the DMS task requires short-term memory of the sensory cue , the delayed go/no-go task is likely to require maintenance of the action plan . This result suggests that the sensory cortex is essential in memory of the sensory cue but not the action plan . This result was also in accord with a recent study finding that the activity of the visual cortex was necessary for the encoding of the stimulus but not for maintenance of the action plan in the memory-guided visual discrimination task ( Goard et al . , 2016 ) . Much of the information about the neural substrates for sensory WM comes from studies of the visual system in nonhuman primates . Relatively fewer studies deal with the storage of information in the auditory modality , perhaps in part due to the difficulties associated with training nonhuman primates to perform auditory tasks ( Bigelow et al . , 2014; Cohen et al . , 2005; Fritz et al . , 2005; Munoz-Lopez et al . , 2010; Scott et al . , 2012 ) . In the present study , we have taken advantage of the relative ease and speed with which rodents can be trained on auditory DMS task to invest the causal role of AC in auditory WM . Future studies are needed to clarify the similarities and differences in neural circuitry and functional principle of auditory and visual WM . WM-related activities during the delay period of WM task have been observed in distributed cortical and subcortical structures , including the PFC ( Fuster and Alexander , 1971; Goldman-Rakic , 1996; Miller et al . , 1996; Romo et al . , 1999; Sreenivasan et al . , 2014; Ungerleider et al . , 1998 ) , parietal cortex ( Chafee and Goldman-Rakic , 1998; Harvey et al . , 2012; Shadlen and Newsome , 2001 ) , and superior colliculus ( Kopec et al . , 2015 ) . The relative importance of these different regions in auditory WM remains to be investigated . Furthermore , these regions are anatomically connected to the AC . Such inter-areal interactions are likely to be essential for memory maintenance in the WM task . An important future direction would be to examine the neural circuits underlying WM . Adult male C57BL/6 mice , aged 8–12 weeks at the start of the experiment , were used for this study . All experiments were performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the US National Institutes of Health . The protocol was approved by the Animal Care and Use Committee of East China Normal University , Shanghai , China . After the start of behavioral training , mice were placed on a restricted water schedule . Water was available only during task performance and immediately after the task . Each mouse’s body weight was measured to ensure that weight loss did not exceed 20% of pre-water restriction weight . Sample sizes were similar to others used in the field . No statistical method was used to predetermine sample size . A separate group of mice were trained to perform a delayed go/no-go auditory discrimination task . Each trial consisted of a sample tone , delay , and response cue . During the sample tone period ( 200 ms ) , a target ( 3 kHz ) or nontarget ( 12 kHz ) tone stimulus was presented ( the corresponding trial was referred to as a ‘go’ trial or ‘no-go’ trial , respectively ) , followed by a 1 . 5 s delay period . After the response cue was presented , licking in ‘go’ trials within the 1 s response window was rewarded ( hit ) . No licking in response to a target or nontarget tone in the response window was regarded as miss or correct rejection ( CR ) , respectively . Licking to a nontarget tone in the response window was regarded as false alarm ( FA ) . The intertrial interval was 10 s . The performance , hit , and false alarm rates were calculated as described in the DMS task . In the go/no-go auditory discrimination experiments , each trial consisted of a starting cue , delay , and sample tone . Trials began with the presentation of the starting cue , followed by a 1 . 5 s delay period . A sample tone stimulus was subsequently presented . The sample tone was either a target tone ( 3 kHz , ‘go’ ) or a nontarget tone ( 12 kHz , ‘no-go’ ) . Licking in response to a target tone in the response window ( 1 s ) was rewarded ( hit ) . No licking in response to a target or nontarget tone in the response window was regarded as miss or correct rejection ( CR ) , respectively . Licking to a nontarget tone in the response window was regarded as false alarm ( FA ) . The intertrial interval was 10 s . The performance , hit , and false alarm rates were calculated as described in the DMS task . The mean lick peri-stimulus time histograms ( PSTHs ) of well-trained mice while performing the WM task delayed go/no-go auditory discrimination task , and go/no-go auditory discrimination task are shown in Figure 1—figure supplement 1 . Once the mice were well trained , they readily performed the task with little licking during the delay period , and these lickings were not punished . Once all experiments were completed , mice were deeply anesthetized with sodium pentobarbital ( 100 mg/kg ) and then perfused transcardially with saline followed by 4% paraformaldehyde . The brains were removed from the skull and kept in 4% paraformaldehyde at 4°C overnight , then transferred to PBS . 50 μm coronal slices were cut and placed in PBS . They were then incubated with DAPI for 10–15 min . Slices were washed again in PBS , mounted , and coverslipped . Fluorescence images were then obtained with a confocal microscope . All statistical analyses were performed in MATLAB . Datasets were tested for normality , and appropriate statistical tests were applied as described in the text ( e . g . , t-test for normally distributed data , Wilcoxon rank-sum test for nonparametric data ) . Unless otherwise stated , data were reported as mean ± s . e . m .	Working memory is the ability to hold information in your head for a few seconds while making decisions , planning or applying logical reasoning to problem solving . It is a fundamental component of cognition , and yet it remains unclear where working memory is stored in the brain . The prefrontal cortex – the front lobe of the brain – is likely the main hub of working memory , since it is responsible for executive functions , such as decision making and planning . This idea is supported by experiments showing sustained brain activity in the prefrontal cortex during working memory tasks . Lesions in that part of the brain also lead to profound deficits in working memory . However , there is increasing evidence that other parts of the brain which process sensory information also participate in retaining working memory . The auditory cortex , which processes sound , is one such candidate . To find out whether the auditory cortex has a role to play in working memory , Yu , Hu , Shi et al . trained mice to lick a water spout after hearing the same sound twice in a row , 1 . 5 seconds apart , and then measured the activities of the mice’s neurons . This showed that neurons in the auditory cortex were active not only when the mice were presented with sound cues , but also for a short time during the delay period between sounds . Yu , Hu , Shi et al . then manipulated this neurons to inactivate them for a fraction of a second after the first sound , which resulted in the animals’ working memory was impaired . However , suppressing the activity of the auditory cortex cells in the later stages of the sound delay period had no effect on working memory . These results indicate that although the auditory cortex may not be involved in storing information for the entire working memory process , it is crucial for encoding of auditory information . In summary , this work uncovers how neurons in the auditory cortex underlie working memory . Further research focusing on these neurons could explain how working memory deteriorates with age , or why it is impaired in people with learning difficulties .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2021	The causal role of auditory cortex in auditory working memory
Epithelial fusion underlies many vital organogenic processes during embryogenesis . Disruptions to these cause a significant number of human birth defects , including ocular coloboma . We provide robust spatial-temporal staging and unique anatomical detail of optic fissure closure ( OFC ) in the embryonic chick , including evidence for roles of apoptosis and epithelial remodelling . We performed complementary transcriptomic profiling and show that Netrin-1 ( NTN1 ) is precisely expressed in the chick fissure margin during fusion but is immediately downregulated after fusion . We further provide a combination of protein localisation and phenotypic evidence in chick , humans , mice and zebrafish that Netrin-1 has an evolutionarily conserved and essential requirement for OFC , and is likely to have an important role in palate fusion . Our data suggest that NTN1 is a strong candidate locus for human coloboma and other multi-system developmental fusion defects , and show that chick OFC is a powerful model for epithelial fusion research . Fusion of epithelia is an essential process during normal human development and its dysregulation can result in birth defects affecting the eye , heart , palate , neural tube , and multiple other tissues ( Ray and Niswander , 2012 ) . These can be highly disabling and are among the most common human birth defects , with prevalence as high as 1 in 500 ( Ray and Niswander , 2012; Morrison et al . , 2002; Nikolopoulou et al . , 2017 ) . Fusion in multiple embryonic contexts displays both confounding differences and significant common mechanistic overlaps ( Ray and Niswander , 2012 ) . Most causative mutations have been identified in genes encoding transcription factors or signalling molecules that regulate the early events that guide initial patterning and outgrowth of epithelial tissues ( Ray and Niswander , 2012; Nikolopoulou et al . , 2017; Patel and Sowden , 2019; Kohli and Kohli , 2012 ) . However , the true developmental basis of these disorders is more complex and a major challenge remains to fully understand the behaviours of epithelial cells directly involved in the fusion process . Ocular coloboma ( OC ) is a structural eye defect that presents as missing tissue or a gap in the iris , ciliary body , choroid , retina and/or optic nerve . It arises from a failure of fusion at the optic fissure ( OF; also referred to as the choroid fissure ) in the ventral region of the embryonic eye cup early in development ( Patel and Sowden , 2019; Onwochei et al . , 2000; Gregory-Evans et al . , 2004 ) . OC is the most common human congenital eye malformation and is a leading cause of childhood blindness that persists throughout life ( Morrison et al . , 2002; Williamson and FitzPatrick , 2014 ) . No treatments or preventative measures for coloboma are currently available . The process of optic fissure closure ( OFC ) requires the coordinated contributions of various cell types in the fusion environment along the proximal-distal ( PD ) axis of the ventral eye cup ( reviewed in Patel and Sowden , 2019; Onwochei et al . , 2000 ) . In all vertebrates studied so far , these include epithelial cells of both the neural retina ( NR ) and retinal pigmented epithelium ( RPE ) , and periocular mesenchymal ( POM ) cells of neural crest origin ( Patel and Sowden , 2019; O’Rahilly , 1966; Hero , 1990; Hero , 1989; Gestri et al . , 2018 ) . As the eye cup grows , the fissure margins come into apposition along the PD axis and POM cells are gradually excluded . Through unknown mechanisms , the basal lamina that surround each opposing margin are either breached or dissolved and epithelial cells from each side intercalate and then subsequently reorganise to form a continuum of NR and RPE , complete with a continuous basal lamina . The function , requirement and behaviour of these epithelial cells in the fusing tissue , and their fates after fusion , are not well understood . Some limited epidemiological evidence suggests environmental factors may contribute to coloboma incidence ( Gregory-Evans et al . , 2004; Hornby et al . , 2003 ) . However , the disease is largely of genetic origin , with as many 39 monogenic OC-linked loci so far identified in humans and the existence of further candidates is strongly supported by evidence in gene-specific animal models ( Patel and Sowden , 2019 ) . Most known mutations cause syndromal coloboma , where the eye defect is associated with multiple systemic defects . A common form of syndromal coloboma is CHARGE syndrome ( MIM 214800 ) for which coloboma , choanal atresia , vestibular ( inner-ear ) and heart fusion defects are defining phenotypic criteria ( Verloes , 2005 ) . Palate fusion defects and orofacial-clefting are common additional features of CHARGE ( ~20% of cases ) and in other monogenic syndromal colobomas ( e . g . from deleterious mutations in YAP1 , MAB21L1 , and TFAP2A [Rainger et al . , 2014; Williamson et al . , 2014; Lin et al . , 1992] ) , suggestive of common genetic mechanisms and aetiologies , and pleiotropic gene function . Isolated ( i . e . non-syndromal ) OC may be associated with microphthalmia ( small eye ) , and the majority of these cases are caused by mutations in a limited number of transcription-factor encoding genes that regulate early eye development ( e . g . PAX6 , VSX2 and MAF [Patel and Sowden , 2019; Williamson and FitzPatrick , 2014] ) , implying that abnormal growth of the eye prevents correct OF margin apposition and that fusion defects are a secondary or an indirect phenotype . Indeed , none of these genes have yet been implicated with direct functional roles in epithelial fusion . However , many isolated coloboma cases also exist without microphthalmia , suggesting that in these patients , eye growth occurs normally but the fusion process itself is defective . These OCs are highly genetically heterogeneous and known loci are not recurrent among non-related patients ( Rainger et al . , 2017 ) . Furthermore , despite large-scale sequencing projects , over 70% of all cases remain without a genetic cause identified ( Rainger et al . , 2017 ) . The most effective and informative models for studying OFC so far have been mouse ( Mus musculus ) and zebrafish ( Danio rerio ) . Both have significant experimental advantages , including powerful genetics and robust genomic data . In particular , live-cell imaging with fluorescent zebrafish embryos has proven to be useful in revealing some intricate cell behaviours at the fissure margin during fusion ( Gestri et al . , 2018 ) . However , both models are restrictive for in-depth molecular investigations due to their limited temporal windows of fusion progression and the number of cells actively mediating fusion and subsequent epithelial remodelling . Here , we present accurate staging and anatomical detail of the process of chick OFC . We show the expansive developmental window of fusion , and the sizable fusion seam available for experimentation and analysis . We take advantage of this to perform transcriptional profiling at key discrete stages during fusion and show significant enrichment for known human OFC genes , and reveal multiple genes not previously associated with OFC . Our analyses also identified specific cellular behaviours at the fusion plate and found that apoptosis was a prominent feature during chick OFC . Furthermore , we reveal Netrin-1 as a mediator of OFC that is essential for normal eye development in evolutionarily diverse vertebrates , and that has a specific requirement during fusion in multiple developmental contexts . This study presents the chick as a powerful model system for further OFC research , provides strong evidence for a novel candidate gene for ocular coloboma , and directly links epithelial fusion processes in the eye with those in broader embryonic tissues . The eye is the foremost observable feature in the chick embryo and grows exponentially through development ( Figure 1a , Figure 1—figure supplement 1 ) . The optic fissure margin ( OFM ) was first identifiable as a non-pigmented region at the ventral aspect of the eye that narrowed markedly in a temporal sequence as the eye increased in size ( Figure 1a ) . To gain a clearer overview of gross fissure closure dynamics we first analysed a complete series of resected flat-mounted ventral eye tissue from accurately staged embryos at Hamburger Hamilton stages ( HH . St ) 25 through to HH . St34 ( n > 10 per stage; Figure 1—figure supplement 1 ) . The OFM was positioned along the proximal-distal ( P-D ) axis of the eye , from the pupillary ( or collar ) region of the iris to the optic nerve . Progressive narrowing of the OFM was observed between HH . St27 to HH . St31 , characterised by the appearance of fused OFM in the midline that separated the non-pigmented iris from the posterior OFM ( Figure 1—figure supplement 1 ) . Both these latter regions remained unpigmented throughout development and we found they were associated , respectively , with the development of the optic nerve and the pecten oculi - a homeostasis-mediating structure that extends out into the vitreous from the optic nerve head and is embedded in the proximal OFM ( Figure 1—figure supplements 1 and 2 ) ( Wisely et al . , 2017 ) . The distal region of the pecten was attached to blood vessels that invade the eye globe through the open region of the iris OFM . This iris region remained open throughout development and well after hatching ( Figure 1—figure supplement 2 ) . A recent study reported that the proximal chick OFM closes via the intercalation of incoming astrocytes and the outgoing optic nerve ( Bernstein et al . , 2018 ) , in a process that does not reflect the epithelial fusion seen during human OFC ( e . g . mediated by epithelial cells of the RPE and neural retina ) ( O’Rahilly , 1966; Bernstein et al . , 2018 ) . To assess the utility of the chick as a model for human OFC and epithelial fusion , we therefore focused our study on OFC progression in the distal and medial eye . Using serial sections from memGFP ( Rozbicki et al . , 2015 ) and wild-type embryos , we then unambiguously identified open fissure and fused seam regions of the medial-distal OFM ( Figure 1b ) . The fused seams were defined by epithelial continuum in both the developing retinal pigmented epithelia ( RPE ) and neural retina ( NR ) layers . We also identified the fusion plates undergoing active fusion using sections and z-stack confocal microscopy ( Figure 1c ) . Serial sectioning at stages HH . St25-34 provided qualitative data for the identification of fusion plates during the progression of chick OFC ( Table 1 ) . We then combined these data with fusion seam length measurements taken from flat mounted fissures to provide a robust quantitative framework of fusion progression ( Table 2 ) . In all analyses , we observed no evidence for fusion in the medial or distal OFM at stages before HH . St27 ( Figure 1; Figure 1—figure supplement 1; Table 1 ) . Fusion was first initiated between HH . St27-28 as confirmed by the definitive appearance of joined epithelial margins at a single fusion point ( FP ) . By HH . St29 , the fused area had expanded to generate a fused seam of 0 . 56 mm ( SD: ± 0 . 12 mm; Figure 1d–e ) with two fusion plates , FP1 and FP2 at the distal and proximal limits , respectively . The position of FP1 became fixed at approximately 0 . 5 mm ( SD: ± 0 . 04 mm ) from the developing pupillary region of the iris in all subsequent developmental stages ( Table 2 , n = 60 fissures analysed ) , and the region between FP1 and the iris remained fully open throughout ocular development ( Figure 1—figure supplement 2 and Table 1 ) . In contrast , the location of FP2 became progressively more proximal until HH . St34 ( Table 2 ) , when FP2 was no longer distinguishable from the pecten ( by flat mount or cryosections ) . This total expansion created a fused epithelial seam of ~1 . 7 mm at its maximum length ( SD: ± 0 . 07 mm , Figure 1e ) . In summary , we observed four distinct phases of fusion ( Figure 1f ) : ( 1 ) pre-fusion when the entire OFM is open ( up to HH . St27 ) ; ( 2 ) fusion initiation at HH . St27-28 in the medial OFM with the appearance of a single medial FP; ( 3 ) active fusion as two FPs separate with the expansion of a fused seam along the P-D axis ( HH . St29-33 ) ; and ( 4 ) complete fusion as the entire OFM is fully fused in the medial OFM ( by HH . St34 ) . The process is active between HH . St27-HH . St34 and proceeds over ~66 hr . By defining fusion progression and the location of the fusion plates during chick OFC , we could then accurately assess the cellular environment within these regions . Immunostaining for the basement membrane ( BM ) ( or basal lamina ) marker Laminin-B1 on cryo-sectioned fissure margins ( Figure 2a ) indicated that fusion occured between cells of the RPE and neural retinal , as observed in human OFC ( O’Rahilly , 1966 ) . Fusion between opposing margins was defined by a reduction of Laminin-B1 at the edges of the directly apposed fissures , followed the appearance of a continuum of BM overlying the basal aspect of the neural retina . Periocular mesenchymal cells were removed from between the fissure margins as fusion progressed . Using a histological approach , we then provided evidence that both the RPE and NR directly contribute cells to the fusion plate ( Figure 2b ) . We also observed that within the fusion plates there was marked epithelial remodelling of both cell types , beginning after apposition of the OFM edges . In contrast , at the fused seam we observed NR and RPE cells were realigned into apical-basal orientation and were indistinguishable from regions outside of the OFM , indicating that the fusion process was complete . To determine whether the expanding seam between FP1 and FP2 was a result of active directional fusion ( e . g . ‘zippering’ ) , or was driven by localised cell-proliferation within the OFM seam ( e . g . pushing forward static fusion plates ) , we used phospho-Histone-H3A ( PH3A ) as a marker for S-phase nuclei in mitotic cells and revealed there was no significant enrichment within the fusion seam ( Figure 2—figure supplement 1 ) . These results suggested that localised cell-proliferation within the seam was not a major mechanism for seam expansion during chick OFC , and further work is required to elucidate the precise mechanisms that drive seam expansion . We then sought to establish whether axonal ingression was a feature of chick OFC in the distal-medial OFM . Using Neurofilament-145 immunofluorescence , we found a complete absence of axonal processes in open , fusing , and fused regions of the distal-medial chick OFM ( Figure 2—figure supplement 2 ) . In contrast , at the same stages we found marked enrichment for axons within the proximal OFM and pecten region , providing further evidence that these regions of the chick optic fissure are distinct ( Bernstein et al . , 2018 ) . Programmed-cell death has been previously associated with epithelial fusion in multiple developmental contexts but the exact requirements for this process remain controversial ( Ray and Niswander , 2012 ) . Even within the same tissues differences arise between species - for example , apoptotic cells are clearly observed at the mouse fusion plate during OFC ( Hero , 1990 ) but are not routinely found in zebrafish ( Gestri et al . , 2018 ) . We therefore asked whether apoptosis was a major feature of chick OFC . Using HH . St30 eyes undergoing active fusion , we performed immunofluorescence staining for the pro-apoptotic marker activated Caspase-3 . We consistently identified apoptotic foci within RPE and NR at both fusion plates , in the adjacent open fissure margin , and at the nascently fused seam with both cryo-section and whole-mount samples ( Figure 2c; Figure 2—figure supplement 2 ) . Foci were not found consistently in other regions of the eye or ventral retina ( not shown ) . By quantifying the number of positive A-Casp-3 foci at FP2 , we found that apoptosis was specifically enriched in the active fusion environment but was absent from fused seam >120 µm and from open regions > 250 µm beyond FP2 ( Figure 2d ) , indicating that apoptosis is a specific feature of OFC in the chick eye . We took advantage of the size and accessibility of the embryonic chick eye to perform transcriptomic profiling with the objectives of: ( i ) assessing the utility of the chick as a genetic model for human OFC by expression of chick orthologues for known disease genes; and ( ii ) to identify novel genes that are required for OFC . Using HH . st25-26 eyes ( pre-fusion; approx . embryonic day E5 ) , segmental micro-dissection of the embryonic chick eye was first performed to obtain separate OFM , ventral eye , dorsal eye and whole eye samples ( Figure 3—figure supplement 1 ) . We took care to not extract tissue from the pecten or optic nerve region of the developing OFM to ensure we obtained transcriptional data for the distal and medial OFM only . Cognate tissues were pooled , RNA was extracted , and region-specific transcriptomes were determined using total RNAseq and analysed to compare mean transcripts per million ( TPM ) values ( Figure 3—source data 1 ) . Pseudoalignment to the Ensembl chicken transcriptome identified 30 , 265 expressed transcripts across all tissue types . To test whether this approach was sensitive enough to reveal domain-specific expression in the developing chick eye , we compared our RNAseq expression data for a panel of genes with clear regional specific expression from a previous study of mRNA in situ analyses in the early developing chick eye cup ( Peters and Cepko , 2002 ) . Markers of the early dorsal retina ( Efnb1 , Efnb2 , Vsx2 , Tbx5 , Aldh1A1 ) clustered as dorsal-specific in our RNAseq data , whereas known ventral markers ( Crx , Maf1 , Pax2 , Aldh6 [Ald1a3] , Vax1 , and Rax1 ) were strongly expressed in our fissure and ventral transcriptomes ( Figure 3—figure supplement 1 ) , which validated this approach to reveal OFC candidate genes . We then repeated the analysis , collecting OFM , ventral tissue and whole eye and included stages HH . st27-28 ( ~E6; during initiation ) and HH . st28-30 ( ~E7; during active fusion ) as discrete time-points ( Figure 3—figure supplement 1 ) . Dorsal tissue was not extracted for these stages . Correlation matrices for total transcriptomes of each sample indicated one of the HH . st25-26 fissure samples as an outlier , but otherwise that there was close correlation between all the other samples ( Pearson’s correlation coefficient >0 . 9; Figure 3—figure supplement 1 ) . Quantitative analyses identified 14 , 262 upregulated genes and 14 , 125 downregulated genes in the fissure margin at the three time points ( Figure 3a; fissure versus whole eye . False discovery rate ( FDR ) adjusted p-value<0 . 05 ) . The largest proportion of these differential expressed genes ( DEGs ) were observed at HH . st25-27 , most likely reflecting the periocular tissue between the fissure margins . Remarkably few DEGs were shared between stages . We used fold change ( FC ) analysis to identify biologically-relevant differential gene expression ( Log2FC ≥1 . 5 or ≤−1 ) in the fissure compared to whole eye , we found 1613 , 2971 and 1491 DEGs at pre-fusion , initiation , and active fusion , respectively ( Figure 3—source data 2 ) . Refining our analysis to identify only those DEGs common across all stages revealed 12 genes with increased expression in the fissure and 26 with decreased expression ( Figure 3b; Table 3 ) . Of these upregulated fissure-specific genes , causative mutations have previously been identified in orthologues of PAX2 , SMOC1 , ALDH1A3 , and VAX1 in human patients with coloboma or structural eye malformations ( Patel and Sowden , 2019; Williamson and FitzPatrick , 2014 ) , and some of these genes , such as PAX2 and inhibitors of BMP expression , induce coloboma phenotypes when overexpressed in the developing ventral chick eye ( Gregory-Evans et al . , 2004; Sehgal et al . , 2008 ) . In addition , targeted manipulations of orthologues of both CHRDL1 and CYP1B1 have recently been shown to cause coloboma phenotypes in Xenopus and zebrafish , respectively ( Pfirrmann et al . , 2015; Williams et al . , 2017 ) . The remaining fissure-specific genes ( NTN1 , RTN4RL1 , TFEC , GALNT6 , CLYBL and RGMB ) had not been previously associated with OFC defects to the best of our knowledge . Clustering for relative expression levels of the RNAseq data at active fusion stages ( HH . St28-30 ) revealed three independent clusters ( 2 , 3 , and 5 ) where expression profiles matched Fissure >ventral > whole eye ( Figure 3c ) . We hypothesised that analysis of these clusters would reveal genes with fusion-specific functions during OFC . Of the three clusters with this profile , ontology analyses showed significant enrichment for sensory organ development and eye development processes ( FDR q < 0 . 001 , 10 genes ) and for adhesion processes ( Figure 3—figure supplement 1; FDR q < 0 . 05 , 25 genes; Biological adhesion [GO:0022610] and cell adhesion [GO:0022610] ) , of which 17 genes had mean TPM values > 10 . Within this group , multiple candidates for roles during OFC fusion were revealed , such as several transmembrane proteins , Integrin-A2 , Cadherin-4 , Collagen 18A1 and FLRT3 ( Figure 3d ) . However , of these NTN1 was the highest expressed and most fissure-specific ( mean TPM values: Fissure = 204; ventral = 35; and whole eye = 4 ) . We used RNAscope , colorimetric in situ hybridisation , and immunostaining with NTN1-specific antibodies to determine the precise location of Netrin-1 in the chick eye ( Figure 4 and Figure 4—figure supplement 1 ) . We observed highly specific expression in both neuroepithelial retina and RPE cells at the fissure margins during active fusion at HH . St29-30 ( Figure 4a ) . This was consistent at both fusion plates ( FP1 and FP2 ) , and in both locations NTN1 expression was markedly reduced in the fused seam compared to expression in the adjacent open margins . Immunofluorescence revealed that , consistent with NTN1 mRNA , NTN1 protein was specifically localised to the basal lamina at the opposing edges of the OFM , and to both RPE and neuroepithelial retina cells in this region ( Figure 4b–c , Figure 4—figure supplement 1 ) . To test the significance of our findings to other vertebrates , we first asked whether this localisation was conserved to the human OFM . Immunofluorescence analysis for NTN1 ( hNTN1 ) in human embryonic fissures during fusion stages ( Carnegie Stage CS17 ) displayed remarkable overlap with our observations in chick , with protein signal localised specifically to open and fusion plate regions of OFM at the NR and RPE ( Figure 4d ) , and an absence of hNTN1 in fused seam . Consistent with the protein localisation , RNAseq analysis on laser-captured human fissure tissue showed a 32x fold increase in hNTN1 expression compared to dorsal eye ( Patel and Sowden; manuscript in preparation ) . Microarray analyses had previously observed enrichment for Ntn1 in the mouse fissure during closure stages ( Brown et al . , 2009 ) , so we then analysed Ntn1 protein localisation in equivalent tissues in the mouse optic fissure ( fusion occurs around embryonic day E11 . 5 and is mostly complete by E12 . 5 ( Hero , 1990 ) . We observed consistency in both cell-type and positional localisation of Ntn1 protein ( Figure 4e ) , and that Ntn1 protein was not detected in the fused seam at E12 . 5 ( immunoreactivity for NTN1 was observed in the proximal optic nerve region at this stage; Figure 4—figure supplement 2 ) . Our results suggested that Netrin-1 has an evolutionarily conserved role in OFC and prompted us to test if NTN1 is essential for this process . We therefore analysed mouse embryos of WT and Netrin-null ( Ntn1-/-; Yung et al . , 2015 ) littermates at embryonic stages after OFC completion ( E15 . 5-E16 . 5 ) ( Hero , 1990 ) and observed highly penetrant ocular coloboma in Ntn1-/- mutants ( >90%; n = 10/11; Figure 4f ) . Mutant eyes analysed at earlier stages of eye development ( E11 . 5 ) when fusion is first initiated ( Hero , 1990 ) were normal ( n = 4 Ntn1-/- embryos; 8x eyes analysed in total ) , with fissure margins positioned directly in appositional contact each other ( Figure 4—figure supplement 2 ) . We also observed variably penetrant orofacial and palate fusion defects in mutant mice ( Figure 4g;~36%; n = 4/11 Ntn1-/- embryos ) , indicating that NTN1 may also have an important role in fusion during palatogenesis and craniofacial development . Finally , we then tested whether Netrin deficiency would cause similar ocular defects in other vertebrates and generated germline netrin-1 mutant zebrafish by creating a nonsense mutation in the first exon of ntn1a using CRISPR/Cas9 gene editing ( Figure 4—figure supplement 3 ) . We inter-crossed heterozygote G0 fish ( ntn1a+/- ) and observed several G1 embryos displaying bilateral ocular defects including coloboma and microphthalmia ( Figure 4—figure supplement 3 ) . DNA sequencing of the targeted ntn1a locus confirmed 100% ( n = 3 ) of the phenotypic embryos were homozygous , whereas ocular defects or colobomas were not observed in any heterozygous ( n = 6 ) or wild-type ( n = 12 ) embryos . A recent study applied morpholino ( MO ) translation-blocking knockdown approaches to target ntn1a in zebrafish embryos and observed bilateral ocular colobomas in all fish injectected ( Richardson et al . , 2019 ) , with normal early eye development and appropriately apposed fissure margins obvious prior to fusion . We were also able induce colobomas using MOs designed to target the translational start site of ntn1a ( Figure 4—figure supplement 3 ) . Bilateral colobomas were observed in 31/71 ( 43 . 7% ) of MO injected embryos with no ocular phenotypes observed in control injections ( n = 40 ) . In combination , these results are in agreement with our data presented in chicken and mouse OFMs and that Netrin-1 is also essential for zebrafish eye development and is likely to have a specific role in tissue fusion . It also confirms an evolutionarily essential requirement for Netrin in ocular development , including OFC , in diverse vertebrate species . Our study provides strong evidence that Netrin-1 is essential for OFC in the developing vertebrate eye and is required for normal orofacial development and palate fusion . The transient and specific NTN1 expression at the fusion plate , and the subsequent reduction/loss in fused OFM , suggests NTN1 has a direct role in the fusion process . Indeed , Netrin1-deficient mouse eyes displayed highly penetrant colobomas but their fissure margins were normally apposed during fusion initiation , arguing against a broad failure of early eye development . In further support for a direct role in epithelial fusion was previously published work showing fusion failure during development of the vestibular system of both chick and mice where NTN1-expression was manipulated ( Yung et al . , 2015; Salminen et al . , 2000; Nishitani et al . , 2017 ) . In this developmental context , otic epithelia must fuse normally for the correct formation of the semicircular canal structures . Although we and others ( Richardson et al . , 2019 ) found coloboma in zebrafish knockdown experiments of ntn1a , we observed coloboma with microphthalmia in the context of complete knockout of ntn1a . This more severe phenotype in the complete absence of ntn1a implies there could be a more general requirement for Netrin-1 during early eye development , or could reflect teleost-specific eye developmental processes not shared among higher vertebrates ( Martinez-Morales et al . , 2017 ) . Further work is required to elucidate the precise role of Netrin-1 during OFC and broader eye development among different species . Taken in combination , these findings strongly implicate NTN1 as a multipotent factor required for tissue fusion in multiple distinct developmental contexts . In humans , variants near NTN1 have been associated with cleft lip in human genome wide association studies ( Leslie et al . , 2016; Leslie et al . , 2015 ) . While these are not monogenic disease mutations , this observation adds additional further relevance for future genetic studies of patients with coloboma . It is also consistent with our observations in Netrin-1 knock-out animals having a high penetrance of both coloboma and cleft palate phenotypes . Therefore , we propose that NTN1 should be included as a candidate gene in diagnostic sequencing of patients with human ocular coloboma , and should also be carefully considered for those with other congenital malformations involving defective fusion . Coloboma in association with additional fusion defects of the inner ear are two of the key clinical classifications for a diagnosis of CHARGE syndrome ( Verloes , 2005 ) . Further phenotypes commonly associated with the syndrome are septal heart defects and orofacial clefting , both with aetiologies likely to involve fusion defects ( Ray and Niswander , 2012 ) . CHARGE syndrome cases are predominantly caused by heterozygous loss-of-function pathogenic variants in the chromodomain helicase DNA-binding protein 7 ( CHD7 ) gene ( Vissers et al . , 2004 ) . Mice lacking Chd7 display CHARGE syndrome-like phenotypes and exhibit abnormal expression of Ntn1 ( Hurd et al . , 2007; Hurd et al . , 2012 ) . In addition , ChIP-seq analyses have shown direct binding of Chd7 to the promoter region of Ntn1 in mouse neural stem cells ( Engelen et al . , 2011 ) . Given the amount of tissue available in the chick model , it would be possible and intriguing to confirm whether CHD7 directly regulates NTN1 expression in ovo in the chick optic fissure . There is also emerging evidence that CHD7 and the vitamin A derivative retinoic acid ( RA ) indirectly interact at the genetic level during inner ear development ( Yao et al . , 2018 ) . Defective RA signalling also leads to significant reduction of Ntn1 expression in the zebrafish OFM ( Lupo et al . , 2011 ) , implicating a possible genetic network involving RA and CHD7 , where NTN1 could directly mediate developmental fusion mechanisms from these hierarchical influences . Netrin-1 is well-studied for its canonical roles in guidance of commissural and peripheral motor axons and growth-cone dynamics , with attraction or repulsion mediated depending on the co-expression of specific receptors ( reviewed in Lai Wing Sun et al . , 2011; Larrieu-Lahargue et al . , 2012 ) . We found that axonal processes were absent from the chick fissure margin during fusion stages , suggesting that the normal function of NTN1 may be to prevent axon ingression into the OFM to permit fusion . However , the phenotypic evidence from both the palate and vestibular system strongly support the argument that NTN1 has a non-guidance mechanistic role during OFC . Netrin orthologues have been recently associated with the regulation of cell migration and epithelial plasticity in the apparent absence of co-localised canonical Netrin-1 receptors ( Manhire-Heath et al . , 2013; Lee et al . , 2014; Yan et al . , 2014 ) . In contrast , netrin acting together with its receptor neogenin combined to mediate close adhesion of cell layers in the developing terminal end buds during lung branching morphogenesis ( Srinivasan et al . , 2003 ) . Although we observed strong NTN1 expression in cells lining the chick OFM , and similar localisation of Netrin-1 protein in chick , human and mouse , we did not observe reciprocal expression of any canonical NTN1 receptors in our RNAseq datasets ( e . g . UNC5 , DCC or Neogenin; Figure 4—figure supplement 4 ) . Indeed , the Netrin repulsive cue UNC5B was the most significantly downregulated DEG in fissure versus whole eye in our data and was also downregulated in human OFM ( Sowden and Patel; manuscript in preparation ) . Therefore , it will be vitally important for future studies to elucidate interaction partners of Netrin in fusing tissues , or to reveal if Nerin-1 can act autonomously in these contexts and to provide deeper insight into its mechanistic function during fusion . The chick is one of the earliest established models for developmental biology and has provided many key insights into human developmental processes ( Stern , 2018 ) . Despite this , and extensive historical study of eye development in chicken embryos , the process of chick OFC has not been well analysed until now . Indeed , the first study appeared only recently and specifically defined aspects of tissue fusion at the proximal ( optic nerve and pecten ) region of the OF ( Bernstein et al . , 2018 ) , and did not observe complete fusion of epithelia in these regions . Indeed , closure of the proximal OF was characterised by intercalation of pecten and the lack of true epithelial continuum of neuroepithelial retina and RPE . By focusing on the epithelial fusion events in the distal and medial eye , our study complements the Bernstein et al study ( Bernstein et al . , 2018 ) to provide a comprehensive framework of OFC progression in the chick . Indeed , taken together , our analyses clearly define three distinct and separate anatomical regions in the developing chick OFM: the iris , the medial OFM , and the pecten . In addition , we present the spatial and temporal sequence of chick OFC at the anatomical and molecular level , and provide strict criteria for staging the process - based on a combination of broad embryonic anatomy , ocular , and fissure-specific features . Fusion initiated at the medial OFM at HH . St27/28 and continued until HH . St34 , with predominantly distal to proximal directionality . In addition , we found that closure of the medial OFM is a true epithelial fusion process that occurs over a large time window of approximately 60 hr , involving two fusion plates , and that closes over 1 . 5 mm of complete fusion seam . This temporal window , the number of directly contributing cells , and the accurate staging of its progression allows unique opportunities for further experimentation . Importantly , one whole chick optic fissure ( from HH . St29 onwards ) can simultaneously provide data for unfused , fusing , and post-fused contexts . In addition , our transcriptional profiling , including the identification of OFM-specific genes in the chick that include multiple human coloboma orthologues , builds on previous work that illustrate the chick as an excellent model for human eye development and the basis of embryonic malformations ( Wisely et al . , 2017; Vergara and Canto-Soler , 2012; Trejo-Reveles et al . , 2018 ) . These features , in combination with recent advances in chick transgenics and genetic manipulations ( Davey et al . , 2018 ) , project the chick as a powerful to analyse cell behaviours during OFC and epithelial fusion . For example , the stable multi-fluorescent Cre-inducible lineage tracing line ( the Chameleon chicken [Davey et al . , 2018] ) will be valuable to determine how the fissure-lining cells contribute to the fusing epithelia , while the very-recent development of introducing gene-targeted or gene-edited primordial germ cells into sterile hosts for germ-line transmission ( Taylor et al . , 2017 ) provides a rapid and cost-effective way to develop stable genetic lines to interrogate specific gene function ( Davey et al . , 2018; Woodcock et al . , 2017 ) . Thus , our study illustrates the powerful utility of the chick as a model for investigating OFC and for the discovery of novel candidate genes for coloboma , and is perfectly timed to coincide with major new developmental biology techniques in avian systems to place the chick model as a powerful addition to OFC and fusion research . This study provides the first detailed report of epithelial fusion during chick OFC and illustrates the power of the embryonic chick eye to investigate the mechanisms guiding this important developmental process further and to provide insights into human eye development and broader fusion contexts . We clearly define the temporal framework for OFC progression and reveal that fusion is characterised by loss of epithelial cell types and a coincidental increase in apoptosis . We reveal the specific expression of orthologues of known coloboma-associated genes during chick OFC , and provide a broad transcriptomic dataset that can be used to improve the identification of candidate genes from human patient exome and whole-genome DNA sequencing datasets . Finally , we identify that NTN1 is specifically and dynamically expressed in the fusing vertebrate fissure - consistent with having a direct role in epithelial fusion , and is essential for OFC and palate development . We propose that NTN1 should therefore now be considered as a new candidate for ocular coloboma and congenital malformations that feature defective epithelial tissue fusion . Hy-Line Eggs were incubated at 37°C at day 0 ( E0 ) , with embryo collection as stated throughout the text . Whole embryos were staged according to Hamburger Hamilton ( Hamburger and Hamilton , 1992; Hamburger and Hamilton , 1951 ) . Heads were removed and either ventral eye tissue was resected and flat-mounted and imaged immediately , or whole heads were placed in ice cold 4% paraformaldehyde ( PFA ) in pH 7 . 0 phosphate buffered saline ( PBS ) , overnight and then rinsed twice in PBS . OFMs used for fusion progression measurements ( flat mounts ) were mounted in glycerol between a coverslip and glass slide , without fixation . Whole embryo , flat mounted OFMs , and dissected eye images for were captured on a Leica MZ8 light microscope and measurements were processed using FIJI ( NCBI/NIH open source software [Schindelin et al . , 2012] ) . For cryosections , resected ventral chick eyes were equilibrated in 15% Sucrose-PBS then placed at 37°C in 7% gelatin:15% Sucrose , embedded and flash-frozen in isopentane at −80°C . Sections were cut at 20 µm . Immunofluorescence was performed on chick fissure sections as follows: 2 × 30 min rinse in PBS , followed by 2 hr blocking in 1% BSA ( Sigma ) in PBS with 0 . 1% Triton-X-100 [IF Buffer 1] . Sections were incubated overnight at 4°C with primary antibodies diluted in 0 . 1% BSA in PBS with 0 . 1% Triton-X-100 [IF Buffer 2] . Slides were then washed in 3 × 20 min PBS , followed by incubation for 1 hr with secondary antibodies ( Alexa Fluor conjugated with 488 nm or 594 nm fluorophores; 1:800–1000 dilution , Thermo Fisher ) , and mounted with ProLong Antifade Gold ( Thermo Fisher ) with DAPI . Alexa Fluor Phalloidin ( 488 nm; Thermo-Fisher #A12379 ) was added at the secondary antibody incubation stages ( 1:50 dilution ) . Human foetal eyes were obtained from the Joint Medical Research Council UK ( grant # G0700089 ) /Wellcome Trust ( grant # GR082557 ) Human Developmental Biology Resource ( http://www . hdbr . org/ ) . For Netrin-1 immunostaining in human and mouse tissues , cryosections were antigen retrieved using 10 mM Sodium Citrate Buffer , pH 6 . 0 and blocked in 10% Goat serum +0 . 2% Triton-X100 in PBS , then incubated overnight at 4°C with primary antibody ( Abcam #ab126729; 1: 300 ) in block . Secondary antibody staining and subsequent processing were the same as for chick ( above ) . For anti-NTN1 immunostaining in chick tissues , cryosections were hydrated in phosphate buffer ( PB ) pH7 . 2 , antigen retrieved using 1% SDS in PB and blocked 2% bovine serum albumen +0 . 2% Tween-20 in PB ( blocking buffer ) . Primary antibody was diluted in blocking buffer and incubated at room temperature for 4 days . Secondary antibody staining and subsequent processing were as stated above , but PB was used instead of PBS . For whole-mount immunofluorescence we followed the protocol from Ahnfelt-Rønne et al ( Ahnfelt-Rønne et al . , 2007 ) , with the exception that we omitted the TNB stages and incubated instead with IF Buffer 1 ( see above ) overnight and then in IF Buffer two for subsequent antibody incubation stages , each for 24 hr at 4°C . No signal amplification was used . Antibodies were used against Phospho-Histone H3A and Netrin-1 . Imaging was performed using a Leica DM-LB epifluorescence microscope , or a Nikon C1 inverted confocal microscope and Nikon EZ-C1 Elements ( version 3 . 90 Gold ) software . All downstream analysis was performed using FIJI . Image analysis for proliferation in the OFM on flat-mounts was performed by counting Phospho-Histone H3A positive foci using a region of interest grid with fixed dimensions of 200 µm2 and throughout the entire confocal Z-stack . To quantitate apoptotic foci at the OFM , we used Activated-Casp3 immunofluorescence on serial cryosections of HH . St29-30 OFMs and collected confocal images for each section along the P-D axis . Image analysis was performed by counting A-Casp3 positive foci at the OFM in sequential sections using a region of interest with fixed dimensions of 100 µm2 . For histology and subsequent haematoxylin and eosin staining , resected eyes processed and image captured according to Trejo-Reveles et al ( Trejo-Reveles et al . , 2018 ) . RNAscope was performed on HH . St29 cryosections using a probe designed specific to chicken NTN1 according to Nishitani et al ( Nishitani et al . , 2017 ) . For colourimetric in situ hybridisation , a ribprobe was for NTN1 was designed using PCR primers to amplify a 500 bp product from cDNA prepared from chick whole embryos at HH . St28-32 ( Oligonucleotide primers: Fwd 5’-ATTAACCCTCACTAAAGGCTGCAAGGAGGGCTTCTACC-3’ and Rev 5’-TAATACGACTCACTATAGGCACCAGGCTGCTCTTGTCC-3’ ) . The PCR products were purified and transcribed into DIG-labelled RNA using T7 polymerase ( Sigma-Aldrich ) and used for In Situ hybridization on cryosectioned chick fissure margin tissue ( prepared as described above for immunofluorescence ) or whole embryos using standard protocols ( described in J . Rainger's doctoral thesis - available on request ) . To obtain Ntn1-/- mouse embryos ( Ntn1tm1 . 1Good , RRID:MGI:5888900 ) , we performed timed matings with male and female heterozygotes and took the appearance of a vaginal plug in the morning to indicate embryonic day ( E ) 0 . 5 . Embryos were collected at E11 . 5 and E16 . 6 and genotyped according to Yung et al ( Yung et al . , 2015 ) . As with this previous report we observed ratios within the expected range for all three expected genotypes ( 28 total embryos: 13x Ntn1+/-; 10x Ntn1-/-; 5x WT – 46%; 35%; 18% , respectively ) . Embryos were fixed in 4% paraformaldehyde overnight and then rinsed in PBS and imaged using a Leica MZ8 light microscope . Ntn1-/- and C57Bl/6J animals were maintained on a standard 12 hr light-dark cycle . Mice received food and water ad lib and were provided with fresh bedding and nesting daily . For zebrafish work , we designed gene-editing sgRNA oligos alleles to target ntn1a: 5´-GGTCTGACGCGTCGCACGTG-3´ . We then generated founder ( G0 ) animals by zygotic microinjection of CRISPR/Cas9 components according to previous work ( Dutta et al . , 2015; Varshney et al . , 2015; Jao et al . , 2013 ) . G0 animals were genotyped and used for crosses to generate G1 embryos which were scored for coloboma phenotypes and genotyped individually ( Figure 4—figure supplement 3 ) . All experiments were conducted in agreement with the Animals ( Scientific Procedures ) Act 1986 and the Association for Research in Vision and Ophthalmology Statement for the Use of Animals in Ophthalmic and Vision Research ( USA ) . Morpholinos were designed and generated by Gene Tools LLC ( Oregon ) to target the translation initiating site of ntn1a: 5′-CATCAGAGACTCTCAACATCCTCGC-3′ , and a Universal control MO sequence was used as a control: 5′-ATCCAGGAGGCAGTTCGCTCATCTG-3′ . One cell stage embryos were injected with 2 . 5 ng or 5 . 0 ng of ntn1a or control morpholino and allowed to develop to OFC stages ( ≥48 hpf ) . Oligos used for ntn1a genotyping by sanger sequencing were: 5′-TTACGACGAGAACGGACACC-3′ and 5′-GGAGGTAATTGTCCGACTGC-3′ . For RNA seq analysis , we carefully dissected regions of ( i ) fissure-margin , ( ii ) ventral eye , and ( iii ) dorsal eye , and ( iv ) whole eye tissue from ≥10 individual embryos for each HH stage range ( Figure 3—figure supplement 1 ) . Samples were collected and pooled for each tissue type and stage to obtain n = 3 technical replicate RNA pools per tissue type per stage . Total RNA was extracted using Trizol ( Thermo Scientific ) . Whole-transcriptome cDNA libraries were then prepared for each pool following initial mRNA enrichment using the Ion RNA-Seq Core Kit v2 , Ion Xpress RNA-Seq Barcodes , and the Ion RNA-Seq Primer Set v2 ( Thermo Scientific ) . cDNA quality was confirmed using an Agilent 2100 Bioanalyzer . Libraries were pooled , diluted , and templates were prepared for sequencing on the Ion Proton System using Ion PI chips ( Thermo Scientific ) . Quantitative transcriptomics was performed using Kallisto psuedoalignment ( Bray et al . , 2016 ) to the Ensembl ( release 89 ) chicken transcriptome . Kallisto transcript counts were imported into R using tximport ( Soneson et al . , 2015 ) and differentially expressed transcripts identified using Limma ( Ritchie et al . , 2015 ) . Genes not expressed in at least three samples were excluded . To identify the relationships between samples , Log2 transformed counts per million were then calculated using edgeR ( Robinson et al . , 2010 ) and Spearman’s rank correlation was used to identify the similarities in genome-wide expression levels between samples . All RNAseq data files are submitted to the NCBI Gene Expression Omnibus database ( http://www . ncbi . nlm . nih . gov/geo ) with the accession number GSE84916 . Bar graphs display means ± SD or 95% confidence intervals as indicated . Sample sizes were n ≥ 3 , unless stated otherwise . Statistical analyses were performed using Prism 8 ( GraphPad Software Inc ) . Data were assessed for normal distribution by Shapiro-Wilk test where appropriate . Significance was evaluated by unpaired Student’s t-test , where p≤0 . 05 was deemed significant . Asterisk indicate significance in Figure 1 as p≤0 . 05 . p≤0 . 01 , **p≤0 . 001 .	Our bodies are made of many different groups of cells , which are arranged into tissues that perform specific roles . As tissues form in the embryo they must adopt precise three-dimensional structures , depending on their position in the body . In many cases this involves two edges of tissue fusing together to prevent gaps being present in the final structure . In individuals with a condition called ocular coloboma some of the tissues in the eyes fail to merge together correctly , leading to wide gaps that can severely affect vision . There are currently no treatments available for ocular coloboma and in over 70% of patients the cause of the defect is not known . Identifying new genes that control how tissues fuse may help researchers to find what causes this condition and multiple other tissue fusion defects , and establish whether these may be preventable in the future . Much of what is currently known about how tissues fuse has come from studying mice and zebrafish embryos . Although the extensive genetic tools available in these ‘models’ have proved very useful , both offer only a limited time window for observing tissues as they fuse , and the regions involved are very small . Chick embryos , on the other hand , are much larger than mouse or zebrafish embryos and are easier to access from within their eggs . This led Hardy et al . to investigate whether the developing chick eye could be a more useful model for studying the precise details of how tissues merge . Examining chick embryos revealed that tissues in the base of their eyes fuse between five and eight days after the egg had been fertilised , a comparatively long time compared to existing models . Also , many of the genes that Hardy et al . found switched on in chick eyes as the tissues merged had previously been identified as being essential for tissue fusion in humans . However , several new genes were also shown to be involved in the fusing process . For example , Netrin-1 was important for tissues to fuse in the eyes as well as in other regions of the developing embryo . These findings demonstrate that the chick eye is an excellent new model system to study how tissues fuse in animals . Furthermore , the genes identified by Hardy et al . may help researchers to identify the genetic causes of ocular coloboma and other tissue fusion defects in humans .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2019	Detailed analysis of chick optic fissure closure reveals Netrin-1 as an essential mediator of epithelial fusion

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.